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
Skeletal muscle myoblasts ( iMyoblasts ) were generated from human induced pluripotent stem cells ( iPSCs ) using an efficient and reliable transgene-free induction and stem cell selection protocol . Immunofluorescence , flow cytometry , qPCR , digital RNA expression profiling , and scRNA-Seq studies identify iMyoblasts as a PAX3+/MYOD1+ skeletal myogenic lineage with a fetal-like transcriptome signature , distinct from adult muscle biopsy myoblasts ( bMyoblasts ) and iPSC-induced muscle progenitors . iMyoblasts can be stably propagated for >12 passages or 30 population doublings while retaining their dual commitment for myotube differentiation and regeneration of reserve cells . iMyoblasts also efficiently xenoengrafted into irradiated and injured mouse muscle where they undergo differentiation and fetal-adult MYH isoform switching , demonstrating their regulatory plasticity for adult muscle maturation in response to signals in the host muscle . Xenograft muscle retains PAX3+ muscle progenitors and can regenerate human muscle in response to secondary injury . As models of disease , iMyoblasts from individuals with Facioscapulohumeral Muscular Dystrophy revealed a previously unknown epigenetic regulatory mechanism controlling developmental expression of the pathological DUX4 gene . iMyoblasts from Limb-Girdle Muscular Dystrophy R7 and R9 and Walker Warburg Syndrome patients modeled their molecular disease pathologies and were responsive to small molecule and gene editing therapeutics . These findings establish the utility of iMyoblasts for ex vivo and in vivo investigations of human myogenesis and disease pathogenesis and for the development of muscle stem cell therapeutics . The technologies for reprogramming human somatic cells into induced pluripotent stem cells ( iPSCs ) ( Takahashi et al . , 2007 ) and for inducing specific differentiated cell types are providing extraordinary opportunities for investigating mechanisms of human tissue differentiation , the molecular pathology of diseases , and therapeutic development . Much of the research on iPS cell-type induction has focused on optimizing the production of differentiated cells to provide a platform for investigations of disease pathologies ( Ardhanareeswaran et al . , 2017; Hashimoto et al . , 2016; Georgomanoli and Papapetrou , 2019; Heslop and Duncan , 2019; van Mil et al . , 2018 ) . Less attention has been given to generation of lineage-specific human stem cells and progenitors to enable studies of tissue and organ development , genetic and epigenetic disease mechanisms , and stem cell therapeutics . The goal of our study has been to isolate and propagate myogenic stem cells from human iPSC cultures in response to gene-free myogenic induction and cell growth selection and to establish the utility of these myoblast stem cells for molecular investigations of human myogenesis and muscular dystrophies . Here , we report an efficient and reliable transgene-free myogenesis protocol to generate human skeletal muscle stem cells ( iMyoblasts ) from healthy control ( Ctrl ) and patient iPSCs . This protocol efficiently produces a stably committed and expandable population of PAX3+/MYOD1+ iMyoblasts that differentiate as regenerative stem cells ex vivo in cell culture and in vivo in muscle xenografts in irradiated and injured mouse tibialis anterior ( TA ) muscle , which maintain a renewable PAX3+ iMyoblast population . iMyoblast muscle xenografts undergo fetal-to-adult MYH isoform switching demonstrating their plasticity to respond to maturation signals provided by the host adult muscle . Finally , we show that iMyoblasts generated from Facioscapulohumeral Muscular Dystrophy ( FSHD ) Type 1 ( FSHD1 ) , Limb-Girdle Muscular Dystrophy ( LGMD ) R7 and R9 ( formerly LGMD2G and 2I ) , and Walker Warburg Syndrome ( WWS ) patient iPSCs model the molecular pathologies of these diseases . We developed a two-step protocol to isolate iMyoblasts from cultures of Ctrl and patient iPSCs , using transgene-free iPSC myogenic induction in combination with reserve cell selection ( Figure 1A ) . Human iPSC lines for these studies were generated by reprogramming bMyoblasts and fibroblasts isolated from muscle biopsies of adult FSHD1 and Ctrl subjects ( Homma et al . , 2012; Jones et al . , 2012 ) , or dermal fibroblasts from subjects with early onset FSHD1 , LGMDR7 , LGMDR9 , and WWS ( Kava et al . , 2013 ) . The first step of the iMyoblast protocol was transgene-free iPSC myogenesis induction using commercially available reagents ( Caron et al . , 2016; Amsbio , Skeletal Muscle differentiation Kit ) ( Figure 1—figure supplement 1A ) . This three-stage iPSC myogenesis protocol induces cultures of Ctrl and disease iPSCs to sequentially upregulate expression of muscle master regulators , PAX3 ( S1 stage ) and MYOD1 ( S2 stage ) , and the muscle differentiation marker MYH8 ( S3 Stage ) ( Caron et al . , 2016 ) , as assayed by qPCR ( Figure 1—figure supplement 1B ) . Gene expression was also assayed in the FSHD1 and Ctrl Embryonic Stem Cell ( ESC ) lines originally used to develop and optimize this induction protocol to assure that iPSCs and ESCs respond similarly to this transgene-free myogenesis induction protocol ( Figure 1—figure supplement 1B ) . These studies established that Ctrl and disease iPSC and ESC lines robustly upregulated expression of PAX3 , MYOD1 , and MYH8 on the order of 1000-fold during S1 , S2 , and S3 stages of myogenic induction , validating the consistency and efficiency of the transgene-free induction protocol . Immunofluorescence ( IF ) assays showed that 90% of cells in S2 stage cultures were MYOD1+ , and 80% of cells in S3 stage cultures were MF20+ and predominantly mononucleated iMyocytes ( Figure 1A ) , similar to the first myogenic cells to differentiate in the embryo ( Lee et al . , 2013 ) . PAX7 expression was detected at 100-fold lower levels than PAX3 during S1 induction ( Figure 1—figure supplement 1B ) , consistent with earlier findings that PAX7+ cells are a minor cell population induced by transgene-free myogenesis protocols ( Chal et al . , 2015; van der Wal et al . , 2018 ) . The second step – for isolation of iMyoblasts – utilized reserve cell selection , as adapted from a protocol previously employed to isolate quiescent myogenic cells generated during differentiation of C2C12 myotube cultures by growth factor stimulation ( Yoshida et al . , 1998; Laumonier et al . , 2017 ) . iMyoblast reserve cells were recovered by activation of proliferation of undifferentiated cells resident in differentiated S3 muscle cultures using the same growth-factor-rich medium used to maintain proliferative cultures of adult biopsy myoblasts ( bMyoblasts ) . This myoblast growth medium promotes the efficient recovery of a proliferative , myogenic cell population of MYOD1+ cells , referred to as iMyoblasts ( Figure 1A ) . The iMyoblast protocol has been successfully applied to the isolation of iMyoblast lines from Ctrl iPSCs as well as classic and early onset FSHD1 , WWS , LGMDR7 , and LGMDR9 iPSCs . To validate the iMyoblast technology and establish its utility for disease studies , subsequent experiments were conducted in parallel with FSHD1 and Ctrl iMyoblasts , FSHD1 and Ctrl bMyoblasts , and also WWS , LGMDR7 , and R9 iMyoblasts . In growth factor-rich medium , bMyoblasts and iMyoblasts expressed muscle master regulatory genes PAX3 and MYOD1 whereas adult bMyoblasts expressed PAX7 in addition to PAX3 and MYOD1 ( Figure 1B ) . iMyoblasts proliferated in myoblast growth medium with 12 hr cell doubling times and could be expanded as primary lines for more than 12 passages ( >30 population doublings ) while cell-autonomously maintaining expression of PAX3 and MYOD1 and the commitment to differentiate in response to growth factor free medium , as assayed by expression of MYH8 and CKM ( Figure 1C ) . During their differentiation , iMyoblasts fused to form multinucleated iMyotubes ( Figure 1A ) . Proliferating FSHD1 and Ctrl iMyoblasts expressed cell surface markers typical of fetal and adult myogenic cells , including CD82 ( Alexander et al . , 2016; Pakula et al . , 2019 ) as well as CD56 , CD318 ( Uezumi et al . , 2016 ) , ERBB3 , and NGFR ( Hicks et al . , 2018; Figure 1D ) . FAC-sorted CD82+/CD56+ and CD82+/CD56− iMyoblasts both differentiated and fused to form MF20+ iMyotubes , validating the commitment of CD82+ iMyoblasts to differentiate ( Figure 1E ) . Finally , iMyoblasts retained their dual commitment to both differentiate and generate reserve cells that could be recovered by growth medium stimulation as MYOD1+ iMyoblasts ( tMyoblasts ) that can fuse and differentiate as MF20+ and MEF2C + iMyotubes ( Figure 1—figure supplement 2 ) . Single-cell RNA sequencing ( scRNA-Seq ) was used to define the iMyoblast transcriptome and compare it to transcriptome signatures of adult muscle bMyoblasts , and S1 and S2 stage cells undergoing iPSC myogenic induction . For these studies , iMyoblasts were generated by iPSC induction and reserve cell propagation from S3 cultures . iPSCs were reprogrammed from parental CD56+ muscle biopsy cells of six subjects , including two classic FSHD1 subjects ( 15A and 30A ) , one early onset FSHD1 subject ( 17A ) , and their unaffected Ctrl family members ( 15V , 17U , and 30W ) . These same six subjects were the source of S1 and S2 stage cells derived from iPSCs and unsorted adult muscle biopsy cells . Normalization of Unique Molecular Identifier ( UMI ) counts , cell-cycle estimation , dimension-reduction , and cell cluster identification were performed using Seurat ( Figure 2A ) . The Uniform Manifold Approximation and Projection ( UMAP ) plots grouped Ctrl and FSHD1 cells together in each of the five main clusters ( Figure 2B ) , as expected as the FSHD1 disease genes are expressed in differentiated myotubes and not in proliferating progenitors . Transcriptomes of differentiated iMyotubes , bMyotubes , and S3 stage muscle were not investigated in this study . UMAP grouped Ctrl and FSHD iMyoblasts into clusters that were distinct from S1 , S2 , bMyoblast ( bMyo ) , and biopsy mesodermal cell ( bMes ) clusters . Each cell type cluster included subsets of cells expressing genes of the different stages of the cell cycle and cells contributed by all six subjects , consistent with a reliable and sensitive UMAP segregation ( Figure 2B ) . Each of the five clusters had a distinct gene expression signature of myogenic regulatory genes , differentiation genes , and cluster marker genes that further validated their myogenic identities ( Figure 2C ) . The iMyoblast cluster expressed PAX3 and MYOD1 , the bMyoblast cluster expressed PAX3 and MYOD1 as well as PAX7 and MYF5 , and the bMes non-myogenic cluster expressed PDGFRA , a mesodermal cell marker ( Evseenko et al . , 2010; Joe et al . , 2010; Uezumi et al . , 2010; Ding et al . , 2013 ) . bMyo and bMes cells expressed NFIX , a regulator of the switch from embryonic to fetal myogenesis ( Messina et al . , 2010 ) , also expressed by iMyoblasts at lower levels ( Supplementary file 1 ) . Cells in the S1 cluster expressed PAX3 at higher levels than cells in the iMyoblast cluster and expressed LIN28A , which encodes an RNA binding protein controlling self-renewal and differentiation ( Shyh-Chang and Daley , 2013 ) . S2 cells had heterogeneous morphology ( Figure 2—figure supplement 1 ) and could be subdivided into a proximal S2A subcluster expressing PAX3 and a more distal S2B subcluster expressing MYOD1 , MYOG , and MYH8 muscle differentiation genes , and G1 cell cycle markers , consistent with their developmental progression from pre- to post-differentiation stages . These findings establish iMyoblasts as a myogenic cell with a transcriptome distinct from bMyoblasts and S1 and S2 stage myogenic cells . Differences in gene expression between iMyoblasts , bMyoblasts , bMes , S1 , S2A , and S2B cell classes were quantitated and subjected to pathway analysis using edgeR ( Robinson et al . , 2010 ) . This analysis was based on pseudo-bulk expression profiles ( Tung et al . , 2017 ) derived from scRNA-Seq data by summing counts of all cells from the same cell class and donor to avoid spurious results due to pseudoreplication ( Hurlbert , 1984 ) . The 15 pairwise comparisons among the six cell classes each identified at least 4600 differentially expressed coding and non-coding genes at a false discovery rate ( FDR ) <0 . 05 ( Supplementary file 1 ) . It is possible that some of these may be attributable to technical batch effects , as cells of the same type ( i . e . , S1 , S2 , iMyoblast , or bMyoblast ) from all donors were multiplexed during the single-cell encapsulation and library construction . This caution does not apply to comparisons of S2A versus S2B and bMyo versus bMes , although these may be biased toward small p-values since these clusters are defined based on the same transcriptomic data that is being compared between clusters ( Zhang et al . , 2019 ) . Tests of differential expression between FSHD1 and Ctrl samples from the same cell cluster are not subject to the biases above , but their power is limited by the small number of subjects and by low expression of DUX4 and its targets , as expected for proliferating FSHD1 cells . For this reason , we did not focus on these comparisons , though as a caution we note that the single gene , AC004556 . 1 , that had FDR<0 . 05 in these comparisons appears to be an annotation artifact: a common variant in the gene MRPL23 that happens to occur in these FSHD1 subjects but not the Ctrl subjects caused reads from MRPL23 to be assigned to the gene AC004556 . 1 on an unlocalized scaffold instead . A gene expression dot plot was generated from scRNA-Seq data to illustrate graphically the quantitative differences in gene expression and cell expression frequency across the six cell classes for each of the six subjects ( Figure 3 ) . This analysis focused on a manually curated set of differentially-expressed genes chosen based on their known developmental and regulatory functions in myogenesis . Some of these curated genes showed cell cluster-specific expression but many were expressed by multiple cell clusters , likely reflecting their shared developmental histories and myogenic functions . However , this dot plot illustrates that each cluster population has a distinct gene expression profile shared by cells from all six subjects in each cluster . These data further establish that iMyoblasts have a myogenic transcriptome that includes extracellular matrix ( ECM ) components , signaling molecules , and transcriptional regulators distinct from bMyoblasts and iPSC-induced S1 and S2 stage myogenic cells . Pathway analysis was also used to compare the transcriptomes of iMyoblasts with myogenic cells in other cell classes based on biological function . Gene ontology ( GO ) and KEGG pathway enrichment analyses were performed using the edgeR functions goana and kegga ( Young et al . , 2010 ) , applying more stringent cutoffs on differential expression , p-value<1E−06 and \|log2 ( FC ) \|>1 , and with enrichment analyses performed separately for upregulated and downregulated genes . The top-ranked GO and KEGG categories for each of the 15 pairwise comparisons , sorted by p-value for enrichment , are listed in Supplementary file 2 , as summarized below . The top-ranked categories for the iMyoblast versus bMyoblast comparison included categories of known myogenic genes , including ECM , focal adhesion , and migration/chemotaxis ( Gillies and Lieber , 2011; Csapo et al . , 2020; Thorsteinsdóttir et al . , 2011; Rayagiri et al . , 2018 ) , signaling ( Chal and Pourquié , 2017 ) , and transcription ( Berkes and Tapscott , 2005; Buckingham and Relaix , 2015 ) . Both the upregulated and downregulated genes were significantly enriched for ECM genes , including distinct collagen gene isoforms , with COL4A1 , COL4A2 , COL4A5 , COL4A6 , COL8A1 , COL11A1 , and COL13A1 UP in iMyoblasts compared to bMyoblasts , and COL6A1 , COL6A2 , COL6A3 , COL1A2 , COL5A2 , COL7A1 , COL8A2 , and COL22A1 UP in bMyoblasts compared to iMyoblasts . Other ECM genes UP in iMyoblasts included AGRN , QSOX1 , DSP , ECM1 , SLIT2 , EXT1 , EXTL1 , and EXTL3 , and those UP in bMyoblasts included DRAXIN , NFASC , EVL , and ELN . iMyoblasts and bMyoblasts also differentially expressed members of matrix processing enzyme gene families , including MMP , ADAMTS , and ADAM , and members of matrix regulatory protein gene families ITG , KRT , CDH , SEMA , and LAM , all of which have established regulatory functions during embryonic and adult myogenesis . Signaling was among the top-ranked GO and KEGG categories and included receptor regulatory activity , receptor ligand activity , and signaling receptor activator activity ( enriched among genes UP in bMyoblasts compared to iMyoblasts ) , and PI3K-Akt , MAPK , Rap1 , and Ras pathways ( enriched among genes UP in iMyoblasts vs . bMyoblasts ) . These pathways include differentially expressed FGF , WNT , and FZD gene family members . Additionally , TGFB , PDGFA , EPHA2 and EPHB2 , NOG , IGF2BP1 and IGF2BP3 , and HMGA2 were UP in iMyoblasts , while GDNF , VEGFA , BMP4 , BMP7 , WISP1 , SULF1 , and GREM2 were UP in bMyoblasts . Transcription categories included DNA-binding transcription activators , for which specific genes UP in bMyoblasts included KLF4 , SOX8 , SIX2 , PITX3 , SCX , SNAI1 , SNAI2 , and SMAD1 , and genes UP in iMyoblasts included GATA3 , GATA6 , HAND2 , MEIS2 , GLI2 , NOTCH1 , ETS1 , and ETS2 . The top-ranked KEGG categories for genes UP in bMyoblasts compared to iMyoblasts included mineral absorption , complement and coagulation cascades , arachidonic acid metabolism , and retinoic acid metabolism , whose functions in adult myogenesis are currently unknown . The results above focus on differences between iMyoblasts and bMyoblasts , but similarities between these cell types can be seen by contrasting each with clusters S1 and S2A , cells in earlier stages of myogenic induction . In all four pairwise comparisons between earlier stages ( S1 or S2A ) and committed iMyoblasts or bMyoblasts stages , genes UP in the committed myogenic stages are enriched for positive regulation of cell migration and TGFβ signaling , while genes UP in the earlier stage are enriched in steroid and cholesterol biosynthesis categories , likely to inhibit replicative stress and replication check point activation leading to cell cycle arrest ( Replogle et al . , 2020 ) . These findings validate the identity of iMyoblasts as a bona fide PAX3/MYOD1 myogenic cell , distinct both from S1 and S2 cells at early developmental stages of iPSC induction and from adult PAX7/PAX3/MYF5/MYOD1 bMyoblasts . The differentiation of Ctrl , FSHD1 , and LGMDR7 iMyoblasts was compared in cultures using growth factor-free N2 medium to induce myotube differentiation . iMyotube cultures from these iMyoblasts upregulated CKM and MYH8 muscle genes with similar kinetics , as determined by qPCR assays ( Figure 4A ) . iMyotubes expressed low levels of adult MYH1 compared to bMyotubes ( Figure 4A ) , consistent with their identity as a fetal/embryonic lineage . Ctrl and FSHD1 iMyotubes similarly upregulated the expression of muscle genes in both N2 and Opti-MEM growth factor-free media ( Figure 4—figure supplement 1A ) . Expression levels of myogenic regulators PAX3 , MYOD1 , and MYOG and CKM and MYH8 differentiation genes varied by cell line but were not specifically impacted by whether iMyoblasts were derived from iPSCs reprogrammed from fibroblasts or bMyoblast parental cells ( Figure 4—figure supplement 1B ) . The effects of myogenic signaling modulators on iMyotube and bMyotube differentiation were investigated by comparing N2 growth factor-free medium to N2 media supplemented with different combinations of myogenic signaling regulators previously shown to enhance myotube differentiation ( Hicks et al . , 2018; Tanoury , 2020; Selvaraj et al . , 2019 ) . These media included N2 + SB medium , supplemented with a TGFβ inhibitor SB431542 ( SB ) ; N2 + SB + P + C medium , supplemented with SB , corticosteroid Prednisolone ( P ) , and a GSK3 inhibitor/Wnt signaling activator , CHIR99021 ( C ) ; and N2+ SB + De + Da + F medium , supplemented with SB , the corticosteroid Dexamethasone ( De ) , α gamma-Secretase/Notch signaling inhibitor DAPT ( Da ) , and an adenyl cyclase activator Forskolin ( F ) ( Figure 4B ) . All three supplemented N2 media significantly increased expression of MYH7 , MYH8 , and CKM in Ctrl and FSHD1 iMyotubes , but not adult MYH1 , which as previously shown was expressed at high levels by bMyotubes except in N2+ SB + De + Da + F medium , which inhibited bMyoblast fusion and differentiation markers , but not MYOD1 and PAX3 expression ( Figure 4B and Figure 4—figure supplement 2B ) . Increased muscle RNA expression in iMyotubes was correlated with increased networks of multinucleated myotubes , most prevalent in N2 + SB + P + C medium ( Figure 4B and Figure 4—figure supplement 2A ) . The expression of myogenic regulators MYOD1 and PAX3 was variable , but their expression was not differentially affected by these media , showing that their effects are on differentiation and not myogenic commitment . By contrast , bMyoblast expression of muscle RNAs was increased in response to N2+ SB medium , particularly for FSHD1 bMyoblasts , likely by reducing DUX4-mediated toxicity . However , N2 + SB + P + C medium showed lower-level muscle RNA expression and N2 + SB + De + Da + F medium completely blocked muscle RNA expression and bMyoblast fusion ( Figure 4B and Figure 4—figure supplement 2B ) . These findings reveal that iMyoblasts and bMyoblasts , and FSHD and Ctrl cells , respond differently to specialized differentiation media , reflecting their underlying differences in operative signaling mechanisms and toxicity responses . To investigate whether iMyoblasts have utility for human disease modeling , we compared expression of the FSHD disease gene , DUX4 , in FSHD1 iMyoblasts and bMyoblasts . DUX4 is a primate-specific member of the double homeobox ( DUX ) family of transcription factor genes of eutherian mammals , located in the D4Z4 retrotransposon repeat array near the telomere of chromosome 4 ( Gabriëls et al . , 1999 ) . DUX4 developmentally functions to coordinate zygotic genome activation and male germline differentiation ( DeSimone et al . , 2017 ) . In other tissues of healthy individuals , the DUX4 locus is maintained in a highly condensed and CpG hypermethylated chromatin state and transcription of DUX4 is repressed . FSHD is caused by DUX4 misexpression in skeletal muscle in response to germline deletions and rearrangements that contract the D4Z4 locus on chromosome 4 to have ten or fewer repeats ( FSHD1 ) , or by mutations in chromatin-modifying genes such as SMCHD1 and DNMT3B in combination with semi-short D4Z4 repeat lengths ( FSHD2 ) . These genetic disruptions lead to chromatin decondensation and CpG hypomethylation of the D4Z4 repeat locus , resulting in low-frequency DUX4 transcription that activates a battery of more than 100 DUX4-regulated germline target genes in FSHD1 bMyotubes nuclei and in FSHD1 patient muscle biopsies ( Geng et al . , 2012; Snider et al . , 2010; Lemmers et al . , 2012; Yao et al . , 2014; van den Boogaard et al . , 2016; Lemmers et al . , 2010 ) . Clinical disease requires that DUX4 transcripts from the terminal D4Z4 unit be polyadenylated using a poly ( A ) site distal to the repeat array , associated with ‘disease permissive’ 4qA haplotypes ( Lemmers et al . , 2010 ) . Misexpression of DUX4 and its target genes in muscles of FSHD patients leads to muscle toxicity and degeneration , resulting in clinical disease ( DeSimone et al . , 2017; Lemmers et al . , 2010 ) . Expression of DUX4 and its target genes has previously been shown to be upregulated during the differentiation of patient biopsy-derived FSHD1 bMyoblasts , leading to myotube death ( Jones et al . , 2012; DeSimone et al . , 2017; Lemmers et al . , 2010 ) . The expression of DUX4 target genes MBD3L2 , TRIM43 , LEUTX , and ZSCAN4 was compared in Ctrl and FSHD1 iMyoblasts and bMyoblasts undergoing myotube differentiation in N2 serum-free differentiation medium . FSHD1 iMyoblasts and bMyoblasts both upregulated the expression of DUX4 target genes by >1000-fold compared to Ctrl iMyoblasts and bMyoblasts over 6 days of differentiation ( Figure 5A ) . bMyoblast target gene upregulation was delayed by 24 hr compared to FSHD1 iMyoblasts following differentiation induction with N2 medium . However , iMyoblasts and bMyoblasts upregulated differentiation genes MYH8 , MYH1 , and CKM with similar kinetics , suggesting that DUX4 transcription is more stringently repressed in bMyoblasts . DUX4 target gene expression was upregulated to similar levels in differentiating FSHD1 iMyoblasts derived from iPSCs reprogrammed from FSHD1 biopsy fibroblasts ( Figure 4—figure supplement 1B ) or FSHD1 biopsy bMyoblasts , indicating that parental somatic cell type used for iPSC reprogramming does not impact DUX4 regulation during iMyoblast differentiation . DUX4 expression also was assayed in FSHD1 and Ctrl iMyoblasts undergoing myotube differentiation using a DUX4-GFP reporter ( Rickard et al . , 2015 ) . FSHD1 iMyotubes expressed GFP in 4/100 nuclei , in contrast to undetectable expression in Ctrl iMyotubes ( Figure 5B ) . These findings show that FSHD iMyotubes sporadically upregulate DUX4 in myotube nuclei as shown previously for bMyotubes using DUX4 IHC assays ( Chen et al . , 2016 ) . However , nuclear frequency in iMyoblasts , as detected by the DUX4-GFP reporter , is 10× higher than in bMyoblasts , further indicating that DUX4 transcription may be more stringently repressed in bMyoblasts . The effects of media supplements on the expression of DUX4 target genes was assayed in FSHD1 and Ctrl cultures of iMyotubes , bMyotubes , and S3 iMyocytes using NanoString digital RNA assays ( Figure 5C ) . Findings revealed that iMyotubes expressed highest levels of DUX4 target genes in growth factor-free N2 medium and less so in N2 + SB medium , whereas additional media supplements dramatically reduced the expression of DUX4 target genes , in contrast to their strong enhancement of muscle gene expression , uncoupling DUX4 target gene expression from expression of muscle differentiation genes ( Figures 4B and 5C ) . bMyotubes expressed highest levels of DUX4 target gene and muscle RNAs in N2 + SB medium , less so in N2 and N2 + SB + P + C media , and not at all in N2 + SB + De + Da + F media which also blocked muscle RNA expression and bMyotube fusion ( Figure 4—figure supplement 2 ) . SB , a TGF-β inhibitor , optimally supports for DUX4 target gene and muscle gene expression by FSHD bMyotubes , and N2 + SB + P + C and N2 + SB + De + Da + F media repressed DUX4 target gene expression in both iMyotubes and bMyotubes , likely through the inclusion of corticosteroids previously shown to repress DUX4 ( Pandey et al . , 2015 ) . However , FSHD iMyoblasts and bMyoblasts were responsive to inhibition of DUX4 target gene expression by the p38 inhibitor , Losmapimod , currently in FSHD clinical trials ( Rojas , 2019; Figure 5—figure supplement 1 ) , showing that FSHD1 iMyotubes and bMyotubes share multiple pathways for DUX4 regulation that are suitable for drug development targeting DUX4 expression . DUX4 is an epigenetically regulated disease gene , which lead us to investigate whether iPSC reprogramming impacted DUX4 epigenetic regulation . As shown above , FSHD1 iMyoblasts upregulated DUX4 target gene expression during differentiation similarly to adult biopsy FSHD1 bMyoblasts ( Figure 5A , Figure 6—figure supplement 1A ) . However , we found that DUX4 and its transcriptional target genes were not upregulated during S3 myocyte differentiation in response to specialized differentiation media ( Figure 5C ) , or during earlier S1 and S2 stages of myogenic induction of both FSHD iPSCs and FSHD1 ESCs ( Figure 6A , Figure 6—figure supplement 1C ) . DUX4 and its target gene levels were higher in FSHD1 than Ctrl iPSCs and ESCs ( Figure 6A , Figure 6—figure supplement 1B ) but were 100-fold lower than FSHD1 iMyoblasts or bMyoblasts undergoing myotube differentiation . These findings contradict the earlier findings of Caron et al . , 2016 , who reported that DUX4 is upregulated tenfold during S3 stage differentiation in one of the FSHD1 ESC lines also investigated in our study . Our findings do not exclude a low level of DUX4 upregulation but it is small compared to the 1000-fold DUX4 upregulation we observed during FSHD1 iMyotube differentiation ( Figure 6—figure supplement 1 ) . To investigate whether the 4qA DUX4 locus became methylated during iPSC reprogramming to repress DUX4 expression , we performed bisulfite DNA sequencing ( Jones et al . , 2014 ) . 4qA alleles of FSHD1 bMyoblasts from three FSHD family cohorts were hypomethylated ( approximately 20% CpG methylation ) before iPSC reprogramming whereas Ctrl iMyoblasts are hypermethylated ( approximately 60% CpG methylation ) ( Figure 6B ) , as previously reported ( Jones et al . , 2015a ) . Since the uncontracted D4Z4 arrays for these two FSHD1 subjects have haplotypes not amplified in this assay ( 4qB and 4qA-L ) , methylation is specifically assayed only on contracted 4qA alleles . Our findings showed that DUX4 4qA remained hypomethylated in FSHD1 iPSCs and hypermethylated Ctrl iPSCs , and these methylation states were maintained throughout S1 , S2 , and S3 differentiation and in proliferating and differentiating iMyoblasts and iMyotubes ( Figure 6B and C ) . DUX4 4qA alleles associated with D4Z4 contracted chromosomes of FSHD1 ESCs were also hypomethylated compared to Ctrl ESCs ( Figure 6—figure supplement 2 ) , as assayed using DUX4 bisulfite sequencing with 4qA-specific primers ( Jones et al . , 2014 ) . These findings contradict a recent report , which found that DUX4 is hypermethylated in these same FSHD1 ESC lines ( Dion et al . , 2019 ) . However , unlike the 4qA-specific primers we used , the bisulfite sequencing primers used in this earlier study can amplify all D4Z4 repeat units from both chromosomes , which we found obscures the hypomethylation at the distal 4qA repeat encoding DUX4 ( data not shown ) . Unlike the DUX4 4qA locus , we found that the MYOD1 core enhancer sequences of bMyoblasts became hypermethylated during iPSC reprogramming of parental bMyoblasts and then became demethylated during myogenic induction and differentiation ( Figure 6D ) , concordant with MYOD1 RNA upregulation ( Figure 1A ) , as previously observed in developing mouse embryos ( Brunk et al . , 1996 ) . Therefore , methylation and demethylation machinery is operative in iPSCs and induced myogenic cells , but contracted 4qA alleles associated with FSHD1 are refractory to this machinery and changes in the methylation status of the 4qA DUX4 locus cannot account for the repression of DUX4 expression in FSHD1 iPSCs and ESCs and S3 iMyocytes . To investigate whether DUX4 repression during iPSC reprogramming is mediated by alternative epigenetic mechanisms , we screened a battery of epigenetic drugs for activation of DUX4 in S3 muscle cultures . The DNA demethylating drug 5-azacytidine ( 5-AzaC ) effectively increased DUX4 expression in FSHD1 S3 Myocyte cultures , in a concentration-dependent manner to levels comparable to those of iMyoblasts and bMyoblasts , as assayed by expression of the DUX4 target gene MBD3L2 ( Figure 6E ) . This finding shows that DUX4 repression is mediated by the 5-AzaC sensitive DNA methylation of a DUX4 regulatory locus that becomes inoperative in FSHD1 iMyoblasts and bMyoblasts . To investigate whether Ctrl and disease iMyoblasts xenoengraft and differentiate in vivo , FSHD1 and Ctrl iMyoblasts and bMyoblasts were engrafted into irradiated and BaCl2 injured TA muscles of NSG immune-deficient mice ( Figure 7A ) . Xenoengraftment was assayed by immunostaining with human-specific antibodies and by qPCR with human-specific qPCR primers . iMyoblast and bMyoblast xenografts were localized in humanized ECM domains within the mouse TA , as delineated by immunostaining with human-specific antibodies to lamin A/C , spectrin β1 , and muscle collagen VI ( Figure 7B and Figure 7—figure supplement 1 ) . These domains were predominantly occupied by human nuclei and muscle fibers and were largely devoid of mouse nuclei and fibers , as identified by immunostaining with nuclear lamin A/C and sarcolemmal spectrin-β1 human-specific antibodies ( Figure 7B ) . Variably sized spectrin β1+ fibers were detectable within 2 weeks following xenoengraftment and were associated with lamin A/C+ nuclei , with fibers oriented in parallel with residual peripheral mouse fibers along the sarcolemma matrix remaining after barium chloride destruction of mouse fibers . The numbers of spectrin β1+ fibers increased from 2 to 4 weeks and were higher in sections of bMyoblast xenografts than iMyoblast xenografts ( Figure 7B , right ) . FSHD1 iMyoblast and bMyoblast xenografts had fewer spectrin β1+ fibers than Ctrl xenografts , likely reflecting DUX4 and DUX4 target gene expression that results in fiber death ( Figure 7B ) . FSHD1 and Ctrl iMyoblast and bMyoblast xenografts expressed muscle genes MYH8 and CKM at comparable levels at 2 and 4 weeks post engraftment ( Figure 7C ) , further validating muscle differentiation in xenografts . Muscle RNA expression was lower in FSHD1 xenograft muscles and xenografts were smaller and more variable in size ( Figure 7B and C and Figure 7—figure supplement 1 ) . Dystrophin+ mouse muscle fibers persisted in more peripheral regions of TA muscle ( data not shown ) and did not have centralized nuclei , validating the effectiveness of hindlimb irradiation for blocking mouse satellite cell regeneration . DAPI+/lamin A/C− mouse nuclei were also present within human muscle xenografts , but were not directly associated with the sarcolemma of human fibers and included CD31+ mouse endothelial cells rebuilding the vascular system of xenograft muscle ( data not shown ) . The engraftment efficiencies of iMyoblasts and bMyoblasts were assessed using NanoString digital RNA analysis of engrafted TA muscles by taking the ratio of human- and mouse-specific probes . Assays of human and mouse RPL13A , a housekeeping gene , showed that its expression in bMyoblast xenografts is 20–60% human , compared to 5–20% human in iMyoblast xenografts . Assays of human and mouse MYH2 , a muscle gene , show that its expression in bMyoblast xenografts is 40–60% human , compared to 20–25% human in iMyoblast xenografts . The lower percentage of xenoengraftment using RPL13A assays likely reflects an abundance of non-muscle mouse cells in TA muscles compared to engrafted human cells , which are predominantly muscle . Overall , however , these findings show that iMyoblasts xenoengraft efficiently for investigations of muscle maturation and regeneration as described below . iMyoblasts and bMyoblasts differentially express a diversity of embryonic and adult ECM , matrix modifying , and regulatory genes that likely impact fetal-like iMyoblast xenoengraftment into the adult TA muscle ( Supplementary file 2 ) , providing a basis for enhancement of iMyoblast engraftment efficiencies . To investigate whether iMyoblast xenograft muscle undergoes embryonic/fetal to adult fiber type maturation in the host mouse TA fast-twitch muscle ( Kammoun et al . , 2014 ) , we compared expression of embryonic , adult fast , and adult slow Myosin Heavy Chain ( MYH ) isoforms ( Schiaffino et al . , 2015 ) ex vivo in cell culture and in vivo in xenografts using a custom NanoString panel of human-specific muscle RNA probes ( Figure 8 ) . Ex vivo , cultured FSHD1 and Ctrl iMyotubes and bMyotubes expressed high levels of MYH3 ( embryonic/fetal ) isoform and intermediate levels of MYH7 ( cardiac/slow twitch ) and MYH8 ( embryonic fetal/adult type 2A slow twitch ) isoforms . iMyotubes expressed low levels of MYH1 ( late fetal/adult fast twitch 2× ) and MYH2 ( fetal/adult fast twitch 2× ) compared to their high expression in bMyotubes . However , 2 and 4 weeks iMyoblast xenografts downregulated expression of MYH3 , MYH7 , and MYH8 isoforms as much as 100-fold , whereas expression of adult fast MYH1 and MYH2 isoforms was upregulated >100-fold . These findings show that iMyoblast xenografts undergo switching from embryonic/fetal to adult fast-twitch MYH isoforms . Similarly , but to a lesser extent , bMyoblast xenografts underwent MYH isoform switching , as evidenced by upregulation of MYH1 and MYH2 , and downregulation of MYH3 , MYH7 , and MYH8 ( Figure 8 ) , while the expression of the MYH4 isoform remained low and unchanged . These findings , therefore , establish that FSHD1 and Ctrl iMyoblast xenografts have regulatory plasticity in response to environmental signals from the adult host mouse TA fast-twitch muscle to promote fast MYH isoform switching and adult fast muscle maturation ( Wang and Kernell , 2001 ) . To further validate the utility of iMyoblasts for muscle disease research , we investigated the potential ex vivo and in vivo modeling applications of iMyoblasts from iPSCs derived from patients with FKRP dystroglycanopathies and LGMDR7 muscular dystrophy ( Figure 9 ) . LGMDR7 muscular dystrophy is an autosomal recessive muscular dystrophy resulting from coding mutations of the TCAP locus encoding Telethonin , a skeletal and cardiac muscle myofibrillar protein that interacts with Titin to maintain sarcomere integrity . The pathogenic TCAP mutation in the patient cell line we used is an 8 bp microduplication in the Telethonin coding sequence ( Cotta et al . , 2014 ) . LGMDR7 iMyoblasts undergo muscle differentiation and expression of muscle genes ex vivo ( Figure 4A ) and in muscle xenografts ( Figure 9A and B ) . Previously we showed that Telethonin expression could be efficiently restored in LGMDR7 iMyoblasts by microhomology-mediated repair using Cas9 and TCAP guide RNAs targeting the microduplication ( Iyer et al . , 2019 ) , establishing the utility of iMyoblasts for CRISPR disease gene editing therapeutics . Additionally , assays of the efficiencies of xenoengraftment showed that S1 stage FSHD1 cells engrafted poorly and did not differentiate well compared to iMyoblast xenografts , as evidenced by qPCR assays for muscle RNAs showing high raw Ct values comparable to unengrafted mouse TA samples ( Figure 9C ) . These data further highlight the efficiency of iMyoblast engraftment and their utility for in vivo studies of muscle maturation and regeneration . WWS and LGMDR9 ( formerly LGMD2I ) FKRP dystroglycanopathies are caused by recessive mutations in the coding sequences of the gene FKRP , disrupting its function in the glycosylation of α-Dystroglycan ( α-DG ) for laminin binding to maintain cellular interactions with the ECM ( Piccolo et al . , 2002 ) . WWS subjects carry loss-of-function FKRP mutations that fully disrupt α-DG glycosylation for laminin binding , causing clinically severe neuronal and muscle developmental damage and death following birth . LGMDR9 FKRP mutations cause partial loss of FKRP enzymatic function , resulting in adolescent onset of muscle weakness without neuronal involvement , and partial disruption of α-DG glycosylation and laminin binding . WWS iMyoblasts xenoengraft efficiently into irradiated and injured TA muscles of NSG mice , as evidenced by IF assays using human-specific lamin A/C and laminin β1 antibodies , and express muscle RNAs at similar abundance to Ctrl iMyoblasts , FSHD1 iMyoblasts , and LGMDR7 iMyoblasts ( Figure 9B and D ) . Ex vivo laminin binding assays of fluorescently labeled laminin to WWS iMyotubes showed almost complete absence of binding compared to Ctrl Myotubes , and LGMDR9 iMyotubes showed intermediate binding ( Figure 9E ) , reflecting the dose-dependent FKRP enzymatic loss of function . Similarly , α-DG glycosylation was assayed biochemically by probing Western blots of iMyotube membrane extracts with IIH6 antibody , which reacts with α-DG glycosylation epitopes , and with α-DG core protein antibody that assays glycosylation based on the reduced mobility of α-DG on SDS gels . WWS iMyotubes were nearly completely deficient in IIH6 reactivity and had smaller , unglycosylated α-DG core protein ( 75 kDa ) compared to glycosylated Ctrl iMyotubes ( 100 kDa ) , whereas LGMDR9 had intermediate IIH6 intensity ( Figure 9F ) , showing that differences in glycosylation between WWS and the less severe LGMDR9 alleles can model the functional and clinical severity of FKRP mutations . Taken together , these data demonstrate the capacity of iMyoblasts in ex vivo and in vivo assays and provide well-developed models for FKRP therapeutic development as well as for muscle regeneration studies , as described below . To investigate whether iMyoblast xenograft muscle had regenerative potential , primary xenografts were generated by xenoengraftment of Ctrl and WWS iMyoblasts and assayed by immunostaining for PAX3 expressing cells ( Figure 10A ) . Both Ctrl and WWS primary xenografts had a significant population of PAX3+ nuclei in humanized regions of the engrafted TA muscle , indicating that xenografts maintain a progenitor cell population . PAX3+ nuclei co-stained with human-specific lamin A/C and were associated with human muscle fibers as detected by immunostaining with human-specific spectrin-β1 whereas other nuclei appeared to be localized more interstitially . Additionally , xenografts included lamin A/C+ nuclei that did not immunostain with PAX3 ( Figure 10B ) . To investigate whether xenografts have regenerative potential , we induced a secondary barium chloride injury in WWS iMyoblast engrafted muscle . PAX3+/lamin A/C+ nuclei were observed in humanized muscle after secondary injury ( Figure 10B ) . WWS iMyoblast secondary injury xenografts were assayed for differentiated muscle fibers using human-specific neonatal myosin heavy chain , collagen VI , and laminin β1 staining identifying human muscle fibers in humanized areas of injury and regeneration ( Figure 10C ) . These findings support the conclusion that iMyoblast xenografts maintain a population of PAX3+ myogenic cells that function to regenerate human muscle in response to injury . Here , we report the invention of an efficient and reliable method to isolate human PAX3-expressing muscle stem cells , iMyoblasts , from iPSCs of Ctrl and muscular dystrophy patients using transgene-free iPSC myogenic induction in combination with reserve stem cell selection , which has not been previously utilized for isolation of iPSC-derived muscle progenitors . Other iPSC myogenesis protocols have been optimized to generate differentiated skeletal muscle from mouse and human ESCs and iPSCs , either by transgene misexpression of muscle master regulatory genes including MYOD1 ( Dekel et al . , 1992; Maffioletti et al . , 2015 ) and PAX7 ( Darabi et al . , 2012; Rao et al . , 2018 ) , or by transgene-free protocols that transition ESCs and iPSCs through an early developmental progression that regulates skeletal myogenesis in the vertebrate embryo ( Caron et al . , 2016; Chal et al . , 2015; van der Wal et al . , 2018; Hicks et al . , 2018 ) rather than through direct induction of PAX3-expressing muscle progenitors as in our protocol . Our approach also differs in its efficient reserve cell selection of highly enriched , stably committed , and expandable populations of iMyoblasts by growth selection , without requiring FAC sorting of induced subpopulations . Our findings demonstrate the utility of our protocol for the isolation of iMyoblasts from iPSCs from healthy control ( Ctrl ) subjects and patients with FSHD1 , LGMDR7 , WWS , and LGMDR9 muscular dystrophies that express the molecular pathologies of these diseases . iMyoblasts can be maintained and expanded for at least 30 population doublings while stably retaining their capacity for myogenic differentiation . iMyoblast cell expansion enables statistically powered cell and molecular studies of human myogenesis and investigations of muscle disease pathology and therapeutic development using both ex vivo and in vivo in muscle xenograft models , as demonstrated in this study . By contrast , adult bMyoblasts often cannot be propagated from biopsies of muscular dystrophy patients who have advanced muscle pathology or have limited growth in culture because of the age and disease-related senescence ( Webster and Blau , 1990 ) . The advantages of iPSC derived iMyoblasts for mechanistic studies of epigenetic disease mechanisms are shown by our discovery of a previously unknown developmental epigenetic repression of the FSHD1 disease gene , DUX4 . Finally , iMyoblasts undergo enhanced differentiation in response to specialized media , which enable 3D modeling of muscle maturation and contractile and electrophysiological activities that will further enhance knowledge of disease mechanisms and enable therapeutic development . iMyoblasts express markers of fetal myoblasts , including PAX3 and CD82 , as well as NFIX , a regulator of the transcriptional switch from embryonic to fetal myoblasts ( Messina et al . , 2010 ) . The iMyoblast transcriptome is distinct from iPSC-induced S1 and S2 myogenic progenitors , which do not express NFIX and differentiate as mononucleated myocytes typical of the earliest stage embryonic muscle , and also is distinct from adult bicep muscle bMyoblasts , which express PAX7 , MYF5 , MRF4 as well as PAX3 and MYOD1 , and higher levels of NFIX . The iMyoblast transcriptome also is distinct from the transcriptomes of iPSC-induced PAX7 Myoblasts , but share the fetal-like gene expression profiles along the developmental trajectory of limb myogenic cells described by Xi et al . , 2020 . iMyoblasts also may be related to adult PAX3 myogenic cells identified in muscles such as diaphragm or to stress resistant PAX3 myogenic cells ( Der Vartanian et al . , 2019; Scaramozza et al . , 2019 ) , noting that , in humans , PAX3 and PAX7 may be functionally redundant , as suggested by recent genetic studies showing that PAX7 is not an essential gene for satellite cell production and muscle regeneration ( Marg et al . , 2019 ) . In addition to PAX3 and MYOD1 , iMyoblasts express an array of other transcription factors , including GATA3 , GATA6 , HAND2 , MEIS2 , GLI2 , NOTCH1 , ETS1 , and ETS2 , that most certainly play a role to regulate the isoform-specific expression of ECM , focal adhesion , migration/chemotaxis , and cell signaling genes of the iMyoblast transcriptome . The fetal-like phenotype of iMyoblasts ex vivo is further reflected by the expression of embryonic/fetal MYH8 during iMyotube differentiation , in contrast to MYH1 expression by adult bMyoblasts . Significantly , iMyoblast xenoengraftment into the mouse TA muscle generated muscle that downregulated expression of embryonic MYH8 and upregulated expression of adult fast MYH1 during fiber differentiation , demonstrating the transcriptional plasticity of iMyoblasts to respond to in vivo signals in the fast fiber type TA muscle to mediate adult MYH isoform switching . The signals in xenograft muscle that initiate adult switching may be controlled by physiological mechanisms such as innervation ( Xi et al . , 2020 ) or by conversion of engrafted iMyoblasts to a more adult-like bMyoblast progenitor phenotype . In either case , the capacity of iMyoblasts to engraft directly into adult muscle to produce xenograft muscle that undergoes adult MYH isoform switching in combination with the capacity of iMyoblasts for extensive cell expansion and efficient CRISPR gene editing ( Iyer et al . , 2019 ) supports their potential utility for development of muscle stem cell therapies , which to date have been unsuccessful . iMyoblasts have growth and differentiation characteristics of self-renewing stem cells based on their dual commitment to differentiation and iMyoblast reserve cell renewal ex vivo and the capacity of iMyoblast xenograft muscles to regenerate human muscle and PAX3+ cells in response to secondary injury . The stem cell characteristics of iMyoblasts likely reflect their isolation using reserve cell selection from cultures of highly differentiated S3 stage muscle following transgene-free induction using the myogenic induction protocol , which was optimized to generate differentiated cultures of S3 iMyocytes without transitioning iPSCs through earlier mesodermal and somite developmental stages ( Caron et al . , 2016 ) . A realized consequence was that this differentiation protocol also generated reserve cells that could be isolated from differentiated cultures by growth factor activation . The mechanisms that control reserve cell generation are not yet understood , but likely are based on the initiation of MYOD1 autoregulation in earlier stage cells during the S1 and S2 stages of iPSC induction , as shown in previous studies of MYOD1 overexpression in somatic cells ( Weintraub et al . , 1989 ) . iMyoblasts maintain low level MYOD1 and PAX3 expression cell-autonomously , which enables their myogenic potential over 30 population doublings while retaining their commitment to differentiate as iMyotubes and replenish PAX3 iMyoblasts as reserve cells . In vivo , this enables iMyoblasts to xenoengraft and generate differentiated muscle and a population of PAX3 expressing cells that are retained in xenografts in response to secondary injury . The regulatory mechanisms that control the distinct growth , differentiation and self-renewal capacities of iMyoblasts are not yet understood . iMyoblasts express genes well-known to regulate MYOD1 function , myoblast differentiation and self-renewal , including ID genes ( Benezra et al . , 1990 ) , Wnt and BMP antagonists DKK1 ( Jones et al . , 2015b ) and GREM1 ( Fabre et al . , 2020 ) , and RGS4 ( Yilmaz et al . , 2016 ) , TXNRD1 ( Mercatelli et al . , 2017 ) , UCHL1 ( Gao et al . , 2017 ) , NR2F2 ( Lee et al . , 2017 ) , FOXC2 ( Lagha et al . , 2009 ) , and HMGA2 and IGF2BP myogenic regulators ( Li et al . , 2012 ) . Studies to investigate the myogenic and stem cell functions of these iMyoblast genes and others identified in our studies are now approachable using CRISPR gene editing . Overall , our findings establish a protocol for isolation of iMyoblasts and establish its suitability for ex vivo and in vivo investigations of human myogenesis and muscle disease pathogenesis . Our expectation is that iMyoblasts will enable development and validation of multiple disease corrective modalities , including gene editing and muscle stem cell therapeutics , to ameliorate disease pathology and disabilities associated with muscular dystrophies . Human iPSCs were generated from CD56+ myoblasts or CD56− fibroblasts enriched from bicep muscle biopsies , or skin fibroblasts at the UMASS Medical School Transgenic Animal Modeling Core using CytoTune-iPS Sendai Reprogramming Kit ( Thermo Fisher Scientific ) . Isolated iPSC clones were characterized by pluripotency identification ( OCT4 staining ) , in vivo teratomas formation , and karyotyping assay . iPSC lines were routinely maintained on Matrigel ( Corning ) with StemMACS iPS-Brew XF ( Miltenyi Biotec ) . The cells were passaged every 4 days using StemMACS Passaging Solution XF ( Miltenyi Biotec ) and the Rock inhibitor Y27632 ( 10 μM , STEMCELL Technologies ) for 24 hr to improve the survival rates . 15AM iPSCs , 15VM iPSCs , 17AM iPSCs , 17AF iPSCs , 17UM iPSCs , 30AM iPSCs , 30WM iPSCs , 54574/75 iPSCs , 54585 iPSCs , and LGMDR7 iPSCs were reprogrammed at UMass Medical School . FP and WWS iPSCs were provided by Anne Bang at Sanford Burnham Prebys Medical Discovery Institute and 54574/75 and 54585 fibroblasts were provided by Steven Moore at the University of Iowa . iPSC cell line identity was confirmed by STR profiling . iMyoblast isolation iPSCs , typically plated on six-well plates , were induced for myogenic differentiation using media prepared by Genea Biocells and now commercially available as a skeletal muscle differentiation kit ( Amsbio ) following the manufacturer’s specifications ( Caron et al . , 2016 ) . After 6–7 days in S3 differentiation medium ( SKM03 ) , culture plates were rinsed with 2 ml 1× phosphate-buffered saline ( PBS ) and cells detached from plates in 0 . 5 ml TrypLE Express at 37°C and diluted with 4 . 5 ml HMP growth medium ( Ham’s F-10 supplemented with: 20% FBS , 1% chick embryo extract rich with growth factors [Emerson Lab] ) , 1 . 2 mM CaCl2 , 1% antibiotic/antimycotic ( Gibco ) ( optional ) . The cell suspension was pipetted 5–10 times to disperse cells and clear plates of residual cells , and then 0 . 5–1 ml of this cell suspension was plated onto 10 cm gelatin-coated dishes in 10 ml HMP medium , cells were cultured at 5% CO2 at 37°C and fed daily with fresh HMP medium to support the growth of iMyoblasts . After 2–3 culture passages , iMyoblast cells were recovered and maintained as frozen stocks for investigations . 15AM iMyoblasts , 15VM iMyoblasts , 17AM iMyoblasts , 17AF iMyoblasts , 17UM iMyoblasts , 30AM iMyoblasts , 30WM iMyoblasts , 54574/75 iMyoblasts , 54585 iMyoblasts , LGMDR7 iMyoblasts , LGMDR9 FP iMyoblast , and WWS iMyoblasts were generated at UMass Medical School . Identity for cell lines used in single-cell RNA-sequencing was confirmed by SNP analysis . FKRP ( for WWS and LGMDR9 ) and TCAP ( for LGMDR7 ) mutations were confirmed by DNA sequencing and aDG expression . All cell lines have been tested for mycoplasma and have tested negative . FAC sorted CD56+ and unsorted bMyoblasts recovered from muscle biopsies ( Homma et al . , 2012; Jones et al . , 2012 ) and iMyoblasts isolated from iPSC muscle cultures were routinely maintained on gelatin-coated 10 or 15 cm dishes in HMP growth medium and passaged at 70–90% confluence . To induce differentiation , cultures were grown to 95% confluence , then washed with PBS and cultured in serum-free Opti-MEM or N2 medium ( DMEM/F12 supplemented with 1% N2 supplement , 1% ITS , and 1% L-glutamine ) ( Barberi et al . , 2007; Chal et al . , 2016 ) for 2–7 days at 37°C/5% CO2 . The fusion index was calculated as the percentage of nuclei in MF20-positive fibers ( ≥2 nuclei ) to total nuclei . 15Abic biopsy myoblast , 15Vbic biopsy myoblast , 17Abic biopsy myoblast , 17Ubic biopsy myoblast , 30Abic biopsy myoblast , and 30Wbic biopsy myoblast cell lines were generated at UMass Medical School . Cells were fixed on plates with 2% paraformaldehyde for 20 min at 37°C . After rinsing plates three times with PBS , the cells attached to plates were treated with blocking/permeabilizing solution ( PBS containing 2% bovine albumin , 2% goat serum , 2% horse serum , and 0 . 2% Triton X-100 ) for 30 min at room temperature and then incubated with primary antibodies in PBS at 4°C overnight , washed three times in PBS , and incubated with the corresponding secondary antibodies for 1 hr at room temperature . Plates were washed two times in PBS and cells stained for 5 min with DAPI ( Sigma-Aldrich ) to stain nuclei and fluorescence images were taken using a Nikon Eclipse TS 100 inverted microscope . Single-cell suspensions of iMyoblasts and bMyoblasts cultures dissociated with TrypLE Express Enzyme ( Thermo Fisher Scientific ) were washed with PBS , filtered with a 40-μm strainer , and incubated with antibodies suspended in PBS for 30–60 min on ice in the dark . Cells were then washed in PBS and resuspended in PBS and 0 . 2% fetal calf serum , and flow cytometry was performed at UMass Medical School Flow Cytometry Core . A BD FACS Aria IIu was used for quantification and a BD FACS C-Aria II Cell Sorter was used for cell sorting . FlowJo software was used for data analysis . DUX4 expression was assayed in bMyoblasts and iMyoblasts expressing DUX4-GFP reporter using lentiviral vector and G418 selection ( Rickard et al . , 2015 ) . Cells were infected using a modified spin-down method ( Springer and Blau , 1997 ) . In brief , 105 cells per well were plated on gelatin coated six-well plates in HMP medium . The next day , cells were incubated with DUX4-GFP lentivirus in HMP medium for 15 min and then centrifuged at 1100×g for 30 min at 32°C . Medium containing virus was replaced with fresh medium and cells were cultured for 48 hr , then treated with 300 μg/ml G418 for 7 days for selection , with daily feedings and passaging at 90% confluence . Genomic DNA was isolated from cell pellets of FSHD1 and Ctrl ESCs ( Genea Biocells ) and iPSCs , S2 cells and iMyoblasts using QIAamp DNA Blood Mini Kit ( QIAGEN ) and bisulfite converted using the EpiTect Kit ( QIAGEN ) following the manufacturer’s specifications . DUX4 4qA and 4qA-L were PCR amplified from bisulfite treated DNA using nested primers ( Jones et al . , 2014 ) and the MYOD1 Core Enhancer with primers that include CpG sites 1 , 2 , and 3 ( Brunk et al . , 1996 ) as shown in Supplementary file 1 . Amplified DNAs were cloned into the pCR2 . 1 TOPO vector , which was transformed into TOP10 Chemically Competent Escherichia coli and selected for kanamycin resistance . Cloned DNA of plasmids from kanamycin-resistant colonies were sequenced using Sanger sequencing ( Sequegen ) , and CpG methylation analysis was analyzed by Bisulfite Sequencing DNA Methylation Analysis ( BISMA ) online software ( Rohde et al . , 2010 ) . For each of the four cell types—cultures at stages S1 , S2 , iMyoblasts , and unsorted primary biopsy cells—the cells from three FSHD1 and three control subjects were detached from plates , dissociated and pooled immediately before loading ~10 , 000 cells on a Chromium platform ( 10× Genomics ) for scRNA-Seq . 3′ Gene Expression v2 libraries ( 10× Genomics ) from four Chromium runs were sequenced on an HS4K instrument at the UMMS Deep Sequencing Core . Cell Ranger version 3 . 1 . 0 ( 10× Genomics ) and STAR 2 . 5 . 1b were used to align reads from FASTQ files to the human reference genome . Gene annotations from GRCh38 . 93 were prepared with cellranger mkgtf as in Cell Ranger , but with filtering modified to also retain gene biotypes ‘processed_transcript’ and ‘bidirectional_promoter_lncRNA . ’ Initial filtering , barcode counting , and UMI counting yielded an estimate of 29 , 049 potential cell barcodes . 92 . 4% of the 562 . 5 million total reads were in cells . SNPs for each subject were genotyped from bulk RNA-seq , performed by Novogene , of S1 and iMyoblast cells for each subject in Python 2 . 7 . 9 using Opossum 0 . 2 ( Oikkonen and Lise , 2017 ) and Platypus 0 . 8 . 1 ( Rimmer et al . , 2014 ) . These genotypes were used to assign cells from pooled scRNA-Seq runs to their subject of origin using Demuxlet ( downloaded on 08/03/2018 ) ( Kang et al . , 2018 ) , and 5 . 7% of the cells were filtered out that did not unambiguously match a single genotype . Further filtering using the Seurat 3 . 1 . 4 package ( Butler et al . , 2018; Stuart et al . , 2019 ) in R 3 . 6 . 2 removed 7 . 0% of the remaining cells that contained ≥12% UMIs mapped to mitochondrial genes , ≤ 1000 or ≥ 5500 detected genes , or ≥40 , 000 detected UMIs . This resulted in 24 , 991 cells , with a median of 2678 genes detected per cell and 7909 UMI per cell ( Figure 2A ) . Cell clustering , cell cycle state estimation , and downstream analyses were performed for these cells in Seurat . Normalization and scaling were performed using SCTransform ( Hafemeister and Satija , 2019 ) with regression against the percentage of mitochondrial gene expression . The 3000 genes with highest variability across cells were used for principal component reduction , and components 1–30 were used to construct a shared nearest neighbor graph for unsupervised cell clustering using the Louvain algorithm . This resulted in 16 cell clusters ( 3 for S1 cells , 4 for S2 cells , 5 for iMyoblast cells , and 4 for muscle biopsy primary cells ) , which were visualized using the UMAP method with resolution 0 . 7 ( McInnes et al . , 2018 ) . Most of the within-cell-type subclusters were merged to reduce to six clusters , keeping separate the bMes cluster that was clearly distinct from the other three primary cell subclusters comprising bMyoblasts ( Figure 2B ) , and the S2A cluster that expressed low MYOD1 and CDH15 and distinguished it from the other three S2 subclusters comprising more differentiated S2B cells , as described in the Results . The top 200 differentially expressed genes from each pairwise cell-type comparison using edgeR ( below ) were reviewed to construct a curated set of genes relevant to myogenesis or cell-type expression ( shown as a Seurat dot plot in Figure 3 ) . Values from the single-cell raw count matrix ( prior to normalization ) were summed for each combination of the six cell types of interest ( S1 , S2A , S2B , iMB , bMB , and bMes ) and the six donors ( 15A , 15V , 17A , 17U , 30A , and 30W ) to obtain a table of pseudobulk counts for these 36 samples ( Tung et al . , 2017; Supplementary file 3 ) . Tests of differential expression were performed on these pseudobulk counts using the R 4 . 0 . 1 package edgeR 3 . 30 . 3 ( Robinson et al . , 2010; Lun et al . , 2016 ) . Lowly expressed genes were filtered out with the function filterByExpr , and counts were normalized using calcNormFactors to yield values for 14 , 103 filtered genes across the 36 samples . We specified a quasi-likelihood negative binomial generalized log-linear model with cell-type and donor as additive factors in the model . We used estimateDisp to estimate the dispersion for all genes , glmQLFit ( robust=T ) to fit a joint model to the data from all cell-types , and glmQLFTest to perform statistical tests of differential expression for each of the 15 contrasts between pairs of cell-types . The FDR was computed with the function topTags . Genes that satisfied ( unadjusted ) p-value<1 . 0E−06 and \|log2 ( FC ) \|>1 were considered differentially expressed and ranked by p-value for each comparison . The results for all genes sorted by p-value are reported in Supplementary file 1 , which also includes the FDR for each gene . Comparisons between the three FSHD1 and three control donors used the same edgeR procedure as above , but with models fit separately for each cell type , including the filterByExpr step . For GO/KEGG analyses , the differentially expressed ( DE ) genes , separated into up and down sets based on the sign of log2 ( FC ) , were used as DE input , and the full set of genes from the edgeR analysis after the filterByExpr step was used as the gene universe . Overrepresentation of GO terms in BP , CC , and MF ontologies ( annotations from org . Hs . eg . db 3 . 11 . 4 ) were computed using the goana function and in KEGG pathways using the kegga function in edgeR , in both cases using log2 ( CPM ) [CPM = counts per million] as a covariate to adjust for potential biases due to gene expression level ( Young et al . , 2010 ) . The top GO and KEGG categories based on overrepresentation p-value , for each comparison between cell types , are summarized in Supplementary file 2 . Immune deficient NOD . Cg-PrkdcscidIL2rγtmiWjl /SzJ ( NSG , Jackson Lab ) mice that lack the ability to produce mature B cells , T cells , and natural killer ( NK ) cells and are highly sensitive to irradiation were used in accordance with the Institutional Animal Care and Use Committee ( IACUC ) at the University of Massachusetts Medical School . NSG mice were anesthetized with ketamine/xylazine and their hindlimbs were subjected to 18 Gy of irradiation using a Faxitron RV-650 or Faxitron CellRad X-ray cabinet to ablate the host mouse satellite cell population . One day after irradiation , mice were anesthetized with isoflurane and TA muscles were injected with 50 μl of 1 . 2% Barium Chloride ( Sigma-Aldrich ) bilaterally to degenerate mouse muscle . Three days after muscle injury , 1×106 bMyoblasts or iMyoblasts were resuspended in 50 μl 1 mg/ml laminin ( Sigma-Aldrich , L2020 ) in PBS and injected bilaterally into the body of TA muscles . Xenoengrafted mice were euthanized 2–4 weeks post engraftment by CO2 asphyxiation followed by cervical dislocation . For IF experiments , TA muscles were isolated and embedded in optimal cutting temperature ( OCT , Tissue Tek ) compound , frozen in liquid nitrogen cooled isopentane and kept at –80°C until cryosectioning . For RNA isolation , xenoengrafted TA muscles were snap-frozen in liquid nitrogen and kept at –80°C until RNA extraction . Frozen TA muscles embedded in OCT were cryosectioned using a Leica CM3050 S Cryostat . Tissue sections 10 μm thick were mounted onto Superfrost Plus glass microscope slides ( Thermo Fisher Scientific ) and kept at –20°C until immunostained . When thawed , the sections were fixed with ice-cold acetone for 10 min at –20°C . We employed the ‘mouse-on-mouse’ ( MOM ) kit ( Vector Laboratories ) to reduce non-specific antibody staining per the manufacturer’s specifications . Antibodies ( key resource table ) were used sequentially then slides were incubated with Hoechst block for 10 min . Following 2× for 5 min PBS washes , the slides were dried and coverslips mounted with Fluorogel . Fluorescent images were taken using a Leica DMR fluorescence microscope . RNA was isolated either from cells in culture or from xenoengrafted TA muscles using the RNeasy Plus Mini Kit ( QIAGEN ) or Aurum Total RNA Fatty and Fibrous Tissue kit ( Bio-Rad ) , respectively , per the manufacturer’s specifications . For qPCR analysis , 2–5 μg of total RNA was converted into cDNA using the SuperScript III First-Strand Synthesis System ( Invitrogen ) . For quantification of housekeeping genes , DUX4 target genes or muscle differentiation marker expression , 20 ng of cDNA was used for each reaction . For quantification of DUX4 expression , 90–150 ng of cDNA was used in each reaction . For NanoString digital RNA quantification , 50 ng or 150 ng of total RNA was used for cell culture and xenografted TA muscle , respectively . An Emerson lab custom muscle NanoString panel with human-specific probes for muscle protein genes , muscle master regulatory genes , developmental transcription factors , signaling genes , and multiple housekeeping genes was used for all analyses on an nCounter Sprint profiler ( NanoString Technologies , Seattle , WA ) . Raw mRNA counts were normalized to a panel of housekeeping genes ( RPL13A , GAPDH , GUUSB , and VCP ) ( Figure 7 ) or just RPL13A ( Figure 5 ) using nSolver software ( NanoString Technologies , Seattle , WA ) . We found that normalizing to either the panel of four housekeeping genes or just to RPL13A gives comparable results . All qPCR and NanoString experiments included three biological replicates and each data point represents an individual biological replicate unless otherwise specified . Statistics qPCR data are shown as the mean± SEM . Statistical differences for qPCR data were evaluated using Student’s t-test and were considered significant when the p-value was less than 0 . 05 ( =p<0 . 05 , =p<0 . 01 , ***=p<0 . 0001 ) . Statistical comparisons were performed using GraphPad Prism software . Statistical methods for scRNA-Seq are described in Single-cell RNA-Seq section above .	Muscular dystrophies are a group of inherited genetic diseases characterised by progressive muscle weakness . They lead to disability or even death , and no cure exists against these conditions . Advances in genome sequencing have identified many mutations that underly muscular dystrophies , opening the door to new therapies that could repair incorrect genes or rebuild damaged muscles . However , testing these ideas requires better ways to recreate human muscular dystrophy in the laboratory . One strategy for modelling muscular dystrophy involves coaxing skin or other cells from an individual into becoming ‘induced pluripotent stem cells’; these can then mature to form almost any adult cell in the body , including muscles . However , this approach does not usually create myoblasts , the ‘precursor’ cells that specifically mature into muscle during development . This limits investigations into how disease-causing mutations impact muscle formation early on . As a response , Guo et al . developed a two-step protocol of muscle maturation followed by stem cell growth selection to isolate and grow ‘induced myoblasts’ from induced pluripotent stem cells taken from healthy volunteers and muscular dystrophy patients . These induced myoblasts can both make more of themselves and become muscle , allowing Guo et al . to model three different types of muscular dystrophy . These myoblasts also behave as stem cells when grafted inside adult mouse muscles: some formed human muscle tissue while others remained as precursor cells , which could then respond to muscle injury and start repair . The induced myoblasts developed by Guo et al . will enable scientists to investigate the impacts of different mutations on muscle tissue and to better test treatments . They could also be used as part of regenerative medicine therapies , to restore muscle cells in patients .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine", "developmental", "biology", "tools", "and", "resources" ]	2022	iMyoblasts for ex vivo and in vivo investigations of human myogenesis and disease modeling
In amyotrophic lateral sclerosis ( ALS ) the large motoneurons that innervate the fast-contracting muscle fibers ( F-type motoneurons ) are vulnerable and degenerate in adulthood . In contrast , the small motoneurons that innervate the slow-contracting fibers ( S-type motoneurons ) are resistant and do not degenerate . Intrinsic hyperexcitability of F-type motoneurons during early postnatal development has long been hypothesized to contribute to neural degeneration in the adult . Here , we performed a critical test of this hypothesis by recording from identified F- and S-type motoneurons in the superoxide dismutase-1 mutant G93A ( mSOD1 ) , a mouse model of ALS at a neonatal age when early pathophysiological changes are observed . Contrary to the standard hypothesis , excitability of F-type motoneurons was unchanged in the mutant mice . Surprisingly , the S-type motoneurons of mSDO1 mice did display intrinsic hyperexcitability ( lower rheobase , hyperpolarized spiking threshold ) . As S-type motoneurons are resistant in ALS , we conclude that early intrinsic hyperexcitability does not contribute to motoneuron degeneration . Glutamate excitotoxicity has long been suggested to contribute to the degeneration of motoneurons in amyotrophic lateral sclerosis . Intrinsic hyperexcitability of motoneurons , which increases discharge probability and thereby calcium inflow , has been assumed to participate in the excitotoxic process ( Ilieva et al . , 2009 ) . However , it was recently suggested that hyperexcitability improves motoneuron survival ( Saxena et al . , 2013 ) . Regardless of its effect , it is still not clear whether spinal motoneurons are hyperexcitable in mutant superoxide dismutase 1 ( mSOD1 ) mice , a standard model of amyotrophic lateral sclerosis ( ALS ) . Indeed , changes in excitability occur very early in mSOD1 mice ( Elbasiouny et al . , 2010 ) . Motoneurons from mSOD1 embryos recorded in culture are hyperexcitable ( Pieri et al . , 2003; Kuo et al . , 2005 ) : they are recruited at lower current and display higher F–I gain . Similarly , Martin et al . ( 2013 ) found , in an in vitro preparation of mSOD1 embryonic cord , that motoneurons are also hyperexcitable: their dendritic tree is reduced causing an increase in input resistance . Investigations in neonates have led to contradictory results . Hypoglossal motoneurons were reported to be hyperexcitable ( F–I gain is increased , van Zundert et al . , 2008 ) . However , Pambo–Pambo et al . ( 2009 ) did not observe any change in spinal motoneuron input resistance , rheobase , or stationary gain suggesting that their excitability was unchanged . In the same line , Quinlan et al . ( 2011 ) found that the excitability of spinal motoneurons is homeostatically maintained despite an increase in their input conductance ( recruitment current and F–I gain unchanged ) . In contrast , Bories et al . ( 2007 ) reported a decrease in input resistance causing the spinal motoneurons to be hypoexcitable . These discrepancies might be due to the location of the mutation on the SOD1 gene , the number of transgenes , or other factors . Until now , however , the fact that the motor unit population is heterogeneous , even in neonates ( Jansen and Flatby , 1990 ) , has never been taken into account . Motoneurons innervate different types of muscle fibers and display different patterns of discharge during the second post-natal week both in rats ( Russier et al . , 2003 ) and in mice ( Pambo–Pambo et al . , 2009 ) . Indeed , for liminal current pulses , the discharge starts at the pulse onset in some motoneurons ( immediate firing pattern ) but is delayed in others ( delayed firing pattern ) . Here , we provide electrical , morphological and molecular evidence that immediate firing spinal motoneurons innervate slow-contracting fibers ( S-type motoneurons ) whereas delayed firing motoneurons innervate fast-contracting fibers ( F-type motoneurons ) . We then investigated whether these two populations are equally affected in neonatal mSOD1 mice . We show that this is not the case: only the immediate firing motoneurons are hyperexcitable . Their rheobase is decreased because of a more hyperpolarized voltage threshold for spiking . In sharp contrast , the excitability of the delayed firing motoneurons is unchanged . Since the F-type motoneurons are vulnerable in ALS whereas the S-type motoneurons are resistant ( Pun et al . , 2006; Hegedus et al . , 2008 ) , the remarkably selective intrinsic hyperexcitality of S-type motoneurons in neonates indicates that intrinsic hyperexcitability is not an early event that triggers degeneration of the motoneurons . In neonatal mice , spinal motoneurons can be sorted according to their firing pattern . Figure 1A illustrates how two motoneurons in a wild-type ( WT ) mouse discharge in response to a long ( 5 s ) square pulse at rheobase , that is , the minimal current pulse that elicits at least one action potential in our protocol ( see ‘Materials and methods’ ) . In the example of Figure 1A1 , the motoneuron did not fire at the onset of a 1 . 6 nA square pulse . Instead the motoneuron depolarized slowly and started to fire only 2 . 9 s after the pulse onset when the membrane potential reached the voltage threshold for spiking ( −35 mV , dashed line ) . Once the firing started , its frequency increased . This motoneuron displayed the delayed firing pattern . A very important feature of this pattern is illustrated in Figure 1—figure supplement 1 . The delay was long at rheobase ( Figure 1—figure supplement 1A ) but it progressively decreased ( Figure 1—figure supplement 1B–C ) to finally disappear ( Figure 1—figure supplement 1D ) as the current intensity increased . 63 out of 94 WT motoneurons ( 67% ) exhibited this delayed firing pattern . The remaining motoneurons displayed the so-called immediate firing pattern: at rheobase , the motoneuron discharged at the pulse onset without any delay ( Figure 1A2 ) . In the immediate firing pattern , the spiking frequency remained constant with little variability . 31 out of the 94 WT motoneurons ( i . e . , 33% ) displayed the immediate firing pattern . 10 . 7554/eLife . 04046 . 003Figure 1 . Electrical and morphological properties of motoneurons displaying the delayed and the immediate firing patterns . ( A1 ) WT motoneuron displaying the delayed firing pattern in response to a 5 s pulse . The current intensity was the minimal intensity necessary to elicit firing in our searching protocol ( rheobase ) . Bottom: injected-current ( square pulses ) , middle: voltage-response and top: instantaneous firing frequency . The horizontal dashed line shows the voltage threshold for spiking ( −50 mV ) . ( A2 ) Response of a WT motoneuron displaying the immediate firing pattern . Same arrangement as in A1 . ( Voltage threshold for spiking: −59 mV ) . ( B1 ) Single action potentials from a delayed ( black line ) and an immediate ( gray line ) firing WT motoneurons elicited by a short square pulse of current . The arrowheads point to the horizontal dotted bar drawn at half action potential amplitudes . ( B2 ) Same records at a longer time base in order to show the after hyperpolarisation ( AHP ) . Dashed lines are the exponential fits of the AHP relaxation . The relaxation time constants are 17 ms and 37 ms for the delayed firing and the immediate firing motoneurons , respectively . Note that the relaxation time constant is longer in the immediate firing motoneuron than in the delayed firing one . C Reconstructed dendritic trees of WT delayed ( C1 ) and immediate ( C2 ) firing motoneurons . The axon was not reconstructed in either case . Reconstructions are projected in the same plane as the slice . DOI: http://dx . doi . org/10 . 7554/eLife . 04046 . 00310 . 7554/eLife . 04046 . 004Figure 1—figure supplement 1 . In delayed firing motoneurons , the delay depends on the intensity of stimulation . ( A–D ) Responses of a delayed firing motoneuron to long lasting pulses ( 5 s ) of increasing amplitude . Bottom: injected-current ( square pulses in nA ) . Top: voltage-response ( in mV ) . Note that the delay decreased when the current intensity increased . DOI: http://dx . doi . org/10 . 7554/eLife . 04046 . 004 On average , the input conductance is smaller , the rheobase is lower and the voltage threshold for spiking is more hyperpolarized in immediate firing motoneurons than in delayed firing ones and this occurs despite of the fact that the resting membrane potential is similar ( Table 1 ) . Moreover , immediate and delayed firing motoneurons display differences in the shape of their action potentials . Immediate firing motoneurons have wider action potentials and a longer relaxation time constant of their after-hyperpolarization ( AHP ) compared to delayed firing motoneurons ( see Figure 1B and Table 1 ) . These results suggest that the two phenotypes ( immediate and delayed firing ) are linked to two different populations of motoneurons . 10 . 7554/eLife . 04046 . 005Table 1 . Electrophysiological propertiesDOI: http://dx . doi . org/10 . 7554/eLife . 04046 . 005WT micemSOD1 micep-valueResting membrane potential ( mV ) Delayed firing−64 ± 3−65 ± 30 . 2−70/−56−70/−59N = 63N = 31Immediate firing−65 ± 3−64 ± 20 . 2−71/−59−70/−60N = 31N = 18p-value0 . 30 . 1Input conductance ( nS ) Delayed firing52 ± 2854 ± 300 . 810/15122/153N = 63N = 31Immediate firing33 ± 2433 ± 160 . 66/986/62N = 31N = 18p-value0 . 00070 . 01Rheobase ( nA ) Delayed firing1 . 2 ± 0 . 61 . 1 ± 0 . 50 . 40 . 3/2 . 80 . 3/2 . 6N = 57N = 30Immediate firing0 . 6 ± 0 . 40 . 3 ± 0 . 20 . 0080 . 05/1 . 60 . 1/0 . 6N = 29N = 16p-value<0 . 0001<0 . 0001Voltage threshold for spiking ( mV ) Delayed firing−33 ± 7−31 ± 100 . 7−47/−17−50/−10N = 58N = 30Immediate firing−44 ± 7−49 ± 60 . 03−50/−41−50/−30N = 30N = 17p-value<0 . 0001<0 . 0001Voltage threshold for spiking–Resting membrane potential ( mV ) Delayed firing31 ± 833 ± 100 . 417/4913/50N = 59N = 31Immediate firing20 ± 714 ± 50 . 0058/316/21N = 30N = 17p-value<0 . 0001<0 . 0001Recruitment current on ramp ( nA ) Delayed firing1 . 1 ± 0 . 61 . 1 ± 0 . 50 . 80 . 1/2 . 80 . 3/2 . 5N = 51N = 29Immediate firing0 . 6 ± 0 . 50 . 3 ± 0 . 30 . 020 . 07/20 . 07/1N = 25N = 15p-value0 . 001<0 . 0001Action potential amplitude ( mV ) Delayed firing89 ± 1387 ± 110 . 566/12171/111N = 29N = 19Immediate firing84 ± 1181 ± 150 . 466/10461/110N = 21N = 13p-value0 . 20 . 2Action potential width ( ms ) Delayed firing1 . 4 ± 0 . 51 . 3 ± 0 . 40 . 60 . 7/2 . 50 . 6/2 . 2N = 29N = 19Immediate firing1 . 7 ± 0 . 41 . 8 ± 0 . 60 . 71 . 1/2 . 90 . 9/3 . 1N = 21N = 13p-value0 . 040 . 01AHP relaxation time constant ( ms ) Delayed firing27 ± 923 ± 50 . 211/5015/34N = 21N = 12Immediate firing42 ± 1248 ± 27121/6019/91N = 11N = 7p-value0 . 0040 . 02 Further supporting the hypothesis of two separate populations , we found that the dendritic trees of delayed and immediate firing motoneurons display different morphologies . We filled motoneurons with neurobiotin in order to investigate the dendritic tree architecture . Only the dendrites that remained in the slice plane were considered for analysis ( see ‘Materials and methods’ ) . In these conditions , the number of primary dendrites ( and thereby dendritic trees ) per motoneuron is similar in immediate and delayed firing motoneurons ( Table 2 ) . This allows us to make relevant morphological comparisons on the reconstructed trees . As exemplified in Figure 1C , the dendritic arborization extends further in delayed firing motoneurons than in immediate firing motoneurons ( compare Figure 1C1 and Figure 1C2 ) . In WT mice , delayed firing motoneurons have on average more branching points ( 44 ± 14 , 26 to 72 , N = 14 ) compared to the immediate firing motoneurons ( 27 ± 13 , 13 to 52 , N = 10 , p = 0 . 007 ) , larger total dendritic length and longer dendritic paths than immediate firing motoneurons ( Table 2 ) . 10 . 7554/eLife . 04046 . 006Table 2 . Morphological propertiesDOI: http://dx . doi . org/10 . 7554/eLife . 04046 . 006WT micemSOD1 micep-valueSoma area ( µm2 ) Delayed firing630 ± 160620 ± 1400 . 2350/1000270/890N = 60N = 31Immediate firing530 ± 180454 ± 1100 . 2260/940250/640N = 30N = 17p-value0 . 0090 . 0002Primary dendritesDelayed firing6 . 4 ± 2 . 06 . 7 ± 1 . 20 . 34/125/9N = 14N = 14Immediate firing6 . 3 ± 2 . 76 . 4 ± 4 . 00 . 93/102/13N = 10N = 5p-value10 . 5Total dendritic length ( mm ) Delayed firing8 . 3 ± 2 . 98 . 7 ± 3 . 80 . 92 . 3/143 . 8/16 . 5N = 14N = 14Immediate firing5 . 3 ± 1 . 53 . 6 ± 0 . 30 . 013 . 0/8 . 03 . 3/4 . 1N = 10N = 5p-value0 . 010 . 0003Dendritic paths ( µm ) Delayed firing296 ± 135293 ± 1390 . 85/68715/840N = 653N = 449Immediate firing252 ± 141181 ± 157<0 . 000111/80311/685N = 281N = 180p-value<0 . 0001<0 . 0001Terminal segments length ( µm ) Delayed firing112 ± 93108 ± 1020 . 54/4573/742N = 618N = 447Immediate firing108 ± 9181 ± 920 . 0015/5454/584N = 296N = 178p-value0 . 60 . 001Note that the ‘overbranching motoneuron’ ( arrowhead on Figure 8B1 ) is excluded for analysis . A body of evidence indicates that both immediate and delayed firing motoneurons are alpha-motoneurons ( i . e . , motoneurons that innervate extrafusal muscle fibers ) and not gamma-motoneurons ( i . e . , motoneurons that innervate intrafusal muscle fibers ) . First , the soma sizes ( Table 2 ) of both immediate and delayed firing motoneurons are in the range of alpha motoneurons ( see supplemental data in Friese et al . ( 2009 ) for the soma size distribution of alpha and gamma-motoneurons in P14 mice ) . Note , however , that on average , immediate firing motoneurons have smaller soma areas than delayed firing motoneurons ( Table 2 ) . Second , since alpha-motoneurons , but not gamma-motoneurons , receive monosynaptic Ia inputs ( Eccles et al . , 1960 ) , we checked whether immediate and delayed firing motoneurons receive such proprioceptive inputs . We recorded motoneurons , characterized their discharge pattern and filled them with neurobiotin . Subsequently , we stained the vesicular glutamate transporter 1 ( VGlut1 ) since VGlut1 is expressed in terminals from primary afferents but not in terminals from excitatory interneurons or descending fibers ( Oliveira et al . , 2003; Friese et al . , 2009 ) . VGlut1 afferents are known to innervate not only alpha-motoneurons but also Renshaw cells in the neonate ( Mentis et al . , 2006 ) . However , since we identified the labelled cells as motoneurons on the basis of antidromic action potential recordings , we are confident that VGlut1 terminals shown on Figure 2A are apposed on alpha-motoneurons and not on Renshaw cells . We observed that both delayed and immediate firing motoneurons received proprioceptive terminals on the soma and proximal dendrites ( Figure 2A , arrows in insets ) . Third , we checked for the expression of the neuronal nuclear antigen ( NeuN ) , a known marker for alpha-motoneurons ( Friese et al . , 2009; Shneider et al . , 2009 ) . Both immediate and delayed firing motoneurons expressed NeuN ( Figure 2B ) . 10 . 7554/eLife . 04046 . 007Figure 2 . immediate and delayed firing motoneurons both receive VGlut1 inputs and express NeuN . ( A ) Vglut1 ( red ) synaptic inputs are apposed to neurobiotin ( green ) filled motoneurons ( examples of appositions pointed by arrowheads in the inserts that show enlargements of the areas surrounded by rectangles ) . The bar scale in all insets is 5 μm . ( B ) NeuN staining ( red ) of neurobiotin ( green ) filled motoneurons . The arrowheads point to the cell bodies of the motoneurons that have been intracellularly filled during the electrophysiological experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 04046 . 007 It has been suggested that the estrogen-related receptor β ( Errβ ) is expressed in S-type but not in F-type motoneurons . Conversely , chondrolectin ( Chodl ) is expressed in a fraction of F-type motoneurons but not in S-type motoneurons ( Enjin et al . , 2010 ) . We therefore tested the expression of these two molecular markers on immediate and delayed firing motoneurons . Figure 3A shows an immediate firing motoneuron expressing Errβ and a delayed firing motoneuron that did not . Remarkably , all 6 investigated immediate firing motoneurons proved to be Errβ-positive whereas all 8 delayed firing motoneurons were Errβ-negative . This molecular distinction matches with differences in the electrical properties ( Figure 3B ) . In situ hybridization allowed us to investigate whether immediate and delayed firing motoneurons expressed Chodl mRNA ( Figure 4A ) . None of the 9 investigated immediate firing motoneurons did . 7 out of the 15 investigated delayed firing motoneurons were Chodl-positive whereas the remaining 8 were Chodl-negative . Altogether , the immediate firing motoneurons were all Errβ-positive and Chodl-negative as expected for S-type . On the other hand , the delayed firing motoneurons were all Errβ-negative and about half of them were Chodl-positive whereas the other half were chondrolectine-negative as expected for F-type motoneurons ( Enjin et al . , 2010 ) . Interestingly , the Chodl-positive motoneurons tended to display the highest rheobases ( Figure 4B ) . 10 . 7554/eLife . 04046 . 008Figure 3 . Immediate firing motoneurons , but not delayed firing motoneurons , express ERRβ . ( A ) Examples of Errβ . staining ( red ) in neurobiotin ( green ) filled motoneurons . The arrowhead in the first row points to the nucleus ( Errβ−positive ) of the recorded motoneuron . The asterisk in the second row indicates that the nucleus of the recorded motoneuron was Errβ−negative . ( B ) Plot of the AHP relaxation time constants against the rheobases for labelled motoneurons . All immediate firing motoneurons were Errβ−positive whereas all delayed firing motoneurons were Errβ−negative . Arrowheads point to the motoneurons illustrated in A . DOI: http://dx . doi . org/10 . 7554/eLife . 04046 . 00810 . 7554/eLife . 04046 . 009Figure 4 . The largest delayed firing motoneurons express chondrolectin mRNA contrary to immediate firing motoneurons . ( A ) Examples of chondrolectin in situ hybridizations ( red ) in neurobiotin filled motoneurons ( green ) . The asterisks in the first and third rows indicate that the cell body of the recorded motoneuron was chondrolectin negative . The arrowhead in the second row indicates that chondrolectin is expressed in this delayed firing motoneuron . ( B ) Plot of the AHP relaxation time constants against the rheobase for the investigated motoneurons . About half of the delayed firing motoneurons ( the ones that display the highest rheobase ) were chondrolectin-positive . All the immediate firing motoneurons were chondrolectin-negative . Arrowheads point to the motoneurons illustrated in A . DOI: http://dx . doi . org/10 . 7554/eLife . 04046 . 009 In addition to identifying them as F- and S-type motoneurons , we set out to directly identify which motoneurons sub-population was vulnerable in ALS . Kaplan et al . ( 2014 ) recently showed that the matrix metalloproteinase-9 ( MMP9 ) is strongly expressed in motoneurons that are the most vulnerable to ALS , that is , those of the large , fast contracting and fatigable ( FF ) motor units . Conversely , the most resistant motoneurons , that is , those that innervate the slow contracting fibers , were devoid of MMP9 . We therefore investigated whether MMP9 was expressed in delayed and immediate firing motoneurons . All 5 investigated immediate firing motoneurons were MMP9 negative ( Figure 5A , first row ) . In contrast , 5 out of the 10 investigated delayed firing motoneurons displayed a strong MMP9 labelling ( Figure 5A , second row ) . Four other delayed firing motoneurons did not express MMP9 ( Figure 5A , third row ) and one was weakly labelled . The delayed firing motoneurons that express MMP9 tend to exhibit the largest input conductances and rheobases ( Figure 5B ) suggesting that they innervate the largest motor units . These results indicate that the immediate firing motoneurons are resistant during ALS , in keeping with our identification as S-type motoneurons . 10 . 7554/eLife . 04046 . 010Figure 5 . The largest delayed firing motoneurons express MMP9 contrary to immediate firing motoneurons . ( A ) Examples of MMP9 labelling in neurobiotin filled motoneurons ( green ) . The asterisks in the first and third rows indicate the cell body of the recorded motoneurons devoid of MMP9 expression . The arrowhead in the second row indicates that MMP9 was expressed in this delayed firing motoneuron . ( B ) Plot of the rheobase against the input conductance for the investigated motoneurons . Half of the delayed firing motoneurons ( the ones that display the highest rheobases and input conductances ) were MMP9-positive . All the immediate firing motoneurons were MMP9-negative . Arrowheads point to the motoneurons illustrated in A . DOI: http://dx . doi . org/10 . 7554/eLife . 04046 . 010 In mSOD1 mice , we observed the same immediate and delayed firing patterns ( Figure 6A ) as in WT mice . 31 out of 49 mSOD1 motoneurons ( 63% ) exhibited the delayed firing pattern ( Figure 6A1 ) whereas the 18 remaining motoneurons ( 37% ) displayed the immediate firing pattern ( Figure 6A2 ) . The proportion of delayed and immediate firing motoneurons is not different between WT and mSOD1 mice ( Fisher's exact test , p = 0 . 5 ) . Moreover , the resting membrane potential , the input conductance , the action potential width and the AHP relaxation time constant of each motoneuron subtype are unchanged in mSOD1 mice compared to WT mice ( Table 1 ) . 10 . 7554/eLife . 04046 . 011Figure 6 . mSOD1 immediate firing motoneurons are selectively hyperexcitable . ( A1-2 ) mSOD1 motoneurons displaying the delayed firing pattern ( A1 ) and the immediate firing pattern ( A2 ) . The current intensity was the minimal intensity necessary to elicit firing ( rheobase ) . Bottom: injected-current ( square pulses ) , middle: voltage-response and top: instantaneous firing frequency . The horizontal dashed line shows the voltage threshold for spiking ( −37 mV for the delayed firing motoneuron and −48 mV for the immediate firing motoneuron ) . ( B1-2 ) Plot of rheobase as a function of input conductance for delayed ( B1 ) and immediate ( B2 ) firing motoneurons . In each plot , open squares are for WT motoneurons whereas red dots are for mSOD1 motoneurons . Linear regressions are indicated by dashed lines . DOI: http://dx . doi . org/10 . 7554/eLife . 04046 . 011 Despite the unchanged input conductance , the excitability of the immediate firing motoneurons , but not of the delayed firing ones , is altered in mSOD1 mice . As expected , the rheobase , that is , a measure of cell excitability , increases with input conductance ( Figure 6B ) . However , in immediate firing motoneurons , the slope of the linear regressions is significantly smaller in mSOD1 mice than in WT mice ( Figure 6B2 , 15 vs 22 mV t test p = 0 . 04 , Chow test p = 0 . 0004 ) . This is not the case for delayed firing motoneurons ( Figure 6B1 , 19 vs 22 mV , t test p = 0 . 5 , Chow test p = 0 . 2 ) . On average the rheobase of immediate firing motoneurons is two times smaller in mSOD1 mice compared to WT mice ( Table 1 ) . On the other hand , the rheobase of delayed firing motoneurons is not significantly affected in mSOD1 mice ( Table 1 ) . However , the resting membrane potential of immediate firing motoneurons is unchanged in mSOD1 mice ( Table 1 ) . The decrease in rheobase in mSOD1 immediate firing motoneurons is instead due to an hyperpolarization of the voltage threshold for spiking ( Table 1 ) . Accordingly , the difference between the voltage threshold for spiking and the resting membrane potential is smaller for the immediate firing motoneurons in mSOD1 mice ( Table 1 ) . As a consequence , a smaller amount of current is required to reach the voltage threshold for spiking in these motoneurons . Motoneuron excitability was also assessed on the basis of their responses to slow triangular ramps of current ( Figure 7—figure supplement 1 ) . The current at which the first action potential was fired ( recruitment current ) during a slow ramp is indeed another way to measure the rheobase . The recruitment current on the slow ramps was very close to the rheobase measured using the current pulses ( Table 1 ) , and again , it was significantly smaller in mSOD1 mice only in the immediate firing motoneurons . Regardless of the way we measured the rheobase ( long pulses or slow ramps ) we found that the immediate firing motoneurons , but not the delayed firing ones , are hyperexcitable in mSOD1 mice . We have previously shown that mixed mode oscillations ( MMOs ) are related to the excitability state of motoneurons ( Iglesias et al . , 2011 ) . MMOs are small oscillations of the membrane potentials between full action potentials ( arrowheads in Figure 7A2 ) ( Manuel et al . , 2009 ) . They create variability in the firing discharge . In most cases they are present only at low current intensity , defining a sub-primary firing range ( Manuel et al . , 2009; Turkin et al . , 2010 ) . We have shown in a previous study ( Iglesias et al . , 2011 ) that MMOs are caused by a relative deficit of sodium current with respect to potassium current , which is due to a slow sodium inactivation . MMOs therefore reflect a low excitability state . In the case of the delayed firing motoneurons , MMOs were observed nearly in all cells ( 49 out of 50 ) recorded in WT animals ( Figure 7A1 ) and in all 31 mSOD1 motoneurons ( Figure 7A2 ) . This is again an indication that their exitability is unaltered by the mutation . In the case of the immediate firing motoneurons , MMOs are encountered in nearly all WT motoneurons ( 13 out of 15 , Figure 7B1 ) but they are absent in most mSOD1 motoneurons ( MMOs are lacking in 8 out of 11 mSOD1 immediate firing motoneurons , Figure 7B2 ) . The proportion of immediate firing motoneurons exhibiting MMOs in mSOD1 is thus significantly reduced compared to WT motoneurons ( Fisher's exact test , p = 0 . 003 ) . The absence of MMOs in the immediate firing motoneurons of mSOD1 mice further indicates that these motoneurons are more excitable than in WT mice . 10 . 7554/eLife . 04046 . 012Figure 7 . Absence of mixed mode oscillations in mSOD1 immediate firing motoneurons . Beginning of the discharge during the injection of slow triangular ramp of current at 0 . 1 nA/s velocity . See Figure 7—figure supplement 1 for the full traces and F–I curves . Arrowheads point to oscillations between full spikes , the signature of mixed mode oscillations . Note that there is no oscillations between spikes in the mSOD1 immediate firing motoneuron ( B2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04046 . 01210 . 7554/eLife . 04046 . 013Figure 7—figure supplement 1 . Responses of a delayed firing motoneuron ( upper row ) and of an immediate firing motoneuron ( lower row ) to a slow current ramp . Same motoneurons as in Figure 7 . ( A1–A2 ) WT delayed firing motoneuron . ( A1 ) Bottom: injected-current ( slow triangular ramp at the 0 . 1 nA/s velocity ) and top: voltage-response . ( A2 ) F–I curves for the ascending ( black dots ) and descending ramps ( red squares ) . Note that the recrutement current occurs at higher intensity than the de-recruitment current . ( B1–B2 ) mSOD1 delayed firing motoneuron . In A2 and B2 the dashed horizontal line points to the frequency reached during the ascending phase ( black dots ) 0 . 5 s after the motoneuron has started to discharge . These frequencies are 28 Hz for the WT delayed firing motoneuron in A2 and 30 Hz for the mSOD1 delayed firing motoneuron in B2 . ( C1–C2 ) WT immediate firing motoneuron . The vertical dashed line is the separation between the subprimary range in which MMOs are present and the primary range ( PR ) in which MMOs are absent during the ascending ramp . The oblique blue dashed line is the linear regression in the primary range . Its slope ( 29 Hz/nA ) is the gain . ( D1–D2 ) mSOD1 immediate firing motoneuron . MMOs are absent , there is only a primary range ( PR ) during the ascending ramp ( gain: 41 Hz/nA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04046 . 013 When a primary range could be observed in immediate firing motoneurons ( Figure 7—figure supplement 1C2 , D2 , dark points ) , its slope was not significantly different between WT and mSOD1 mice ( 35 ± 22 Hz nA−1 , 11–80 Hz nA−1 , N = 12 for WT vs 26 ± 8 Hz nA−1 , 16–41 Hz nA−1 for mSOD1 , N = 9 , p = 0 . 7 ) . Since the gain in the primary range is largely determined by the AHP conductance ( Brownstone et al . , 1992; Kernell , 2006; Manuel et al . , 2006 ) , this result suggests that the AHP conductance is unchanged in mSOD1 mice . In delayed firing motoneurons , a gain could not be measured because it is difficult to identify a linear primary range ( see Figure 7—figure supplement 1A2 , B2 , dark points ) . However , we measured the firing frequency reached 0 . 5 s after the recruitment and we found that it is not different between delayed firing motoneurons of WT ( 30 ± 7 Hz , 17 to 49 Hz , N = 42 ) and mSOD1 mice ( 29 ± 6 Hz , 13 to 42 Hz , N = 26 ) . In mSOD1 mice , similarly to WT mice , delayed firing motoneurons are larger than immediate firing ones ( Figure 8A ) . However the morphology of the dendritic tree is affected specifically in immediate firing motoneurons . Figure 8B shows the relationships between the number of branching points and the total dendritic length for WT and mSOD1 mice: the more branching points ( and therefore branches ) , the longer the total dendritic length . In the case of delayed firing motoneurons , WT and mSOD1 relationships are largely overlapping ( Figure 8B1 , slopes of the linear regressions: 0 . 18 vs 0 . 18 mm/branching point , t test p = 1 , Chow test p = 0 . 7 ) , except for one motoneuron ( arrowhead in Figure 8B1 ) that displays more branching points and a longer total dendritic length than the largest WT motoneurons . This particular motoneuron thereby displays an overbranching of its dendritic tree . If we exclude this outlayer motoneuron , the total dendritic length , the length of dendritic paths ( Figure 8C1 ) and the length of terminal segments are not different in delayed firing motoneurons of mSOD1 and WT mice ( Table 2 ) . 10 . 7554/eLife . 04046 . 014Figure 8 . The dendritic tree of immediate firing motoneurons is shrunk in mSOD1 mice . ( A1-2 ) Reconstructed dendritic trees of mSOD1 delayed ( A1 ) and immediate ( A2 ) firing motoneurons . The axons were not reconstructed . ( B1-2 ) Plots of total dendritic length as a function of number of branching points for delayed firing motoneurons ( B1 ) and immediate firing motoneurons ( B2 ) . In each plot , open squares are for WT motoneurons whereas red dots are for mSOD1 motoneurons . Arrowhead in B1 points at an outlying over-branching mSOD1 delayed motoneuron . Linear regressions are indicated by dashed lines . ( C1-2 ) Distribution of the dendritic paths for delayed ( C1 ) and immediate firing motoneurons ( C2 ) . Distributions between WT ( open columns ) and mSOD1 ( red columns ) dendritic paths are compared for each firing patterns . DOI: http://dx . doi . org/10 . 7554/eLife . 04046 . 014 In sharp contrast , immediate firing motoneurons undergo profound changes in mSOD1 mice . Their total dendritic length is 32% smaller in mSOD1 motoneurons than in WT motoneurons ( Table 2 ) while the number of branching points is not significatively different ( mSOD1 mice: 27 ± 14 , 9–45 , N = 5; WT mice: 27 ± 13 , 13–52 , N = 10 , p = 0 . 8 ) . As a result the relationships between the total dendritic length and the number of branching points are very different ( Figure 8B2 , slopes of the linear regressions: 0 . 02 vs 0 . 1 mm/branching point t test p = 0 . 02 , Chow test p = 0 . 009 ) . Consistently , the dendritic paths and terminal segments are on average 28% and 25% shorter , respectively , in mSOD1 mice ( Table 2 ) . Figure 8C2 shows that the distribution of the dendritic paths is strongly skewed towards small lengths in mSOD1 mice ( Kolmogorov–Smirnov bilateral test , p < 0 . 0001 ) . Altogether , our data indicate that the dendrites of immediate firing motoneurons are shorter in mSOD1 mice than in WT mice . We provide electrical , morphological and molecular evidence that the immediate firing motoneurons are S-type motoneurons whereas the delayed firing ones are F-type motoneurons . First , immediate and delayed firing motoneurons are alpha-motoneurons and not gamma-motoneurons , since both of them receive proprioceptive inputs ( VGlut1 contacts ) , express NeuN and have soma sizes in the alpha-motoneuron range ( Friese et al . , 2009 ) . Second , the immediate firing motoneurons express Errβ but not Chodl as expected for S-type motoneurons ( Enjin et al . , 2010 ) . Conversely , the delayed firing motoneurons do not express Errβ and those likely to innervate the largest fast-contracting motor units express Chodl as seen by Enjin et al . ( 2010 ) . Third , delayed firing motoneurons display , on average , larger input conductances , higher rheobases , and shorter AHPs than the immediate firing ones . This is in keeping with the electrical differences that have been observed between F-type and S-type motoneurons in adult cats ( Burke , 1981; Zengel et al . , 1985 ) , rats ( Beaumont and Gardiner , 2002; Button et al . , 2006 ) and mice ( Manuel and Heckman , 2011 ) . Finally , the delayed firing motoneurons display a larger dendritic tree than the immediate firing motoneurons . Similar differences in the dendritic trees have been reported between F-type and S-type motoneurons in cats ( Burke et al . , 1982; Cullheim et al . , 1987 ) . In most studies of neonate motoneurons , the distinction between the immediate and the delayed firing patterns has been overlooked ( Fulton and Walton , 1986; Vinay et al . , 2000; Miles et al . , 2002 ) . This was probably due to the fact that these studies used pulses of much shorter duration to elicit firing . Indeed , in the delayed firing motoneurons , the long latency of the first spike is only apparent on long-lasting stationary current pulses close to the rheobase ( no spike would have been visible with this current intensity if a shorter pulse duration would have been used ) . If the pulses are short , higher current intensities are required to make the neuron fire , and the first action potential is fired with a shorter delay after the pulse onset . When the current was large enough , the discharge started shortly after the current onset ( see Figure 1—figure supplement 1 ) . In addition , since a slow current contribute to the delayed of firing ( Leroy et al . , SfN abstract 2012 ) , long intervals are needed to allow it to recover their initial state before the next pulse . A rapid repetition of the test pulses might prevent the observation of the delayed firing . However , spinal motoneurons with a delayed discharge have been observed in Pambo–Pambo et al . ( 2009 ) who carefully checked the current intensity at which the motoneurons start to fire in response to a 0 . 5 s square pulse ( see also Russier et al . ( 2003 ) for abducens motoneurons ) . Nonetheless , in these studies the latency of the first action potential was substantially shorter than in the present study , likely because the intensity to bring the cell to fire was slightly higher than the one that would have been required for a 5 s pulse duration . It is noteworthy that Pambo–Pambo et al . ( 2009 ) observed similar proportions of the two firing patterns . In previous ALS studies , data coming from immediate and delayed firing motoneurons were pooled together when comparing motoneurons from mSOD1 mice and WT mice ( Bories et al . , 2007; Amendola and Durand , 2008; Pambo–Pambo et al . , 2009; Quinlan et al . , 2011 ) . Since many morphological and electrophysiological properties differ between these two populations , the statistics for each property ( average values , standard deviations ) are heavily dependent on the proportion of immediate and delayed firing motoneurons present in each sample . If these proportions differ substantially in the WT and mSOD1 samples , this might obfuscate the conclusions . Therefore , one must be careful to separate these two motoneuron subtypes when comparing electrical and morphological properties between WT and mSOD1 neonatal mice . The intrinsic hyperexcitability of mSOD1 immediate firing motoneurons is not due to a decrease in their input conductance . Indeed , the input conductance remains unchanged despite a reduction of the dendritic length of the mSOD1 immediate firing motoneurons . The relatively large overall size of the motoneuron dendritic tree in the second postnatal week probably limits the influence that morphological alterations have on the input conductance ( Elbasiouny et al . , 2010 ) . Furthermore the reduction in dendritic length could be compensated by a decrease of the specific membrane resistivity . The average 5 mV hyperpolarization of the voltage threshold for spiking likely accounts for the reduction of the rheobase and thereby the intrinsic hyperexcitability of mSOD1 immediate firing motoneurons . Quinlan et al . ( 2011 ) observed an increase in the sodium persistent inward current in mSOD1 motoneurons during the second post-natal week . This increase in sodium persistent current can account for both the disappearance of MMOs and the hyperpolarization of the voltage threshold for spiking ( Iglesias et al . , 2011 ) . In a modeling study , Dai et al . ( 2002 ) showed that spiking threshold hyperpolarization occurs when the sodium conductance increases or when the activation curve of the sodium channels is shifted toward hyperpolarized levels . They report that this effect on the spiking threshold occurs without changes in amplitude and width of action potentials , as in the present study . To a lesser extent , a reduction of the delayed rectifier conductance or a depolarization of its activation curve may also hyperpolarize the spiking voltage threshold ( Dai et al . , 2002 ) . In sharp contrast , we did not observe any change in intrinsic excitability in the motoneurons with the delayed firing pattern . This type of firing is due to the specific presence in these motoneurons of two potassium currents that act at two time scales: an A-current and a slowly activating and inactivating potassium current ( Leroy et al . , SfN abstract 2012 ) . These two currents make the F-type motoneurons less excitable than the S-type motoneurons in neonates . Moreover , it is possible that these two currents prevent the spiking threshold to be hyperpolarized and thereby the rheobase to be decreased in mSOD1 neonates . However the delayed firing pattern is transient and disappears with age ( Russier et al . , 2003 ) . In adults , the lower excitability of F-type motoneurons observed in WT animals is mainly due to the fact that they are much larger than the S-type motoneurons . The origin of the hyperexcitability in the immediate firing motoneurons seems to be different from the mechanism at work in late embryonic motoneurons for which hyperexcitability arises from a decrease in input conductance ( van Zundert et al . , 2008; Martin et al . , 2013 ) . Numerical simulations suggest that the decrease in input conductance is due to a shortening of the terminal segments of embryonic spinal motoneurons which are much more electrotonically compact than in neonates ( Martin et al . , 2013 ) . Interestingly , in spinal muscular atrophy , another motoneuron degenerative disease , spinal motoneurons also become hyperexcitable ( Mentis et al . , 2011 ) . In this case , the increased excitability is caused both by an hyperpolarization of the spiking threshold and by a decrease in input conductance . The shortening of the dendritic tree of the immediate firing motoneurons might result from an increase in spontaneous synaptic activity occurring during the embryonic life ( Yvert et al . , 2004 ) . Alterations in both the inhibitory and excitatory synaptic inputs impinging on mSOD1 motoneurons , as well as the properties of their postsynaptic receptors , have been reported in cultured motoneurons derived from embryos ( Carunchio et al . , 2008; Chang and Martin , 2011 ) as well as in motoneurons from neonates ( van Zundert et al . , 2008 ) and adults ( Jiang et al . , 2009; Sunico et al . , 2011; Wootz et al . , 2013 ) . It has been shown that , during a critical developmental period , the morphology of the dendritic tree deeply relies on the synaptic activity received by the dendrites ( Spitzer , 2006; Cline and Haas , 2008 ) . More synaptic inputs can lead to a shortening of the dendrites ( Tripodi et al . , 2008 ) . Such structural homeostatic response of the motoneurons might counterbalance the increase in synaptic activity to ensure that an appropriate level of input is achieved ( Tripodi et al . , 2008 ) . The change in voltage threshold might also be a homeostatic regulation in response to synaptic hyperactivity . Since the frequency of both excitatory and synaptic spontaneous events are increased in early stages ( van Zundert et al . , 2008 ) , it is possible that the net input to S-type motoneurons is shifted towards more inhibition . An excess of inhibition at early age might be compensated by the hyperpolarization of the voltage threshold that increases the motoneuron excitability . As suggested by van Zundert et al . ( 2008 ) synaptic hyperactivity , dendritic shrinkage and intrinsic hyperexcitability might well be causally linked . The differential impact of the disease on the dendritic morphology of our two subpopulations of motoneurons suggests that synaptic alterations might be restricted to one subtype of motoneurons . Demonstrating this point will require , however , a preparation in which it is possible to preserve the integrity of spinal networks and identify the subtype of recorded motoneurons . Our results show that only the S-type motoneurons display an intrinsic hyperexcitability in mSOD1 neonates . We further confirmed that the motoneurons displaying hyperexcitability were ALS resistant thanks to the expression pattern of MMP9 ( Kaplan et al . , 2014 ) . On the other hand , F-type motoneurons vulnerable in ALS , are not hyperexcitable . We can therefore conclude that , contrary to the standard hypothesis ( Ilieva et al . , 2009 ) , intrinsic hyperexcitabilty is not an early event that triggers degeneration of the motoneurons . In mSOD1 mice , degeneration of the neuromuscular junctions of F-type motor units does not start before P50 whereas the S-type motor units do not degenerate ( Pun et al . , 2006; Hegedus et al . , 2008 ) . We may wonder whether the morphological and electrophysiological changes that we have observed perinatally in S-type motoneurons have a long term impact in adulthood and whether these early changes contribute to the survival of S-type motoneurons . The specific dendritic shrinkage of S-type motoneuron increases the size difference between S- and F- types motoneurons , and it has most likely a beneficial effect on S-type motoneurons . Indeed the smaller the total membrane surface , the smaller the metabolic demand to maintain this surface . Given the fact that motoneurons cope to an energetic issue in ALS that is caused by a disruption of the mitochondrial function ( von Lewinski and Keller , 2005; Ilieva et al . , 2009 ) , a lesser energetic demand may increase their chance to survive . Specific hyperexcitability of S-type motoneurons might also contribute to motoneuron survival as recently suggested ( Saxena et al . , 2013 ) . One might question whether an early hyperexcitability of S type motoneurons has a long-term benefit on their survival or whether the hyperexcitability has an impact only if it is still present in adults . We do not know whether S-type motoneurons remain hyperexcitable in adults . However , the whole population of adult spinal motoneurons are , on average , not hyperexcitable just prior to the onset of neuromuscular junctions degeneration ( Delestree et al . , 2014 ) . Some of them even turn out to be hypoexcitable since they lose the capacity to discharge repetitively in response to stationary inputs . However , in Delestree et al . ( 2014 ) , S-type motoneurons ( which represent only a small fraction of motoneurons ) and F-type motoneurons were pooled together and the possibility that S-type motoneurons remain hyperexcitable in adults cannot be ruled out . Unlike neonates , adult S-type and F-type motoneurons cannot be distinguished on the basis of their discharge pattern . Only highly demanding in vivo experiments would allow distinguishing motor unit sub-types on the basis of their contractile properties . A selective intrinsic hyperexcitability of S-type motoneurons that persists in adults would strongly reinforce the assumption that hyperexcitability contributes by itself to the protection of these motoneurons . The experiments were performed in accordance with European directives ( 86/609/CEE and 2010-63-UE ) and the French legislation . They were approved by Paris Descartes University ethics committee ( Permit Number: CEEA34 . BLDI . 065 . 12 . ) . All surgery was performed under sodium pentobarbital anesthesia , and every effort was made to minimize suffering . 6 to 10 day-old high expressor line B6 . Cg-Tg ( SOD1-G93A ) 1Gur/J mice and their non-transgenic littermates of either sex were used ( The Jackson Laboratory , RRID:IMSR_JAX:004435 ) . Genotyping was performed following the protocol given by the Jackson Laboratory . Mice were anesthetized using an intra-peritoneal 0 . 1 ml injection of pentobarbital 10% ( 5 . 5 mg/ml ) . Oblique slices were then prepared from the L3 to L5 spinal segments in order to keep a ventral rootlet in continuity with the cord as described in Lamotte d'Incamps et al . ( 2012 ) . The slices were transferred into artificial cerebrospinal fluid ( ACSF ) containing ( in mM ) : 130 NaCl , 2 . 5 KCl , 2 CaCl2 , 1 MgCl2 , 1 NaH2PO4 , 26 NaHCO3 , 25 glucose , 0 . 4 ascorbic acid , 2 Na-pyruvate , bubbled with 95% O2 and 5% CO2 ( pH 7 . 4 ) . The recording chamber was continuously perfused with ACSF at a rate of 1–2 ml/min , at room temperature . The slices used were those containing a ventral rootlet of sufficient length to be mounted on a suction stimulation electrode: a glass pipette with a tip size adapted to the diameter of the rootlet ( 40–170 µm ) and filled with ACSF . Patch pipettes had an initial open-tip resistance of 3–6 MΩ . The internal solution contained ( in mM ) : 140 K-gluconate , 6 KCl , 10 HEPES , 1 EGTA , 0 . 1 CaCl2 , 4 Mg-ATP , 0 . 3 Na2GTP . The pH was adjusted to 7 . 3 with KOH , and the osmolarity to 285–295 mOsm . An AxoClamp 700B ( Molecular Device , Sunnydale , CA ) amplifier was used for data acquisition . Whole-cell recordings were filtered at 3 kHz , digitized at 10 kHz using a CED 1401 and monitored using the Signal 5 software ( Cambridge Electronic Design Limited ) . Bridge resistance was compensated in current-clamp mode . Liquid junction potential was not corrected in order to readily compare with previous studies . We targeted large cells ( long soma axis >20 μm ) in the ventral horn under visual control using a video-camera ( Scientifica , Uckfield , UK ) and confirmed their motoneuron identity based on the recording of an antidromic action potential following stimulation of the ventral root . Single biphasic stimulation of the ventral rootlet ( 1–50 V , 0 . 1–0 . 3 ms ) was used to elicit antidromic action potential in motoneurons . We retained for analysis motoneurons exhibiting a resting potential equal or below −50 mV and an overshooting action potential . Access resistance ranged from 8 . 5 to 20 MΩ . Motoneurons were discarded from analysis if series resistance or resting potential varied more than 5 MΩ or 10 mV throughout the recording period . These criteria are similar to previous studies ( Pambo–Pambo et al . , 2009; Quinlan et al . , 2011 ) . Three delayed firing motoneurons displayed unusualy large conductances ( 123 , 151 and 153 nS ) . When excluding them our measurements fell in the same range and had the same variability than in previous studies ( Pambo–Pambo et al . , 2009; Quinlan et al . , 2011 ) . However , we chose to maintain these three delayed firing motoneurons in our calculations and analysis . Some of the electrophysiologically characterized motoneurons were filled with neurobiotin in order to study the anatomy of their dendritic tree . The intracellular solution was supplemented with 2% neurobiotin ( Vector Labs , Burlingame , CA ) and the motoneurons were recorded for at least 30 min to allow diffusion of the dye . After carefully removing the electrode from the cell , the slice was bathed in phosphate-buffered saline ( PBS ) with 4% paraformaldehyde for 1 hr . Blocking solution containing 0 . 1% of bovine serum albumin and 0 . 1% Triton X-100 ( Sigma–Aldrich , St . Louis , MO ) in PBS was applied for 1 hr . Slices were incubated overnight at 4°C with streptavidin-Cy3 conjugated antibody ( Sigma–Aldrich ) diluted at 1/500 in the blocking solution , washed three times in PBS and then mounted with Fluoromount ( Sigma–Aldrich ) . Acquisition was performed on a confocal microscope LSM 710 ( Carl Zeiss , Oberkochen , Germany ) and the dendritic tree of the motoneuron was reconstructed using Neurolucida software ( MBF Bioscience Williston , VT , RRID:nif-0000-10294 ) . Because of the slicing procedure , parts of the dendritic trees were missing and motoneurons reconstructions are therefore partial . Analysis of the dendritic trees included only the radial dendrites that remained in the same plane as the slice and did not plunge deeper than 50 µm below the surface of the slice . However , the number of reconstructed primary dendrites ( and thereby dendritic trees ) per motoneuron was similar between immediate and delayed firing motoneurons and we could therefore readily compare those radial dendrites in the two subtypes . A sample of motoneurons was electrophysiologically characterized , filled with neurobiotin and subsequently immunostained for NeuN , VGglut1 , Errβ or MMP9 . Following fixation , the slices were incubated in blocking solution ( see above ) . Then , the slices were incubated overnight at 4°C in the blocking solution supplemented with 1:500 rabbit anti-NeuN ( Cat# ABN78; EMD Millipore , Billerica , MA , RRID:AB_10807945 ) , 1:4000 guinea-pig anti-VGlut1 ( Cat# AB5905; EMD Millipore , RRID:AB_2301751 ) or 1:500 mouse anti-Errβ ( Cat# PP-H6705-00; R&D Systems , Minneapolis , MN , RRID:AB_2100412 ) After three washes in PBS , 1:500 streptavidin-Cy3 and 1:500 of the appropriate secondary antibody were applied in blocking solution during 3 hr at room temperature . We used the following secondary antibodies: anti-rabbit Alexa 488-conjugated ( Cat# 111-545-003; Jackson Immunoresearch , West Grove , PA ) , anti-guinea-pig Alexa 647-conjugated ( Cat# 106-605-003; Jackson Immunoresearch ) and anti-mouse CF633-conjugated ( Cat# SAB4600333; Sigma–Aldrich ) . For MMP-9 labelling , neurobiotin was used at 0 . 2% and the blocking solution contained 3% BSA , 0 . 5% Triton and 5% Horse Serum . Slices were subsequently incubated with 1:500 goat anti-MMP9 ( Cat# M9570; Sigma–Aldrich , RRID:AB_1079397 ) and then with 1:1000 anti-goat Cy3-conjugated ( Cat# 705-165-003; Jackson ImmunoResearch ) and 1:500 streptavidin Cy2-conjugated ( Cat# 016-220-084; Jackson ImmunoReserach ) . All slices were mounted and imaged as described above . In another sample of motoneurons electrophysiologically characterized and filled with neurobiotin ( see above ) , we performed chondrolectin in situ hybridization as described in Enjin et al . ( 2010 ) . Chondrolectin probes ( Genebank number NM_139134 . 3 ) were produced from commercial cDNA ( Source BioScience , Nottingham , UK ) , using T3 RNA polymerase in the presence of digoxigenin-11-UTP ( Roche Diagnostics , Basel , Switzerland ) . Slices were washed with PBT ( PBS supplemented with 0 . 1% Tween-20 , Sigma–Aldrich ) followed by treatment with 0 . 5% Triton X-100 . Slices were then post-fixed in 4% formaldehyde followed by prehybridization in hybridization buffer ( 50% formamide , 5× saline-sodium citrate [SSC] , pH 4 . 5 , 1% sodium dodecyl sulphate [SDS] , 10 mg/ml tRNA [Life Technologies , Carslbad , CA] , 10 mg/ml heparin [Sigma–Aldrich] in PBT ) . The probe ( 300 ng/ml ) was heat-denatured before starting the overnight hybridization ( 20–22 hr ) at 63°C . Overnight hybridization was followed by sequential washes with wash buffer 1 ( 50% formamide , 5× SSC , pH 4 . 5 and 1% SDS in PBT ) followed by buffer 2 ( 50% formamide , 2× SSC , pH 4 . 5 , and 0 . 1% Tween-20 in PBT ) at 63°C to remove unbound probe . Slices were then washed in 0 . 1% Tween-20 Tris-buffered saline followed by incubation in 1% blocking reagent ( Roche Diagnostics ) . Then the slices were incubated overnight at 4°C with 1:5000 diluted anti-digoxigenin alkaline phosphatase-conjugated antibody ( Roche Diagnostics ) . Hybridized probes were vizualized using SIGMAFAST Fast Red TR/Naphthol AS-MX ( Sigma–Aldrich ) . After hybridization , neurobiotin was revealed by washing the slices in PBS followed by PBS-T-G ( PBS , 0 . 25% Triton X-100 , 0 . 25% Gelatin ) . Slices were then incubated with 7 . 5 μg/ml Cy2-conjugated streptavidin ( Jackson Immunoresearch ) diluted in PBS-T-G for 2 hr at room temperature .	Amyotrophic lateral sclerosis ( ALS ) , which is also known as Lou Gherig's disease or motoneuron disease , is a neurodegenerative disorder in which muscles throughout the body gradually waste away due to the death of the neurons that control their activity . The disease often begins with weakness of the arms or legs , but progresses to include difficulties with movements such as swallowing and breathing . Around half of those affected die within 3 or 4 years of diagnosis . Although the causes of the disease are unclear , one leading theory is that the neurons that control muscle activity—motoneurons—are hyperexcitable during early development , and therefore fire too frequently . This causes too much calcium to enter the neurons and , because calcium is toxic to cells in high quantities , leads ultimately to the death of the neurons . But despite the popularity of this idea , and the fact that many therapeutic assays for ALS are based on attempts to reverse this process , there is no direct evidence that early hyperexcitability of motoneurons causes their death in ALS . Leroy et al . have now tested this theory directly by taking advantage of the fact that not all motoneurons are affected by ALS . The large ‘F-type’ motoneurons that control fast-contracting muscle fibres degenerate in ALS , whereas the small ‘S-type’ motoneurons that control slow-contracting muscle fibres do not . A comparison of F-type and S-type motoneurons in a mouse model of ALS revealed that , surprisingly , S-type motoneurons are hyperexcitable in young ALS mice , whereas F-type motoneurons are not . Given that S-type motoneurons are resistant to the effects of ALS , this indicates that early hyperexcitability cannot be the cause of motoneuron degeneration . Previous studies have tended to pool different types of motoneurons together , which might explain why this difference has not been seen before . Further experiments are now required to determine whether the hyperexcitability of S-type motoneurons persists into adulthood , and whether it might even contribute to their survival in ALS .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2014	Early intrinsic hyperexcitability does not contribute to motoneuron degeneration in amyotrophic lateral sclerosis
Birds and other vertebrates display stunning variation in pigmentation patterning , yet the genes controlling this diversity remain largely unknown . Rock pigeons ( Columba livia ) are fundamentally one of four color pattern phenotypes , in decreasing order of melanism: T-check , checker , bar ( ancestral ) , or barless . Using whole-genome scans , we identified NDP as a candidate gene for this variation . Allele-specific expression differences in NDP indicate cis-regulatory divergence between ancestral and melanistic alleles . Sequence comparisons suggest that derived alleles originated in the speckled pigeon ( Columba guinea ) , providing a striking example of introgression . In contrast , barless rock pigeons have an increased incidence of vision defects and , like human families with hereditary blindness , carry start-codon mutations in NDP . In summary , we find that both coding and regulatory variation in the same gene drives wing pattern diversity , and post-domestication introgression supplied potentially advantageous melanistic alleles to feral populations of this ubiquitous urban bird . Vertebrates have evolved a vast array of epidermal colors and color patterns , often in response to natural , sexual , and artificial selection . Numerous studies have identified key genes that determine variation in the types of pigments that are produced by melanocytes ( e . g . , Hubbard et al . , 2010; Manceau et al . , 2010; Roulin and Ducrest , 2013; Domyan et al . , 2014; Rosenblum et al . , 2014 ) . In contrast , considerably less is known about the genetic mechanisms that determine pigment patterning throughout the entire epidermis and within individual epidermal appendages ( e . g . , feathers , scales , and hairs ) ( Kelsh , 2004; Protas and Patel , 2008; Kelsh et al . , 2009; Lin et al . , 2009; Kaelin et al . , 2012; Lin et al . , 2013; Eom et al . , 2015; Poelstra et al . , 2015; Mallarino et al . , 2016 ) . In birds , color patterns are strikingly diverse among different populations and species , and these traits have profound impacts on mate-choice , crypsis , and communication ( Hill and McGraw , 2006 ) . The domestic rock pigeon ( Columba livia ) displays enormous phenotypic diversity among over 350 breeds , including a wide variety of plumage pigmentation patterns that also vary within breeds ( Shapiro and Domyan , 2013; Domyan and Shapiro , 2017 ) . Some of these pattern phenotypes are found in feral and wild populations as well ( Johnston and Janiga , 1995 ) . A large number of genetic loci contribute to pattern variation in rock pigeons , including genes that contribute in an additive fashion and others that epistatically mask the effects of other loci ( Van Hoosen Jones , 1922; Hollander , 1937; Sell , 2012; Domyan et al . , 2014 ) . Despite the genetic complexity of the full spectrum of plumage pattern diversity in pigeons , classical genetic experiments demonstrate that major wing shield pigmentation phenotypes are determined by an allelic series at a single locus ( C , for ‘checker’ pattern ) that produces four phenotypes: T-check ( CT allele , also called T-pattern ) , checker ( C ) , bar ( + ) , and barless ( c ) , in decreasing order of dominance and melanism ( Figure 1A ) ( Bonhote and Smalley , 1911; Hollander , 1938a , 1983b; Levi , 1986; Sell , 2012 ) . Bar is the ancestral phenotype ( Darwin , 1859; Darwin , 1868 ) , yet checker and T-check can occur at higher frequencies than bar in urban feral populations , suggesting a fitness advantage in areas of dense human habitation ( Goodwin , 1952; Obukhova and Kreslavskii , 1984; Johnston and Janiga , 1995; Čanády and Mošanský , 2013 ) . Color pattern variation is associated with several important life history traits in feral pigeon populations . For example , checker and T-check birds have higher frequencies of successful fledging from the nest , longer ( up to year-round ) breeding seasons , and can sequester more toxic heavy metals in plumage pigments through chelation ( Petersen and Williamson , 1949; Lofts et al . , 1966; Murton et al . , 1973; Janiga , 1991; Chatelain et al . , 2014; 2016 ) . Relative to bar , checker and T-check birds also have reduced fat storage and , perhaps as a consequence , lower overwinter adult survival rates in harsh rural environments ( Petersen and Williamson , 1949a; Jacquin et al . , 2012 ) . Female pigeons prefer checker mates to bars , so sexual selection probably influences the frequencies of wing pigmentation patterns in feral populations as well ( Burley , 1977; 1981; Johnston and Johnson , 1989 ) . In contrast , barless , the recessive and least melanistic phenotype , is rarely observed in feral pigeons ( Johnston and Janiga , 1995 ) . In domestic populations , barless birds have a higher frequency of vision defects , sometimes referred to as ‘foggy’ vision ( Hollander and Miller , 1981; Hollander , 1983b; Mangile , 1987 ) , which could negatively impact fitness in the wild . In this study , we investigate the molecular basis and evolutionary history underlying wing pattern diversity in pigeons . We discover both coding and regulatory variation at a single candidate gene , and a polymorphism linked with pattern variation within and between species that likely resulted from interspecies hybridization . To identify the genomic region containing the major wing pigmentation pattern locus , we used a probabilistic measure of allele frequency differentiation ( pFst; Domyan et al . , 2016 ) to compare the resequenced genomes of bar pigeons to genomes of pigeons with either checker or T-check patterns ( Figure 1A ) . Checker and T-check birds were grouped together because these two patterns are sometimes difficult to distinguish , even for experienced hobbyists . Checker birds are typically less pigmented than T-check birds , but genetic modifiers of pattern phenotypes can minimize this difference ( see Figure 1—figure supplement 1 for examples of variation ) . A two-step whole-genome scan ( see Materials and methods; Figure 1B and C , Figure 1—figure supplement 2 ) identified a single ~103 kb significantly differentiated region on Scaffold 68 that was shared by all checker and T-check birds ( position 1 , 702 , 691–1 , 805 , 600 of the Cliv_1 . 0 pigeon genome assembly , Shapiro et al . ( 2013 ) ; p=1 . 11e-16 , genome-wide significance threshold = 9 . 72e-10 ) . The minimal shared region was defined by haplotype breakpoints in a homozygous checker and a homozygous bar bird , and is highly differentiated from the same region in bar ( 63 . 28% mean sequence similarity at informative sites ) . This region is hereafter referred to as the minimal checker haplotype . As expected for the well-characterized allelic series at the C locus , we also found that a broadly overlapping region of Scaffold 68 was highly differentiated between the genomes of bar and barless birds ( p=3 . 11e-15 , genome-wide significance threshold = 9 . 71e-10; Figure 1—figure supplement 2 ) . Together , these whole-genome comparisons identified a single genomic region corresponding to the wing-pattern C locus . To identify genetic variants associated with the derived checker and T-check phenotypes , we first compared annotated protein-coding genes throughout the genome . We found a single , predicted , fixed change in EFHC2 ( Y572C , Figure 1—figure supplement 3 ) in checker and T-check birds relative to bar birds ( VAAST; Yandell et al . , 2011 ) . However , this same amino acid substitution is also found in Columba rupestris , a closely related species to C . livia that has a bar wing pattern . Thus , the Y572C substitution is not likely to be causative for the checker or T-check pattern , nor is it likely to have a strong impact on protein function ( MutPred2 score 0 . 468 , no recognized affected domain; PolyPhen-2 score 0 . 036 , benign; Adzhubei et al . , 2010; Pejaver et al . , 2017 ) . Next , we examined sequence coverage across the checker haplotype and discovered a copy number variable ( CNV ) region ( approximate breakpoints at Scaffold 68 positions 1 , 790 , 000 and 1 , 805 , 600 ) . Based on normalized read-depths of resequenced birds , we determined that the CNV region has one , two , or four copies per chromosome . Bar birds ( n = 12 ) in our resequencing panel always had a total of two copies in the CNV region ( one on each chromosome ) , but most checker ( n = 5 of 7 ) and T-check ( n = 2 of 2 ) genomes examined had additional copies of the CNV ( Figure 2A ) . Using a PCR assay to amplify across the breakpoints in birds with more than one copy per chromosome , we determined that additional copies result from tandem repeats . We found no evidence that the checker haplotype contains an inversion based on mapping of paired-end reads at the CNV breakpoints ( WHAM; Kronenberg et al . , 2015 ) . In addition , we were able to amplify unique PCR products that span the outer CNV breakpoints ( data not shown ) , suggesting that there are no inversions within the CNV region . Consistent with the dominant inheritance pattern of the phenotype , all checker and T-check birds had at least one copy of the checker haplotype . However , the fact that some checker birds had only one copy of the CNV region on each chromosome demonstrates that a copy number increase is not necessary to produce melanistic phenotypes . Pedigree analysis of a laboratory cross also confirmed perfect co-segregation of the checker haplotype and phenotype ( Figure 1—figure supplement 4 , Supplementary file 1 ) . Thus , a checker haplotype on at least one chromosome appears to be necessary for the dominant melanistic phenotypes , but additional copies of the CNV region are not . In a larger sample of pigeons , we found a significant association between copy number and phenotype ( TaqMan assay; pairwise Wilcoxon test , p=2 . 1e-05 ) . Checker ( n = 40 of 55 ) and T-check ( n = 15 of 18 ) patterns are usually associated with expansion of the CNV , but pigeons with the bar pattern ( n = 20 ) never had more than two copies in total ( one copy on each chromosome; Figure 2B ) . Although additional copies of the CNV only occurred in checker and T-check birds , we did not observe a consistent number of copies associated with either phenotype . This could be due to a variety of factors , including modifiers that darken genotypically-checker birds to closely resemble T-check ( Van Hoosen Jones , 1922; Sell , 2012 ) and environmental factors such as temperature-dependent darkening of the wing shield during feather development ( Podhradsky , 1968 ) . Due to the potential ambiguity in categorical phenotyping , we next measured the percent of pigmented area on the wing shield and tested for associations between copy number and the percentage of pigmented wing-shield area . We phenotyped and genotyped an additional 63 birds from diverse domestic breeds as well as 26 feral birds , and found that estimated copy number in the variable region was correlated with the amount of dark pigment on the wing shield ( nonlinear least squares regression , followed by r2 calculation; r2 = 0 . 46 ) ( Figure 2C ) . This correlation was a better fit to the regression when ferals were excluded ( r2 = 0 . 68 , Figure 2—figure supplement 1 ) , possibly because numerous pigmentation modifiers ( e . g . , sooty and dirty ) are segregating in feral populations ( Hollander , 1938a; Johnston and Janiga , 1995 ) . Together , our analyses show that the minimal checker haplotype is associated with increased pigmentation on the wing shield plumage , resulting in qualitative variation between bar and checker ( including T-check ) phenotypes . Furthermore , copy number variation is found only in checker haplotypes , and higher numbers of copies are associated with quantitative pigmentation increases in checker and T-check birds only . The CNV that is associated with wing pattern variation resides between two genes , EFHC2 and NDP . EFHC2 is a component of motile cilia , and mouse mutants have juvenile myoclonic epilepsy ( Linck et al . , 2014 ) . In humans , allelic variation in EFHC2 is also associated with differential fear responses and social cognition ( Weiss et al . , 2007; Blaya et al . , 2009; Startin et al . , 2015; but see Zinn et al . , 2008 ) . However , EFHC2 has not been implicated in pigmentation phenotypes in any organism . NDP encodes a secreted ligand that activates WNT signaling by binding to its only known receptor FZD4 and its co-receptor LRP5 ( Smallwood et al . , 2007; Hendrickx and Leyns , 2008; Deng et al . , 2013; Ke et al . , 2013 ) . Notably , NDP is one of many differentially expressed genes in the feathers of closely related crow subspecies that differ , in part , by the intensity of plumage pigmentation ( Poelstra et al . , 2015 ) . Furthermore , FZD4 is a known melanocyte stem cell marker ( Yamada et al . , 2010 ) . Thus , based on expression variation in different crow plumage phenotypes , and the expression of its receptor in pigment cell precursors , NDP is a strong candidate for pigment variation in pigeons . NDP is a short-range signal ( Niehrs , 2004 ) , so we suspect that this ligand is secreted by melanocytes themselves or by cells in close proximity to them . The CNV in the intergenic space between EFHC2 and NDP in the candidate region , coupled with the lack of candidate coding variants between bar and checker haplotypes , led us to hypothesize that the CNV region might contain regulatory variation that could alter expression of one or both neighboring genes . To test this possibility , we performed qRT-PCR on RNA harvested from regenerating wing shield feathers of bar , checker , and T-check birds . EFHC2 was not differentially expressed between bar and either checker or T-check patterned feathers ( p=0 . 19 , pairwise Wilcoxon test , p-value adjustment method: fdr ) , although expression levels differed slightly between the checker and T-check patterned feathers ( p=0 . 046 , Figure 3A ) . Expression levels of other genes adjacent to the minimal checker haplotype region also did not vary by phenotype ( Figure 3—figure supplement 1 ) . In contrast , expression of NDP was significantly increased in checker feathers – and even higher in T-check feathers – relative to bar feathers ( Figure 3A ) ( bar-checker comparison , p=1 . 9e-05; bar-T-check , p=1 . 0e-08; checker-T-check , p=0 . 0071; pairwise Wilcoxon test , all comparisons were significant at a false discovery rate of 0 . 05 ) . Moreover , when qRT-PCR expression data for checker and T-check feathers were grouped by copy number instead of categorical phenotype , the number of CNV copies was positively associated with NDP expression level ( Figure 3—figure supplement 2 ) . Thus , expression of NDP is positively associated with both increased melanism ( categorical pigment pattern phenotype ) and CNV genotype . The increase in NDP expression could be the outcome of at least two molecular mechanisms . First , one or more regulatory elements in the CNV region ( or elsewhere on the same DNA strand ) could increase expression of NDP in cis . Such changes would only affect expression of the allele on the same chromosome ( Wittkopp et al . , 2004 ) . Second , trans-acting factors encoded within the minimal checker haplotype ( e . g . , EFHC2 or an unannotated feature ) could increase NDP expression , resulting in an upregulation of NDP alleles on both chromosomes . To distinguish between these possibilities , we carried out allele-specific expression assays ( Domyan et al . , 2014; 2016 ) on the regenerating wing shield feathers of birds that were heterozygous for bar and checker alleles in the candidate region ( checker alleles with one , two , or four copies of the CNV ) . In the common trans-acting cellular environment of heterozygous birds , checker alleles of NDP were more highly expressed than bar alleles , and these differences were further amplified in checker alleles with more copies of the CNV ( Figure 3B ) ( p=0 . 0028 for two-sample t-test between 1 vs . 4 copies , p=1 . 84e-06 for generalized linear model regression; ratios of checker:bar expression for 1- and 4-copy checker alleles were significantly different than 1:1 , p≤0 . 002 for each comparison ) . In comparison , transcripts of EFHC2 from checker and bar alleles were not differentially expressed in the heterozygote background ( Figure 3B ) ( p=0 . 55 for two-sample t-test between 1 vs . 4 copies , p=0 . 47 for linear regression; ratios of checker:bar expression for 1- and 4-copy checker alleles were not significantly different than 1:1 , p>0 . 3 for each comparison ) . Checker alleles of NDP were also more highly expressed in feathers from other body regions ( tail and dorsum , Figure 3—figure supplement 3 ) , even though the pigment pattern on these regions is generally similar in bar and checker birds ( e . g . , both phenotypes have a dark band on the tail ) . Together , our expression studies indicate that a cis-acting regulatory change drives increased expression of NDP in pigeons with more melanistic plumage patterns , but does not alter expression of EFHC2 or other nearby genes . Furthermore , because NDP expression increases with additional copies of the CNV , the regulatory element probably resides within the CNV itself . To search for known enhancers in the CNV region , we mapped elements from the VISTA ( Visel et al . , 2007 ) and REPTILE ( He et al . , 2017 ) enhancer datasets to the pigeon genome . We found no hits within the minimal haplotype from the VISTA dataset and 12 hits from the REPTILE dataset ( Supplementary file 2 ) . Of these 12 , one hit was within the CNV region ( Scaffold 68: 1 , 795 , 453–1 , 795 , 511 ) . However , this lone mouse enhancer ( ENSMUSR00000084784 , http://uswest . ensembl . org/Mus_musculus/ ) is not known to regulate EFHC2 or NDP in mice , and is located on a mouse chromosome that is not orthologous to pigeon Scaffold 68 . Further functional work will be required to assess whether this or other sequences in the CNV region act as regulatory elements in C . livia . In humans , mutations in NDP can result in Norrie disease , a recessively-inherited disorder characterized by a suite of symptoms including vision deficiencies , intellectual and motor impairments , and auditory deficiencies ( Norrie , 1927; Warburg , 1961; Holmes , 1971; Chen et al . , 1992; Sims et al . , 1992 ) . Protein-coding mutations in NDP , including identical mutations segregating within single-family pedigrees , result in variable phenotypic outcomes , including incomplete penetrance ( Meindl et al . , 1995; Berger , 1998; Allen et al . , 2006 ) . Intriguingly , barless pigeons also have an increased incidence of vision deficiencies and , as in humans with certain mutant alleles of NDP , this phenotype is not completely penetrant ( Hollander , 1983b ) . Thus , based on the known allelism at the C locus , the nomination of regulatory changes at NDP as candidates for the C and CT alleles , and the vision-related symptoms of Norrie disease , NDP is also a strong candidate for the barless phenotype ( c allele ) . To test this prediction , we used VAAST to scan the resequenced genomes of 9 barless pigeons and found that all were homozygous for a nonsynonymous protein-coding change at the start codon of NDP that was perfectly associated with the barless wing pattern phenotype ( Figure 4 , Figure 1—figure supplement 2 ) . We detected no other genes with fixed coding changes or regions of significant allele frequency differentiation ( pFst ) elsewhere in the genome . We genotyped an additional 14 barless birds and found that all were homozygous for the same start-codon mutation . The barless mutation is predicted to truncate the amino terminus of the NDP protein by 11 amino acids , thereby disrupting the 24-amino acid signal peptide sequence ( www . uniprot . org , Q00604 NDP_Human ) . NDP is still transcribed and detectable by RT-PCR in regenerating barless feathers; therefore , we speculate that the start-codon mutation might alter the normal secretion of the protein into the extracellular matrix ( Gierasch , 1989 ) . In humans , coding mutations in NDP are frequently associated with a suite of neurological deficits . In pigeons , however , only wing pigment depletion and vision defects are reported in barless homozygotes . Remarkably , two human families segregating Norrie disease have only vision defects , and like barless pigeons , these individuals have start-codon mutations in NDP ( Figure 4 ) ( Isashiki et al . , 1995 ) . Therefore , signal peptide mutations might affect a specific subset of developmental processes regulated by NDP , while leaving other ( largely neurological ) functions intact . NDP is critical for retinal vascular formation ( Xu et al . , 2004 ) and hedgehog-dependent retinal progenitor proliferation ( McNeill et al . , 2013 ) in mammals , and we speculate that one or both of these processes is affected by the start codon mutations in pigeons as well . In summary , wing pattern phenotypes in pigeons are associated with the evolution of both regulatory ( checker , T-check ) and coding ( barless ) changes in the same gene , and barless pigeons share a partially-penetrant visual deficiency with human patients who have start-codon substitutions . Future work will test whether the barless ( and human ) start-codon mutations affect extracellular secretion of NDP , and how NDP expression directly or indirectly regulates melanocyte activity . Sharp boundaries define the heavily pigmented areas of checker feathers ( Figure 1 , Figure 1—figure supplement 1 ) , similar to intra-feather patterns in other species that are mediated by both activity of melanocytes and the topological distribution of their progenitors ( Lin et al . , 2013; Chen et al . , 2015 ) . Considerably more is known about the molecular control of plumage structure and color than pigmentation pattern , based in part on experiments to manipulate gene expression in vivo by viral infection and in explants by protein misexpression ( Harris et al . , 2002; Yu et al . , 2002; Harris et al . , 2005; Chen et al . , 2015; Boer et al . , 2017 ) . We expect the identification of NDP as a patterning gene to open new avenues of similar functional experiments to understand how pigment distribution is mediated . Pigeon fanciers have long hypothesized that the checker pattern in the rock pigeon ( Columba livia ) resulted from a cross-species hybridization event with the speckled pigeon ( C . guinea , Figure 5D ) , a species with a checker-like wing pattern ( G . Hochlan , G . Young , personal communication ) ( Hollander , 1983b ) . We estimate that C . livia and C . guinea diverged 4–5 million years ago ( MYA ) : columbid species ( pigeons and doves ) diverge from each other in mitochondrial cytochrome b nucleotide sequence at 1 . 96% per MY ( Weir and Schluter , 2008 ) , and C . livia and C . guinea differ at this gene by 8 . 0% . Divergence date estimates for these two species based on nuclear genome sequences range between 3 . 2 and 6 . 7 MYA ( K . P . J . , unpublished results ) . Despite this divergence time of several MY , inter-species crosses between C . livia and C . guinea can produce fertile hybrids ( Whitman , 1919; Irwin et al . , 1936; Taibel , 1949; Miller , 1953 ) . Moreover , hybrid F1 and backcross progeny between C . guinea and bar C . livia have checkered wings , much like C . livia with the C allele ( Whitman , 1919; Taibel , 1949 ) . Taibel ( 1949 ) showed that , although hybrid F1 females were infertile , two more generations of backcrossing hybrid males to C . livia could produce checker offspring of both sexes that were fully fertile . In short , Taibel introgressed the checker trait from C . guinea into C . livia in just three generations . To evaluate the possibility of an ancient introgression event , we sequenced an individual C . guinea genome to 33X coverage and mapped the reads to the C . livia reference assembly . We calculated four-taxon D-statistics ( ‘ABBA-BABA’ test; Durand et al . , 2011 ) to test for deviations from expected sequence similarity between C . guinea and C . livia , using a wood pigeon ( C . palumbus ) genome as an outgroup ( Supplementary file 3 ) . In this case , the null expectation is that the C candidate region will be more similar between conspecific bar and checker C . livia than either will be to the same region in C . guinea . That is , the phylogeny of the candidate region should be congruent with the species phylogeny . However , we found that the D-statistic approaches one in the candidate region ( n = 10 each for bar and checker C . livia ) , indicating that checker C . livia are more similar to C . guinea than to conspecific bar birds in this region ( Figure 5A ) . The mean genome-wide D-statistic was close to zero ( 0 . 021 ) , indicating that bar and checker sequences are more similar to each other throughout the genome than either one is to C . guinea . This similarity between C . guinea and checker C . livia in the pattern candidate region was further supported by sequence analysis using HybridCheck ( Ward and van Oosterhout , 2016 ) . Outside of the candidate region , checker birds have a high sequence similarity to conspecific bar birds and low similarity to C . guinea ( Figure 5B ) . Within the candidate region , however , this relationship shows a striking reversal , and checker and C . guinea sequences are most similar to each other . In addition , although the genome-wide D-statistic was relatively low , the 95% confidence interval ( CI ) was greater than zero ( 0 . 021 to 0 . 022 ) , providing further evidence for one or more introgression events from C . guinea into checker and T-check genomes . Unlike in many checker and T-check C . livia , we did not find additional copies of the CNV region in C . guinea . This could indicate that the CNV expanded in C . livia , or that the CNV is present in a subset of C . guinea but has not yet been sampled . Taken together , these patterns of sequence similarity and divergence support the hypothesis that the candidate checker haplotype in rock pigeons originated by introgression from C . guinea . While post-divergence introgression is an attractive hypothesis to explain the sequence similarity between checker C . livia and C . guinea , another formal possibility is that sequence similarity between these groups is due to incomplete lineage sorting . In an analogous example , light- and dark-pigmentation alleles of tan probably segregated in the ancestor of Drosophila americana and D . novamexicana , and the light allele subsequently became fixed in the latter species ( Wittkopp et al . , 2009 ) . However , light and dark alleles continue to segregate in D . americana , and the light allele in this species has the same ancestral origin as the one that is fixed in D . novamexicana . Similarly , we wanted to test if the minimal checker haplotype might have been present in the last common ancestor of C . guinea and C . livia , but now segregates only in C . livia . We measured nucleotide differences among different alleles of the minimal haplotype and compared these counts to polymorphism rates expected to accumulate over the 4-5 MY divergence time between C . livia and C . guinea ( Figure 5C , purple bar , see Materials and methods ) . We found that polymorphisms between bar C . livia and C . guinea approached the number expected to accumulate in this region in 4–5 MY ( 59 . 90% sequence similarity at segregating sites , SD = 2 . 6% , 1708 ± 109 mean SNPs , Figure 5C ) , but so did intraspecific comparisons between bar and checker C . livia ( 63 . 28% , SD = 2 . 3% , 1564 ± 99 ) . In contrast , C . guinea and C . livia checker sequences had significantly fewer differences than would be expected to accumulate between the two species ( 90 . 96% , SD = 0 . 13% , 384 ± 6 , p<2 . 2e-16 , t-test ) . These results support an introgression event from C . guinea to C . livia , rather than a shared allele inherited from a common ancestor prior to divergence . Among 11 checker haplotype sequences , we found remarkably high sequence similarity ( 99 . 39% , SD = 0 . 18% , 26 ± 8 mean differences ) , corresponding to a haplotype divergence time of 89 ± 27 thousand years ( KY ) , based on mutation rate . The rock pigeon reference genome contains the checker haplotype , which could bias the discovery of SNPs in our resequenced genomes . We therefore performed de novo assemblies using Illumina shotgun reads from C . guinea and high-coverage bar and checker individuals , then compared nucleotide sequences in regions of the minimal haplotype where all three assemblies overlapped ( 92 , 199 of 102 , 909 bp , or 89 . 6% ) . We found similar patterns of divergence between the de novo assemblies and the resequenced genomes that were mapped to the reference , indicating that that SNP discovery was not heavily biased by our short-read mapping approach ( Figure 5—figure supplement 1 ) . Based on pairwise polymorphisms between the checker reference and the de novo checker assembly ( 11 differences ) , the haplotype divergence time is 42 KY . This figure is more recent than our estimate based on more individuals , but the key results are that both estimates are roughly 2 orders of magnitude more recent than the divergence time between species , and the similarity between checker and C . guinea sequences is characteristic of within-species rather than between-species variation . Lastly , to date the putative introgression event ( s ) , we estimated the age of the minimal checker haplotype based on the pattern of linkage disequilibrium decay ( Voight et al . , 2006 ) . Using a recombination rate calculated for rock pigeon ( Holt et al . , 2018 ) , the checker haplotype originated in C . livia between 429 and 857 years ago , assuming one to two generations per year . The corresponding 95% confidence intervals are 267 to 716 years ago assuming one generation per year and 534 to 1 , 432 years ago assuming two generations per year . Together , these multiple lines of evidence support the hypothesis that the checker haplotype was introduced from C . guinea into C . livia after the domestication of the rock pigeon ( ~5000 years ago ) . The four-taxon D-statistic values approach one at the NDP locus ( Figure 5A ) , indicating that checker C . livia is far more closely related to C . guinea than to bar C . livia at this locus . Additionally , the pairwise differences between C . guinea and checker haplotypes are incompatible with incomplete lineage sorting ( Figure 5C ) , assuming a 4–5 MY species divergence time and no subsequent gene flow . The lack of single nucleotide diversity among checker haplotypes , with only 26 ± 8 mean differences and an estimated gene tree divergence of 89 KY , is unusually low for the diversity typically observed in large , free-living pigeon populations ( Shapiro et al . , 2013 ) . The differences between the mutation-based ( 89 KY ) and LD-based ( 0 . 4 to 0 . 9 KY ) estimates of the checker haplotype age are an expected consequence of crossbreeding and artificial selection given that the former is an estimate of the age of the most recent common ancestor in the source population while the latter is a lower bound estimate for the date of introgression . Inconsistencies of this magnitude are unexpected in the absence of introgression . Additionally , the genome-wide D-statistic comparing C . guinea and bar to C . guinea and checker is low but significantly greater than 0 , indicating that gene flow from C . guinea to checker has been higher than from C . guinea to bar throughout the genome . Notably , the non-zero D-statistic result holds when the NDP locus is excluded from this calculation . These results are expected if the checker haplotypes were recently introduced into C . livia by pigeon breeders , and interbreeding between checker and bar populations has not been completely random . Consistent with this expectation , non-random mating is observed in feral populations , and pigeon breeders often impose color pattern selection on their birds ( Darwin , 1868; Burley , 1977; 1981; Johnston and Johnson , 1989; National_Pigeon_Association , 2010 ) . Finally , the upper bound of the LD-based age estimate of the checker haplotype of 1 , 432 years ago indicates that checker haplotype was introduced into C . livia well after the domestication of rock pigeons . Because the ranges of C . livia and C . guinea overlap in northern Africa ( del Hoyo et al . , 2017 ) , it is possible that introgression events occurred in free-living populations . However , the more likely explanation is that C . guinea haplotypes were introduced into C . livia by pigeon breeders . Once male hybrids are generated , this can be accomplished in just a few generations ( Taibel , 1949 ) . Thus , humans might have intentionally selected this phenotype , which is linked to life history traits that are advantageous in urban environments , and then built ideal urban habitats for them to thrive ( Jerolmack , 2008 ) . Adaptive traits can arise through new mutations or standing variation within a species , and a growing number of studies point to adaptive introgressions among vertebrates and other organisms ( Hedrick , 2013; Martin and Orgogozo , 2013; Harrison and Larson , 2014; Zhang et al . , 2016 ) . In some cases , introgressed loci are associated with adaptive traits in the receiving species , including high-altitude tolerance in Tibetan human populations from Denisovans ( Huerta-Sánchez et al . , 2014 ) , resistance to anticoagulant pesticides in the house mouse from the Algerian mouse ( Song et al . , 2011; Liu et al . , 2015 ) , and beak morphology among different species of Darwin’s finches ( Lamichhaney et al . , 2015 ) . Among domesticated birds , introgressions are responsible for skin and plumage color traits in chickens and canaries , respectively ( Eriksson et al . , 2008; Lopes et al . , 2016 ) . Alleles under artificial selection in a domesticated species can be advantageous in the wild as well , as in the introgression of dark coat color from domestic dogs to wolves ( Anderson et al . , 2009 ) ( however , color might actually be a visual marker for an advantageous physiological trait conferred by the same allele; Coulson et al . , 2011 ) . In this study , we identified a putative introgression into C . livia from C . guinea that is advantageous both in artificial ( selection by breeders ) and free-living urban environments ( sexual and natural selection ) . A change in plumage color pattern is an immediately obvious phenotypic consequence of the checker allele , yet other traits are linked to this pigmentation pattern . For example , checker and T-check pigeons have longer breeding seasons , up to year-round in some locations ( Lofts et al . , 1966; Murton et al . , 1973 ) , and C . guinea breeds year-round in most of its native range as well ( del Hoyo et al . , 2017 ) . Perhaps not coincidentally , NDP is expressed in the gonad tissues of adult C . livia ( MacManes et al . , 2017 ) and the reproductive tract of other amniotes ( Paxton et al . , 2010 ) . Abrogation of expression or function of NDP or its receptor FZD4 is associated with infertility and gonad defects ( Luhmann et al . , 2005; Kaloglu et al . , 2011; Ohlmann et al . , 2012; Ohlmann and Tamm , 2012 ) . Furthermore , checker and T-check birds deposit less fat during normally reproductively quiescent winter months . In humans , expression levels of FZD4 and the co-receptor LRP5 in adipose tissue respond to varying levels of insulin ( Karczewska-Kupczewska et al . , 2016 ) , and LRP5 regulates the amount and location of adipose tissue deposition ( Loh et al . , 2015; Karczewska-Kupczewska et al . , 2016 ) . Therefore , based on its reproductive and metabolic roles in pigeons and other amniotes , NDP is a viable candidate not only for color pattern variation , but also for the suite of other traits observed in free-living ( feral and wild ) checker and T-check pigeons . Indeed , the potential pleiotropic effects of NDP raise the possibility that reproductive output and other physiological advantages are secondary or even primary targets of selection , and melanistic phenotypes are honest genetic signals of a cluster of adaptive traits controlled by a single locus . Adaptive cis-regulatory change is also an important theme in the evolution of vertebrates and other animals ( Shapiro et al . , 2004; Miller et al . , 2007; Wray , 2007; Carroll , 2008; Chan et al . , 2010; Wittkopp and Kalay , 2011; O'Brown et al . , 2015; Signor and Nuzhdin , 2018 ) . This theme is especially prominent in studies of color variation in Drosophila , in which regulatory variation impacts both the type and pattern of pigments on the body and wings ( Gompel et al . , 2005; Prud'homme et al . , 2006; Rebeiz et al . , 2009 ) . In some cases , the evolution of multiple regulatory elements of the same gene can fine-tune phenotypes , such as mouse coat color and trichome distribution in fruit flies ( McGregor et al . , 2007; Linnen et al . , 2013 ) . In cases of genes that have multiple developmental roles , introgression can result in the simultaneous transfer of multiple advantageous traits ( Rieseberg , 2011 ) . The potential role of NDP in both plumage and physiological variation in pigeons could represent a striking example of pleiotropic regulatory effects . Wing pigmentation patterns that resemble checker are present in many wild species within and outside of Columbidae including Patagioenas maculosa ( Spot-winged pigeon ) , Spilopelia chinensis ( Spotted dove ) , Geopelia cuneata ( Diamond dove ) , Gyps rueppelli ( Rüppell’s vulture ) , and Pygiptila stellaris ( Spot-winged antshrike ) . Based on our results in pigeons , NDP and its downstream targets can serve as initial candidate genes to ask whether similar molecular mechanisms generate convergent patterns in other species . Animal husbandry and experimental procedures were performed in accordance with protocols approved by the University of Utah Institutional Animal Care and Use Committee ( protocols 10–05007 , 13–04012 , and 16–03010 ) . Blood samples were collected in Utah at local pigeon shows , at the homes of local pigeon breeders , from pigeons in the Shapiro lab , and from ferals that had been captured in Salt Lake City , Utah . Photos of each bird were taken upon sample collection for our records and for phenotype verification . Tissue samples of C . rupestris , C . guinea , and C . palumbus were provided by the University of Washington Burke Museum , Louisiana State University Museum of Natural Science , and Tracy Aviary , respectively . Breeders outside of Utah were contacted by email or phone to obtain feather samples . Breeders were sent feather collection packets and instructions , and feather samples were sent back to the University of Utah along with detailed phenotypic information . Breeders were instructed to submit only samples that were unrelated by grandparent . DNA was then extracted from blood , tissue , and feathers as previously described ( Stringham et al . , 2012 ) . Feather and color phenotypes of birds were designated by their respective breeders . Birds that were raised in our facility at the University of Utah or collected from feral populations were assigned a phenotype using standard references ( Levi , 1986; Sell , 2012 ) . BAM files from a panel of previously resequenced birds were combined with BAM files from eight additional barless birds , 23 bar and 23 checker birds ( 22 feral , 24 domestics ) , a single C . guinea , and a single C . palumbus . SNVs and small indels were called using the Genome Analysis Toolkit ( Unified Genotyper and LeftAlignAnd TrimVariants functions , default settings; McKenna et al . , 2010 ) . Variants were filtered as described previously ( Domyan et al . , 2016 ) and the subsequent variant call format ( VCF ) file was used for pFst and ABBA-BABA analyses as part of the VCFLIB software library ( https://github . com/vcflib ) and VAAST ( Yandell et al . , 2011 ) as described previously ( Shapiro et al . , 2013 ) . pFst was first performed on whole-genomes of 32 bar and 27 checker birds . Some of the checker and bar birds were sequenced to low coverage ( ~1X ) , so we were unable to confidently define the boundaries of the shared haplotype . To remedy this issue , we used the core of the haplotype to identify additional bar and checker birds from a set of birds that had already been sequenced to higher coverage ( Shapiro et al . , 2013 ) . These additional birds were not included in the initial scan because their wing pattern phenotypes were concealed by other color and pattern traits that are epistatic to bar and check phenotypes . For example , the recessive red ( e ) and spread ( S ) loci produce a uniform pigment over the entire body , thereby obscuring any bars or checkers ( Van Hoosen Jones , 1922; Hollander , 1938a; Sell , 2012; Domyan et al . , 2014 ) . Although the major wing pattern is not visible in these birds , the presence or absence of the core checker haplotype allowed us to characterize them as either bar or checker/T-check . We then re-ran pFst using 17 bar and 24 checker/T-check birds with at least 8X mean read depth coverage ( Figure 1B ) and found a minimal shared checker haplotype of ~100 kb ( Scaffold 68 position 1 , 702 , 691–1 , 805 , 600 ) , as defined by haplotype breakpoints in a homozygous checker and a homozygous bar bird ( NCBI BioSamples SAMN01057561 and SAMN01057543 , respectively; BioProject PRJNA167554 ) . pFst was also used to compare the genomes of 32 bar and nine barless birds . New sequence data for C . livia are deposited in the NCBI SRA database under BioProject PRJNA428271 with the BioSample accession numbers SAMN08286792- SAMN08286844 . New sequence data for C . guinea and C . palumbus are deposited in the NCBI SRA database under accession numbers SRS1416880 and SRS1416881 , respectively . We genotyped and phenotyped a laboratory intercross that segregates bar and checker patterns in the F2 generation . We generated a pedigree from this family for F2 individuals whose phenotypes we could identify as bar or checker ( n = 62 ) . We could not determine bar or checker phenotypes for all individuals because other pigment patterns that epistatically mask bar and checker – almond ( St locus ) , spread ( S ) , and recessive red ( E ) – are also segregating in the cross . F2 individuals were excluded from the analysis if they had one of these masking phenotypes , but F1 parents were retained if they produced F2 offspring with checker or bar phenotypes . We used primers that amplify within the minimal haplotype ( AV17 primers , see Supplementary file 1 ) to genotype all F2 individuals , their F1 parents ( n = 26 ) , and the founders ( n = 4 ) by Sanger sequencing for the checker haplotype to assess whether the checker haplotype segregated with wing pattern phenotype . The approximate breakpoints of the CNV region were identified at Scaffold 68 positions 1 , 790 , 000 and 1 , 805 , 600 using WHAM in resequenced genomes of homozygous bar or checker birds with greater than 8x coverage ( Kronenberg et al . , 2015 ) . For 12 bar , seven checker , and 2 T-check resequenced genomes , Scaffold 68 gdepth files were generated using VCFtools ( Danecek et al . , 2011 ) . Two subset regions were specified: the first contained the CNV and the second was outside of the CNV and was used for normalization ( positions 1 , 500 , 000–2 , 000 , 000 and 800 , 000–1 , 400 , 000 , respectively ) . Read depth in the CNV was normalized by dividing read depth in this region by the average read depth from the second ( non-CNV ) region , then multiplying by two to normalize for diploidy . Copy number variation was estimated using a custom Taqman Copy Number Assay ( assay ID: cnvtaq1_CC1RVED; Applied Biosystems , Foster City , CA ) for 93 birds phenotyped by wing pigment pattern category and 89 birds whose pigmentation was quantified by image analysis . After DNA extraction , samples were diluted to 5 ng/μL . Samples were run in quadruplicate according to the manufacturer’s protocol . At the time of blood sample collection , the right wing shield was photographed ( RAW format images from a Nikon D70 or Sony a6000 digital camera ) . Using Photoshop software ( Adobe Systems , San Jose , CA ) , the wing shield including the bar ( on the secondary covert feathers ) was isolated from the original RAW file . Images were adjusted to remove shadows and the contrast was set to 100% . The isolated adjusted wing shield image was then imported into ImageJ ( imagej . nih . gov/ ) in JPEG format . Image depth was set to 8-bit and we then applied the threshold command . Threshold was further adjusted by hand to capture checkering and particles were analyzed using a minimum pixel size of 50 . This procedure calculated the area of dark plumage pigmentation on the wing shield . Total shield area was calculated using the Huang threshold setting and analyzing the particles as before ( minimum pixel size of 50 ) . The dark area particles were divided by total wing area particles , and then multiplied by 100 to get the percent dark area on the wing shield . Measurements were done in triplicate for each bird , and the mean percentages of dark area for each bird were used to test for associations between copy number and phenotype using a non-linear least squares regression . Two secondary covert wing feathers each from the wing shields of 8 bar , seven checker , and 8 T-check birds were plucked to stimulate feather regeneration for qRT-PCR experiments . Nine days after plucking , regenerating feather buds were removed , the proximal 5 mm was cut longitudinally , and specimens were stored in RNAlater ( Qiagen , Valencia , CA ) at 4°C for up to three days . Next , collar cells were removed , RNA was isolated , and mRNA was reverse-transcribed to cDNA as described previously ( Domyan et al . , 2014 ) . Intron-spanning primers ( see Supplementary file 1 ) were used to amplify each target using a CFX96 qPCR instrument and iTaq Universal Syber Green Supermix ( Bio-Rad , Hercules , CA ) . Samples were run in duplicate and normalized to β-actin . The mean value was determined and results are presented as mean ± S . E . for each phenotype . Results were compared using a Wilcoxon Rank Sum test and expression differences were considered statistically-significant if p<0 . 05 . SNPs in NDP and EFHC2 were identified as being diagnostic of the bar or checker/T-check haplotypes from resequenced birds . Heterozygous birds were identified by Sanger sequencing in the minimal checker haplotype region ( AV17 primers , see Supplementary file 1 ) . Twelve checker and T-check heterozygous birds were then verified by additional Sanger reactions ( AV54 for NDP and AV97 for EFHC2 , see Supplementary file 1 ) to be heterozygous for the diagnostic SNPs in NDP and EFHC2 . PyroMark Custom assays ( Qiagen , Valencia , CA ) were designed for each SNP using the manufacturer’s software ( Supplementary file 1 ) . Pyrosequencing was performed on gDNA derived from blood and cDNA derived from collar cells from 9 day regenerating wing shield feathers using a PyroMark Q24 instrument ( Qiagen , Valencia , CA ) . Additional pyrosequencing was performed for 9 of the 12 of the original birds from 9 day regenerating dorsal and tail feathers following the same protocol . Signal intensity ratios from the cDNA samples were normalized to the ratios from the corresponding gDNA samples to control for bias in allele amplification . Normalized ratios were analyzed by Wilcoxon Rank Sum tests . We compared the expression ratios of 1-copy checker:bar to 4-copy checker:bar to determine whether additional copies of the CNV were associated with higher checker:bar allele expression . We also compared 1-copy checker:bar expression ratios and four copy checker:bar expression ratios to a 1:1 ratio ( equal expression of both alleles ) using the Wilcoxon Rank Sum test to determine whether the measured checker:bar ratios were significantly different from the null hypothesis of equal expression of bar and checker alleles . The 2-copy checker:bar ratio was not compared in these analyses because there was only one sample . Allele expression ratios were analyzed together for 1 , 2 , and 4-copies using a glm regression to determine whether CNV copy number was associated with increased checker allele expression . Results were considered significant if p<0 . 05 . VISTA ( https://enhancer . lbl . gov/ ) ( Visel et al . , 2007 ) and REPTILE ( He et al . , 2017 ) enhancer datasets were mapped to the pigeon reference genome using bwa-mem ( Li and Durbin , 2009 ) . BAM output files were filtered for high quality orthologous regions and further filtered for alignments within the minimal checker haplotype on Scaffold 68 ( Supplementary file 2 ) . NDP exons were sequenced using primers in Supplementary file 1 . Primers pairs were designed using the rock pigeon reference genome ( Cliv_1 . 0 ) ( Shapiro et al . , 2013 ) . PCR products were purified using a QIAquick PCR purification kit ( Qiagen , Valencia , CA ) and Sanger sequenced . Sequences from each exon were then edited for quality with Sequencher v . 5 . 1 ( GeneCodes , Ann Arbor , MI ) . Sequences were translated and aligned with SIXFRAME and CLUSTALW in SDSC Biology Workbench ( http://workbench . sdsc . edu ) . Amino acid sequences outside of Columbidae were downloaded from Ensembl ( www . ensembl . org ) . Whole genome ABBA-BABA ( https://github . com/vcflib ) was performed on 10 × 10 combinations of bar and checker ( Supplementary file 3 ) birds in the arrangement: bar , checker , C . guinea , C . palumbus . VCFLIB ( https://github . com/vcflib ) was used to smooth raw ABBA-BABA results in 1000 kb or 100 kb windows for whole-genome or Scaffold 68 analyses respectively . For each 10 × 10 combination . We calculated the average Dstatistic across the genome . These were then averaged to generate a whole genome average of D = 0 . 0212 , marked as the dotted line in Figure 5A . Confidence intervals were generated via moving blocks bootstrap ( Kunsch , 1989 ) . Block sizes are equal to the windows above , with D-statistic values resampled with replacement a number of times equal to the number of windows in a sample . In Figure 5A , three representative ABBA-BABA tests are shown for different combinations of bar and checker birds . The checker and bar birds used in each representative comparison are: ARC-STA , SRS346901 and SRS346887; MAP-ORR , SRS346893 and SRS346881; IRT-STA , SRS346892 and SRS346887 respectively . ARC , MAP , and IRT are homozygous for the checker haplotype . STA and ORR are homozygous for the bar haplotype . VCF files containing Scaffold 68 genotypes for 16 bar , 11 homozygous checker , and 1 C . guinea were phased using Beagle version 3 . 3 ( Browning and Browning , 2007 ) . VCFs were then converted to fasta format using vcf2fasta in vcf-lib ( https://github . com/vcflib ) . HybridCheck ( Ward and van Oosterhout , 2016 ) ( https://github . com/Ward9250/HybridCheck ) was run to visualize pairwise sequence similarities between trios of bar , checker , and C . guinea sequences across Scaffold 68 using default settings . Phased VCF files for 16 homozygous bar , 11 homozygous checker , and 1 C . guinea were subsetted to the minimal checker haplotype region ( positions 1 , 702 , 691–1 , 805 , 600 ) with tabix ( Li , 2011 ) . The vcf-compare software module ( VCFtools , ( Danecek et al . , 2011 ) was used to run pairwise comparisons between bar , checker , and C . guinea birds ( 176 bar-checker , 16 bar-guinea , and 11 checker-guinea comparisons ) as well as among bar and checker birds ( 120 bar-bar and 55 checker-checker comparisons ) . The total number of differences for each group was compared to the number of differences that are expected to accumulate during a 4–5 MY divergence time in a 102 , 909 bp region ( the size of the minimal checker haplotype ) with the mutation rate μ = 1 . 42e-9 ( Shapiro et al . , 2013 ) using the coalescent equation: Time= #SNPs/ ( 2xμx length of the region ) . The observed pairwise differences and the expected number of differences were evaluated with two-sample t-tests and all groups were considered statistically different from the 4–5 MY expectation ( 1169 . 05–1461 . 31 ) . There were 4261 total segregating sites in the minimal haplotype region between all birds used for pairwise comparisons . Means and standard deviations for each group were calculated in R ( R_Development_Core_Team , 2008 ) . To ensure that SNP calling was not biased by using a reference that has the checker haplotype , we performed de novo assemblies of one bar ( SRS346895 ) , one checker ( SRS346878 ) , and one C . guinea ( SRS1416880 ) individual using CLC Genomics Workbench ( Qiagen , Valencia , CA ) . These C . livia individuals were chosen because they had the highest genome-wide mean read depth coverage for each phenotype at 14X ( bar ) and 15X ( checker; the C . guinea sample was sequenced to 33X ) . Whole-genome assemblies were mapped to the reference genome and variants ( single nucleotide variants , structural variants , indels ) were called by SMARTIE-SV ( https://github . com/zeeev/smartie-sv ) , which uses the BLASR aligner ( Chaisson and Tesler , 2012 ) , using default parameters . We identified regions where all three new assemblies intersected with the reference assembly . We then counted SNPs across the minimal haplotype where all three assemblies intersected ( 92 , 199 of 102 , 909 bp; 12 intersecting contigs ranging in length from 678 to 21565 bp , median = 5047 . 5 ) . Variants identified in the de novo assemblies for checker , bar , or C . guinea individuals were manually filtered to remove variants where the alternate allele was ‘N’ or a series of ‘N’ base pairs . Variants spanning multiple base pairs in each individual file were identified and manually split into multiple single nucleotide polymorphisms . Filtered and split tab-delimited variant calls between each de novo assembly and the reference genome were read into R v . 3 . 3 . 2 ( R_Development_Core_Team , 2008 ) . For each variant call file , the start position was extracted . Pairwise comparisons of positions for checker , bar , and C . guinea de novo assemblies were made using the ‘setdiff’ command to generate lists of variants that were only observed in one individual out of any given pair ( checker vs . bar , checker vs . C . guinea , bar vs . C . guinea ) . These lists of positions were then used to subset the original variant call files and assemble lists of pairwise differences . For example , SNPs that differ between checker and bar would include variants that differ from the reference in checker , but not bar , plus variants that differ from the reference in bar , but not checker . Additionally , the ‘intersect’ command was used to identify variants in multiple de novo assemblies . For variants that appeared in more than one de novo assembly , alternative alleles for each assembly were compared . In the majority of cases , both de novo assemblies showed the same alternative allele , and thus did not differ from one another . We found 1458 total SNP positions based on comparison of the three de novo assemblies . In the comparison described above and shown in Figures 5C , 362 SNPs were identified in the same region . This higher number of SNPs was driven by the much larger sample size and haplotype diversity among the 16 bar birds . In order to determine whether the barless allele of NDP is transcribed and persists in collar cells , or is degraded ( e . g . , by non-sense mediated decay ) , we designed a PCR assay to amplify NDP mRNA transcripts . Feathers from four barless , 2 bar , two checker , and 2 T-check birds were plucked to stimulate regeneration . We then harvested regenerated feathers after 9 days , extracted RNA from collar cells , and synthesized cDNA as described above . We then generated amplicons from each sample using intron-spanning primers ( AV200 primers , see Supplementary file 1 ) . Primers were anchored in the exon containing the barless start-codon mutation and the exon 3’ to it , so this assay tested for both the presence of transcripts and consistent splicing among alleles and phenotypes . EFHC2 exonic sequences from resequenced homozygous bar ( n = 16 ) , homozygous check or T-check ( n = 11 ) , and barless ( n = 9 ) Columba livia; C . rupestris ( n = 1 ) ; C . guinea ( n = 1 ) ; and C . palumbus ( n = 1 ) were extracted using the IGV browser ( Thorvaldsdóttir et al . , 2013 ) . Exon sequences for each group were translated using SIXFRAME in SDSC Biology Workbench ( http://workbench . sdsc . edu ) . Peptide sequences were then aligned to EFHC2 amino acid sequences from other species downloaded from ensembl ( http://www . ensembl . org ) using CLUSTALW ( Thompson et al . , 1994 ) in SDSC Biology Workbench . Exon sequences from additional C . livia ( n = 17 checker or T-check , and n = 14 bar ) and C . guinea ( n = 5 ) birds were determined by Sanger sequencing . Recombination frequency estimates were generated from a genetic map based an F2 cross of two divergent C . livia breeds , a Pomeranian Pouter and a Scandaroon ( Domyan et al . , 2016 ) . Briefly , for genetic map construction , genotyping by sequencing ( GBS ) data were generated , trimmed , and filtered as described ( Domyan et al . , 2016 ) , then mapped to the pigeon genome assembly ( Holt et al . , 2018 ) using Bowtie2 ( Langmead and Salzberg , 2012 ) . Genotypes were called using Stacks ( Catchen et al . , 2011 ) , and genetic map construction was performed using R/qtl ( www . rqtl . org ) ( Broman et al . , 2003 ) . Pairwise recombination frequencies were calculated for all markers based on GBS genotypes . Within individual scaffolds , markers were filtered to remove loci showing segregation distortion ( Chi-square , p<0 . 01 ) or probable genotyping error . Specifically , markers were removed if dropping the marker led to an increased LOD score , or if removing a non-terminal marker led to a decrease in length of >10 cM that was not supported by physical distance . Individual genotypes with error LOD scores > 5 ( Lincoln and Lander , 1992 ) were also removed . Pairwise recombination frequencies for markers flanking the candidate region that were retained in the final linkage map were used to estimate the age of the introgression event between C . guinea and C . livia ( Scaffold 68 , marker positions 1 , 017 , 014 and 1 , 971 , 666; Supplementary file 4 ) . The minimal haplotype age was estimated following Voight et al . ( 2006 ) . We assume a star-shaped phylogeny , in which all samples with the minimal haplotype are identical to the nearest recombination event , and differ immediately beyond it . Choosing a variant in the center of the minimal haplotype , we calculated EHH , and estimated the age using the largest haplotype with a probability of homozygosity just below 0 . 25 . Note thatPr[homoz]= e−2rgwhere r is the genetic map distance , and g is the number of generations since introgression / onset of selection . Thereforeg=−100log⁡ ( Pr[homoz] ) 2r The confidence interval around g was estimated assumingN∼Binom ( n=22 , p=0 . 204 ) Here , N is a binomially distributed random variable for the number of samples that have not recombined to a map distance equal to 2 r . Then , Pr[homoz]=N/22 . The probability that a sample has no recombination event within 2 r of the focal SNP is p = ( Pr[homoz \| left]+Pr[homoz \| right] ) /2 is derived from the data . Both left and right of the focal SNP we chose the end of the haplotype at the first SNP which brought Pr[homoz]<0 . 25 .	The rock pigeon is a familiar sight in urban settings all over the world . Domesticated thousands of years ago and still raised by hobbyists , there are now more than 350 breeds of pigeon . These breeds have a spectacular variation in anatomy , feather color and behavior . Color patterns are important for birds in species recognition , mate choice and camouflage . Pigeon fanciers have long observed that color patterns can be linked to health problems , such as lighter birds suffering more often from poor vision . In addition , pigeons with certain pigment patterns are more likely to survive and reproduce in urban habitats . But despite centuries of pigeon-breeding and the abundance of rock pigeons in urban spaces , how pigeons generate such different feather color patterns , is still largely a mystery . Vickrey et al . sequenced the genomes of pigeons with different patterns and found that a gene called NDP played an important role in wing pigmentation . In birds with darker patterns ( called checker and T-check ) the gene NDP was expressed more in their feathers , but the gene itself was not altered . The lightest colored birds ( barless patterned ) , however , had a mutation in the NDP gene itself that led to less pigmentation . The NDP mutation found in barless pigeons is the same as one that is sometimes found in the human version of NDP , where it is linked to hereditary blindness . Vickrey et al . also discovered that the darker patterns most likely arose from breeding of the rock pigeon with a different species , the African speckled pigeon , something pigeon fanciers have suspected for some time . The findings could help to parse out the different functions of the NDP gene in both pigeons and humans . Mutations in the NDP gene in humans typically cause a range of neurological problems in addition to loss of sight , but in barless pigeons , the mutation appears to cause only vision defects . These findings suggest that a specific part of the gene is particularly important for vision in birds and humans , and shed light on the surprisingly complex evolutionary history of the rock pigeon .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "genetics", "and", "genomics" ]	2018	Introgression of regulatory alleles and a missense coding mutation drive plumage pattern diversity in the rock pigeon
Membrane transporters that clear the neurotransmitter glutamate from synapses are driven by symport of sodium ions and counter-transport of a potassium ion . Previous crystal structures of a homologous archaeal sodium and aspartate symporter showed that a dedicated transport domain carries the substrate and ions across the membrane . Here , we report new crystal structures of this homologue in ligand-free and ions-only bound outward- and inward-facing conformations . We show that after ligand release , the apo transport domain adopts a compact and occluded conformation that can traverse the membrane , completing the transport cycle . Sodium binding primes the transport domain to accept its substrate and triggers extracellular gate opening , which prevents inward domain translocation until substrate binding takes place . Furthermore , we describe a new cation-binding site ideally suited to bind a counter-transported ion . We suggest that potassium binding at this site stabilizes the translocation-competent conformation of the unloaded transport domain in mammalian homologues . Glutamate transporters , or excitatory amino acid transporters ( EAATs ) , reside in the plasma membranes of glial cells and neurons , where they catalyze the re-uptake of the neurotransmitters glutamate and aspartate ( L-asp ) ( Danbolt , 2001 ) . EAATs terminate neurotransmission events supporting memory formation and cognition , and also prevent excitotoxicity caused by overstimulation of glutamate receptors . Dysfunction of EAATs is linked to neurological disorders , poor recovery from stroke and traumatic brain injuries ( Yi and Hazell , 2006; Sheldon and Robinson , 2007; Kim et al . , 2013 ) . To maintain steep trans-membrane glutamate gradients , EAATs transport one substrate molecule together with three sodium ions ( Na+ ) and one proton . After their release into the cytoplasm , counter-transport of one potassium ion ( K+ ) resets the transporter for the next cycle ( Zerangue and Kavanaugh , 1996; Levy et al . , 1998; Owe et al . , 2006 ) . Key mechanistic and structural insights into this family of transporters come from studies on an archaeal homologue from Pyrococcus horikoshii , GltPh ( Figure 1—figure supplement 1 ) , which symports L-asp together with three Na+ ions ( Groeneveld and Slotboom , 2010 ) ; however , it shows no dependence on counter-transport of K+ under the conditions tested ( Ryan et al . , 2009 ) . GltPh , like EAATs , is a homo-trimer ( Gendreau et al . , 2004; Yernool et al . , 2004 ) . Each protomer consists of a central scaffolding trimerization domain and a peripheral transport domain containing the substrate and ion binding sites ( Boudker et al . , 2007; Reyes et al . , 2009 ) . When bound to Na+ and L-asp ( ‘fully bound’ from here on ) , each transport domain moves by ∼15 Å across the membrane from an outward- to an inward-facing position , in which the substrate binding site is near the extracellular solution and the cytoplasm , respectively ( Reyes et al . , 2009 ) . Structurally symmetric helical hairpins , HP1 and HP2 , occlude the bound substrate from the solvent and are thought to serve as gates ( Boudker et al . , 2007; Huang and Tajkhorshid , 2008; Shrivastava et al . , 2008; Reyes et al . , 2009; DeChancie et al . , 2010; Focke et al . , 2011; Zomot and Bahar , 2013 ) . Two Na+-binding sites ( Na1 and Na2 ) , neither of which directly coordinates the substrate , were identified crystallographically using thallium ( Tl+ ) ( Boudker et al . , 2007 ) . The location of the third Na+-binding site is being debated ( Holley and Kavanaugh , 2009; Huang and Tajkhorshid , 2010; Larsson et al . , 2010; Tao et al . , 2010; Bastug et al . , 2012; Teichman et al . , 2012 ) . A highly conserved non-helical Asn310-Met311-Asp312 ( NMD ) motif interrupts trans-membrane segment ( TM ) 7 ( see below ) . It lines the back of the substrate- and ion-binding sites and is involved in binding of the ligands ( Rosental et al . , 2006; Tao et al . , 2006; Rosental and Kanner , 2010 ) . The main chain carbonyl oxygen of Asn310 contributes to Na1 site , while the side chain of Met311 protrudes between the substrate , Na1 and Na2 binding sites ( Boudker et al . , 2007 ) . Symport requires that neither the substrate nor the ions alone are efficiently transported ( Crane , 1977 ) . Therefore to traverse the membrane , the transport domains of GltPh and EAATs must be loaded with both Na+ ions and substrate . To complete the transport cycle , the transport domain of GltPh must also translocate readily when it is free of both solutes ( apo ) , while in EAATs it requires binding of a K+ ion . To establish the structural underpinnings of these processes , we determined crystal structures of the outward- and inward-facing states of GltPh in apo and ions-only bound forms ( Tables 1 , 2 and 3 ) . We find that the apo transport domain shows identical structures when facing outward or inward . While ligand-binding sites are distorted , the domain remains compact , suggesting that it relocates across the membrane as a rigid body , similarly to when it is fully bound ( Reyes et al . , 2009 ) . Ion binding to Na1 site , located deep in the core of the transport domain , triggers structural changes that are propagated to the extracellular gate HP2 , at least in part , by the side chain of Met311 in the NMD motif . Consequently HP2 , which in the apo form is collapsed into the substrate binding and Na2 sites , frees the sites , assuming conformations more similar to the conformation observed in the fully bound transporter . We suggest that these Na+-dependent structural changes underlie the high cooperativity of Na+ and substrate binding , which is thought to be one of the key coupling mechanisms ( Reyes et al . , 2013 ) . Furthermore , in the structure of Na+-bound outward-facing GltPh we observe opening of HP2 tip , which may facilitate L-asp access to its binding site and prevent the inward movement of the Na+-only bound transport domain , as previously suggested ( Focke et al . , 2011 ) . Remarkably , soaks of apo GltPh crystals in Tl+ reveal new cation-binding sites within the apo-like protein architecture . One such site overlaps with the substrate-binding site . Because binding of a cation to this site would compete with binding of Na+ and the transported substrate , it is well suited to serve as a binding site for a counter-transported ion . We propose that the closed translocation-competent conformation of the transport domain free of Na+ and substrate is intrinsically stable in GltPh but not in EAATs , in which K+ binding at the newly identified site is required , coupling transport cycle completion to K+ counter-transport . 10 . 7554/eLife . 02283 . 005Table 1 . X-ray crystallographic data and refinement statistics for GltPh-R397A and GltPh-K55C-A364CHg ( GltPhin ) structures deposited at the PDBDOI: http://dx . doi . org/10 . 7554/eLife . 02283 . 005GltPhinapoTl+-bound ( apo conf . ) alkali-freeTl+-bound ( bound conf . ) Data collection Space groupC2221C2221C2221C2221 Cell dimensions a , b , c ( Å ) 109 . 93 , 201 . 81 , 207 . 14106 . 98 , 196 . 56 , 206 . 50106 . 95 , 196 . 84 , 207 . 48110 . 83 , 200 . 43 , 206 . 40 α , β , γ ( ° ) 90 . 00 , 90 . 00 , 90 . 0090 . 00 , 90 . 00 , 90 . 0090 . 00 , 90 . 00 , 90 . 090 . 00 , 90 . 00 , 90 . 00 Resolution ( Å ) 100 . 0–3 . 25 ( 3 . 31–3 . 25 ) 100 . 0–3 . 75 ( 3 . 81–3 . 75 ) 100 . 0–3 . 50 ( 3 . 56–3 . 50 ) 100 . 0–4 . 0 ( 4 . 14–4 . 0 ) Rsym or Rmerge10 . 9 ( 88 . 6 ) 14 . 0 ( 94 . 4 ) 8 . 0 ( 88 . 1 ) 16 . 3 ( 75 . 2 ) I/σI12 . 3 ( 1 . 2 ) 8 . 95 ( 1 . 1 ) 13 . 5 ( 1 . 2 ) 7 . 9 ( 1 . 3 ) Completeness ( % ) 98 . 7 ( 88 . 1 ) 99 . 7 ( 99 . 8 ) 94 . 4 ( 92 . 7 ) 65 . 2 ( 6 . 5 ) Redundancy5 . 6 ( 2 . 8 ) 3 . 8 ( 3 . 7 ) 3 . 3 ( 3 . 2 ) 3 . 4 ( 3 . 5 ) Refinement Resolution ( Å ) 15 . 0–3 . 2515 . 0–3 . 7515 . 0–3 . 515 . 0–4 . 0 No . reflections34534215652544611105 Rwork/Rfree22 . 2/25 . 823 . 0/25 . 726 . 3/27 . 825 . 8/29 . 6 No . atoms Protein9121911490888985 Ligand/ion3939 B-factors Protein108 . 5141 . 8144 . 2137 . 2 Ligand/ion135 . 3170 . 8214 . 1102 . 3 R . m . s . deviations Bond lengths ( Å ) 0 . 0100 . 0130 . 0050 . 012 Bond angles ( ° ) 1 . 6801 . 8611 . 1161 . 407PDB code4P194P1A4P3J4P6HGltPh-R397AApoNa+-boundNa+/aspartate-boundData collection Space groupP21P31P31 Cell dimensions a , b , c ( Å ) 112 . 37 , 424 . 42 , 113 . 99110 . 58 , 110 . 58 , 306 . 92116 . 96 , 116 . 96 , 313 . 52 α , β , γ ( ° ) 90 . 00 , 119 . 40 , 90 . 0090 . 00 , 90 . 00 , 120 . 0090 . 00 , 90 . 00 , 120 . 00 Resolution ( Å ) 100 . 0–4 . 00 ( 4 . 14–4 . 00 ) 50 . 0–3 . 39 ( 3 . 51–3 . 39 ) 100 . 0–3 . 50 ( 3 . 63–3 . 50 ) Rsym or Rmerge7 . 8 ( 62 . 2 ) 14 . 0 ( >100 ) 8 . 4 ( >100 ) I/σI9 . 3 ( 1 . 3 ) 13 . 8 ( 1 . 4 ) 10 . 6 ( 0 . 4 ) Completeness ( % ) 67 . 9 ( 13 . 0 ) 87 . 3 ( 12 . 0 ) 98 . 1 ( 96 . 6 ) Redundancy1 . 8 ( 2 . 0 ) 11 . 8 ( 8 . 6 ) 4 . 5 ( 4 . 2 ) Refinement Resolution ( Å ) 20 . 0–4 . 012 . 0–3 . 4115 . 0–3 . 50 No . reflections520684836655613 Rwork/Rfree24 . 9/26 . 628 . 4/29 . 324 . 3/26 . 8 No . atoms Protein352771758018192 Ligand/ionN/A654/12 WaterN/A66 B-factors Protein139 . 5152 . 097 . 1 Ligand/ionN/A145 . 184 . 7/86 . 9 WaterN/A102 . 6144 . 6 R . m . s . deviations Bond lengths ( Å ) 0 . 0100 . 0100 . 015 Bond angles ( ° ) 1 . 3931 . 4681 . 735PDB code4OYE4OYF4OYG10 . 7554/eLife . 02283 . 006Table 2 . Completeness of datasets corrected for anisotropyDOI: http://dx . doi . org/10 . 7554/eLife . 02283 . 006Tl+-bound GltPhin ( bound conformation ) Na+-bound GltPh-R397AResolution range ( Å ) Completeness ( % ) Resolution range ( Å ) Completeness ( % ) 100 . 0–8 . 6299 . 350 . 00–7 . 3099 . 68 . 62–6 . 8499 . 97 . 30–5 . 79100 . 06 . 84–5 . 97100 . 05 . 79–5 . 06100 . 05 . 97–5 . 4399 . 95 . 06–4 . 60100 . 05 . 43–5 . 0499 . 94 . 60–4 . 27100 . 05 . 04–4 . 7469 . 64 . 27–4 . 02100 . 04 . 74–4 . 5039 . 24 . 02–3 . 82100 . 04 . 50–4 . 3123 . 63 . 82–3 . 6598 . 64 . 31–4 . 1414 . 43 . 65–3 . 5163 . 04 . 14–4 . 006 . 53 . 51–3 . 3912 . 0Apo GltPh-R397AResolution range ( Å ) Completeness ( % ) 100 . 0–8 . 6285 . 08 . 62–6 . 8475 . 66 . 84–5 . 9775 . 55 . 97–5 . 4375 . 35 . 43–5 . 0475 . 25 . 04–4 . 7475 . 84 . 74–4 . 5075 . 34 . 50–4 . 3175 . 44 . 31–4 . 1451 . 74 . 14–4 . 0013 . 010 . 7554/eLife . 02283 . 007Table 3 . X-ray crystallographic data and refinement statistics for GltPh-R397A and GltPh-K55C-A364CHg structures not deposited at the PDBDOI: http://dx . doi . org/10 . 7554/eLife . 02283 . 007GltPh-R397AGltPhinTl+-bound ( apo conf . ) Tl+/Na+ ( apo conf . ) Tl+/k+ ( apo conf . ) Data collection Space groupP21C2221C2221 Cell dimensions a , b , c ( Å ) 115 . 18 , 428 . 53 , 116 . 61108 . 11 , 198 . 86 , 206 . 34106 . 59 , 198 . 48 , 205 . 82 α , β , γ ( ° ) 90 . 00 , 119 . 49 , 90 . 0090 . 00 , 90 . 00 , 90 . 0090 . 00 , 90 . 00 , 90 . 00 Resolution ( Å ) 30 . 0–5 . 0 ( 5 . 18–5 . 00 ) 100 . 0–4 . 0 ( 4 . 07–4 . 00 ) 100 . 0–4 . 15 ( 4 . 22–4 . 15 ) Rsym or Rmerge10 . 9 ( >100 ) 15 . 0 ( 92 . 2 ) 13 . 9 ( 94 . 1 ) I/σI13 . 8 ( 1 . 9 ) 8 . 9 ( 1 . 5 ) 9 . 2 ( 1 . 5 ) Completeness ( % ) 86 . 4 ( 75 . 1 ) 99 . 9 ( 100 ) 94 . 5 ( 90 . 2 ) Redundancy5 . 5 ( 5 . 8 ) 3 . 9 ( 3 . 9 ) 4 . 0 ( 3 . 9 ) Refinement Resolution ( Å ) 20 . 0–5 . 015 . 0–4 . 015 . 0–4 . 15 No . reflections347471818415419 Rwork/Rfree22 . 0/26 . 528 . 2/31 . 728 . 3/31 . 2 No . atoms Protein3510791359135 Ligand/ionN/AN/AN/A WaterN/AN/AN/A B-factors Protein223 . 00183 . 6194 . 4 Ligand/ionN/AN/AN/A WaterN/AN/AN/A R . m . s . deviations Bond lengths ( Å ) 0 . 0080 . 0060 . 008 Bond angles ( ° ) 1 . 1861 . 2661 . 440 To determine the structure of apo GltPh , we used R397A mutant that shows a drastically decreased affinity for substrate ( Figure 1A ) . When fully bound , GltPh-R397A crystallizes in the outward-facing state , like wild type GltPh , except that L-asp coordination is slightly altered because the mutant is missing the key coordinating side chain of Arg397 ( Figure 1B , Figure 1—figure supplement 2; Bendahan et al . , 2000; Boudker et al . , 2007 ) . These results suggest that R397A is suitable to capture the apo and ions-only bound outward-facing states for their structural characterization . However , removal of Arg397 may affect local electrostatics , potentially altering ion binding; thus these studies should be interpreted with caution . Apo GltPh-R397A also crystallized in an outward-facing conformation that is similar to the structure reported for a close GltPh homologue ( Jensen et al . , 2013 ) . To obtain an apo inward-facing state , we used GltPh-K55C-A364C mutant trapped in the inward-facing state upon cross-linking with mercury ( Reyes et al . , 2009 ) ( GltPhin , Figure 1—figure supplement 1 ) . The positions and orientations of the transport domains relative to the trimerization domains remain essentially unchanged in the apo and fully bound forms of GltPh-R397A and GltPhin ( Figure 2 ) . In contrast , the conformations of the transport domains themselves differ significantly . Most remarkably , the apo conformations of the transport domain are nearly identical in the outward- and inward-facing states ( Figure 3A , Figure 3—figure supplement 1 , Figure 3—figure supplement 2A ) and are therefore independent of the transport domain orientations and crystal packing environments . 10 . 7554/eLife . 02283 . 008Figure 1 . Substrate binding to GltPh-R397A . ( A ) Raw binding heat rates measured by isothermal titration calorimetry ( top ) and binding isotherms ( bottom ) obtained for GltPh-R397A ( left ) and wild type GltPh ( right ) at 25°C in the presence of 100 mM NaCl . The solid lines through the data are fits to the independent binding sites model with the following parameters for GltPh-R397A and wild type GltPh , respectively: enthalpy change ( ΔH ) of −3 . 2 and −14 . 3 kcal/mol; the apparent number of binding sites ( n ) of 0 . 8 and 0 . 7 per monomer; dissociation constant ( Kd ) of 6 . 6 µM and 27 nM . Note that L-asp binding to the wild type transporter is too tight at 100 mM NaCl to be accurately measured in this experiment . The binding Kd has been estimated to be ∼1 nM ( Boudker et al . , 2007 ) . ( B ) L-asp binding site in GltPh-R397A ( left ) and wild type GltPh ( right ) . L-asp and residues coordinating the side chain carboxylate are shown as sticks with carbon atoms colored light brown and blue , respectively . Potential hydrogen bonds ( distances less than 3 . 5 Å ) between the L-asp side chain carboxylate and transporter residues are shown as dashed lines . Note that Y317 , which forms cation-π interactions with guanidium group of R397 in wild type GltPh , interacts directly with L-asp in GltPh-R397A . DOI: http://dx . doi . org/10 . 7554/eLife . 02283 . 00810 . 7554/eLife . 02283 . 009Figure 1—figure supplement 1 . Alternating access mechanism in GltPh . ( A ) The transporter consists of a rigid trimerization domain ( light brown ) and a dynamic transport domain ( blue , with HP1 yellow and HP2 red ) . The apo outward-facing transporter binds the substrate ( blue square ) and Na+ ions ( pink circles ) within the transport domain under the tip of HP2 . The transport domain moves into the inward-facing state , in which the ligands are occluded from the cytoplasm by the tips of HP1 and HP2 , and then released . ( B ) GltPh trimer viewed from the extracellular medium . ( C ) Single protomers of the outward-facing GltPh ( PDB code 2NWX ) ( left ) , and GltPhin ( PDB code 3KBC ) ( right ) . Cα atoms of the cross-linked cysteine residues and Hg2+ ion are shown as spheres . The bar next to the structures indicates approximately the thickness of the membrane , separating the extracellular ( Ext ) and intracellular ( Int ) solutions . DOI: http://dx . doi . org/10 . 7554/eLife . 02283 . 00910 . 7554/eLife . 02283 . 010Figure 1—figure supplement 2 . Structure of GltPh-R397A bound to Na+ and L-asp . Stereo view of the averaged 2Fo-Fc electron density map contoured at 1σ ( grey mesh ) around residues in L-asp binding site of GltPh-R397A . DOI: http://dx . doi . org/10 . 7554/eLife . 02283 . 01010 . 7554/eLife . 02283 . 011Figure 2 . Apo protomer structures . ( A ) GltPh protomers in the outward-facing state ( left ) and a GltPhin protomer ( right ) viewed from within the plane of the membrane . Shown are superimpositions between apo ( colors ) and fully bound protomers ( grey ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02283 . 01110 . 7554/eLife . 02283 . 012Figure 3 . Structures of the apo transport domain . ( A ) Superimposition of the nearly identical apo transport domains in the outward- and inward-facing states . HP1 , HP2 , and NMD motif are colored yellow , red , and green , respectively . The remainder of the domain is blue . ( B ) Superimposition of the fully bound ( light colors , PDB accession number 2NWX ) and apo GltPhin ( dark colors ) transport domains . ( C ) The NMD motif and adjacent TM3 . Met311 is shown as sticks , and the light blue spheres indicate the Cα positions for T92 and S93 . ( D ) The HP2-TM8a structural modules in the fully bound ( pink ) and apo ( red ) transport domains superimposed on TM8a and HP2b to emphasize the re-orientation of the HP2a . ( E ) The Na+ and L-asp binding sites in the fully bound ( left ) and apo forms ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02283 . 01210 . 7554/eLife . 02283 . 013Figure 3—figure supplement 1 . Apo protomer structures . Stereo views of the averaged 2Fo-Fc electron density maps contoured at 1σ around HP2 tip of the apo GltPh-R397A ( top ) and the NMD motif and TM3 of the apo GltPhin ( bottom ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02283 . 01310 . 7554/eLife . 02283 . 014Figure 3—figure supplement 2 . Structural comparison of the transport domain in various states . Plotted are root mean square deviations ( R . M . S . D . s ) of the main chain atoms calculated per residue using VMD for the transport domains in ( A ) apo outward- and inward-facing states , ( B ) outward-facing apo and fully bound forms , ( C ) outward-facing apo and Na+-bound forms and ( D ) outward-facing Na+-bound and fully loaded forms . The loop regions were excluded from the calculations . Notably , there are no significant structural differences between the apo transport domains in the outward- and inward-facing states . Comparison of the apo and fully bound forms of the domain shows differences in TM3 , near NMD motif in TM7 and in HP2 . Most of these differences are also observed when apo transport domain is compared to Na+-bound form . In contrast , differences between the Na+-bound and fully loaded forms are confined mostly to the tip of HP2 . DOI: http://dx . doi . org/10 . 7554/eLife . 02283 . 014 The conformational differences between fully bound and apo forms of the transport domain include a concerted movement of HP2 and TM8a , which form the extracellular surface of the domain , and local rearrangements at the ligand binding sites , involving HP2 , the NMD motif and TM3 ( Figure 3B–E , Figure 3—figure supplement 2B , Figure 4 ) . In HP2 , the last helical turn of HP2a unwinds , and HP2a together with the loop region at HP2 tip collapse into the substrate and Na2 binding sites . Within the NMD motif , the side chain of Asn310 rotates away from TM3 and partially fills the empty Na1 site , while the side chain of Met311 undergoes an opposite movement , flipping away from the binding sites ( Figure 4 ) . Finally , TM3 bends away from the NMD motif , particularly around Thr92 and Ser93 ( Figure 3B , C ) . Notably , these residues together with the side chain of Asn310 form one of the proposed third Na+-binding sites ( Huang and Tajkhorshid , 2010; Bastug et al . , 2012 ) . Thus , all known ligand-binding sites are distorted in the apo forms ( Figure 4 ) . 10 . 7554/eLife . 02283 . 003Figure 4 . Remodeling of L-asp and Na+ binding sites in the apo conformations . Close-up views of the fully bound ( left ) and apo ( right ) transport domains at L-asp binding site ( top ) , Na1 and Na2 sites ( middle ) , and one of the proposed locations for the third Na+ binding site ( Huang and Tajkhorshid , 2010; Bastug et al . , 2012 ) ( dashed circle ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02283 . 00310 . 7554/eLife . 02283 . 004Figure 4—figure supplement 1 . Transport domain remains compact . Surface representation of the transport domain in fully bound and apo forms . DOI: http://dx . doi . org/10 . 7554/eLife . 02283 . 004 The overall structures of the apo transport domain remain as closed and compacted as in the fully bound forms ( Figure 4—figure supplement 1 ) . Therefore , we propose that the unloaded transport domains traverse the membrane as rigid bodies as deduced previously for the fully loaded transport domains ( Reyes et al . , 2009 ) . In GltPh , cooperative binding of Na+ ions and L-asp is central to tightly coupled transport of the solutes ( Reyes et al . , 2013 ) . Our structures of the apo and fully bound GltPh suggest that binding of L-asp and Na+ at the Na2 site is coupled because the same structural element , the tip of HP2 , contributes to both sites and is restructured upon binding . Thus , structural changes in HP2 upon binding of either L-asp or Na+ ion should greatly favor binding of the other . Met311 in the NMD motif is the only residue that is shared between the Na1 site and the substrate and Na2 sites and also undergoes a conformational change upon ligand binding . To examine whether the structural changes in HP2 upon binding of L-asp and Na+ at the Na2 site could occur independently from those in the NMD motif upon Na+ binding at the Na1 site , we modeled transport domains with HP2 in the bound conformation and the NMD motif in the apo conformation , or vice versa ( Figure 5A ) . In both models , the side chain of Met311 clashes with residues in HP2 , suggesting that the conformational changes in HP2 and the NMD motif must be concerted . 10 . 7554/eLife . 02283 . 015Figure 5 . Met311 is key to the allosteric coupling . ( A ) Structural models combining HP2 bound to L-asp and Na+ at Na2 site with apo conformation of the NMD motif ( left ) , and apo conformation of HP2 with the NMD motif bound to Na+ at Na1 site ( right ) . Met311 and clashing residues in HP2 are shown as sticks and transparent spheres . ( B ) The dependence of L-asp dissociation constant , Kd , on Na+ activity plotted on a log–log scale for mutants within the context of GltPhin ( left ) and unconstrained GltPh ( right ) . The data were fitted to straight lines with slopes shown on the graph or to arbitrary lines for clarity . Dashed lines and corresponding slopes correspond to published dependences for GltPhin and GltPh ( Reyes et al . , 2013 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02283 . 015 We then mutated bulky Met311 to either another bulky residue , leucine , or to a smaller residue , alanine , which is not expected to experience similar clashes . For these mutants , generated in the context of unconstrained wild type GltPh and inward cross-linked GltPhin , we measured the dependence of L-asp dissociation constant on Na+ concentration ( Figure 5B ) . While this dependence is very steep for the wild type GltPh constructs ( Reyes et al . , 2013 ) and nearly as steep for the M311L mutants , it is substantially shallower for the M311A mutants . The most parsimonious interpretation of these results is that M311A mutation reduces binding cooperativity between the substrate and Na+ ions . However , it is also possible , though we think unlikely , that the mutation abrogates ion binding at one or more Na+-binding sites in the tested concentration range ( 1–100 mM ) . Mutating the equivalent methionine to smaller residues in EAAT3 also resulted in less steep dependence of the ionic currents on Na+ concentration ( Rosental and Kanner , 2010 ) . Based on these results , we hypothesize that Met311 is key to the allosteric coupling between the Na1 , L-asp and Na2 sites . Consistently , bulky methionine or leucine residues are found at this position in ∼85% of glutamate transporter homologues . However , it should be noted that methionine is conserved in the Na+-coupled GltPh and EAATs , while a characterized proton-coupled homologue has leucine at this position ( Gaillard et al . , 1996 ) . Hence , it is possible that the methionine thioether , which is proximal to both Na1 and Na2 sites , plays a direct role in Na+ binding . Our hypothesis further predicts that binding of an ion at Na1 site should prime the transporter to accept its substrate . Therefore , we crystallized GltPh-R397A in the presence of 400 mM Na+ , but in the absence of L-asp . We also soaked crystals of apo GltPhin in Tl+ , an ion with strong anomalous signal that seems to mimic some aspects of Na+ in GltPh and EAATs ( Boudker et al . , 2007; Tao et al . , 2008 ) . The obtained outward- and inward-facing structures pictured the transport domains in conformations overall similar to those observed in the fully bound transporter: straightened TM3 , Met311 pointing toward the binding sites , extended helix in HP2a and HP2 tip raised out of the substrate binding site ( Figure 6A–D ) . Indeed , the structure of Tl+-bound GltPhin is indistinguishable from the fully bound GltPhin and both Na1 and Na2 sites are occupied by Tl+ ions ( Figure 6A ) . The structure of Na+-bound GltPh-R397A differs significantly from the fully bound GltPh-R397A only at the tip of HP2 ( Figure 3—figure supplement 2 , also see below ) . The coordinating residues at the Na1 site are correctly positioned and the site is likely occupied by a Na+ ion . The Na2 site still shows a distorted geometry: the last helical turn of HP2a points away from the site due to the altered conformation of the tip of HP2 ( Figure 6C ) . Collectively , our results demonstrate that binding of the coupled ions , notably at the Na1 site , is sufficient to trigger isomerization of the transport domain from the apo conformation to the bound-like conformation . The energetic penalty associated with this isomerization likely explains why Na+ ions alone bind weakly to the transporter ( Reyes et al . , 2013 ) . This experimental observation contrasts with highly favorable calculated binding energies ( approximately −10 kcal/mol for Na1 ) that were obtained using fully bound protein conformation and where the reference ion-free state is the same as the bound state ( Larsson et al . , 2010; Bastug et al . , 2012; Heinzelmann et al . , 2013 ) . 10 . 7554/eLife . 02283 . 016Figure 6 . Structures of ions-only bound transport domain . ( A ) Superimposition of the fully bound transport domains ( grey ) and Tl+-bound GltPhin transport domain in the bound-like conformation ( colors ) , with the averaged anomalous difference Fourier map contoured at 8σ ( cyan mesh ) . ( B ) Superimposition of the fully bound ( grey ) and Na+-only bound GltPh-R397A ( colors ) transport domains . ( C ) Na+ and L-asp binding sites with fully-bound structure shown in white and Na+-bound structure in colors . Hinge glycine residues are shown as spheres . The modeled Na+ ion in Na1 site is pink . ( D ) Superimposition of the HP2-TM8 in the fully bound transport domain ( grey ) and in GltPh-R397A bound to Na+ only ( colors ) , showing similar conformations of HP2a . ( E ) WebLogo representation of the consensus sequence and relative abundance of residues in HP2 tip . ( F ) Surface representation of the transport domain of GltPh-R397A bound to Na+ only showing access to the substrate-binding site . L-asp was placed into the binding site for reference . DOI: http://dx . doi . org/10 . 7554/eLife . 02283 . 01610 . 7554/eLife . 02283 . 017Figure 6—figure supplement 1 . Na+ only bound GltPh-R397A . Stereo view of the averaged 2Fo-Fc electron density map contoured at 1σ around HP1 and HP2 . DOI: http://dx . doi . org/10 . 7554/eLife . 02283 . 01710 . 7554/eLife . 02283 . 018Figure 6—figure supplement 2 . Superimposition of the transport domains bound to Na+ and L-asp ( light grey ) , Na+ and L-TBOA ( dark grey ) and Na+ only ( colors ) . Ligands are omitted for clarity , except that L-TBOA is shown as spheres in the right panel . Nostably , the observed additional opening of HP2 is necessary to accommodate L-TBOA . DOI: http://dx . doi . org/10 . 7554/eLife . 02283 . 01810 . 7554/eLife . 02283 . 019Figure 6—figure supplement 3 . Sequence alignment for the HP2 tip region of GltPh and human EAAT sub-types 1–5 . DOI: http://dx . doi . org/10 . 7554/eLife . 02283 . 019 The structure of the Na+-only bound GltPh-R397A shows HP2 in a conformation overall similar to that observed in the fully bound transporter , but with an opened tip ( Figure 6B–D , F , Figure 6—figure supplement 1 ) . This opening is smaller than the opening observed previously in the structure of GltPh in complex with the blocker L-threo-β-benzyloxyaspartate ( Figure 6—figure supplement 2; Boudker et al . , 2007 ) , and it is hinged at two well-conserved glycine residues at positions 351 and 357 ( Figure 6C ) . Interestingly , among the nine amino acids forming the tip in GltPh ( residues 351 to 359 ) , five are glycines in the consensus sequence generated for the glutamate transporter family , although not all are present in each homologue ( Figure 6E , Figure 6—figure supplement 3 ) . We suggest that the glycines support the structural flexibility of the HP2 tip in all members of the family , but that the structural specifics of the tip opening may vary among homologues . To test whether the trans-membrane movement of the transport domain is possible when the tip of HP2 is opened , we modeled the open tip conformation in the context of the previously reported early transition intermediate structure ( Figure 7; Verdon and Boudker , 2012 ) . In this structure , the transport domain tilts towards the trimerization domain but does not yet undergo a significant translation toward the cytoplasm . We find that such intermediate state with the opened tip of HP2 can be achieved without major steric clashes , while further progression of the transport domain to the inward-facing position could be impeded because the tip is likely to clash with TM5 in the trimerization domain ( Figure 7B ) . Also in the inward-facing state HP2 is packed against the trimerization domain and cannot open in a manner observed in the outward-facing state . Consistently , HP2 is closed in GltPhin bound to Tl+ ( Figure 6A ) . 10 . 7554/eLife . 02283 . 020Figure 7 . Modeled Na+-bound early transition intermediate between the outward- and inward-facing states . ( A ) Surface representations of the protomer in the fully bound intermediate state ( PDB code 3V8G ) ( left ) , and the modeled Na+-bound intermediate with an open HP2 tip ( right ) viewed from the extracellular space ( top ) . The model reveals no clashes , suggesting that the observed opening of HP2 is structurally compatible with the intermediate orientation of the transport domain . The arrows indicate the point of access to the domain interface with potentially increased solvent accessibility . ( B ) Side views of thin cross-sections of the closed fully bound ( left ) and open Na+-only bound ( right ) intermediate state . The protomers are sliced normal to the membrane plane , as indicated by the dashed lines in A . DOI: http://dx . doi . org/10 . 7554/eLife . 02283 . 020 Therefore , opening of the HP2 tip upon Na+ binding in the outward-facing state may serve as a structural mechanism preventing uncoupled uptake of Na+ ions . We suggest that the structural changes in the NMD motif and HP2 that are triggered upon Na+ binding at the Na1 site may lead to the loss of direct interactions between the tip of HP2 and the rest of the transport domain , resulting in tip opening . Subsequent binding of L-asp and Na+ at the Na2 site is then required to provide compensatory interactions , allowing HP2 tip to close . Similar conformational behavior has been observed for transporters with the LeuT fold: when bound to Na+ ions only , substrate binding sites are open to the extracellular solution , and substrate binding is required for occlusion ( Weyand et al . , 2008; Krishnamurthy and Gouaux , 2012 ) . We do not see a transition into an open conformation in the inward-facing GltPhin bound to Tl+ ions ( Figure 6A ) . This may be because Tl+ ions do not faithfully mimic Na+ ions and fail to induce an open state or it may be because Na+ bound inward-facing state is , indeed , closed . This latter possibility does not contradict the requirements of symport because the measured dissociation constant for Na+ ions in the inward-facing state ( 250 mM ) ( Reyes et al . , 2013 ) , is far above Na+ concentration in the cytoplasm ( 10 mM ) and therefore , Na+-bound inward-facing state is not expected to be significantly populated . We and others have proposed that transition intermediates mediate fluxes of polar solutes , including anions , because potentially hydrated cavities form in such intermediates at the interface between the trimerization and transport domains ( Stolzenberg et al . , 2012; Verdon and Boudker , 2012; Li et al . , 2013 ) . Interestingly , because the tip of HP2 forms part of this interface in the fully bound intermediate state of the transporter , opening of the tip in the Na+-only bound form may increase solvent accessibility to the interface ( Figure 7A ) . While soaking apo GltPhin crystals in Tl+ solutions , we observed that only in approximately one third of crystals Tl+ ions bound to the Na1 and Na2 sites , inducing transition from apo- to bound-like conformation as described above . In the majority of the crystals , we observed no conformational changes of the transport domain and Tl+ ions incorporated at two previously uncharacterized sites ( Figure 8 ) , within the small cavities that remain under the collapsed HP2 ( Figure 8—figure supplement 1 ) . One site , termed Na2' , involves residues of HP2 and TM7a that form the Na2 site , but in a different ion coordinating geometry due to the conformational difference in HP2 ( Figure 8B ) . The second site , termed Ct , overlaps with the L-asp binding site and is formed by the side chains of highly conserved Asp394 and Thr398 in TM8 and main chain carbonyl oxygen atoms of HP1 and HP2 ( Figure 8B , C ) . Tl+ soaks of the outward-facing apo GltPh-R397A also showed no conformational changes of the transport domain , with Tl+ binding at the Ct site , but not at the Na2' site ( Figure 8A ) . The ion selectivity of the Ct site remains ambiguous , because neither 300 mM K+ nor 10 mM Na+ efficiently inhibited incorporation of Tl+ ( 150 mM ) at this site in GltPhin ( Figure 8—figure supplement 2 ) . Crystals deteriorated at higher Na+ concentrations . In contrast , the Na2' site seems to show a preference for Na+ , which even at low concentration ( 10 mM ) interfered significantly with Tl+ binding . 10 . 7554/eLife . 02283 . 021Figure 8 . New cation binding sites . ( A ) Superimpositions of GltPhin ( left ) and outward-facing GltPh-R397A ( right ) transport domains in the apo form ( grey ) with Tl+-bound apo-like conformations ( colors ) . Averaged anomalous difference Fourier maps are contoured at 8σ ( cyan mesh ) . ( B ) Modeled Tl+ ions bound to the Ct and Na2’ sites ( left ) , and L-asp aspartate and Na+ ions bound to the Na1 and Na2 sites in the fully bound transport domain ( right ) . ( C ) Close-up view of Tl+ in the Ct site of apo-like GltPhin and L-asp in the fully bound transporter . DOI: http://dx . doi . org/10 . 7554/eLife . 02283 . 02110 . 7554/eLife . 02283 . 022Figure 8—figure supplement 1 . Transport domain internal cavities . Internal cavities in the apo GltPhin structure ( left ) . Cavities were calculated using solvent radius of 1 . 4 Å and colored by local electrostatic potential with red and blue being negative and positive , respectively . The same structure superimposed with the Tl+-bound apo-like GltPhin ( right ) . The Tl+-bound structure is shown in grey and Tl+ ions at the Ct and Na2' sites are shown as spheres . DOI: http://dx . doi . org/10 . 7554/eLife . 02283 . 02210 . 7554/eLife . 02283 . 023Figure 8—figure supplement 2 . Specificity of the new cation binding sites in apo-like GltPhin . ( A ) Averaged anomalous difference Fourier maps contoured at 8σ ( cyan mesh ) for Tl+-soaked ( 150 mM ) GltPhin in the presence of 300 mM K+ ( left ) and 10 mM Na+ ( right ) showing a decrease of the Tl+ signal at Na2' site in the presence of Na+ . ( B ) Means of the anomalous difference Fourier peak heights in the three protomers and their associated standard deviations at the Ct and Na2' sites in GltPhin crystals . Soaking conditions are listed below the graphs . DOI: http://dx . doi . org/10 . 7554/eLife . 02283 . 023 The functional relevance of these sites is speculative at present . However , it is remarkable that the Ct site is positioned exactly at the same place as the amino group of the bound L-asp and share several coordinating moieties . Therefore , binding of a cation at the Ct site and binding of the substrate are mutually exclusive . Because the Ct site is observed only in the apo-like conformation , cation binding at this site would also inhibit the transition into the bound-like conformation upon Na+ binding at the Na1 site . Finally , the Ct site is observed in both the inward- and outward-facing states , suggesting that the apo-like transport domain could carry the ion across the membrane . These are the exact properties expected for the K+-binding site in EAATs . Moreover , it has been previously proposed that K+ binds to EAATs at a similar position ( Holley and Kavanaugh , 2009 ) . Most remarkably , in an insect K+-independent dicarboxylate transporter , an asparagine to aspartate mutation at the position equivalent to Asp394 in GltPh changes the transporter substrate specificity to amino acid glutamate , and also leads to dependence on K+ counter-transport ( Wang et al . , 2013 ) . Therefore this aspartate plays a key role in both binding the amino group of substrate and coupling to K+ counter-transport . Consistently , Asp394 in GltPh coordinates both the amino group of the bound substrate and Tl+ in the Ct site . Notably , while Tl+ mimics , to some extent , Na+ ions in GltPh and EAATs , it is a better mimic of K+ ions in EAATs ( Boudker et al . , 2007; Tao et al . , 2008 ) . To examine whether a complete removal of Na+ and K+ ions had an effect on the structure of GltPhin , we soaked apo GltPhin crystals ( typically grown in the presence of K+ ) in alkali-free buffer . Interestingly , we observed a small , but reproducible structural change in several crystals examined: HP1 and TM7a that form the transport domain cytoplasmic surface moved slightly towards TM8 , with the tip of HP1 detaching from that of HP2 ( Figure 9 , Figure 9—figure supplement 1 ) . This movement is observed clearly in one protomer ( chain B in 4P3J ) , in which these helices are not involved in crystal packing contacts . It is reminiscent of the isomerization of the structurally symmetric HP2 and TM8a on the extracellular side of the domain observed upon the transition from bound to apo forms ( Figure 9B ) . It was suggested previously that HP1 participates in intracellular gating in GltPh ( Reyes et al . , 2009; DeChancie et al . , 2010 ) . Indeed , the observed movement of HP1 generates a small opening , leading to the substrate and Ct sites ( Figure 9—figure supplement 2 ) , and it is reminiscent of the movement observed in molecular dynamics simulations ( DeChancie et al . , 2010; Zomot and Bahar , 2013 ) . However , this conformational difference is too small to be interpreted unambiguously . 10 . 7554/eLife . 02283 . 024Figure 9 . Movements of the HP1-TM7a structural module . ( A ) Superimposition of GltPhin transport domains when bound to Tl+ in the apo-like conformation ( grey ) and when prepared in an alkali-free solution ( colors ) . ( B ) The transport domains of the fully bound GltPh ( light colors ) and alkali-free GltPhin ( dark colors ) superimposed on TM6 . Arrows indicate movements of the structurally symmetric HP1-TM7a and HP2-TM8a modules . ( C ) Per residue main chain R . M . S . D . values calculated for the structures of the inward-facing transport domains bound to Tl+ in apo-like conformation and alkali-free shown in A and superimposed on HP2 . The bars above the plot represent secondary structure elements colored as in A . DOI: http://dx . doi . org/10 . 7554/eLife . 02283 . 02410 . 7554/eLife . 02283 . 025Figure 9—figure supplement 1 . Alkali-free inward-facing GltPhin . ( A ) Stereo view of the 2Fo-Fc ( 1σ ) and Fo-Fc ( 3σ ) omit maps obtained after molecular replacement using Tl+-bound GltPhin and refinement of a model with HP1 and TM7a omitted , in the protomer showing the detachment of HP1 . The model shown is that of Tl+-bound GltPhin . It is clear that HP1 and TM7a do not fit well into the electron density . ( B ) Stereo view of the 2Fo-Fc ( 1σ ) map around HP1 and TM7a obtained after refinement of a complete model . The model was generated by moving HP1 and TM7a as a rigid-body to fit into the electron density , with no further manual rebuilding . DOI: http://dx . doi . org/10 . 7554/eLife . 02283 . 02510 . 7554/eLife . 02283 . 026Figure 9—figure supplement 2 . Surface representation of the alkali-free inward-facing GltPhin transport domain in this protomer after refinement . A dashed circle indicates a small opening between HP1 and HP2 . DOI: http://dx . doi . org/10 . 7554/eLife . 02283 . 026 Our apo and ions-only bound structures reveal a remarkable structural plasticity of GltPh transport domain that is likely a conserved feature in the glutamate transporter family . In addition to the large trans-membrane rigid-body movements of the transport domain between outward- and inward-facing orientations , local conformational changes within the domain accompany binding and release of the transported substrate and ions ( Figure 10 ) . These local changes provide a structural explanation of how Na+ gradients are harnessed to drive concentrative substrate uptake , supporting two previously proposed coupling mechanisms ( Focke et al . , 2011; Reyes et al . , 2013 ) : allosterically coupled binding of the substrate and symported Na+ ions , and opening of HP2 upon Na+ binding , which impedes the inward trans-membrane movement of the Na+-only bound transport domain . 10 . 7554/eLife . 02283 . 027Figure 10 . Proposed transport cycle for GltPh and EAATs . Ion binding to the Na1 site of the outward-facing apo transport domain triggers isomerization into bound-like conformation , formation of the L-asp and Na2 binding sites and HP2 opening , impeding translocation of the domain . Closure of HP2 , coupled to L-asp and Na2 binding , allows translocation . After the release of the ligands into the cytoplasm by as yet an unknown gating mechanism , the domain is in a compact apo state , and returns to the extracellular side . Notably , binding of cations to the inward-facing state does not lead to a crystallographically observed gate opening that would impede translocation . However , Na+ affinity in this state is only ∼250 mM ( Reyes et al . , 2013 ) , and it will remain largely unbound when facing the cytoplasm . Hence , uncoupled Na+ transport should be limited . In EAATs , an open conformation of the gates might be more favored in the apo state , and K+ binding at the Ct site might be required to stabilize translocation-competent conformation of the apo transport domain . DOI: http://dx . doi . org/10 . 7554/eLife . 02283 . 027 The apo transport domains in the outward- and inward-facing states are essentially identical and as compact as when they are fully bound , consistent with previous spectroscopic experiments ( Focke et al . , 2011 ) . Therefore , the apo transport domain is likely able to transition readily between the cytoplasmic and extracellular orientations . Consistently , previous spectroscopic studies showed that the transport domains continuously sample the outward- and inward-facing positions with nearly equal probabilities either when bound to Na+ and L-asp or when free of the solutes ( Akyuz et al . , 2013; Erkens et al . , 2013; Georgieva et al . , 2013; Hanelt et al . , 2013 ) . Moreover , these transitions are more frequent in the apo transporter , consistent with a lack of large energetic barriers ( Akyuz et al . , 2013 ) . In GltPh , the compact translocation-competent apo conformation of the transport domain is stabilized by interactions between the collapsed HP2 , and HP1 , TM7 , and TM8 . In EAATs , by contrast , we speculate that these interactions are insufficient and that K+ binding to the Ct site is required to stabilize the translocation-competent closed conformation that can return to the outside , ensuring coupling between substrate uptake and counter-transport of K+ ion . Local structural differences in EAATs in the vicinity of the Ct site may underlie the higher affinity and specificity of this site for K+ ion . In conclusion , we have shown structurally that ion binding and unbinding events in GltPh and , by analogy , in EAATs control the conformational state of the transporter , determining its competence to bind substrate and undergo transitions between the outward- and inward-facing states . Studies establishing the location of the Na3 binding site; the potential role of the Ct site in binding K+; and the gating mechanism in the inward-facing state will be necessary to verify and refine our proposed mechanisms . R397A mutation was introduced by PCR into GltPh containing seven point mutations to histidine ( Yernool et al . , 2004 ) , referred as wild type GltPh for brevity . Proteins were produced in Escherichia coli DH10b strain ( Invitrogen , Inc . , Grand Island , NY ) as fusions with a thrombin cleavage site , and a metal-affinity octa-histidine at their carboxyl-terminus . Proteins were purified by nickel-affinity chromatography , digested with thrombin to remove the affinity tag , and purified by size exclusion chromatography ( SEC ) in appropriate buffers as described below . Protein concentrations were determined by measuring the absorbance at 280 nm using an extinction coefficient of 26 , 820 M−1 . cm−1 . Diffraction data were collected at the National Synchrotron Light Source beamlines X25 and X29 ( Brookhaven National Laboratory ) . Data from crystals soaked in Tl+ were collected at a wavelength of 0 . 97 Å . Data were processed using HKL2000 ( Otwinowski and Minor , 1997 ) , and further analyzed using the CCP4 program suite ( Collaborative Computational Project , 1994 ) . Anisotropy correction was performed as described previously ( Strong et al . , 2006 ) . Briefly , resolution limits along the a , b , and c axes were determined using the UCLA–MBI Diffraction Anisotropy server ( http://services . mbi . ucla . edu/anisoscale/ ) and applied as cutoffs to truncate the dataset obtained after processing of diffraction images . After scaling in HKL2000 , structure factors were anisotropically scaled using PHASER ( McCoy et al . , 2007 ) , and a negative B factor correction was applied to these structure factors using CAD . Initial phases were determined by molecular replacement with PHASER ( McCoy et al . , 2007 ) , using the structure of GltPh either in the outward-facing state ( PDB code 2NWX ) or the inward-facing state ( PDB code 3KBC ) as the search model . Refinement was carried out by rounds of manual model building in COOT ( Emsley and Cowtan , 2004 ) and refinement in REFMAC5 with TLS ( Winn et al . , 2001 ) . With the exception of the analysis of the data from alkali-free GltPhin crystals , where protomers in the trimer were clearly not identical , the electron density maps and the anomalous difference Fourier maps were three or sixfold averaged in real space . Strict non-crystallographic symmetry constrains were also applied during structural refinement in REFMAC5 when necessary . Structures of the transport domain were superimposed and R . M . S . D . s calculated using VMD software ( Humphrey et al . , 1996 ) . All structural figures were prepared using Pymol ( DeLano Scientific , LLC ) ( DeLano , 2008 ) . ITC experiments were performed as described previously ( Reyes et al . , 2013 ) . Briefly , GltPh mutant proteins were purified by SEC in 10 mM HEPES/Tris , pH 7 . 4 , 200 mM choline chloride , 0 . 5 mM n-dodecyl-β-D-maltopyranoside and concentrated to 4 mg/ml . The protein was diluted to 40 μM in buffer containing 20 mM HEPES/Tris , pH 7 . 4 , 200 mM choline chloride , 1 mM n-dodecyl-β-D-maltopyranoside and various NaCl concentrations . ITC experiments were performed using a small cell NANO ITC ( TA instruments , Inc . , New Castle , DE ) at 25°C . Protein samples were placed into the instrument cell and titrated with L-asp solution prepared in the same buffer . The isotherms were analyzed using the NanoAnalyze software ( TA instruments , Inc . , New Castle , DE ) , and fitted to independent binding sites model . Fluorescence-based binding assays were performed as described previously ( Reyes et al . , 2013 ) . In brief , 100 μg/ml of protein in 20 mM HEPES/Tris , pH 7 . 4 , 200 mM choline chloride , 0 . 4 mM n-dodecyl-β-D-maltopyranoside , 200 nM styryl fluorescent dye RH421 ( Invitrogen , Inc . , Grand Island , NY ) were titrated with L-asp in the presence of various concentrations of NaCl at 25°C . Fluorescence experiments were carried out using a QuantaMaster ( Photon International Technology , Inc . , Edison , NJ ) . The RH421 dye was excited at 532 nm , and the fluorescence was collected at 628 nm . Fluorescence emissions were measured after at least 1000 s equilibration . The data were analyzed using SigmaPlot12 ( Systat software , Inc . , San Jose , CA ) . Fractional fluorescence changes were corrected and normalized with respect to the dilution factors and maximal fluorescence changes , respectively . Corrected fluorescence changes were plotted as a function of ligand concentration and fitted to the Hill equation . Sodium activity was calculated as γ × [Na+] , where γ is the activity coefficient . The activity coefficient is calculated with the Debye-Hückel equation as described ( Reyes et al . , 2013 ) . All the experiments were performed at least in triplicate . Sodium:dicarboxylate symporter family sequences were harvested from PFAM database ( PF00375 ) ( Finn et al . , 2008 ) , parsed to remove incomplete sequences and sequences with over 70% identity and aligned in ClustalW ( Larkin et al . , 2007 ) . The alignment was manually adjusted and the final dataset containing 463 aligned sequences was used to generate a consensus sequence using WebLogo ( Crooks et al . , 2004 ) . To model the structures of the transport domains with HP2 in the bound conformation and the NMD motif in the apo conformation , and vice versa , we superimposed the structures of the fully bound and apo forms of the transport domains using TM6 , HP1 , and TM7 . We then generated new coordinates files combining the coordinates of TM7 , including the NMD motif , from the bound form and the coordinates for HP2 from the apo form or vice versa . In both of these models , we observed steric clashes between Met311 and residues in HP2 . To construct a model of the intermediate state with an open tip of HP2 , we superimposed the structure of Na+-only bound GltPh-R397A and the intermediate state ( PDB accession code 3V8G ) using TM6 , HP1 , and TM7 . We then replaced HP2 in the structure of the intermediate with HP2 from Na+-only bound GltPh-R397A . We moved slightly the side chain of Lys55 that was involved in a minor steric clash with the HP2 tip . We observed no major steric clashes in the resulting model .	Molecules of glutamate can carry messages between cells in the brain , and these signals are essential for thought and memory . Glutamate molecules can also act as signals to build new connections between brain cells and to prune away unnecessary ones . However , too much glutamate outside of the cells kills the brain tissue and can lead to devastating brain diseases . In a healthy brain , special pumps called glutamate transporters move these molecules back into the brain cells , where they can be stored safely . However , when brain cells are damaged—by , for example , a stroke or an injury , —the glutamate stored inside spills out , killing the surrounding cells . This leads to a cascade of dying cells and leaking glutamate , which causes even more damage and slows the recovery . Glutamate transporters ensure that there are more glutamate molecules inside cells than outside . However , it requires energy to maintain this gradient in the concentration of glutamate molecules . The transporters get this energy by moving three sodium ions into the cell with each glutamate molecule , and moving one potassium ion out of the cell . However , it is not clear how these transporters ensure that they move the glutamate molecules and the sodium ions at the same time . Now , Verdon , Oh et al . have uncovered the 3D structure of a glutamate transporter homologue at each step of the transport process . These structures reveal that , on the outside of the cell membrane , sodium ions attach to the so-called ‘transporter domain’ and make it better able to bind glutamate . The transporter domain then carries the sodium ions and glutamate through the cell membrane and releases them into the cell . Verdon , Oh et al . suggest that a potassium ion then binds to the empty transport domain , stabilizing it into a more compact shape that easily makes the return trip to the outside of the cell . Most experiments on glutamate transporters , including the work of Verdon , Oh et al . , are carried out on model proteins taken from bacteria . An important challenge for the future will be to obtain structural information on human glutamate transporters , as these could be therapeutic targets for the treatment of various neurological conditions .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics", "neuroscience" ]	2014	Coupled ion binding and structural transitions along the transport cycle of glutamate transporters
TCF/LEF factors are ancient context-dependent enhancer-binding proteins that are activated by β-catenin following Wnt signaling . They control embryonic development and adult stem cell compartments , and their dysregulation often causes cancer . β-catenin-dependent transcription relies on the NPF motif of Pygo proteins . Here , we use a proteomics approach to discover the Chip/LDB-SSDP ( ChiLS ) complex as the ligand specifically binding to NPF . ChiLS also recognizes NPF motifs in other nuclear factors including Runt/RUNX2 and Drosophila ARID1 , and binds to Groucho/TLE . Studies of Wnt-responsive dTCF enhancers in the Drosophila embryonic midgut indicate how these factors interact to form the Wnt enhanceosome , primed for Wnt responses by Pygo . Together with previous evidence , our study indicates that ChiLS confers context-dependence on TCF/LEF by integrating multiple inputs from lineage and signal-responsive factors , including enhanceosome switch-off by Notch . Its pivotal function in embryos and stem cells explain why its integrity is crucial in the avoidance of cancer . TCF/LEF factors ( TCFs ) were discovered as context-dependent architectural factors without intrinsic transactivation potential that bind to the T cell receptor α ( TCRα ) enhancer via their high mobility group ( HMG ) domain ( Waterman and Jones , 1990; Giese et al . , 1992 ) . They facilitate complex assemblies with other nearby enhancer-binding proteins , including the signal-responsive CRE-binding factor ( CREB ) and the lineage-specific RUNX1 ( also called Acute Myeloid Leukemia 1 , AML1 ) . Their activity further depends on β-catenin , a transcriptional co-factor activated by Wnt signaling , an ancient signaling pathway that controls animal development and stem cell compartments , and whose dysregulation often causes cancer ( Clevers , 2006 ) . The context-dependence of TCFs is also apparent in other systems , for example in the embryonic midgut of Drosophila where dTCF integrates multiple signaling inputs with lineage-specific cues during endoderm induction ( Riese et al . , 1997 ) . The molecular basis for this context-dependence remains unexplained . In the absence of signaling , T cell factors ( TCFs ) are bound by the Groucho/Transducin-like Enhancer-of-split ( Groucho/TLE ) proteins , a family of co-repressors that silence TCF enhancers by recruiting histone deacetylases ( HDACs ) ( Turki-Judeh and Courey , 2012 ) and by ‘blanketing’ them with inactive chromatin ( Sekiya and Zaret , 2007 ) . TLEs are displaced from TCFs by β-catenin following Wnt signaling , however this is not achieved by competitive binding ( Chodaparambil et al . , 2014 ) but depends on other factors . One of these is Pygopus ( Pygo ) , a conserved nuclear Wnt signaling factor that recruits Armadillo ( Drosophila β-catenin ) via the Legless/BCL9 adaptor to promote TCF-dependent transcription ( Kramps et al . , 2002; Parker et al . , 2002; Thompson et al . , 2002 ) . Intriguingly , Pygo is largely dispensable in the absence of Groucho ( Mieszczanek et al . , 2008 ) , which implicates this protein in alleviating Groucho-dependent repression of Wg targets . Pygo has a C-terminal plant homology domain ( PHD ) and an N-terminal asparagine proline phenylalanine ( NPF ) motif , each essential for development and tissue patterning ( Mosimann et al . , 2009 ) . Much is known about the PHD finger , which binds to Legless/BCL9 ( Kramps et al . , 2002 ) and to histone H3 tail methylated at lysine 4 via opposite surfaces ( Fiedler et al . , 2008; Miller et al . , 2013 ) that are connected by allosteric communication ( Miller et al . , 2010 ) . By contrast , the NPF ligand is unknown , but two contrasting models have been proposed for its function ( Figure 1 ) . 10 . 7554/eLife . 09073 . 003Figure 1 . Two models of Pygo function . ( A ) The co-activator model ( Kramps et al . , 2002; Hoffmans et al . , 2005 ) : the NPF ligand ( X , orange ) is a transcriptional co-activator recruited to dTCF enhancers exclusively during Wnt signaling through the Pygo-Legless/BCL9 adaptor chain ( Stadeli and Basler , 2005 ) , co-operating with other transcriptional co-activators recruited to the C-terminus of Armadillo ( such as chromatin remodelers and modifiers , black ) in stimulating Wg-induced transcription . ( B ) The Armadillo-loading model ( Townsley et al . , 2004 ) : the NPF ligand ( X , orange ) mediates constitutive tethering of Pygo to dTCF enhancers prior to Wg signaling , jointly with PHD-mediated recognition of H3K4me1 ( marking poised enhancers; Kharchenko et al . , 2011 ) or H3R2me2aK4me1 ( marking silenced enhancers in the process of being activated; Kirmizis et al . , 2007 ) , priming these enhancers for Wg responses via its ability to capture Armadillo ( once available during Wg signaling , indicated by grey ) through the Legless/BCL9 adaptor . In both models , the homology domain 1 ( HD1 ) of Lgs/BCL9 binds to the Pygo PHD finger , while HD2 binds to the N-terminus of the Armadillo Repeat Domain ( light grey ) of Armadillo/β-catenin ( Kramps et al . , 2002; Sampietro et al . , 2006; Fiedler et al . , 2008; Miller et al . , 2013 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09073 . 003 Here , we use a proteomics approach to discover that the NPF ligand is an ancient protein complex composed of Chip/LDB ( ( Lin-11 Isl-1 Mec-3- ) LIM-domain-binding protein ) and single-stranded DNA-binding protein ( SSDP ) , also called SSBP . This complex controls remote Wnt- and Notch-responsive enhancers of homeobox genes in flies ( Bronstein and Segal , 2011 ) , and remote enhancers of globin and other erythroid genes in mammals , integrating lineage-specific inputs from LIM-homeobox ( LHX ) proteins and other enhancer-binding proteins ( Love et al . , 2014 ) . Using nuclear magnetic resonance ( NMR ) spectroscopy , we demonstrate that Chip/LDB-SSDP ( ChiLS ) binds directly and specifically to Pygo NPFs , and also to NPF motifs in Runt-related transcription factors ( RUNX ) proteins and Osa ( Drosophila ARID1 ) , whose relevance is shown by functional analysis of Drosophila midgut enhancers . Furthermore , we identify Groucho as another new ligand of ChiLS by mass spectroscopy . We thus define the core components of a Wnt enhanceosome assembled at TCF enhancers via Groucho/TLE and RUNX , primed for timely Wnt responses by ChiLS-associated Pygo . The pivotal role of ChiLS in integrating the Wnt enhanceosome provides a molecular explanation for the context-dependence of TCFs . To identify the NPF ligand of Drosophila Pygo , we inserted various tags into its low-complexity linker that separates NPF from PHD ( Figure 2—figure supplement 1A ) , and used stably transfected S2 cell lines expressing wild-type ( wt ) or NPF-mutant versions , for tandem-affinity purification of associated proteins and identification by mass spectrometry . We thus discovered Chip , SSDP and three LIM domain proteins—Beadex and CG5708 ( both LIM-only proteins , LMO ) and Apterous ( an LHX ) —amongst the top hits specifically associated with wt but not mutant Pygo ( Figure 2A ) . These proteins are known to form a complex: Chip dimerizes through DD ( dimerization domain ) and binds to SSDP through LDB/Chip conserved domain ( LCCD ) and to LIM domains through LIM-interacting domain ( LID ) . The latter allow ChiLS to associate with enhancers , either directly through LHX ( e . g . , Apterous ) , or indirectly through LMO adaptors that bind to bHLH ( e . g . , Achaete/Scute ) and GATA factors ( e . g . , Pannier ) ( Bronstein and Segal , 2011; Love et al . , 2014 ) . Indeed , LMOs displace LHXs from ChiLS by virtue of their high expression level and/or high affinity for LID ( Milan and Cohen , 1999; Ramain et al . , 2000; Matthews et al . , 2008 ) , and are thus capable of switching from LHX to GATA/bHLH . We also found ChiLS components associated with Pygo2 in stably transfected HEK293T cell lines , and with recombinant triple-NPF baits in lysates from mouse brains and colorectal cancer cell lines ( Figure 2—figure supplement 1B–D ) . 10 . 7554/eLife . 09073 . 004Figure 2 . Pygo NPF binds to ChiLS . ( A ) Top proteins associated with wt but not NPF-mutant Pygo in S2 cells ( unweighted spectral counts >95% probability are given ) ; bold , ChiLS and its ligands . ( B–D ) Western blots of coIPs from transfected HEK293T cells , showing NPF-dependent coIP of ( B , C ) endogenous LDB1 with HA-Pygo2 or ( D ) wt vs truncated LDB1-Flag with HA-Pygo2 ( left ) , and LDB1-Flag +/− SSDP-Flag with HA-Pygo2 ( right; increasing amounts of LDB1-Flag indicated above panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09073 . 00410 . 7554/eLife . 09073 . 005Figure 2—figure supplement 1 . Baits used for mass spectrometry analysis . ( A ) Cartoon of tandem-tagged Drosophila Pygo bait , generated for expression in stably transfected S2 cells ( continuously selecting with 5 μg ml−1 puromycin ) , and subsequent tandem purification ( TAP ) of associated proteins using α-Flag followed by Strep-tactin pull-down; red arrowheads indicate wt and mutant Pygo baits after electrophoresis on a stained polyacrylamide gel prior to excision for mass spectrometry analysis . ( B ) Cartoon of tandem-tagged human Pygo2 bait ( subcloned in V51 pGLUE ) , for expression in stably transfected HEK293T cells ( continuously selecting with 2 μg ml−1 puromycin ) , for TAP purification of associated proteins with streptavidin followed by Calmodulin resin , essentially adopting a protocol previously described ( Angers et al . , 2006 ) . ( C ) Cartoon of triple-NPF bait from mouse Pygo1 , generated by trimerizing an extended NPF motif ( amino acids 49-77 ) , or a corresponding NPF > AAA triple mutant , for Ni-NTA pull-down experiments with lysates from fresh mouse brains ( strain C57Bl6J ) , or from SW480 or COLO320 colorectal cancer cell ( CCC ) lines ( obtained from Marc de la Roche; de la Roche et al . , 2014 ) . Interacting proteins were eluted with 6M urea , and samples were prepared for analysis using a filter-aided sample preparation method , essentially as described ( Wisniewski et al . , 2009 ) . ( D ) Hits identified with TAP-Pygo and TAP-Pygo2 baits in both S2 and HEK293T cells , respectively ( see A , B ) , and overlap with hits identified by triple-NPF immunoprecipitations from mouse brain and CCC lysates ( see C ) , which include ChiLS components . Note that EHD paralogs were only found in the lysate-based approaches ( amongst the top hits , due to the cytoplasmic abundance of these proteins ) but not with full-length Pygo or Pygo2; spectrins were also commonly found , but the significance of this is unclear . DOI: http://dx . doi . org/10 . 7554/eLife . 09073 . 005 Co-immunoprecipitation ( coIP ) assays in transfected HEK293T cells revealed that only Chip but none of the other hits coIPed with Pygo . Furthermore , wt but not mutant HA-Pygo2 coIPs with endogenous LDB1 , and vice versa ( Figure 2B , C ) . The conserved proline cluster upstream of NPF is also required for binding , consistent with transcription assays in S2 cells that indicated the function of this cluster ( Stadeli and Basler , 2005 ) . Testing truncations of LDB1 for coIP with Pygo2 , we found that LID is dispensable for binding , whereas dimerization seems important since there is little interaction with Pygo2 in the absence of DD ( Figure 2D ) . Importantly , LDB1 coIPs with Pygo2 in a dose-dependent way upon co-expression , whereas SSDP hardly coIPs with Pygo2 in the absence of exogenous LDB1 , even at high SSDP excess ( Figure 2D ) . The reverse could not be established since LDB1 overexpressed on its own is highly unstable ( Figure 2D ) , being targeted for proteasomal degradation in the absence of SSDP ( Xu et al . , 2007b ) . This reinforces the notion that Pygo NPF binds to Chip/LDB . To test direct binding of NPF to ChiLS , we purified the NPF-interacting DD-LCCD fragments of Chip and LDB1 after bacterial expression . Both have a strong tendency to aggregate if expressed on their own , but become soluble if co-expressed with SSDP1-92 ( i . e . , the LisH domain-containing N-terminus of the fly protein , without its unstructured tail; this LisH domain is nearly identical to its human counterpart , with two residues only semi-conserved ) . Expressed by itself , SSDP1-92 is soluble and elutes as a single peak after gel filtration , regardless of concentration ( Figure 3A ) . Size exclusion chromatography coupled to multi-angle light scattering ( SEC-MALS ) revealed an average apparent molecular mass of 43 kDa for the protein in this peak , which corresponds to a tetramer ( expected molecular mass 42 . 7 kDa ) . Evidently , SSDP1-92 forms a stable tetramer in solution . We note that LisH domains typically form dimers , but some tetramerize ( e . g . , that of TBL1 , a subunit of a HDAC-recruiting co-repressor; Oberoi et al . , 2011 ) . 10 . 7554/eLife . 09073 . 006Figure 3 . Stoichiometry of the ChiLS complex . SEC-MALS of ( A ) Lip-SSDP1-92 , or co-expressed ( B ) Lip-SSDP1-92 + MBP-Chip205-436 or ( C ) Lip-SSDP1-92 + MBP-LDB156-285; solid black lines , elution profile as detected by the RI detector; grey circles , molecular mass; cartoons in panels indicate stoichiometries consistent with the measured molecular masses . DOI: http://dx . doi . org/10 . 7554/eLife . 09073 . 00610 . 7554/eLife . 09073 . 007Figure 3—figure supplement 1 . Gel filtration of ChiLS . ( A ) Bi-cistronic expression vector ( with T7 promoter ) used for bacterial expression of ChiLS complex , encoding 6xHis-MBP-Chip205-436 or 6xHis-MBP-LDB156-285 followed by 6xHis-Lipoyl-SSDP1-92 with its own ribosome-binding site ( RBS ) . ( B ) Elution profile of MBP-Chip205-436 , Lipoyl-SSDP1-92 ( after co-expression with vector shown in A ) and incubation with excess 6xHis-Lipoyl-Pygo67-107 ( Pygo Nterm ) , revealing three main species , as indicated above graph , corresponding to free Pygo Nterm , Chip-SSDPlow and Chip-SSDPhigh complexes , as shown by polyacrylamide gel electrophoresis ( inset ) ; note that the high-molecular weight shoulder ( Chip-SSDPhigh ) corresponds to a dimer of the major ∼190 kDa low-molecular weight species ( Chip-SSDPlow ) , whereby the latter corresponds to an MBP-Chip dimer bound to an SSDP tetramer ( see Figure 3B ) . Chip-SSDPlow and Chip-SSDPhigh complexes cannot be further separated by subsequent anion exchange chromatography , and do not seem to interconvert , as judged by iterative G-200 gel chromatography . Both complexes bind to 15N-Pygo-NPF ( see Figure 4 ) , however , Pygo Nterm dissociates from ChiLS during gel filtration ( see inset ) , indicating a low affinity to ChiLS ( estimated to be mid-μM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09073 . 007 On co-expression , Chip205-436 and SSDP1-92 form a stable complex that elutes consistently as one main peak with a higher-mass shoulder after gel filtration ( Figure 3—figure supplement 1 ) , regardless of concentration . The peaks of these two species are not baseline-resolved during SEC-MALS , but their average masses of ∼190 kDa and ∼380 kDa suggest that the shoulder is a dimer of the major ∼190 kDa species ( Figure 3B ) . The latter thus corresponds to an MBP-Chip dimer bound to an SSDP tetramer ( expected 181 kDa ) , the only possible stoichiometry that fits the observed molecular mass . The same architecture is also found for the human LDB1-SSDP complex ( Figure 3C ) , revising previous models of ChiLS ( e . g . , Bronstein and Segal , 2011 ) . To monitor direct NPF-dependent binding , we incubated purified Chip-SSDP with wt and mutant 15N-labeled Lipoyl-tagged ( Lip ) Pygo67-107 ( 15N-Nterm ) and recorded heteronuclear single-quantum correlation ( HSQC ) spectra by NMR . We thus observed clear line broadenings with wt ( Figure 4A ) but not with F99A mutant 15N-Nterm ( bearing a point mutation in NPF; Figure 4B ) . Incubation of wt 15N-Nterm with purified SSDP1-92 did not produce any spectral changes ( Figure 4C ) , confirming that SSDP does not bind NPF ( Figure 2D ) . We were unable to test binding to Chip205-436 alone , owing to its aggregation . LDB1-SSDP also binds to wt but not F78A mutant Pygo2 ( Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 09073 . 008Figure 4 . Direct NPF-dependent binding of Pygo by ChiLS . Overlays of HSQC spectra of 50 μM 15N-labeled wt or F99A mutant Pygo67-107 alone ( red ) or probed with ( A , B ) MBP-Chip205-436- Lip-SSDP1-92 or ( C ) Lip-SSDP1-92 alone ( blue ) ; interacting residues are labeled , with NPF in red ( binding to P is not detectable by HSQCs ) . The HSQC obtained with 50 μM of minimal 15N-labeled Pygo87-102 is indistinguishable from that shown in ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09073 . 00810 . 7554/eLife . 09073 . 009Figure 4—figure supplement 1 . Direct NPF-dependent binding of Pygo2 by ChiLS . Overlays of BEST-TROSY spectra of 50 μM 15N-labeled ( A , C ) wt or ( B ) F78A mutant human Pygo237-93 alone ( red ) or probed with 100 μM ( A , B ) MBP-LDB156-285- Lipoyl-SSDP1-92 ( blue ) or ( C ) MBP-LDB220-246- Lipoyl-SSDP1-92 ( blue ) ; interacting residues are labeled , with NPF motif in red ( binding to P is not detectable by BEST-TROSY ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09073 . 009 Assignments of double-labeled protein allowed us to determine that the ChiLS ‘interaction footprints’ span 10 NPF-spanning residues in Pygo , and 25 residues in Pygo2 . A minimal 16-mer without the proline cluster produces a comparable interaction ( Figure 5A ) , showing that this cluster is dispensable in this binding assay . Its requirement in cell-based assays may reflect the need for a rigid spacer between NPF and the upstream nuclear localization signal . Notably , the residues immediately flanking the NPF are conserved ( Figure 5A ) and may contribute to binding ( see ‘Discussion’ ) . 10 . 7554/eLife . 09073 . 010Figure 5 . ChiLS binds to RUNX NPFs . ( A ) Summary of NMR binding assays of 15N-labeled NPF fragments probed with ChiLS; +–+++ , estimates of binding affinities , based on minimal ChiLS concentrations required for line-broadening; − , no binding ( see also Figure 4 ) ; top , preferred NPF context in strong binders ( numbering of positions as in de Beer et al . , 2000 ) . ( B ) Schematic of RUNX orthologs , with DNA-binding domain ( RD , black ) , region IIIC , activation and inhibitory domains ( AD , ID ) , NPF ( or GPF ) and WRPY indicated . ( C–E ) Western blots as in Figure 2 , showing coIP between co-expressed proteins as indicated above panels; for Runt-N and Runt-C , see ( B ) ; mGFP ( control ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09073 . 01010 . 7554/eLife . 09073 . 011Figure 5—figure supplement 1 . Specific recognition of RUNX NPFs by ChiLS . ( A , B ) Overlays of HSQC spectra of 50 μM 15N-labeled 6xHis-Lipoyl-tagged ( A ) wt or ( B ) NPF > AAA triple-mutant Runt-NPF228-257 alone ( red ) or probed with 300 μM MBP-Chip205-436-Lipoyl-SSDP1-92 ( blue ) , as indicated in panels; interacting residues are labeled in ( A ) . ( C , D ) Overlays of HSQC spectra of 50 μM 15N-labeled 6xHis-Lipoyl-tagged ( C ) wt or ( D ) F > A mutant RUNX2-NPF350-378 alone ( red ) or probed with 300 μM MBP-LDB156-285- Lipoyl-SSDP1-92 ( blue ) , as indicated in panels . Note that a triple-NPF mutation ( NPF > AAA ) is required to block binding of Runt to ChiLS ( B ) while a single F > A mutant Runt still binds to ChiLS ( at this concentration ) . Sequences of 15N-labeled NPF peptides are given in the panels ( red , NPF motifs; bold , mutated residues ) . RUNX3 binding was undetectable with NPF-containing peptides from human RUNX3 ( amino acids 227–249; SQPQTPIQGTSELNPFSDPRQFD ) or mouse Runx3 ( amino acids 214-269 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09073 . 01110 . 7554/eLife . 09073 . 012Figure 5—figure supplement 2 . Specific recognition of Osa NPF by ChiLS . Overlays of ( A ) HSQC or ( B ) BEST-TROSY spectra of 50 μM 15N-labeled 6xHis-Lipoyl-tagged ( A ) wt or ( B ) F1215A mutant Osa1195-1221 alone ( red ) , or probed with 300 μM MBP-Chip205-436- Lipoyl-SSDP1-92 ( blue ) , as indicated in panels; interacting residues are labeled in ( A ) . Inset in ( A ) shows purified 6xHis-Lipoyl-Osa1195-1220 and MBP-Chip205-436- Lipoyl-SSDP1-92 complex separated by polyacrylamide gel electrophoresis . Note also that the NPF-containing peptide we tested for ARID1B was negative , however , this large protein contains further matches to the NPF consensus sequence that might bind ChiLS ( to be tested in future studies ) . The same applies for its ARID1A paralog . DOI: http://dx . doi . org/10 . 7554/eLife . 09073 . 01210 . 7554/eLife . 09073 . 013Figure 5—figure supplement 3 . Direct binding of RUNX RD by ChiLS . Overlays of BEST-TROSY spectra of 15N-labeled 6xHis-Lipoyl ( blue ) , 6xHis-Lipoyl-RUNX2-RD102-234 ( black ) and 6xHis-Lipoyl-RUNX2-RD102-234 + 300 μM MBP-LDB156-285- Lipoyl-SSDP1-92 ( red ) , as indicated in panels . Note that the resonances affected by incubation with LDB1-SSDP are predominantly RD-specific ( black ) whereas most Lip residues ( marked by blue ) are unaffected . DOI: http://dx . doi . org/10 . 7554/eLife . 09073 . 013 To identify other NPF-containing proteins as putative ChiLS ligands , we conducted genome-wide database searches for matches to NPFDD-like motifs . We thus found NPFs in several enhancer-binding proteins ( Figure 5A ) , notably in RUNX and Osa/ARID1 ( the DNA-binding subunit of SWI/SNF chromatin remodeling complexes; Wu and Roberts , 2013 ) , and also in MACC1 ( metastasis-associated in colon cancer 1 ) whose molecular function is unknown ( Stein et al . , 2009 ) . NPF-containing fragments from MACC1 , RUNX2 , Runt and Lozenge ( fly RUNX proteins ) and Osa tested positive in NMR binding assays , while those from ARID1B and RUNX3 were negative—somewhat curious in the latter case , given that its NPF motif resembles that of Runt ( Figure 5A ) . Testing of NPF mutants confirmed NPF-specific binding ( Figure 5—figure supplements 1 , 2 ) . Note that ARID1B ( and its paralog ARID1A ) are large proteins with multiple putative NPFs , whose binding remains to be tested . Given the linkage between RUNX and TCF in the TCRα enhancer , we decided to pursue this interaction further . In Runt , the NPF is within a short conserved sequence block abutting its Runt domain ( the DNA-binding domain; Figure 5B ) , called region IIIC which is conserved in Runt orthologs of other invertebrates and functionally relevant in Drosophila ( Walrad et al . , 2010 ) . In vertebrate RUNX2/3 , the NPFs are further downstream , at the start of a conserved sequence block ( Figure 5B ) . Interestingly , AML1 paralogs do not exhibit an NPF motif at this position , although we note a near-invariant GPFQT/A motif further downstream in all three RUNX paralogs , which could potentially also bind to ChiLS . CoIP assays revealed that full-length Runt coIPs efficiently with ChiLS ( Figure 5C ) , more so than Pygo ( Figure 5D ) , pointing to a second ( stronger ) interaction between Runt and ChiLS . Indeed , the N-terminal half of Runt ( without NPF ) coIPs with ChiLS ( and vice versa ) , as does an NPF-mutant Runt , while the C-terminal half of Runt does not ( Figure 5C ) . Evidently , the affinity of the Runt NPF for ChiLS ( estimated to be high μM; Figure 5—figure supplement 1 ) is too low for Runt to remain associated with ChiLS during coIP . Rather , this interaction depends on the Runt domain to which ChiLS binds directly , as can be shown by NMR ( Figure 5—figure supplement 3 ) . Efficient coIP with ChiLS was also observed for AML1 , Runx2 and Runx3 ( Figure 5E ) , consistent with the previously reported association of Runx1 with Ldb1 in differentiated mouse erythroleukemic cells ( Meier et al . , 2006 ) . RUNX are context-dependent enhancer-binding proteins that control transcription of master-regulatory genes in Drosophila and mammals , co-operating with TCFs ( e . g . , in the TCRα enhancer; Figure 6A ) , but also with other signaling inputs including TGF-β/SMAD and Notch ( Canon and Banerjee , 2000; Chuang et al . , 2013 ) . Indeed , the midgut enhancers from the homeotic ( HOX ) genes Ultrabithorax ( Ubx ) and labial each contains two putative RUNX binding sites , linked to functional signal response elements such as dTCF binding sites and CREs ( Figure 6A ) . We exploited these enhancers , to test whether Runt directly controls Ubx and labial during endoderm induction ( Figure 6B ) . This seemed possible , given that runt is required for normal Ubx expression in the embryonic visceral mesoderm ( Tremml and Bienz , 1989 ) . 10 . 7554/eLife . 09073 . 014Figure 6 . Runt acts through the labial midgut enhancer . ( A ) Cartoon of UbxB , lab550 and TCRα enhancers , with the following binding sites ( responding to signals , as indicated in brackets ) : green , dTCF ( Wg ) ; purple , SMAD ( Dpp ) ; light-blue , Ets ( Ras ) ; blue , CRE ( Ras ) ; in UbxB , the SMAD binding site also mediates Wg-mediated repression by Brinker ( red ) , and CRE mediates Osa-mediated repression ( see text ) ; black , RUNX; orange , Labial; residue numbers between binding sites are given ( right , total length; note that UbxB and lab550 extend beyond these modules which however contain all known functional binding sites ) ; capital letters , matches to binding site consensus sequences ( C/G G C/G G G T C/G for RUNX; Melnikova et al . , 1993 ) . ( B ) Cartoon of endoderm induction , color-coded as in ( A ) ; V , midgut constrictions at parasegment boundaries . ( C , D ) 14 hr old embryos stained with α-Labial; arrow marks incipient second midgut constriction , lacking in runt mutants ( which only form first and third constrictions , arrowheads ) . ( E–K ) 12–14 hr old embryos bearing wt or mutant lab550 as indicated on the right , stained with α-β-galactosidase; high magnification views are imaged at different focal planes , to highlight Wg-dependent expression gradients ( Wg sources indicated by asterisks ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09073 . 01410 . 7554/eLife . 09073 . 015Figure 6—figure supplement 1 . Runt acts through the Ubx midgut enhancer . ( A–E ) 14–15 hr old wt or runt mutant embryos , bearing wt or mutant UbxB enhancers ( Thuringer et al . , 1993 ) , with or without mesodermal GAL4-mediated overexpression of Runt ( using 24B . GAL4 ) as indicated in panels , fixed and stained with α-β-galactosidase ( α-β-gal ) antibody; Wg sources ( in parasegment 8 of the visceral mesoderm; see Figure 6B ) are indicated by asterisks . The expression of the mutant enhancer ( UbxBR ) is reduced , and seen in fewer cells ( C ) , similarly to the expression of the wt enhancer in runt mutant embryos ( B ) , as in the case of lab550 ( Figure 6F , G , J ) , although in both cases , mutating the Runt binding sites has a somewhat stronger effect on expression than loss of Runt ( possibly because of a maternal contribution in the runt mutant embryos , which is difficult to rule out ) . Note also that this minimal enhancer mediates expression posteriorly to parasegment 7 ( the Ubx expression domain in the visceral mesoderm ) , partially escaping repression by high Wg signaling levels near the Wg source ( Yu et al . , 1998 ) , which depends on displacement of SMAD by the Groucho-recruiting Brinker co-repressor ( Saller and Bienz , 2001 ) . ( F , G ) ∼12 hr old wt or runt mutant embryos , fixed and stained with α-Wg antibody; Wg sources in the visceral mesoderm are indicated by arrows . DOI: http://dx . doi . org/10 . 7554/eLife . 09073 . 015 Staining runt mutant embryos with α-Labial antibody , we discovered that these mutants lack the middle midgut constriction , the signature phenotype found in all mutants with defective endoderm induction ( including wg and pygo; Bienz , 1994; Thompson et al . , 2002 ) . Furthermore , staining is significantly weaker than in the wt , and exhibits no gradient ( Figure 6C , D ) , mimicking the phenotype seen in wg mutants ( Bienz , 1994; Thompson et al . , 2002 ) ( although Wg is expressed in runt mutants; Figure 6—figure supplement 1 ) . Thus , runt is required for endoderm induction , likely for the Wg response of labial . Next , we mutated the two Runt binding sites in the labial enhancer ( lab550R ) , to test their function in transgenic lines with single copies of chromosomally integrated LacZ reporters . lab550 produces a gradient of LacZ staining ( Figure 6E ) recapitulating endogenous Labial expression ( Tremml and Bienz , 1992 ) ( Figure 6C ) . This LacZ staining is much reduced in runt mutants , with no discernible gradient ( Figure 6F ) , as in wg mutants . Similarly , lab550R mediates relatively weak and even LacZ expression , limited to a narrow band of cells ( Figure 6G ) , which phenocopies the LacZ pattern from an enhancer with mutant dTCF binding sites ( lab550L; Figure 6H ) . Importantly , lab550 is strongly activated posteriorly if Runt is expressed throughout the endoderm ( Figure 6I ) , while neither lab500R nor lab550L are Runt-responsive ( Figure 6J , K ) . Thus , Runt acts through lab550 , apparently cooperating with dTCF to render lab550 signal-responsive . Likewise , the Runt binding sites in the Ubx midgut enhancer are also functional targets of Runt ( Figure 6—figure supplement 1 ) , implicating Runt in the Wg-dependent indirect autoregulation of Ubx at the top of the inductive cascade ( Figure 6B ) . Recall that neither Pygo nor ChiLS bind to DNA , nor to TCFs , raising the question how these proteins are recruited to TCF enhancers . To answer this , we used the same approach as described ( Figure 2 ) , to identify new ChiLS-interacting factors by mass spectrometry . Using SSDP as bait , we found Chip as the top hit ( Figure 7A ) , confirming efficient complex formation between the two proteins ( van Meyel et al . , 2003; Xu et al . , 2007b ) , comparable to that in the other two samples with co-overexpressed Chip and SSDP . 10 . 7554/eLife . 09073 . 016Figure 7 . Groucho binds to ChiLS . ( A ) Top proteins associated with >2 baits ( as indicated in table ) in S2 cells , in addition to Legless and Pygo ( unweighted spectral counts as in Figure 2A ) . ( B–D ) Western blots as in Figure 5C , showing coIP between co-expressed wt and truncated proteins as indicated above panels . ( E ) Cartoon of Groucho and TLE3 , with domains indicated ( GP , CcN , SP , semi-conserved elements within linker ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09073 . 016 Looking for hits with similar spectral counts in all three coIPs provided a stringent criterion that eliminated most hits ( many of which were false positives as they scored highly in the Flag but not HA IPs ) . The resulting overlap list contains Beadex , Apterous and CG5708 as the top hits , followed by Groucho ( Figure 7A ) , the only new hit identified with high confidence . Additionally , two LisH domain-containing proteins ( Ebi and CG6617 ) were identified with lower confidence , as well as Pygo and Legless ( Figure 7A ) whose low spectral counts may simply reflect low abundance ( and S2 cells express neither Runt nor Lozenge ) . CoIPs confirmed the interaction between ChiLS and Groucho or TLE3 , and revealed that the WD40 domain is both necessary and sufficient for association , while the Q domain and linker are dispensable ( Figure 7B–E ) . Even though the minimal WD40 domain coIPs only weakly with ChiLS , the same is true for full-length Groucho , while Δlinker interacts very strongly ( Figure 7C ) , suggesting that the linker attenuates the interaction between Groucho/Tranducin-like enhancer protein ( TLE ) and ChiLS , consistent with the known regulatory role of this region ( controlled by phosphorylation; Turki-Judeh and Courey , 2012 ) . Groucho's binding to Chip neither requires its LID , nor SSDP ( M . F . , unpublished ) . The WD40 propeller domain binds two classes of motifs in enhancer-binding proteins ( WRPW/Y or en1 , FxIxxIL ) that ‘plug’ its pore ( Jennings et al . , 2006 ) , but there are no convincing matches to either of these in ChiLS . Its interaction with WD40 may thus be mediated by a degenerate eh1-like or unknown motif , as found in other bona fide Groucho-binding proteins ( Flores-Saaib et al . , 2001; Turki-Judeh and Courey , 2012 ) , or ChiLS may even bind to an alternative WD40 surface ( for a precedent , see He et al . , 2013 ) . Crucially , since Groucho/TLE binds to TCF via its Q domain ( Mieszczanek et al . , 2008; Chodaparambil et al . , 2014 ) , it can thus function as an adaptor between ChiLS and TCF . We note that the WD40 domain also binds to the WRPY motif at the C-terminus of RUNX proteins ( Canon and Banerjee , 2000; Chuang et al . , 2013 ) . chip is required for the function of remote enhancers of the homeobox genes cut ( a Notch and Wg target which patterns the wing margin ) , and of Ubx ( Morcillo et al . , 1996 ) which specifies the middle body region including the third leg and haltere ( the dorsal appendage that substitutes for the hind wing in flies ) . Furthermore , chip is required for Apterous-dependent wing development ( Milan and Cohen , 1999 ) , and for the notum bristles which are specified by Pannier and Achaete/Scute in a Wg-dependent fashion ( Garcia–Garcia et al . , 1999; Ramain et al . , 2000 ) . ssdp mutant clones produce wing defects that phenocopy pygo mutant clones ( van Meyel et al . , 1999 ) . Transcriptional profiling of ssdp mutant wing discs identified numerous negatively-regulated SSDP target genes linked to Apterous and Pannier binding sites ( Bronstein et al . , 2010 ) . Strikingly , the top scoring positively-regulated SSDP target genes were linked to dTCF binding sites ( although this is not mentioned in the text , but see Table S3 in Bronstein et al . , 2010 ) —a strong indication that SSDP promotes primarily dTCF-dependent transcription in this tissue . Clonal analysis in wing discs confirmed that chip and ssdp control dTCF targets including vestigial and wg , similarly to pygo ( Figure 8—figure supplement 1 ) . In haltere discs , ssdp and pygo are required for high levels of Ubx expression , causing similar overgrowth phenotypes in halteres ( Figure 8—figure supplement 2 ) . These similarities between the ssdp and pygo mutant defects in the primordia for the dorsal appendages implicate both genes in the control of Wg-dependent master-regulatory genes . chip has a pioneer-like role in the early embryo , when zygotic transcription starts at the maternal-zygotic transition , enabling expression of segmentation genes along the antero-posterior axis ( Morcillo et al . , 1997 ) . We confirmed that embryos without maternal and zygotic Chip ( chip germ-line clones , glcs ) do not develop , except for rare escapers which show severe expression defects of segmentation genes , including even-skipped ( eve ) ( Morcillo et al . , 1997 ) ( J . M . , unpublished ) . Eve is a master-regulatory homeobox gene expressed in stripes along the embryonic antero-posterior axis where it activates numerous genes including wg ( Kobayashi et al . , 2001 ) and Ubx ( Tremml and Bienz , 1989; Muller and Bienz , 1992 ) . Nothing is known about ssdp function in embryos . Antibody staining of ssdp glcs revealed only residual expression of Eve , Wg and Ubx ( Figure 8A–C ) . In contrast , twi-LacZ ( recapitulating endogenous twist , a mesoderm-specifying bHLH gene activated in the ventral-most zone; Thisse et al . , 1991 ) appears normal in ssdp glcs ( Figure 8D ) . Likewise , hb-LacZ ( recapitulating expression of endogenous hunchback , a gene specifying anterior development; Struhl et al . , 1992 ) is normal in these mutants ( Figure 8E ) . These ssdp defects phenocopy those of chip glcs ( Morcillo et al . , 1997 ) , and they indicate that ChiLS is required for gene activity along the antero-posterior but not the dorso-ventral axis . An important corollary is that Wg is never expressed in embryos lacking ChiLS . In support of this , the cuticles of the rare chip glc escapers show a lawn of ventral denticles ( Morcillo et al . , 1997 ) , as do ssdp glcs ( Figure 8F ) and pygo glcs ( Parker et al . , 2002; Thompson et al . , 2002 ) —the hallmark of Wg signaling failure ( Cadigan and Nusse , 1997 ) . 10 . 7554/eLife . 09073 . 017Figure 8 . ssdp and pygo mutants show similar early embryonic defects . ( A–E ) 3–5 hr old wt and mutant glc embryos as indicated above panels , stained with antibodies as indicated in panels . ( F ) Larval cuticles of wt and mutant glc embryos; denticle belts ( arrows ) signify lack of Wg signaling . DOI: http://dx . doi . org/10 . 7554/eLife . 09073 . 01710 . 7554/eLife . 09073 . 018Figure 8—figure supplement 1 . ChiLS is required for Wg and Notch responses in wing discs . ( A , B ) Single confocal sections through third larval instar wing discs , fixed and stained with α-Wg or α-Vg antibody , as indicated in panels . wg is activated by Notch signaling along the prospective anterior margin of the third instar disc , while vg is expressed in a Wg-dependent fashion in a broad zone straddling Wg expression along the prospective margin and sustains proliferation in the prospective wing blade ( Couso et al . , 1995; Rulifson and Blair , 1995; Kim et al . , 1996; Klein and Arias , 1999; Micchelli et al . , 1997 ) . ( C–E ) Single confocal sections of wing discs as in ( A , B ) , bearing ( C ) pygoS123 , ( D ) ssdpL7 or ( E ) chipe55 mutant clones , stained with α-Vestigial ( Vg ) ( red in merge ) ; clones are marked by absence of GFP ( green in merge ) ; all discs were counterstained with DAPI ( blue in merge ) , to label the nuclei ( as internal control for the focal plane ) . Reduction of Vg expression within mutant clones of all three genes is best detectable in the ventral compartment ( without Apterous expression ) at a distance from the Wg source , as shown in high-magnification insets below panels ( from marked squares , marked in merges; clones demarcated by white lines ) , as previously reported for dTCF mutant clones that are unable to respond to Wg signaling ( Schweizer et al . , 2003 ) . ( F–H ) Single confocal sections of wing discs as in ( C–E ) , bearing ( F ) ssdpL7 , ( G ) chipe55 or ( H ) pygoS123 mutant clones , stained with α-Wg ( red in merge ) ; clones are marked by absence of GFP ( green in merge ) , and discs were counterstained with DAPI; insets below panels are high-magnification views of selected clones ( marked by squares in merges ) revealing ectopic activation of Wg along the edge of the clone ( signifying activation by Notch signaling derived from adjacent wt cells , combined with loss of Wg-mediated repression within the clone; Rulifson and Blair , 1995; Rulifson et al . , 1996 ) . Note also that , in the dorsal disc ( the Apterous territory ) , chipe55 mutant clones barely survive ( indicated by asterisks in the merge in G ) , compared to ssdpL7 mutant clones that appear to grow normally , like pygoS123 mutant clones; however , in the absence of Apterous , i . e . in the ventral compartment , chipe55 mutant clones grow normally ( see clone marked by arrow in E ) . Collectively , our clonal analysis indicates that ChiLS is required for Wg and Notch responses in the prospective wing blade and margin , consistent with previous results ( Morcillo et al . , 1996; Morcillo et al . , 1997; van Meyel et al . , 2003 ) similarly to Pygo ( Belenkaya et al . , 2002; Parker et al . , 2002; Thompson et al . , 2002 ) , although Chip exhibits a far more stringent requirement for proliferation in the Apterous territory than either SSDP or Pygo . DOI: http://dx . doi . org/10 . 7554/eLife . 09073 . 01810 . 7554/eLife . 09073 . 019Figure 8—figure supplement 2 . ChiLS is required for Ubx and Wg expression in haltere discs . ( A , B ) Single confocal sections through third larval instar haltere discs , fixed and stained with α-Ubx or α-Wg antibody , as indicated in panels . Ubx specifies haltere development by attenuating vg and proliferation in haltere discs , partly through interaction with Wg ( Prasad et al . , 2003 ) . ( C , D ) Single confocal sections of haltere discs as in ( A , B ) , bearing pygoS123 or ssdpL7 mutant clones as indicated in panels , stained with α-Ubx antibody ( red in merge ) ; clones are marked by absence of GFP ( green in merge ) ; all discs were counterstained with DAPI ( blue in merge ) , to label the nuclei ( as internal control for the focal plane ) . Note the reduction of Ubx expression in clones marked by red arrows , which results in slight overproliferation and outgrowth of clone ( as judged by the bulging out from the focal plane of the mutant epithelium ) . ( E , F ) Single confocal sections of clone-bearing haltere discs as in ( C , D ) , stained with α-Wg antibody ( red in merge ) ; clones are marked by absence of GFP ( green in merge ) , and counterstained with DAPI . Note the ectopic expression of Wg along the edges within the mutant clones , similarly to the clones shown in Figure 8—figure supplement 1 ( F–H ) , likely as a result of the same Notch-stimulatory and Wg autoinhibitory and inputs . ( G–I ) Patterning defects in halteres , reflecting overgrowth of pygoS123 or ssdpL7 mutant clones due to reduced Ubx and ectopic Wg expression in the disc . DOI: http://dx . doi . org/10 . 7554/eLife . 09073 . 019 We also examined glcs bearing a pygo null allele ( Thompson et al . , 2002 ) , which revealed considerable defects in Eve and Ubx expression , and almost complete absence of Wg ( Figure 8A–C ) , as previously reported for a different null pygo allele ( Parker et al . , 2002 ) . These Wg-independent defects are not detectable with weaker pygo truncation alleles that merely lack the PHD finger ( Kramps et al . , 2002; Thompson et al . , 2002 ) , confirming that the ( ChiLS-dependent ) activation of segmentation genes along the antero-posterior axis resides in the NPF-containing N-terminus of Pygo . NPF is a versatile endocytosis motif that binds to structurally distinct domains ( Mahadev et al . , 2007 ) , including eps15 homology ( EH ) domains in epsin15 homology domain ( EHD ) proteins ( Kieken et al . , 2010 ) . Indeed , we consistently identified EHDs in lysate-based pull-downs with triple-NPF baits ( Figure 2—figure supplement 1D ) . EHDs are predominantly cytoplasmic , and do not interact with nuclear Pygo upon co-expression ( M . G . , unpublished ) , nor are any of the Drosophila EHDs required for Wg signaling in S2 cells ( Stadeli and Basler , 2005 ) . ChiLS is the first nuclear NPF-binding factor . NPF binding to ChiLS appears to depend on the same residues as NPF binding to EHD domains ( de Beer et al . , 2000 ) , that is , on the aromatic residue at +2 ( Figure 4 ) , the invariant P at +1 , N ( or G ) at 0 and NPF-adjacent residues ( Figure 5A ) , including negative charges at +3 and +4 ( whereby a positive charge at +3 abolishes binding to EHD; Kieken et al . , 2010 ) . Indeed , an intramolecular interaction between the +3 side-chain and that of N predisposes NPF to adopt a type 1 β-turn conformation , which increases its affinity to the EHD pocket , while the −1 residue undergoes an intermolecular interaction with this pocket ( de Beer et al . , 2000 ) . ChiLS also shows a preference for small residues at −1 and −2 , similarly to N-terminal EHDs ( Paoluzi et al . , 1998 ) although RUNX seems to differ at −1 and −2 from Pygo and MACC1 ( F/L A/E/D vs S A , respectively ) . Groucho/TLE is recruited to TCF via its Q domain , which tetramerizes ( Chen et al . , 1998; Chodaparambil et al . , 2014 ) . Intriguingly , the short segment that links two Q domain dimers into a tetramer ( Chodaparambil et al . , 2014 ) is deleted in a dTCF-specific groucho allele that abolishes dTCF binding and Wg responses ( Mieszczanek et al . , 2008 ) , suggesting that TCF may normally bind to a Groucho/TLE tetramer . Groucho/TLE uses its second domain , the WD40 propeller , to bind to other enhancer-binding proteins on Wnt-responsive enhancers ( Turki-Judeh and Courey , 2012 ) , most notably to the C-terminal WRPY motif of RUNX proteins ( common partners of TCFs in Wnt-responsive enhancers; Figure 6A ) . This interaction can occur simultaneously with the WD40-dependent binding to ChiLS ( Figure 7F ) , given the tetramer structure of Groucho/TLE . In turn , RUNX uses its DNA-binding Runt domain to interact with HMG domains of TCFs ( Kahler and Westendorf , 2003; Ito et al . , 2008 ) , and to recruit ChiLS ( Figure 5—figure supplement 3 ) . RUNX thus appears to be the keystone of the Wnt enhanceosome since it binds to the enhancer directly while undergoing simultaneous interactions with Groucho/TLE ( through its C-terminal WRPY motif ) , TCF and ChiLS ( though its Runt domain ) . In line with this , Runt has pioneering functions in the early Drosophila embryo , shortly after the onset of zygotic transcription ( Canon and Banerjee , 2000 ) , and in the naïve endoderm ( Figure 6 ) as soon as this germlayer is formed , in each case prior to the first Wg signaling events . RUNX paralogs also have pioneer-like functions in specifying cell lineages , that is , definitive hematopoiesis in flies and mammals ( Chuang et al . , 2013 ) . Our model predicts that ChiLS , once tethered to the enhanceosome core complex , recruits Pygo via NPF to prime the enhancer for Wnt responses ( Figure 9 ) . Given the dimer-tetramer architecture of ChiLS ( Figure 3C ) , its binding to Pygo can occur simultaneously to its NPF-dependent binding to RUNX . In turn , tethering Pygo to the Wnt enhanceosome may require Pygo's binding to methylated histone H3 tail ( Fiedler et al . , 2008; Miller et al . , 2013 ) , similarly to Groucho/TLE whose tethering to enhancers depends on binding to hypoacetylated histone H3 and H4 tails ( Sekiya and Zaret , 2007 ) . Interestingly , Pygo's histone binding requires at least one methyl group at K4 ( Fiedler et al . , 2008; Miller et al . , 2013 ) —the hallmark of poised enhancers ( Kharchenko et al . , 2011 ) . Indeed , Drosophila Pygo is highly unorthodox due to an architectural change in its histone-binding surface that allows it to recognize asymmetrically di-methylated arginine 2 ( Miller et al . , 2013 ) —a hallmark of silent chromatin ( Kirmizis et al . , 2007 ) . Thus , the rare unorthodox Pygo proteins ( Miller et al . , 2013 ) may recognize silent enhancers even earlier , long before their activation , consistent with the early embryonic function of Pygo , prior to Wg signaling ( Figure 8 ) . Overcoming the OFF state imposed on the enhancer by Groucho/TLE ( Mieszczanek et al . , 2008 ) involves Pygo-dependent capturing of β-catenin/Armadillo , which recruits various transcriptional co-activators to its C-terminus ( Mosimann et al . , 2009 ) . Although these include CREB-binding protein ( CBP ) , a histone acetyl transferase , its tethering to TCF enhancers is likely to co-depend on CRE-binding factors ( CREB , c-Fos ) and SMAD ( Bienz , 1997; Waltzer and Bienz , 1999 ) which synergize with Armadillo to activate these enhancers ( Bienz , 1997; Waltzer and Bienz , 1999 ) —similarly to the interferon-β enhanceosome where CBP recruitment also co-depends on multiple enhancer-binding proteins ( Panne et al . , 2007 ) . The ensuing acetylation of the Wnt enhancer chromatin could promote the eviction of Groucho/TLE whose chromatin anchoring is blocked by acetylation of histone H3 and H4 tails ( Chodaparambil et al . , 2014 ) , thus initiating the ON state . Osa antagonizes Wg responses throughout development , and represses UbxB through its CRE ( Collins and Treisman , 2000 ) , which also mediates repression in response to high Wg signaling ( Yu et al . , 1998 ) ( Figure 6B ) . Osa could therefore terminate enhancer activity , by displacing HAT-recruiting enhancer-binding proteins such as CREB and c-Fos from CREs and by cooperating with repressive enhancer-binding proteins such as Brinker ( a Groucho-recruiting repressor that displaces SMAD from UbxB; Saller and Bienz , 2001; Saller et al . , 2002 ) to re-recruit Groucho/TLE to the enhancer , thereby re-establishing its OFF state ( Figure 9 ) . Notably , Osa binds Chip , to repress various Wg and ChiLS targets including achaete-scute and dLMO ( Heitzler et al . , 2003; Milan et al . , 2004 ) . Therefore , ChiLS is not only a coincidence detector of multiple enhancer-binding proteins and NPF proteins , but also a switch module that exchanges positively- and negatively-acting enhancer-binding proteins ( through LID ) and NPF factors , to confer signal-induced activation , or re-repression ( Figure 9 ) . Its stoichiometry and modularity renders it ideally suited to both tasks . We note that the interferon-β enhanceosome does not contain a similar integrating module ( Panne et al . , 2007 ) , perhaps because it is dedicated to a single signaling input . ChiLS is essential for activation of master-regulatory genes in the early embryo ( Morcillo et al . , 1997 ) ( Figure 8 ) , similarly to DNA-binding pioneer factors such as Zelda ( in the Drosophila embryo ) or FoxA ( in the mammalian endoderm ) which render enhancers accessible to enhancer-binding proteins ( Zaret and Carroll , 2011 ) . Moreover , ChiLS maintains HOX gene expression throughout development ( Figure 5; Figure 8—figure supplement 2 ) , enabling Wg to sustain HOX autoregulation , a mechanism commonly observed to ensure coordinate expression of HOX genes in groups of cells ( Bienz , 1994 ) . Another hallmark of pioneer factors is that they initiate communication with the basal transcription machinery associated with the promoter . Chip is thought to facilitate enhancer-promoter communication ( Morcillo et al . , 1997 ) , possibly by bridging enhancers and promoters through self-association ( Cross et al . , 2010 ) ( Figure 3 ) . Indeed , Ldb1 occupies both remote enhancers and transcription start sites ( e . g . , of globin genes and c-Myb; Love et al . , 2014 ) , likely looping enhancers to the basal transcription machinery at promoters ( Deng et al . , 2012; Stadhouders et al . , 2012 ) which requires self-association ( Krivega et al . , 2014 ) , but possibly also other factors ( such as cohesin , or mediator; Levine et al . , 2014 ) . We note that the chromatin association of Ldb1 has typically been studied in erythroid progenitors or differentiated erythroid cells ( Love et al . , 2014 ) , following activation of erythoid-specific genes ( Stadhouders et al . , 2012 ) . It would be interesting ( if technically challenging ) to examine primitive cells , to determine whether ChiLS is associated exclusively with poised enhancers prior to cell specification or signal responses . Previous genetic analysis in Drosophila has linked chip predominantly to Notch-regulated processes ( Bronstein and Segal , 2011 ) . Likewise , groucho was initially thought to be dedicated to repression downstream of Notch ( Preiss et al . , 1988 ) , before its role in antagonizing TCF and Wnt responses emerged ( Roose and Clevers , 1999 ) . Moreover , Lozenge facilitates Notch responses in the developing eye , and in hematocytes ( Canon and Banerjee , 2000; Terriente-Felix et al . , 2013 ) . Indeed , the first links between Groucho/TLE , RUNX and nuclear Wnt components came from physical interactions ( Roose and Clevers , 1999 ) , as in the case of ChiLS ( Figure 2 ) . Our work indicates that these nuclear Notch signaling components constitute the Wnt enhanceosome . Although our most compelling evidence for this notion is based on physical interactions , the genetic evidence from Drosophila is consistent with a role of ChiLS in Wg responses ( Bronstein et al . , 2010 ) ( Figure 6; Figure 8—figure supplements 1 , 2 ) . Indeed , mouse Ldb1 has been implicated in Wnt-related processes , based on phenotypic analysis of Ldb1 knock-out embryos and tissues ( Mukhopadhyay et al . , 2003; Mylona et al . , 2013 ) . Notably , Ldb1 has wide-spread roles in various murine stem cell compartments that are controlled by Wnt signaling ( Xu et al . , 2007a; Dey-Guha et al . , 2009; Li et al . , 2011; Salmans et al . , 2014 ) . An interesting corollary is that the Wnt enhanceosome may be switchable to Notch-responsive , by NPF factor exchange and/or LMO-mediated enhancer-binding protein exchange at ChiLS ( Figure 9 ) . Hairy/Enhancer-of-split ( HES ) repressors could be pivotal for this switch ( Delidakis et al . , 2014 ) : these bHLH factors are universally induced by Notch signaling , and they bind to ChiLS enhancers to re-recruit Groucho/TLE via their WRPW motifs ( Turki-Judeh and Courey , 2012 ) . HES repressors may thus be capable of re-establishing the OFF state on Wnt enhancers in response to Notch . Notably , restoring a high histone-binding affinity in Drosophila Pygo by reversing the architectural change in its histone-binding surface towards human renders it hyperactive towards both Wg and Notch targets ( Miller et al . , 2013 ) even though pygo is not normally required for Notch responses in flies . Humanized Pygo may thus resist the Notch-mediated shut-down of the Wnt enhanceosome , owing to its elevated histone affinity that boosts its enhancer tethering , which could delay its eviction from the enhanceosome by repressive NPF factors . The apparent Notch-responsiveness of the Wnt enhanceosome supports our notion that orthodox Pygo proteins ( as found in most animals and humans ) confer both Wnt and Notch responses ( Miller et al . , 2013 ) . Previous genetic studies have shown that the components of the Wnt enhanceosome ( e . g . , TCF , RUNX , ChiLS and LHX ) have pivotal roles in stem cell compartments , as already mentioned ( see also Folgueras et al . , 2013; Lien and Fuchs , 2014 ) , suggesting a universal function of this enhanceosome in stem cells . It is therefore hardly surprising that its dysregulation , that is , by hyperactive β-catenin , is a root cause of cancer , most notably colorectal cancer but also other epithelial cancers ( Clevers , 2006 ) . Indeed , genetic evidence implicates almost every one of its components ( as inferred from the fly counterparts ) in cancer: AML1 and RUNX3 are tumour suppressors whose inactivation is prevalent in myeloid and lymphocytic leukemias ( Mangan and Speck , 2011 ) , and in a wide range of solid tumors including colorectal cancer ( Chuang et al . , 2013 ) , respectively . Likewise , ARID1A is a wide-spread tumor suppressor frequently inactivated in various epithelial cancers ( Wu and Roberts , 2013 ) . Furthermore , many T-cell acute leukemias can be attributed to inappropriate expression of LMO2 ( Rabbitts , 1998 ) . Intriguingly , AML1 and ARID1A behave as haplo-insufficient tumor suppressors , consistent with the notion that these factors compete with activating NPF factors such as Pygo2 , RUNX2 and possibly MACC1 ( predictive of metastatic colorectal cancer; Stein et al . , 2009 ) for binding to ChiLS , which will be interesting to test in future . The case is compelling that the functional integrity of the Wnt enhanceosome is crucial for the avoidance of cancer . 6xHis-MBP-Chip205-436 , 6xHis-MBP-LDB156-285 and 6xHis-Lip-SSDP1-92 were co-expressed with a bi-cistronic expression vector ( including N-terminal Tobacco Etch virus ( TEV ) protease sites for removal of tags ) in E . coli BL21-CodonPlus ( DE3 ) -RIL cells ( Stratagene , La Jolla , California , United States ) and purified by Ni-NTA resin and size exclusion chromatography , as described ( Fiedler et al . , 2008 ) . 6xHis-Lip-tagged NPF-containing fragments ( Figures 4 , and 5 ) were purified similarly , after labeling in minimal media for NMR ( Miller et al . , 2010 , 2013 ) . Drosophila S2 cells were grown in Lonza serum-free medium and transfected with bait plasmids ( Figure 2—figure supplement 1A ) using Fugene HD and subsequently grown under continuous selection with 5 μg ml−1 puromycin . For tandem-affinity purification of Pygo-associated proteins , ∼2 × 109 S2 cells ( grown as suspension culture ) or twenty 175 cm2 flasks of subconfluent HEK293T cells stably transfected with Pygo baits were used for each experiment . Cells were lysed in 30 ml lysis buffer ( 12 . 5 mM Tris–HCl pH 7 . 4 , 6 . 25% glycerol , 125 mM NaCl , 0 . 625 mM EDTA , 3 . 1 mM NaF , 1 . 25 mM Na3PO4 , 0 . 125% Triton-X-100 ) , and sonicated 4 × 10 s at 50% intensity with a Branson 250 Sonifier . Cell lysates were cleared by centrifugation for 20 min at 15 , 000 rpm , and incubated ( while rotating ) for 1 hr with α-Flag affinity resin ( Sigma ) in the cold room . Immunoprecipitates were washed 4x with 1 ml lysis buffer , and subsequently eluted with 4 consecutive 500 μl elutions of lysis buffer supplemented with 200 µg ml−1 3xFlag-Peptide ( Sigma ) . Eluates were subjected to α-Strep pull-down with 20 µl packed volume of StrepTactin ( IBA Lifesciences ) . Beads were then washed 3x with 2 ml lysis buffer , and subsequently boiled in 50 µl 2× LDS sample buffer . Proteins were resolved on 4–12% Bis-Tris SDS-polyacrylamide gels . These were stained with Imperial Protein Stain ( Thermoscientific ) , and gel lanes were cut into 1–2 mm slices for in situ digestion with trypsin . The analytical column outlet was directly interfaced via a modified nano-flow electrospray ionisation source , with a hybrid linear quadrupole ion trap mass spectrometer ( Orbitrap LTQ XL , ThermoScientific , San Jose , United States ) . LC-MS/MS data were searched against a protein database ( UniProt KB ) with the Mascot search engine program ( Matrix Science , UK ) ( Perkins et al . , 1999 ) . MS/MS data were validated using the Scaffold programme ( Proteome Software Inc . , United States ) . 100 μl SSDP , 6xHis-MBP-Chip–6xHis-Lip-SSDP or 6xHis-MBP-LDB1–6xHis-Lip-SSDP samples were resolved on a Superdex S-200 or Superose 6 HR 10/300 analytical gel filtration column ( GE Healthcare ) at 0 . 5 ml min−1 in 25 mM phosphate buffer , 150 mM NaCl , pH 6 . 7 before light scattering and concentration determination using refractive index ( RI ) or UV absorbance in a standard SEC-MALS configuration ( containing a Wyatt Heleos II 18 angle light scattering instrument coupled to a Wyatt Optilab rEX online RI detector ) . Protein concentration was determined from the excess differential refractive index based on 0 . 186 RI increment for 1 g ml−1 protein solution . Concentrations and observed scattered intensities at each point in the chromatograms were used to calculate absolute molecular mass from the intercept of the Debye plot , using Zimm's model as implemented in Wyatt's ASTRA software . The stoichiometries of ChiLS indicated by the model-free RI measurements ( using dn/dc 0 . 186 for 1g ml−1 protein ) were further confirmed by using appropriate UV extinction coefficients and UV absorbance as the concentration measurement , which produced essentially identical masses to those from RI . They were independent of protein concentration ( in the range of 0 . 1–10 mg ml−1 ) . [1H , 15N]fast-HSQC spectra of 15N-labeled proteins in 25 mM phosphate buffer , 150 mM NaCl were recorded with 600 MHz 1H frequency ( at 25°C ) , and 13C/15N double-labeled samples were used for backbone resonance assignments . Datasets were acquired , processed and analyzed as described ( Miller et al . , 2013 ) . The following plasmids were recloned in pCMVtag2b , for transfecting HEK293T cells ( with PEI at a ratio of 1:3 . 5 , DNA:PEI ) : Chip ( Morcillo et al . , 1997 ) , SSDP ( van Meyel et al . , 1999 ) , Groucho ( Jennings et al . , 2006 ) , LDB1 ( from Luc Sabourin ) , AML1 , Runx2-P1 , Runx3-P1 ( from Anna Kilbey and Karen Blyth ) , monomeric GFP ( mGFP , from John James ) . HA-Pygo , HA-hPygo2 ( Thompson et al . , 2002 ) and 6xMyc-TLE3 ( Hanson et al . , 2012 ) were also used . Internal deletions were generated by standard procedures in the same vectors , and verified by sequencing . Cell culture , lysate preparation and coIPs were done essentially as described ( Thompson et al . , 2002 ) . The following antibodies and antibody-coupled resins were used: α-LDB1 ( Epitomics ) ; α-β-actin ( Abcam ) ; α-Flag M2 , α-HA ( Sigma ) . All Drosophila strains used are described in Flybase . The following new transgenic lines bearing mutant enhancers were generated: UbxBR was derived from mutating UbxB ( in a ry+ vector; Thuringer et al . , 1993 ) , and 2 independent transformant lines were isolated by standard procedures . Likewise , lab550R and lab550L were derived from mutating lab550 ( in a ry+ vector; Tremml and Bienz , 1992 ) , and 5 independent transformat lines were isolated . The following mutations were made ( numbers refer to binding site number , from 5′ to 3′ ) : UbxB Runt1 , AACCTCG > CTCTAGA; UbxB Runt2 , TCTGGTA > CTCTAGA; lab550 Runt1 , TTTGGTT > AAGATCT; lab550 Runt2 , TGTGGTC > AAGATCT; lab550 dTCF1 , TTACAAA > GCCGGCA; lab550 dTCF2 , CATCAAT > GGGCCCT; lab550 dTCF3 , CATCAAC > CTCGAGC; lab550 dTCF4 , GTTGATG > GaGTACTG ( ‘a’ denotes a one-base insertion in this mutant dTCF binding site ) ; only dTCF1 and dTCF2 binding sites are shown in Figure 6A ( depicting the 5′ portion of the lab550 enhancer ) while dTCF3 and dTCF4 are near the 3′ end of lab550 . ssdp and pygo mutant wing disc clones were generated with vg . GAL4 UAS . flp as described ( Vegh and Basler , 2003; de la Roche and Bienz , 2007; Fiedler et al . , 2008 ) , but hs . flp was used for chip mutant clones ( which were generated by heat-shocking late second or early third instar larvae for 30 min at 37°C ) . The GAL4 drivers used for overexpressing Runt ( Tracey et al . , 2000 ) are described in Flybase ( 24B . GAL4 , for mesodermal expression; 48Y . GAL4 , for endodermal expression ) . Paraformaldehyde-fixed embryos were stained with α-Labial ( Tremml and Bienz , 1992 ) , α-Eve ( Azpiazu et al . , 1996 ) , α-Ubx , α-Wg ( Developmental Studies Hybridoma Bank ) , α-β-galactosidase ( Promega ) as described ( Thompson et al . , 2002 ) . DIC optics were used for imaging embryos on a Zeiss Axiophot . Paraformaldehyde-fixed imaginal discs were stained with α-Vg ( Kim et al . , 1996 ) , α-Ubx , α-Wg ( Developmental Studies Hybridoma Bank ) , rabbit or mouse α-GFP ( Sigma ) as described ( Thompson et al . , 2002; Fiedler et al . , 2008 ) . All discs were counterstained with DAPI , to control for the focal plane , and single confocal images were acquired at identical settings with a Zeiss Confocal Microscope .	In animals , cells have to be able to communicate with neighboring cells in order to generate and maintain the different tissues and organs . One ancient method of cell communication that is used in all animals is the Wnt signaling pathway . In this pathway , a cell secretes a protein called Wnt , which binds to a Wnt receptor present on the surface of another cell . This triggers a cascade of signals inside the second cell that leads to the activation of proteins called TCF factors . These proteins bind to regions of DNA called enhancers to trigger the expression of particular genes that control the development of the animal . Hyperactive Wnt signaling in humans can result in cancer , so Wnt signaling is tightly controlled to avoid this . One of the proteins that regulates Wnt signaling is called Groucho and it interacts with TCF to prevent it from activating genes in the absence of a Wnt signal . However , when Wnt is present , a protein called Pygo overcomes this repression by Groucho to activate TCF , but it is not clear how this works . Fiedler , Graeb , Mieszczanek et al . discovered that Pygo directly binds to a protein complex called Chip/LDB-SSDP ( or ChiLS for short ) . ChiLS is able to associate with TCF enhancers through its association with Groucho . Fiedler , Graeb , Mieszczanek et al . observed that ChiLS can also interact with a number of other proteins that control body formation . This enables ChiLS to integrate multiple signals that regulate the activity of TCF factors . Fiedler , Graeb , Mieszczanek et al . named this complex the ‘Wnt enhanceosome’ because it serves to activate the expression of genes in response to Wnt signaling . Fiedler , Graeb , Mieszczanek et al . analyzed the role of the Wnt enhanceosome during the development of the fly wing and the embryo's midgut . Many genes that are required to form these organs were switched on by the Wnt enhanceosome . This study shows that ChiLS and Pygo are core components of a large complex of proteins that regulate animal development . The next challenge is to study how the components of this complex work together to regulate the enhancers in response to different signals .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2015	An ancient Pygo-dependent Wnt enhanceosome integrated by Chip/LDB-SSDP
Adaptation is a key component of efficient coding in sensory neurons . However , it remains unclear how neurons can provide a stable representation of external stimuli given their history-dependent responses . Here we show that a stable representation is maintained if efficiency is optimized by a population of neurons rather than by neurons individually . We show that spike-frequency adaptation and E/I balanced recurrent connectivity emerge as solutions to a global cost-accuracy tradeoff . The network will redistribute sensory responses from highly excitable neurons to less excitable neurons as the cost of neural activity increases . This does not change the representation at the population level despite causing dynamic changes in individual neurons . By applying this framework to an orientation coding network , we reconcile neural and behavioral findings . Our approach underscores the common mechanisms behind the diversity of neural adaptation and its role in producing a reliable representation of the stimulus while minimizing metabolic cost . The range of firing rates that a sensory neuron can maintain is limited by biophysical constraints and available metabolic resources . Yet , these same neurons represent sensory inputs whose strength varies by orders of magnitude . Seminal work by Barlow ( 1961 ) and Laughlin ( 1981 ) demonstrated that sensory neurons in early processing stages adapt their response threshold and gain to the range of inputs that they recently received . A particularly striking example of such gain modulation at the single cell level has been shown in the fly H1 neuron ( Brenner et al . , 2000 ) . Gain adaptation has been observed in other early sensory circuits ( Blakemore and Campbell , 1969; Fairhall et al . , 2001; Solomon and Kohn , 2014; Wark et al . , 2007 ) , such as in the retina ( Kastner and Baccus , 2014 ) , auditory hair cells ( Nagel and Doupe , 2006; Wen et al . , 2009 ) and is also present in later sensory stages ( Adibi et al . , 2013; Wainwright , 1999 ) . Moreover , cortical neurons acquire this property during development ( Mease et al . , 2013 ) . The work of Laughlin and Barlow was instrumental in uncovering a principle of neural adaption as maximizing information transfer . However , the natural follow-up question concerns the decoding of such neural responses after they have been subject to adaptation . Indeed , such changes in neural gains may result in profound changes of the mapping of neural responses to stimuli in a history-dependent manner . This raises the issue of how such adapting responses are interpreted by downstream sensory areas ( Seriès et al . , 2009; Webster , 2011 ) . One possibility , of course , is that downstream areas do not change their decoding strategy , thus introducing systematic biases in perception . This has been interpreted as the source of perceptual illusions such as the tilt after-effect or the waterfall illusion ( Barlow and Hill , 1963; Wainwright , 1999; Clifford , 2014; He and MacLeod , 2001; Schwartz et al . , 2009 ) . Such illusions are classically triggered by long presentations of particularly strong or repetitive stimuli ( Maffei et al . , 1973 ) . However , adaptation deeply affects neural responses even at short time scales or after only one repetition of the same stimulus ( Patterson et al . , 2013 ) . Adaptation could make it impossible to recognize a visual object independently of the stimuli presented previously . An example is given in Figure 1 where we present successive visual patterns to a population of randomly connected leaky integrate-and-fire ( LIF ) neurons . For simplicity and for the sake of illustration , the network takes a 7-dimensional time-varying input interpreted as a spatio-temporal sequence of digital numbers ( Figure 1b , top row ) . An optimal linear decoder was trained to reconstruct the patterns from the spike counts during the presentation of the patterns . Not surprisingly , the decoder could reconstruct the patterns accurately , regardless of their place in the sequence ( Figure 1b , 2nd row ) . We then tested the network in the presence of spike-based adaptation in the LIF neurons . Spike-based adaptation was induced by temporarily hyperpolarizing the neurons after each spike . The time scale of this adaptation was chosen to be long enough to cover several visual patterns . When subjected to this spike-time dependent adaptation , the responses became strongly history dependent , resulting in a highly inaccurate decoding ( Figure 1b , 3rd row ) . This would suggest that activity in downstream areas and perceptual interpretations should be based not only on the current sensory responses , but also on the recent history of neural activity ( Fairhall et al . , 2001; Borst et al . , 2005 ) . In this study , we show that this is not necessarily the case . Recurrent connections can be tuned such that spike-dependent adaptation will not impair the stability of the representation ( Figure 1b , bottom row ) . We will start from an objective function quantifying the efficiency of a population of spiking neurons in representing a time varying sensory stimulus , ϕ ( t ) . We will then show that appropriate recurrent connections between the neurons , namely connections that maintain a tight balance between the excitation and inhibition received by each neuron , will minimize this objective function and thus , maximize the efficiency of the neural code . For the sake of illustration , we hereby assume that the stimulus is unidimensional and positive , as for luminance or color saturation ( see Materials and methods for multidimensional stimuli ) , and the stimulus has arbitrary units . The stimulus will be decoded from the firing activity of the neurons by summing their responses with their respective readout weights , wi . ( 1 ) ϕ^ ( t ) =∑iwiri ( t ) The neural response , ri ( t ) , is defined as the spike train integrated at a short time scale , ( 2 ) r˙i=−1τri+oiwhere oi⁢ ( t ) corresponds to the spike train of neuron i . The readout weight of neuron i is denoted as wi and it is a fixed parameter . One may choose to include a wide range of readout weights in the network . The output estimate , ϕ^ ( t ) , can be interpreted as a postsynaptic integration of the output spike trains of the population , weighted by synaptic weights w . We wish to construct a network that will minimize the difference between ϕ ( t ) and ϕ^⁢ ( t ) , ensuring an accurate representation of the stimulus . Additionally , we wish to impose , not only accuracy , but also cost efficiency in the neural representation . For biological neurons , spiking comes with inherent metabolic costs . For example , resources are expended after each spike and neurons or neural populations may need some time to recover from a period of strong activity . Albeit many different types of cost can be incorporated into our approach , here we summarize these constraints as a cost term representing the sum of all squared firing rates . Thus , we define an objective function composed of two terms , one representing the precision of the representation , and the other the cost of neural activity ( Boerlin et al . , 2013 ) : ( 3 ) E ( t ) = ( ϕ ( t ) −ϕ^ ( t ) ) 2+μ∑ifi ( t ) 2 Where fi ( t ) is the firing history of neuron i and the parameter µ weights the relative contributions of error and metabolic costs . The firing history , fi ( t ) , is defined as the spike train integrated with a time constant , τa . ( 4 ) f˙i=−1τafi+oi Typically , the adaptation time scale is assumed to be significantly longer than the decoder time scale ( τa≫τ ) . A short τ ( e . g . of the order of 10 ms ) ensures that fast changes in the stimulus can be represented accurately . However , the metabolic cost of spiking accumulates and recovers at slower time scales ( e . g . τa corresponds to hundreds of ms ) . The underlying assumption is that the dynamics allowing metabolic resources to be replenished are slower than the time scale at which neural populations transmit information . The sum of squared firing history will encourage , not only low activity at the level of the population , but also low activity in single neurons . As a result , neurons will share the burden of the representation . From these assumptions , we derive a prescription for the voltage dynamics of leaky integrate-and-fire ( LIF ) neurons performing a greedy minimization of the objective function , E ( see Materials and methods for full derivation ) . Our framework revolves around the assumption that a neuron spikes only when doing so reduces the decoding error . This condition can be expressed in terms of the objective function as E⁢ ( t ) n⁢o⁢s⁢p⁢i⁢k⁢e>E⁢ ( t ) s⁢p⁢i⁢k⁢e where a spike is justified if the objective is minimized relative to having no spike at that time step . Using Equation 3 , we obtain a new expression from this inequality that embodies a condition for spiking and that we interpret as a voltage expression and a threshold ( see Materials and methods for derivation details ) such that V⁢ ( t ) >t⁢h⁢r⁢e⁢s⁢h⁢o⁢l⁢d and voltage is: ( 5 ) Vi⁢ ( t ) =1wi2+μ⁢ ( wi⁢ ( ϕ⁢ ( t ) -ϕ^⁢ ( t ) ) -μ⁢fi⁢ ( t ) ) Taking the derivative of the voltage expression produces the voltage equation below: ( 6 ) τ⁢V˙i=-Vi+gi⁢wi⁢ ( τ⁢ϕ˙+ϕ ) -τ⁢gi⁢∑jΩi⁢j⁢oj-κi⁢fi Where gi is the gain of neuron i , ( 7 ) gi=1/ ( wi2+μ ) and the lateral connections are given by Ωi⁢j ( 8 ) Ωi⁢j=wi⁢wj+μ⁢δi⁢jwhere δi⁢j is the delta function ( equals one only if j=i , zero otherwise ) and κi=μ⁢gi⁢ ( 1-ττa ) . The form of the voltage equation is amenable to being interpreted as a set of currents to a neuron embedded in a recurrent network with all-to-all connectivity . Neurons are connected by mutually inhibitory synapses determined by their decoding weights . The final term corresponds to an adaptation current that depresses the voltage as a function of its recent activity ( see Figure 2c ) . This indicates that spike-frequency adaptation in single neurons is part of the solution to the cost-accuracy tradeoff . However , we will show that it cannot work alone; it needs to be associated with appropriately tuned recurrent connections . It is easier to interpret the network function if we consider that the membrane potentials are effectively proportional to the global coding error penalized by the past activity of the neuron , as seen in Equation 5 . A neuron that reaches the firing threshold is guaranteed to contribute a decrease of the error term in the objective function ( Equation 3 ) . As a whole , the population performs a greedy minimization of the objective function , or , in other terms , a greedy maximization of the coding efficiency . Finally , we note that since the integrated excitatory input , gi⁢wi⁢ϕ⁢ ( t ) , is cancelled as precisely as possible ( except for the cost penalty ) by the recurrent inhibition , -gi⁢wi⁢ϕ^⁢ ( t ) , the network can be considered as balancing feedforward excitation and recurrent inhibition ( see Equation 5 ) . The second ingredient for population efficiency ( in addition to spike-based adaptation ) is thus to maintain a tight E/I balance in the network . In other words , we show that a memoryless decoder will be able to reconstruct the stimulus from the output spike trains of an E/I balanced population of adapting neurons . This is shown in the bottom row of Figure 1b . Before we investigate the network dynamics and performance , we first describe the properties of single neurons and the relationship between their gain and their coding precision . Let us first consider the case without quadratic cost ( i . e . μ=0 ) . In that case , each neuron effectively has identical voltage and spiking dynamics . Neurons are differentiated only by their gain , gi , and their decoding weight , wi ( Figure 2 ) . The strength of the feedforward gain is inversely related to the strength of the output weight for each neuron . As a result , neurons with the smallest decoding weights ( and thus , the highest precision in representing the input ) tend to respond most strongly to the stimulus ( Figure 2a ) . We will refer to these costly but reliable neurons as 'strongly excitable’ . In contrast , neurons with large decoding weights and small input weights ( thus 'low gain’ neurons ) bring less precision to the estimate but are metabolically efficient . We will refer to these neurons as 'weakly excitable’ . Note that if μ=0 , the cost is not taken into account by the network . Thus , it will always favor precision over cost . In that case , only the most excitable neuron ( with the smallest decoding weight ) will respond to the stimulus while completely inhibiting the other neurons . However , with the addition of a cost ( μ>0 ) , adaptive currents contribute to the voltage dynamics , penalizing neurons with large firing rates . Moreover , the feedforward gain , gi⁢wi , does not necessarily decrease monotonically with the decoding weight ( Figure 2b ) . For very small decoding weights , \|wi\|2≪μ , neurons with decoding weights smaller in magnitude than μ are penalized . These neurons would simply be too costly to participate meaningfully in the cost/accuracy tradeoff solved by the population . Model neurons in isolation ( i . e . without any contribution from recurrent connections ) would respond to a step-like input with a rate that decreases exponentially in time before reaching a plateau ( Figure 2c ) , a classic signature of activity-dependent suppression . The time constant of this adaptation is determined by τa , while the strength of this adaptation increases with the gain . Highly excitable neurons adapt strongly , while less excitable neurons adapt weakly . However , these intrinsic properties of single neurons will be deeply affected by the dynamics introduced by recurrent connections . To gain a better understanding of population adaptation , we investigate how inhibitory connections orchestrate the relative contributions of different neurons over the duration of a long stimulus . We first illustrate the effect of recurrent connections with an example network composed of only two neurons ( Figure 3 ) . The two neurons are reciprocally connected with inhibitory connections , as prescribed in the derivation ( schematized in Figure 3a ) . They receive a constant stimulus , but have different input weights . During sustained stimulation , the response of each neuron fluctuates dynamically despite the fact that the stimulus is constant ( Figure 3b ) . This would be expected given their spike-time dependent adaptation . However , if one removes the recurrent connections and plots the response of one neuron as a function of the other ( Figure 3c , right ) , we discover that the population response wanders from the iso-coding line , ϕ^=w1⁢r1+w2⁢r2 ( i . e . the manifold in activity space where the stimulus would be decoded properly ) . In contrast , the intact network with its recurrent connections coordinates the two neurons such that the weighted sum of their responses remains accurate . The movement of the activity along the manifold defined by the constant stimulus and the decoding weights reflects a progressive redistribution of activity to satisfy the unfolding cost-accuracy tradeoff , as the cost slowly accumulates ( Figure 3d ) . While to a naive observer , the high gain neuron may appear to adapt while the low gain neuron has a sustained response and a longer delay , in fact both contribute to population adaptation because both neurons coordinate and adapt their activity to limit the metabolic cost of the representation while maintaining its accuracy . Recurrent connections deeply affect the dynamics of each neuron . For example , the inhibition from the strongly excitable neuron is responsible for the response delay of the weakly excitable neuron . Within a network with many neurons ( Figure 4 ) , recurrent connections interact with the intrinsic properties of the neurons in a similar manner as in the previous example . The first neurons to be recruited are strongly excitable and provide an initially very precise representation of the signal . These neurons inhibit the less excitable neurons , preventing them from firing early in this stimulation period . As the cost accumulates , however , the response of the high gain neurons decays due to spike-frequency adaptation . This is compensated by weakly excitable neurons that become disinhibited , fire , and then adapt in their turn . The less excitable a neuron is , the later it will be recruited , resulting in strong response delays . The dynamic response properties of individual neurons are thus dominated by network interactions and are markedly different from their intrinsic adaptive properties ( Figure 2 ) . Because the disinhibition of weakly excitable neurons automatically compensates for the decay in strongly excitable neural responses , the stimulus representation remains stable during the whole period ( Figure 4d ) . However , note that its precision degrades as more low gain neurons contribute to the representation . As a result , the bias and standard deviation of the representation increases as imposed by the global cost/accuracy tradeoff . To illustrate what coordinated adaptation would predict for tuning curves measured experimentally , we constructed a population of neurons that code for visual orientation; V1 simple cells . The input to the network takes the form of a two-dimensional signal with a cosine and a sine of the presented orientation ( see Materials and methods ) . Each neuron has a preferred orientation that is given by the combination of input weight strengths in the two input dimensions . In turn , the network orientation estimate can be decoded from the population ( see Materials and methods ) . The lateral connections derived from the model maximally inhibit neurons with similar preferred orientations and excite neurons with orthogonal orientations ( see schematic in Figure 5a ) due to the choice of decoding weights which can be positive or negative , or some combination . To observe the effects of adaptation on a diverse population of neurons , we constructed our network so that neurons have equally spaced preferred orientations and a partner neuron that shares the same preference but has a different gain . There is a high gain and a low gain neuron among each pair of neurons that code for the same orientation . Figure 5b illustrates the spiking response of the network to a prolonged oriented stimulus . As seen in the simpler model from Figure 4 , high gain neurons respond first , then adapt . As the responses of those strongly excitable neurons decay , weakly excitable neurons are recruited , maintaining the representation . This results in systematic changes in the tuning curves from early in the response to later in the response ( Figure 6 ) . Highly excitable neurons are suppressed relative to their early responses ( Figure 6 , top ) . In contrast , weakly excitable neurons see their tuning curves widen when the adapting stimulus is similar to their preferred orientation ( Figure 6 , bottom ) . At the flank of the adapting orientation , low gain neurons see an increase in their responsiveness . Here , the network interactions override the intrinsic adapting currents in the weakly excitable neural population . In other words , the disinhibition from strongly excitable neurons combined with the constant feedforward drive to these low gain neurons results in facilitated activity rather than the suppressed activity one would expect to be caused by adaptation . Finally , the tuning curves for the most excitable neurons are broader than those for weakly excitable neurons . These neurons are more likely to fire first in response to oriented stimuli that are near their preferred orientation and prevent the low gain neurons from doing the same . The same qualitative effects are observed in a more realistic network where the preferred orientations of different neurons and their decoding weights are taken from a random distribution ( Figure 7 ) , rather than regularly spaced with two levels of excitability . The tuning curves are more heterogeneous not because of noise but because of the randomness of the decoding weights . Tuning curves can be either facilitated or suppressed by adaptation . When the adapted stimulus falls on the flank of the tuning curve , it can be accompanied by a shift toward or away from the adapting stimulus . The effect of adaptation on single neurons is variable not because of noise ( we did not introduce any ) but because of local heterogeneity in the competition they receive from other neurons , itself due to the random choices of weights . In fact , adaptation in one neuron would be impossible to predict quantitatively without observing the rest of the network . We have stressed the accuracy of the stimulus representation in the face of time-varying activity due to adaptation . While this kind of activity could be interpreted as leading to a stable percept in spite of adaptation , we acknowledge that perceptual errors and biases are abundant in the natural world . Our network is capable of emulating these errors and it is able to do so in a manner that is consistent with experimental findings . The network is designed to negotiate the tradeoff between accuracy and efficiency and it will prioritize the production of a stable representation if µ is small . If µ is large , the network will favor cost over accuracy . Thus , strong adaptation and a prolonged stimulus presentation can produce a representation that degrades over time . This degradation can lead to a bias in the decoder . In Figure 8 , an oriented , strong , adapting stimulus is presented for 2 seconds followed by a test orientation ( this is schematized in Figure 8a ) . An example of the resulting network activity is shown in Figure 8b . Before adaptation takes hold , the adapting stimulus activates the high gain neurons that have preferences at or near the stimulus orientation . Because the adapting stimulus is strong , high gain neurons with similar preferences are quickly recruited . As the stimulus persists , the most strongly activated high gain neurons fatigue and the low gain neurons with matching preferences are recruited . After the presentation of the adapting orientation , a weaker peripherally oriented test stimulus is delivered . The response distribution and dynamics are markedly different . Instead of a widely-tuned response , the weaker test stimulus produces a more narrowly distributed response . The decoded orientation is offset from the test stimulus orientation , indicating a bias in the perceived orientation . A classical study of perceptual bias is the tilt illusion ( Gibson and Radner , 1937; Clifford , 2014 ) . In the tilt illusion , the orientation of a test grating is perceived incorrectly after adaptation to a differently oriented stimulus . Experimental studies report that the perceived orientation is often repulsed away from the adapted orientation , the effect being maximal for adapting stimuli tilted around 15–20 degrees from the test stimulus . If the adapting stimulus is oriented around 60 degrees from the test stimulus , a repulsive effect is observed instead . This effect has been confirmed in the visual cortex ( Jin et al . , 2005; He and MacLeod , 2001 ) . Our model replicates this effect ( Figure 8c , d ) . The test stimulus is decoded at an orientation that is repulsed from its actual orientation away from the adaptor when the adaptor is approximately 15 degrees from vertical ( Figure 8c , middle ) . However , when the adaptor is obliquely oriented from the test orientation , the test stimulus is perceived to be oriented in a direction that is attracted to the adaptor ( Figure 8c , right ) . Test stimuli within a range of 0–45 degrees difference from the adaptor orientation are repulsed whereas test stimuli with a greater than 45 degree difference from the adaptor orientation are attracted ( Figure 8d ) . In accordance with experimental findings , the repulsion effect has a greater amplitude than the attraction effect . Our model suggests that diverse adaptation properties within a population can be an asset . The variability of adaptation effects has been observed in V1 neurons ( Jeyabalaratnam et al . , 2013; Nemri et al . , 2009; Ghisovan et al . , 2009 ) . A heterogeneous population of neurons is able to better distribute the cost to maximize efficiency in different contexts . Studies in the retina show that retinal ganglion cells with different adaptive properties complement each other such that sensitizing cells can improve the encoding of weak signals when fatiguing cells adapt ( Kastner and Baccus , 2011 ) . This arrangement is particularly advantageous for encoding contrast decrements which would be difficult to distinguish from the prior stimulus distribution if only suppressive adaptation prevailed . At the same time , these heterogeneities contribute to complex dynamics in the neural spike trains ( Dragoi et al . , 2000; Okun et al . , 2015; Nirenberg et al . , 2010; Mohar et al . , 2013; Wissig and Kohn , 2012 ) , obscuring the relationship between neural activity and neural coding for an observer of single neuron activity . We make the prediction that neurophysiological studies where single neuron activity is recorded may exhibit an experimental bias that results in highly responsive neurons being overrepresented in the sample . Moreover , our study challenges the notion that tuning is a static characteristic of neurons . Experiments increasingly reveal that neurons change their tuning dynamically with changing stimulus statistics ( Hollmann et al . , 2015; Hong et al . , 2008; Hosoya et al . , 2005; Nagel and Doupe , 2006; Smirnakis et al . , 1997; Solomon and Kohn , 2014; Wark et al . , 2007; Wark et al . , 2009 ) . In the visual cortex , it has been shown that the tilt after effect is not only an effect of response suppression but that it also has the effect of shifting the tuning curves of neurons away from their preferred orientations ( Jin et al . , 2005; Ghisovan et al . , 2009; Dragoi et al . , 2000 ) . While it may be possible to predict some aspect of the tuning change from measurements of intrinsic neuron properties , our study shows that a great deal of the change may be a network effect rather than an intrinsic neuronal effect . Thus , the extent of adaptation for a single neuron may be difficult to predict without considering the properties of the rest of the network ( Fairhall , 2014 ) . Such unpredictable adaptation could be a problem for the interpretation by downstream readouts , however , we show that when the network is considered as a whole , the adaptive effects in one neuron can be compensated for by another neuron that reports to the same readout . In other words , the apparently complex adaptation at the single neuron level is not an impediment to the network but rather an indicator of the manner in which the signal is encoded by the network as a whole . Our model applies at the level of relatively densely connected , and thus local , populations . Observing the organized transfer of responses between neurons through adaptation and E/I balance would require one to record a significant proportion of these neurons locally ( neurons that are likely to be interconnected directly or through interneurons ) . Recent experimental techniques render such recordings possible ( Buzsáki , 2004 ) , bringing an experimental validation of this framework within grasp . These recordings could be compared before and after adaptation , over the duration of prolonged stimuli , or over many repetitions of the same stimulus . What we expect to see is a generalization of the effect illustrated in Figure 4b , c to larger neural populations . First of all , there should exist a decoder of neural activity , independent of stimulus history that can detect the stimulus despite large changes in neural activity over time . Second of all , shuffling the neural responses , for example between the early and latter part of the responses to a prolonged stimulus , should have detrimental effects on such stable decoding . And finally , over the course of adaptation , the activity of the different neurons should not vary independently . For example , if we performed a dimensionality reduction ( such as a principle component analysis ) of the neural population activity during a prolonged stimulus presentation , we might be able to observe that neural responses over time are constrained on a subspace where the stimulus representation is stable . Another , more direct way of testing our framework would be to activate or inactivate a part of the neural population . This could be done optogenetically , for example ( Okun et al . , 2015 ) . The network used in Figure 1 is a generic recurrent network of 400 neurons with random recurrent and feedforward weights . The feedforward weights are a 7 × 400 matrix of values drawn from a uniform distribution in the [−1 , 1] range . The recurrent weights are drawn from a Gaussian distribution with mean = 0 , std = 0 . 87 ( close to 1 ) and are a 400 × 400 matrix , however , all neurons had an autapse that was the sum of the negative squares of its feedforward weights . The network was trained on 100 stimulus examples of 300 ms each that were generated randomly from a uniform distribution of arbitrary input values between 0 and 4 . An optimal linear decoder was obtained from this training by taking the inverse of the responses and multiplying them by the stimulus training examples: decoder = pseudoinverse ( r⁢ ( t ) ) ϕ⁢ ( t ) . The trained network was then presented with a sequence of 8 digitized patterns for 200 ms each separated by 100 ms of no stimulus input . To demonstrate the effect of adaptation , the trained network was run on the same stimulus sequence and with the same linear decoder but this time the spiking threshold was dynamically regulated by past spiking activity such that the threshold was 1+μ⁢fi⁢ ( t ) , where f˙i⁢ ( t ) =-1τa⁢fi⁢ ( t ) +oi⁢ ( t ) . For the example of the balanced network with adaptation , the network was derived using the framework described below using the following parameters: τa=2000⁢m⁢s , μ=0 . 02 , τ=5⁢m⁢s . We provide here a brief description of the network structure and the objective function it minimizes . A detailed and closely related version of this derivation is found in Boerlin et al . ( 2013 ) . The innovation in our present study is the incorporation of a variable for spiking history in the derivation . We consider a spiking neural network composed of N neurons that encodes a set of M sensory signals , ϕ=[ϕ1 , … , ϕM] . Estimates of these input signals , ϕ^=[ϕ^1 , … , ϕ^M] , are decoded by applying a set of decoding weights , [wi1 , wi2 , . . . , wim] , to the filtered spike train of neuron i so that ϕ^m⁢ ( t ) =∑iNwi⁢m⁢ri⁢ ( t ) ( see Equation 1 ) . The filtered spike train , ri⁢ ( t ) , corresponds to a leaky integration of its spikes , oi⁢ ( t ) , while the spike history , fi⁢ ( t ) , filters the spike train on a longer time scale so that τa>τ . ( 9 ) oi ( t ) =∑kδ ( t−tik ) ( 10 ) r˙i=−1τri+oi ( 11 ) f˙i=−1τafi+oiwith tik the spike time of the kt⁢h spike in neuron i and τ the time scale of the decoder . As we will see , τ is the membrane time constant of the model neurons and τa is the adaptation time constant . The decoding weights wi⁢m are chosen a priori . They determine the selectivity and gain of the model neurons . We want to construct a neural network that represents the signals most efficiently , given the fixed decoding weights . Efficiency is defined as the minimization of an objective function composed of two terms , one penalizing coding errors , and the other penalizing firing rates: ( 12 ) E⁢ ( t ) =∥ϕ⁢ ( t ) -ϕ^⁢ ( t ) ∥2+μ⁢∑iNfi2 µ is a positive constant regulating the cost/accuracy tradeoff . In order to minimize this objective function , we define a spiking rule that performs a greedy minimization . Thus , neuron i fires as soon as this results in a minimization of the cost , that is as soon as Espike in i ( t ) <Eno spike in i ( t ) . A spike in neuron i contributes a decaying exponential kernel , h⁢ ( u ) , to its firing rate so that ( 13 ) ri⁢ ( u ) →ri⁢ ( u ) +h⁢ ( u-t ) ( 14 ) ϕ^m⁢ ( u ) →ϕ^m⁢ ( u ) +h⁢ ( u-t ) ( 15 ) fi⁢ ( u ) →fi⁢ ( u ) +h~⁢ ( u-t ) where h~⁢ ( u ) is a more slowly decaying exponential kernel than h⁢ ( u ) . The spiking condition , Espike in i ( t ) <Eno spike in i ( t ) , can be expressed as: ( 16 ) ‖ϕ−ϕ^+wihei‖2+μ∑n≠iNfn2+μ ( fi+h~ ) 2<‖ϕ−ϕ^‖2+μ∑nNfn2 The i-th element in the Euclidean basis vector , 𝒆i , is one while all other entries are zero . Algebraically rearranging this expression leads to the following spiking rule: neuron i spikes if: ( 17 ) gi ( ∑mMwim ( ϕm ( t ) −ϕ^m ( t ) ) −μfi ) >12 ( 18 ) gi=1/ ( ∥𝒘i∥2+μ ) With gi being the 'gain’ of neuron i . We interpret the left-hand side of Equation 17 as the membrane potential , Vi⁢ ( t ) , of neuron i , and the right-hand side as its firing threshold . Membrane potentials are normalized such that each neuron has a threshold equal to 1/2 and reset potential equal to −1/2 . The membrane potential dynamics are obtained by taking the derivative of the voltage expression with respect to time ( where κi=μ⁢gi⁢ ( 1-ττa ) ) : ( 19 ) τ⁢V˙i=-Vi+gi⁢∑mMwi⁢m⁢ ( τ⁢ϕ˙m+ϕm ) -κ⁢fi-τ⁢gi⁢∑mM∑jNwi⁢m⁢wj⁢m⁢oj-μ⁢τ⁢gi⁢oi Notice that the lateral connection between neuron i and neuron j is equal to τ⁢gi⁢∑mwi⁢m⁢wj⁢m . Thus , the lateral connections measure to what extent the feed-forward connections of two neurons are correlated , and they remove those correlations to obtain the most efficient code . The orientation-coding network follows the same derivation as outlined for the generic network model . It has two input dimensions and 200 neurons . There are two subpopulations of neurons such that 100 neurons are high gain neurons and the remaining 100 neurons are low gain neurons . Both populations span the unit circle evenly such that one low gain and one high gain neuron share the same preferred orientation . More precisely , we endowed each neuron with a decoding vector , wi: ( 20 ) [wi⁢1 , wi⁢2]=[γi⁢c⁢o⁢s⁢ ( 2⁢Θi ) , γi⁢s⁢i⁢n⁢ ( 2⁢Θi ) ] , -π2<Θi<π2 γi equals three for high gain neurons and nine for low gain neurons and Θi is the preferred orientation of neuron i . Feedforward inputs correspond to two time-varying inputs , ϕ1⁢ ( t ) =C⁢ ( t ) ⁢c⁢o⁢s⁢ ( θ⁢ ( t ) ) , ϕ2=C⁢ ( t ) ⁢s⁢i⁢n⁢ ( θ⁢ ( t ) ) , where C⁢ ( t ) is the stimulus magnitude and θ⁢ ( t ) is the stimulus orientation at time t ( -π2<θ<π2 ) . The orientation estimate , θ^⁢ ( t ) , is decoded from the population as ( 21 ) θ^⁢ ( t ) =a⁢r⁢c⁢t⁢a⁢n⁢ ( ϕ^2⁢ ( t ) ϕ^1⁢ ( t ) ) The spiking threshold includes an additional term , η⁢gi , that ensures that neurons with opposing preferences will not be activated to spike so easily by the excitation from opposing neurons . Tuning curves in Figures 6 and 7 were generated by presenting the network with a full range of stimulus orientations Each orientation was presented for 250 ms after reinitializing the network . Neuron responses were centered on their preferred orientation and the mean was taken for each subpopulation . Tuning curves after adaptation were made by lining up neuron responses to an adapting stimulus that corresponded with its preferred orientation . Adapting stimuli were presented for 1 . 5 s . Standard deviations were computed on these centered data . The random gain network was identical to the above with the exception that the feedforward weight gains , γ , were randomly selected from a uniform distribution with values ranging from 3 to 9 . The tilt illusion was generated by presenting the network with an adaptor orientation ( duration of 2 s ) and a subsequent test orientation ( 250 ms ) . The perceived angle was decoded from the mean network output over the 250 ms presentation of the test stimulus . The adaptor had a stimulus magnitude of 25 while the test stimulus had a magnitude of 5 .	Humans see , hear , feel , taste and smell the world as spiking electrical signals in the brain encoded by sensory neurons . Sensory neurons learn from experience to adjust their activity when exposed repeatedly to the same stimuli . A loud noise or that strange taste in your mouth might be alarming at first but soon sensory neurons dial down their response as the sensations become familiar , saving energy . This neural adaptation has been observed experimentally in individual cells , but it raises questions about how the brain deciphers signals from sensory neurons . How do downstream neurons learn whether the reduced activity from sensory neurons is a result of getting used to a feeling , or a signal encoding a weaker stimulus ? The energy saved through neural adaptation cannot come at the expense of sensing the world less accurately . Neural networks in our brain have evidently evolved to code information in a way that is both efficient and accurate , and computational neuroscientists want to know how . There has been great interest in reproducing neural networks for machine learning , but computer models have not yet captured the mechanisms of neural coding with the same eloquence as the brain . Gutierrez and Denève used computational models to test how networks of sensory neurons encode a sensible signal whilst adapting to new or repeated stimuli . The experiments showed that optimal neural networks are highly cooperative and share the load when encoding information . Individual neurons are more sensitive to certain stimuli but the information is encoded across the network so that if one neuron becomes fatigued , others receptive to the same stimuli can respond . In this way , the network is both responsive and reliable , producing a steady output which can be readily interpreted by downstream neurons . Exploring how stimuli are encoded in the brain , Gutierrez and Denève have shown that the activity of one neuron does not represent the whole picture of neural adaptation . The brain has evolved to adapt to continuous stimuli for efficiency at both the level of individual neurons and across balanced networks of interconnected neurons . It takes many neurons to accurately represent the world , but only as a network can the brain sustain a steady picture .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "neuroscience" ]	2019	Population adaptation in efficient balanced networks
When females mate with more than one male , the males’ paternity share is affected by biases in sperm use . These competitive interactions occur while female and male molecules and cells work interdependently to optimize fertility , including modifying the female’s physiology through interactions with male seminal fluid proteins ( SFPs ) . Some modifications persist , indirectly benefiting later males . Indeed , rival males tailor their ejaculates accordingly . Here , we show that SFPs from one male can directly benefit a rival’s sperm . We report that Sex Peptide ( SP ) that a female Drosophila receives from a male can bind sperm that she had stored from a previous male , and rescue the sperm utilization and fertility defects of an SP-deficient first-male . Other seminal proteins received in the first mating ‘primed’ the sperm ( or the female ) for this binding . Thus , SP from one male can directly benefit another , making SP a key molecule in inter-ejaculate interaction . In many animal species , females mate with more than one male . This polyandry lays the foundation for differential fertilization success of sperm from the different males ( Parker , 1970; Parker , 1979 ) within a female , whose ‘cryptic female choice’ can bias the relative use of these sperm ( Eberhard , 1996 ) . This , in turn can drive the evolution of male reproductive traits including optimal sperm numbers , morphology , and seminal protein sequences ( Almeida and Desalle , 2008; Birkhead , 1998; Pitnick and Miller , 2000 ) . Against the backdrop of these conflicts , male and female molecules and/or cells must also work together to ensure reproductive success . How efficiently sperm interact with the egg and instigate successful fertilization or embryo support ( where relevant ) is key to successful fertility . Accordingly , males have evolved molecular mechanisms that trigger physiological changes in females that increase the reproductive success of the mating pair . Seminal fluid proteins ( SFPs ) are crucial regulators of these changes . SFPs are produced within glandular tissues in the male reproductive tract and are transferred to females along with sperm during mating ( Avila et al . , 2010; Poiani , 2006; Ravi Ram and Wolfner , 2007b; Ravi Ram et al . , 2005; Ravi Ram and Wolfner , 2009; Wigby et al . , 2009 ) . Within a mated female , SFPs mediate an array of post-mating responses such as , in insects , changes in egg production , elevated feeding rates , higher activity or reduced sleep levels , long-term memory , activation of the immune system and reduced sexual receptivity ( Avila et al . , 2011; Bath et al . , 2017; Scheunemann et al . , 2019; Isaac et al . , 2010; Domanitskaya et al . , 2007; Chapman et al . , 2003; Schwenke et al . , 2016 ) . The ability of a male’s SFPs to induce long-term changes in the mated female enhances that male’s reproductive success . For example , the seminal Sex Peptide ( SP ) of male Drosophila binds to his sperm stored in the female , persisting there for approximately 10 days ( Peng et al . , 2005 ) . This binding of SP to sperm is aided by the action of a network of other SFPs , the ‘LTR-SFPs’ ( Ravi Ram and Wolfner , 2009; Singh et al . , 2018; Findlay et al . , 2014 ) . The active region of SP is then gradually cleaved from sperm in storage , dosing the females to maintain high rates of egg laying , decreased receptivity to remating ( Peng et al . , 2005 ) , increased food intake , and slower intestinal transit of the digested food to facilitate maximum absorption and production of concentrated faeces ( Avila et al . , 2011; Apger-McGlaughon and Wolfner , 2013; Carvalho et al . , 2006; Gioti et al . , 2012; Cognigni et al . , 2011 ) . However , induction of these changes can also indirectly benefit his rival , as the female’s physiology will have already been primed for reproduction by her first mate’s SFPs . Such indirect benefits to the second male have been suggested to explain the tailoring of the ejaculate by males that mate with previously mated females ( Wigby et al . , 2009; Garbaczewska et al . , 2013; Sirot et al . , 2011; Neubaum and Wolfner , 1999 ) . For example , the Drosophila seminal protein ovulin increases the number of synapses that the female’s Tdc2 ( octopaminergic ) neurons make on the musculature of the oviduct above the amount seen in unmated females ( Rubinstein and Wolfner , 2013 ) . This is thought to sustain high octopaminergic ( OA ) signaling on the oviduct musculature of mated female , allowing increased ovulation to persist in mated female , even after ovulin is no longer detectable in the female . Therefore , males mating with previously mated females need transfer less ovulin than males mated to virgin females , presumably because it may be less necessary , as they benefit from the ovulation stimulating effect of ovulin from the prior mating . In another example , prior receipt of Acp36DE can rescue sperm storage of a male that lacks this SFP ( Avila and Wolfner , 2009; Chapman et al . , 2000 ) . The benefits to the second male described above are indirect consequences of the first male’s SFPs' effects on female’s physiology . The second male is thus the lucky beneficiary of the first male’s SFPs' actions . However , it is unknown whether a male could directly benefit from a rival’s SFPs , for example , whether the latter could associate with and improve the success of another male’s sperm . There was some suggestion that this might occur from the phenomenon of ‘copulation complementation’ ( Xue and Noll , 2000 ) , in which a female Drosophila singly-mated to a male lacking SFPs did not produce progeny unless she remated to a male who provided SFPs . That finding suggested that something from the second mating allowed the first male’s sperm to be used . However , the molecular basis for this phenomenon was unknown . The relevance of such ‘complementation’ to male reproductive fitness was strengthened by several sperm competition studies , that suggested that a male’s reproductive success could benefit from a rival’s SFPs . For example , Avila et al . , 2010 reported that the sperm of SP-null males were better at defensive sperm competition than the sperm of control males . Specifically , females mated to SP-null males and then subsequently remated to a wildtype ( wt ) competitor produced significantly more progeny from the first male ( P1 ) relative to the P1 of control males who had mated to females before the wt competitor . The higher P1 of SP-null males in this situation likely occurred because at the time of the second mating , the mates of SP-null males contained more of his sperm compared to the sperm retained from control males . This is because SP is required for efficient release of sperm from storage ( Avila et al . , 2010 ) . The higher P1 of SP-null males suggested that SP received from a second male might promote release of both his sperm and of the stored sperm from the previous SP-null male . However , this had never been tested . Here , we report that Drosophila SP received from a second male can bind to a prior male’s SP-deficient sperm and restore his fertility , including sperm release from storage and changes in the female’s behavior . We also show that although LTR-SFPs are normally required for SP to bind sperm , sperm from an SP-deficient mating can bind SP from a subsequent male , even if he lacks LTR-SFPs . This suggests that the LTR-SFPs from the first mating ‘primed’ the sperm ( or the female ) , allowing sperm-binding by subsequently-received SP . Our results reveal direct benefits that previously stored sperm from the first ( or prior ) male can receive from the second ( or last ) male’s ejaculate during the course of successive matings . Our results also establish SP as a crucial long-term molecule that facilitates this inter-ejaculate interaction , and SP-sperm binding as the molecular mechanism that underlies the reported ‘copulation complementation' ( Xue and Noll , 2000 ) in Drosophila . In matings with wt males , SP binds to sperm with which it enters the female . We wondered if sperm stored by mates of SP-null males , that lack bound SP , could become decorated with SP from a second male even if he did not provide sperm . If so , this would mean that SP from a second ( spermless ) male can bind to sperm from a prior male , already present in the female's reproductive tract ( Figure 1 . Cartoon ) . To test whether SP from a second male can bind to SP-deficient sperm stored by mates of SP-null males , we first confirmed that no SP was detectable on sperm stored in females that had singly-mated to SP-null males ( Figure 1A ) . We then examined whether SP was detected on such sperm if the female subsequently remated to a spermless male ( who provided SP ) . We observed SP bound to the stored sperm following such rematings at either 1d ( Figure 1B ) or 4d ( Figure 1C ) after the original SP-less mating . We confirmed these findings with western blotting . Sperm stored in seminal receptacles of females that had mated to SP-null males and subsequently remated to spermless males were dissected and probed for the presence of SP . Consistent with our immunostaining data , SP was detected in samples of SP-null male’s sperm from females that had remated to spermless males at 1d or 4d after the start of first mating ( ASFM; Figure 1D , lanes 7 and 8 ) . Thus , SP from a second male can bind to SP-deficient sperm stored from a prior male . To see if mating order was important , we carried out the reciprocal cross , that is , testing if SP deposited by a first male ( spermless , in this scheme ) could bind to sperm that were subsequently introduced by a second ( SP-null ) male ( Figure 2 . Cartoon ) . Spermless males transfer SP to the female tract after mating ( Kalb et al . , 1993 ) , but we did not detect any SP in females mated to spermless males by 1d after the start of mating ( ASM; Figure 1D . lane 4 ) . We saw no SP signal in samples isolated from females that had mated to spermless males , and then subsequently to SP-null males at 1d ASFM ( Figure 1D . lane 5 ) . Our immunofluorescence data were consistent with our western blots: we saw no SP-sperm binding in females that mated first with a spermless male and a day later with SP-null male ( Figure 2B ) . Therefore , if SP entered the female without sperm , it was unavailable to bind to sperm from a subsequent SP-deficient male . We hypothesized that we did not see SP bound to sperm in this second ( reciprocal ) crossing scheme because by the time of the second mating SP from the spermless male was no longer present in the female at 1d ASFM , since it could not be retained without binding to sperm ( Peng et al . , 2005 ) and no sperm were being supplied by these first males . To circumvent this , we attempted to remate females that had previously mated to spermless males as soon as 3–6 hr ASFM . However , few females remated , likely due to the recent experience of copulation , or to the effects of pheromones from the previous mating ( Shao et al . , 2019; Laturney and Billeter , 2016 ) . In the few females that did remate , no SP-sperm binding was observed ( Figure 2—figure supplement 1 ) . Since the simplest explanation for these results was that SP transferred without sperm had disappeared from females by the time of the second mating , we performed western blotting to determine how long SP persists in the reproductive tract of females in absence of sperm . We probed for SP in proteins from the SR and bursa of females at 0 min , 35 min , 1 hr , and 3 hr ASM after mating . We detected SP in the bursa protein samples at 0 min , 35 min , and 1 hr ASM . ( Figure 2D . lanes 5 , 7 , 9 ) . However , SP was undetected in bursa or seminal receptacles of females at 3 hr ASM ( Figure 2D . lane 11 , 12 ) . Thus , we could not determine whether SP from mating with a spermless male could bind a second male’s sperm , because SP received from the first mating was lost from the female reproductive tract before a second mating could occur . Xue and Noll , 2000 reported that a similar cross ( females mated first to spermless males and then to Prd males ) also gave no progeny ( showed no copulation complementation ) , which they proposed to be due to inactivation or early loss of SFPs in the absence of sperm . Our results , showing that SP can bind to stored sperm from a prior male , provide the molecular explanation for their observation . SP is needed for efficient sperm release and utilization from the female sperm storage organs ( Avila et al . , 2010 ) . We tested whether SP from a second male could restore the use of a first male’s sperm . Females mated to spermless males have no progeny ( Figure 3A ) . Females singly-mated to SP-null males have significantly reduced numbers of progeny ( Figure 3A . SE of diff = 8 . 043; p*=<0 . 001 ) relative to females mated to control males ( Figure 3A ) , likely because lack of SP prevents the increase in egg production ( Chapman et al . , 2003; Liu and Kubli , 2003; Chen et al . , 1985 ) and release of sperm from storage ( Avila et al . , 2010 ) . However , females mated to SP-null males and then remated to spermless males at 1d ( Figure 3B; p=0 . 2487 ) and 4d ( Figure 3C; p=0 . 8618 ) ASFM had progeny levels similar to those of females that had mated to control ( SP+ ) males and were subsequently remated to spermless males at the same time points . Thus , SP from a second ( SOT ) male could rescue the fertility defects that resulted from the lack of SP from an SP-null first male . Reducing the likelihood of mated females to remate is another crucial postmating response regulated by SP ( Liu and Kubli , 2003; Chen et al . , 1988 ) . Females that do not receive SP generally fail to exhibit this reluctance , and remate readily . We tested whether SP from a second male could delay the receptivity of females that had previously mated to SP-null males . Females singly-mated to SP-null males or spermless males show a significantly higher tendency to remate at 1d ASM ( Figure 3D; p=<0 . 001 ) or 4d ASM ( Figure 3E; p=<0 . 001 ) relative to females mated to wt ( CS ) males ( Figure 3D and E ) . In contrast , females mated to SP-null males and then remated to spermless males at 1d ASFM ( Figure 3D; p=0 . 43 ) showed receptivity similar to mates of control males at 1d after the start of second mating ( ASSM ) . The effect , however , did not persist as long as after a mating to a wt male . At 4d ASSM ( Figure 3E; p=<0 . 001 ) doubly-mated females exhibited higher receptivity relative to females mated to wt males but lower than those mated to spermless males . This could be either because less SP from the second ( spermless ) mating is able to bind to stored sperm from the previous mating and thus SP levels have been more depleted by 4 days ASSM than after a control mating where the sperm-SP enter the female together . Alternatively , the active portion of SP received from a rival male , bound to first male’s sperm might be released from the sperm at a higher rate . We performed western blots to determine how long SP received from the second ( spermless ) male persists in the reproductive tract of females previously mated to SP-null males . Protein was extracted , and probed for SP , from females singly-mated to CS males and those doubly-mated to SP-null males and spermless males at 1d ASFM , or at 2 hr , 1d or 4d ASM/ASSM , respectively . SP signals were detected in females mated to CS males at 2 hr , 1d or 4d ASM ( Figure 3F . lanes 3 , 4 , 5 ) . SP was detected in females mated to SP-null males and then remated to spermless males at 2 hr and 1d ASSM ( Figure 3F . lanes 6 , 7 ) but not ( or very weakly ) at 4d ASSM ( Figure 3F . lane 8 ) . Taken together , our results show that SP from a second male can rescue the receptivity defects that resulted from the first male’s of lack of SP but that sufficient SP for such an effect is not retained for as long as in a control situation ( e . g . a mating with a wt male ) . SP is also needed for release of sperm from storage within the mated female ( Avila et al . , 2010 ) . Thus , females mated to SP-null males retain significantly more sperm in their seminal receptacle at 4d ASM . To test whether SP acquired from a spermless male in a second mating could also rescue this defect , we counted sperm in storage after a single mating with SP-null; ProtB-eGFP males and after mates of SP-null; ProtB-eGFP males had remated with spermless males . As expected , females mated to control ( SP+; ProtB-eGFP ) males had fewer sperm in their seminal receptacle ( average of 192; Figure 3G and J ) relative to mates of SP-null; ProtB-eGFP males , which had significantly higher sperm counts , indicating poor release of stored sperm ( Figure 3H and J; p*=<0 . 01; average of 304 at 4d ASM ) . However , mates of SP-null; ProtB-eGFP males that had remated with spermless males retained sperm in numbers similar to those observed in females mated to control males ( average of 212; Figure 3I and J; p=ns ) . We also counted sperm stored in seminal receptacle of females mated to SP-null; ProtB-eGFP males at 5d ASM ( Figure 3J . average of 313 ) to make sure that the evident decline in sperm counts or release of stored sperm in doubly mated females ( SP-null; ProtB-eGFP mates remated to spermless males and assayed at 5d ASFM or 4d ASSM ) was not dependent on days after mating , but rather on receipt of SP from spermless males . Thus , SP from a second male can rescue the sperm release defects of prior mating to a male that lacked SP . In the experiments described above SP was provided by a spermless second male , but in nature females are much more likely to encounter a male who has his own sperm , capable of binding his SP . To test whether SP from a male with sperm can still bind to sperm from another male , we modified our experimental protocol such that females were mated to SP-null males as described earlier , but rather than spermless males , we now used ProtB-dsRed males ( Manier et al . , 2010 ) as the second male ( Figure 4I . Cartoon ) . These second males have a full suite of SFPs and sperm , and their sperm-heads are labeled with ProtB-dsRed . This allowed us to distinguish between sperm received from SP-null males ( blue heads ) and those received from ProtB-dsRed males ( red heads ) . Females were frozen at 2 hr ASSM and sperm dissected from their seminal receptacles were probed for SP . We observed anti-SP staining along the entire sperm ( head and tail ) from ProtB-dsRed males ( Figure 4B ) . Sperm received from the SP-null males ( blue heads ) were also stained with anti-SP along their length ( head and tail; Figure 4B ) . Therefore , a control ( wt ) male with a complete suite of SFPs and sperm of his own can also provide SP to bind to SP-deficient sperm from another male . The likelihood of finding an SP-null male in nature is very low . However , multiple-mating has been shown to deplete SFP reserves ( Hihara , 1981 ) , so it is possible that inter-ejaculate interaction could occur if the first male had depleted his SFP reserves . To test whether this could occur , we performed a crossing scheme in which we substituted multiply-mated control ( CS ) males with exhausted seminal reserves ( Hihara , 1981 ) for the SP-null males used in Figure 4A . We carried out western blotting to determine the levels of SP in accessory glands ( AG ) of such multiply mated ( CS ) males and the amount of SP in their mates at 2 hr ASM . We observed relatively weak SP signals in the AG of multiply-mated males ( Figure 4—figure supplement 1 . A , lane 4 ) and a very faint SP signal in females mated to these males ( Figure 4—figure supplement 1 . A , lane 5 ) compared to relatively strong SP signal in virgin ( unmated ) males and the females mated to these males ( Figure 4—figure supplement 1 . A , lanes 2 , 3 respectively ) . Our immunofluorescence data showed no ( or extremely weak ) SP-sperm binding in sperm dissected from the seminal receptacle of females mated to SFP-depleted males ( Figure 4—figure supplement 1 . C ) . Females mated to SFP-depleted CS males were then subsequently remated at 4d ASFM ( long enough to have lost any SP signal from their first multiply-mated , mates ) to ProtB-dsRed males . Sperm dissected from the seminal receptacles of these females at 2 hr ASSM were probed for SP ( Figure 4II . Cartoon ) . There was no detectable SP signal on sperm stored in females singly-mated to SFP-depleted CS males at 4d ASM ( Figure 4C ) . However , we observed anti-SP staining along the entire sperm ( head and tail ) received by the doubly-mated female from the SFP-depleted CS male ( blue heads; Figure 4D ) and ProtB-dsRed males ( red+ blue heads; Figure 4D ) . Thus , in a normal mating , the amount of SP that a male transfers is sufficient to bind not only his own sperm but also to remaining sperm from a rival . Moreover , SP from an unmated control male can bind to previously stored sperm of a male that had his SFP reserves depleted prior to mating with the female . SP binding to sperm requires the action of a network of other SFPs- ‘LTR-SFPs’ ( Ravi Ram and Wolfner , 2009; Findlay et al . , 2014 ) . Most of the known LTR-SFPs bind to sperm transiently ( CG1656 , CG1652 , CG9997 and Antares ) ( Singh et al . , 2018 ) , while others do not bind to sperm ( CG17575 or seminase; LaFlamme et al . , 2012 ) the latter facilitate the localization of other LTR-SFPs , and SP , to the seminal receptacle . However , no LTR-SFPs are detectable on sperm or in female RT at 1d ASM ( Figure 5 ) . We wondered whether LTR-SFPs were required from the second male in order to bind his SP to the first male’s sperm . We carried out experiments similar to those previously described , in which females were first mated to SP-null males and then remated to spermless males at 1d ASFM . Sperm from the seminal receptacles of these females at 2 hr ASSM were immunostained for the presence of LTR-SFPs that had been received from second ( spermless ) males . Females mated to CS males and frozen at 2 hr ASM served as positive controls for the sperm-binding of LTR-SFPsCG1656 ( Figure 5A ) , CG1652 ( Figure 5E ) and CG9997 ( Figure 5I ) . Females singly mated to SP-null males and frozen at 2 hr ASM exhibited normal sperm-binding of LTR-SFPs CG1656 ( Figure 5B ) , CG1652 ( Figure 5F ) and CG9997 ( Figure 5J ) , confirming that loss of SP affects neither the transfer nor the sperm-binding of other LTR-SFPs ( Singh et al . , 2018 ) . By 1d ASM , stored sperm from females singly-mated to SP-null males showed no signal for the LTR-SFPs , CG1656 ( Figure 5C ) , CG1652 ( Figure 5G ) and CG9997 ( Figure 5K ) , as expected given the transient sperm-binding of these proteins ( Singh et al . , 2018 ) . Thus , by the time these females remated with spermless males ( 1d ASFM ) , all known LTR-SFPs received from the first ( SP-null ) male were undetectable on sperm . Interestingly , although females that mated to SP-null males and then to spermless males showed SP signal on their sperm ( as in Figure 1 ) at 2 hr ASSM , we detected no signal of LTR-SFPs , CG1656 ( Figure 5D ) , CG1652 ( Figure 5H ) and CG9997 ( Figure 5L ) on those sperm at 2 hr ASSM . This could be because LTR-SFPs from the second male could not enter the sperm storage organs in the absence of sperm or , alternatively , that their binding sites on sperm had been modified prior to the second mating ( perhaps by the action of LTR-SFPs received from the first mating ) to make them incapable of binding . We verified these observations with western blots . Consistent with the immunofluorescence data in Figure 5A–L , LTR-SFP signals for CG1656 , CG9997 , CG1656 and Antares were detected in sperm dissected from females mated to CS and SP-null males at 2 hr ASM ( Figure 5M . lanes 3 , 4 ) . No LTR-SFP signals were detected in sperm dissected from females mated to SP-null males at 1d ASM ( Figure 5M . lane 5 ) or in sperm dissected from females mated to SP-null males , remated to spermless males at 1d ASFM , and frozen 2 hr ASSM ( Figure 5M . lane 7 ) . However , as expected SP signals were detected in sperm dissected from females that mated to SP-null males , remated to spermless males at 1d ASFM and frozen 2 hr ASSM ( Figure 5M . blot probed for SP , lane 7 ) . Thus , sperm no longer detectably bind new LTR-SFPs after they have bound LTR-SFPs from their own ( SP-null ) male . That LTR-SFPs are needed for SP-sperm binding , and that SP from spermless male binds the first male’s sperm , further suggests that the first male’s sperm ( or the female RT ) had already been primed with its own LTR-SFPs during storage in the female tract . Unlike the four LTR-SFPs assessed above , the two other LTR-SFPs , CG17575 and seminase , do not bind to sperm , yet are crucial for SFP-sperm binding . In the absence of CG17575 or seminase , SP fails to bind to sperm ( Ravi Ram and Wolfner , 2009; LaFlamme et al . , 2012 ) . To determine if these proteins were required for a second male’s SP binding to a first male’s sperm , we first crossed females to SP-null males and then to CG17575-null or seminase-null males at 1d ASFM ( Figure 6 . Cartoon ) . In this situation , CG17575 and seminase had entered the female with the first male’s sperm , but by the time of the second mating , were undetectable in the female ( Figure 6—figure supplement 1 ) . We examined whether in this situation SP transferred by CG17575-null ( or seminase-null ) males would still bind to the SP-null sperm stored in the female . We made use of ProtB-eGFP labeled SP-null males to differentiate between sperm received from first ( cyan ( DAPI+ eGFP ) sperm heads ) and second ( blue ( DAPI ) sperm heads ) males . Immunostaining and western blots for detection of SP on sperm dissected from females mated to SP-null; ProtB-eGFP males and then remated to seminase-null ( Figure 6 . A and C , lane 4 ) or CG17575-null ( Figure 6 . B and C , lane 5 ) males showed that SP received from the second male bound to sperm ( head and tail ) received from SP-null; ProtB-eGFP males . Sperm dissected from females singly-mated to SP-null; ProtB-eGFP males gave no SP signal , as expected ( Figure 6 . D , lane 3 and E ) and sperm dissected from females singly-mated to seminase-null ( Figure 6 . D , lane 4 and F ) or CG17575-null ( Figure 6 . D , lane 5 and G ) males also showed no SP-sperm binding , as expected , due to lack of the LTR-SFP . Therefore , sperm no longer require even CG17575 or seminase from the second male’s ejaculate , after they have received the LTR-SFPs from their own ( SP-null ) male . Xue and Noll , 2000 reported that sperm transferred to females by Prd mutant males that lack the entire suite of SFPs were capable of fertilizing a few eggs to yield progeny , but only after the females were subsequently remated to spermless males . They coined the term ‘copulation complementation’ to describe this phenomenon , and proposed that SFPs from the second male might interact with the first male’s sperm to yield this result . Consistent with this idea , several reports suggested that first-males that provided sperm but lacked particular SFPs ( Avila et al . , 2010; Ravi Ram and Wolfner , 2007b; Mueller et al . , 2008; Wong et al . , 2008; Hopkins et al . , 2019 ) can have higher paternity shares in competitive situations: they were better competitors , compared to control males , in defensive sperm competition assays ( Avila et al . , 2010; Avila and Wolfner , 2009; Fricke et al . , 2009 ) . These reports align and are consistent with our observations that SP from a second male can bind to and assist the sperm from a prior SP-deficient male . A simple explanation for these results , based on the findings that we report here , is that the deficiency of these SFPs in the first male led to impaired release/use , and thus retention , of his sperm , and this was rescued by receipt of the second male’s SFPs , as we have shown here , for SP . SP is the only SFP thus far known to persist within the Drosophila female ( for 10–14 days post-mating ) , eliciting long-term post mating responses through gradual release of its C-terminal portion ( Peng et al . , 2005 ) . The long-term persistence of SP on sperm made it an excellent candidate to examine for interaction with rival sperm . Here , we report that SP subsequently received from a spermless male binds to a first male’s sperm ( SP-null ) . This association is apparent even if the second mating occurs at 1d or as long as 4d ASFM , indicating that binding of SP to the first male’s sperm occurs irrespective of how long sperm have been in the storage organs . It remains unclear how SP received from spermless ( second ) male enters the sperm storage organs , where sperm from the first mating had been stored . However , Manier et al . , 2010 reported that 60–90 min after the start of a second mating , 26% of the resident sperm ( received from the previous mating ) are moved from storage back into the bursa where they mix with the second male’s ejaculate before moving back into the storage . Therefore , it is possible that SP received from the spermless male binds to the first male’s sperm that relocated to the bursa , and the newly SP-bound sperm are then transferred back into storage in the seminal receptacle . In the absence of sperm , or if SP is not bound to sperm , females do not maintain post-mating responses and fail to efficiently release sperm from storage resulting in fewer sperm available for fertilization and fewer progeny ( Avila et al . , 2010; Ravi Ram and Wolfner , 2009; Chapman et al . , 2003; Liu and Kubli , 2003 ) . We observed that these defects were rescued when SP was received by females in a remating with spermless males . Thus , the second male’s SP bound to the first male’s sperm is functional . The rescue of the phenotype , however , was not as long lasting as in a normal single mating with SP transfer , wearing off by 4d postmating rather than the normal ~10 d . This could be because only fewer sperm relocated from storage to the bursa ( Manier et al . , 2010 ) , so they may not carry sufficient SP back into storage to associate with SP-null sperm . Consistent with this , the levels of SP that we see stained in these situations are lower than those in a wild type mating . We did not know whether the amount of SP that is transferred during mating is more than the available binding sites on sperm . Here , we observed that an unmated control male does transfer enough SP to bind his own as well as pre-stored sperm ( SP-null ) in a previously mated female . Consistent with our findings , several reports suggest that in response to potential threats of sperm competition and conflicts , males adjust the levels of SFPs and transfer high amounts of SP , yet less ovulin , to previously mated females ( Wigby et al . , 2009; Sirot et al . , 2011 ) . Rubinstein and Wolfner , 2013 demonstrated that ovulin induces ovulation , acting through octopamine ( OA ) neuronal signaling and increases the number of synapses that the female’s Tdc2 neurons make on the musculature of the oviduct . Persistence of this latter effect could benefit rivals too , so second-mating males may thus be able to mitigate the levels of ovulin in their ejaculate . But the question remains that if SP from one male’s ejaculate can bind to and assist another’s sperm , why do males not lower the amount of SP transferred while mating ? A potential explanation is that a male would still benefit by transferring enough SP to ensure that his own sperm remains saturated with SP , even at a cost of part of his SP binding to another male’s sperm . SP binds to sperm through its N-terminal region , and this region remains bound to sperm long-term ( Peng et al . , 2005 ) . The bound N-terminal region of SP on sperm stored in a mated female does not allow any further binding of SP coming from rival male’s ejaculate . Therefore under what circumstances might SP-mediated copulation complementation occur in nature ? In polygynous males , SFPs are depleted faster than sperm ( Hihara , 1981 ) . This could result in a situation in which a female who mated with a male with low levels of SFPs might not receive enough SP to saturate his sperm . In these circumstances , SP received from another male would help compensate for the lower amount of SP from the depleted first male’s ejaculate . SFP depletion would , of course , not only affect the levels of SP , but also all the other crucial LTR-SFPs . However , while other LTR-SFPs enable SP to bind sperm , it is the quantity of bound SP that correlates with the duration of post-mating responses . In line with this hypothesis , we subjected control males to recurrent matings ( providing six virgins over the span of 2 days ) , with an intent to exhaust their SFPs . We observed that sperm stored by subsequent ( 7th ) females mated to these multiply-mated males had undetectable SP signals . However , when these females were remated to unmated control males , strong SP signals were detected on both the SFP-depleted sperm received from the previous mating and the newly received rival sperm . Therefore , our results support the idea that in nature males who have multiply-mated might get some help from the SFPs of subsequent , less depleted , males . Interestingly , this inter-ejaculate interaction might also confer an advantage to the second male . More of the second male’s SP will be retained in the female reproductive tract , for even longer , if it binds to previously-stored sperm in addition to his own sperm . This could allow the post-mating responses in polyandrous females to be maintained for longer than in singly-mated females . Binding of SP to sperm is facilitated by a network of LTR-SFPs ( Ravi Ram and Wolfner , 2009 ) . Two LTR-SFPs , CG17575 and seminase , do not themselves bind to sperm , whereas other LTR-SFPs bind sperm transiently ( CG1652 , CG1656 , CG9997 , antares ) . CG17575 and seminase localize the other LTR-SFPs , and SP , to sperm storage organs ( Ravi Ram and Wolfner , 2009; Singh et al . , 2018; LaFlamme et al . , 2012; Ravi Ram and Wolfner , 2007a; Ravi Ram et al . , 2006 ) . We found that SP from a second male ( spermless or control ) can associate with sperm from the first male ( SP-null ) even if it enters the female in absence of its own LTR-SFPs . This suggested that SP-null sperm ( or the mated female RT ) had already received modifications ( ‘priming’ ) from its own LTR-SFPs that were required for SP binding . This further suggests that once primed , a sperm can bind SP from a rival’s ejaculate without the need for additional LTR-SFPs , and can restore its own post-mating dynamics . Thus , we find that a critical SFP from one male can associate and offer direct benefits to sperm from another male , restoring the SP function to the previously stored sperm . Our work shows that SP is a crucial candidate for copulation complementation in Drosophila , and that sperm in storage ( or the female RT ) are primed for SP binding by the first male’s LTR-SFPs . Therefore , despite potential competition between males , there could be subtle cooperation between males as well . In addition , the allocation of resources by , and effects on , rival males that mate to polyandrous females , should be viewed in light of not only sexual conflicts , but also both direct and indirect effects of SFPs . Additionally , our findings raise some intriguing questions for further study . First , our experiments , like those of Xue and Noll , 2000 , were no-choice situations . SP from the second male had the opportunity to bind sperm only from one prior ( SP-null ) male . It will be interesting in the future to determine whether SP from a later male shows any preference to bind sperm from a more-related male , relative to sperm from a less-related one . Second , if there is such preference , its molecular mechanism is currently unknown and would be an important topic for further elucidation; we do not know what mechanisms distinguish self- from non-self sperm – whether molecular , temporal , or both . Third , although our data show that a male can benefit from the SP of a subsequent male , whether this is a true cooperation , or rather an accident of there being sufficient SP from the second male to bind to SP-deficient sperm in the female is unclear . It is possible that it is advantageous for a male to transfer large amounts of SP so as to coat his own sperm efficiently , even if this has the unintended consequence of there being sufficient SP to also bind to ( and benefit ) an SP-deficient rival’s sperm . Alternatively , if related males are mating in proximity to each other , there may have been selection for such SP binding from a rival , if sperm from a male who was depleted of SFPs by prior mating bound SP that was likely from his relative . Each of these will be an intriguing topic for future investigation . Spermless males , [sons of tudor , ( SOT ) that lack sperm but produce and transfer a complete suite of SFPs] were the progeny of bw sp tud1 females ( Boswell and Mahowald , 1985 ) mated to control , Canton S ( CS ) males . Sex peptide null mutant males ( Δ325/Δ130; which have sperm and the entire suite of SFPs except for SP ) ( Liu and Kubli , 2003 ) were generated by crossing the SP knockout line ( Δ325/TM3 , Sb ry ) to a line carrying a deficiency for the SP gene ( Δ130/TM3 , Sb ry ) . Control males were the TM3 siblings of SP-null mutants . Matings were conducted with wild type D . melanogaster females ( CS ) . To determine sperm numbers , we generated a line carrying the SP-null mutation and Protamine B-eGFP tagged sperm ( ProtB-eGFP/Y; Δ325/Δ130 ) by series of crosses between the SP knockout line ( Δ325/TM3 , Sb ry ) and ProtB-eGFP ( X ) ; TM3/TM6 ( Manier et al . , 2010 ) . The TM3 siblings of these males , ( SP+; ProtB-eGFP ) served as controls . Sperm-heads of these control males were tagged with ProtB-eGFP , but the males had normal levels of SP ( Figure 3—figure supplement 1 ) . ProtB-ds Red males with Protamine B-dsRed tagged sperm heads ( Manier et al . , 2010 ) served as additional controls . All flies were reared under a 12:12 hr light-dark cycle at 22 ± 1°C on standard yeast-glucose medium . Mating experiments were carried out by single-pair mating 3–5 day old virgin CS females to 3- to 5-day-old unmated males of genotypes indicated in the text and remating the same female 1 day or 4 days after the start of first mating ( ASFM ) to age matched unmated males of the genotypes indicated in the text . Xue and Noll , 2000 reported copulation complementation in females mated to Prd males ( which produce sperm but lack SFPs ) remated to spermless males that produce SFPs . We followed a similar scheme but to focus on SP specifically , we used SP-null males as the first male . As described in Results , we then remated these females to spermless males , which make SFPs but not sperm . We attempted to do the reciprocal experiment , where females were mated to spermless males and then remated to SP-null males , but consistent with what was reported by Xue and Noll , 2000 , we could not detect copulation complementation in this direction for technical reasons: SP from the spermless male did not persist long enough in the mated female to interact with the second male’s sperm ( see Results ) . We carried out rematings at three time points , 3–6 hr , 1d , and 4d AFSM . We assessed results at 2 hr after the start of the second mating ( ASSM ) . The reproductive performances of singly-mated or doubly-mated females were assayed by analyzing fertility ( numbers of progeny eclosed over ten days ) ( Kalb et al . , 1993 ) . Briefly , the fertility assays were carried out with ( A ) . ‘Single matings’: Females were singly mated to ( i ) spermless males , ( ii ) SP-null males , or their ( iii ) TM3 siblings ( genetically-matched control males ) in three individual sub-batches , and ( B ) . ‘Rematings’: Females were mated to SP-null males or their TM3 siblings ( SP+ ) and were then subsequently remated to spermless males at 1d and 4d ASFM . Matings that lasted 15 mins or more were considered successful . At the end of a mating , males were removed from the vials and females were allowed to lay eggs for 10 days after the start of mating ( ASM ) in the first batch and after the start of second mating ( ASSM ) in the second batch . Females were transferred to fresh food vials every 3 days . Flies emerging from each vial were counted . Fertility is represented as total number of progeny produced by each female over a period of 10 days . The differences in fertility were analyzed through one-way Analysis of Variance ( ANOVA ) followed by Tukey’s multiple comparison tests for single-matings and Mann Whitney U-tests for rematings . All assays were repeated more than two times and comprised of two technical replicates , with each group consisting of a minimum sample size of 15–20 . To determine the propensity of females to remate , receptivity assays ( Chapman et al . , 2003 ) were set for females singly mated to SP-null , spermless or CS males and females mated to SP-null males and then subsequently remated to spermless males at 1d ASFM . For the assay , females from singly-mated and doubly-mated groups were then provided with ( CS ) males at 1d and 4d ASM or ASSM , respectively . We determined the number of females that mated within 1 hr from when the CS male was introduced within the vial . Assays were repeated more than two times , with each group consisting of a minimum sample size of 15–20 . The data were analyzed by Fisher exact tests and Chi-squared group analyses . To study the effect of first male’s sperm and rival male’s SP binding on sperm utilization and release , we generated SP-null males whose sperm-heads are labelled with ProtB-eGFP ( Manier et al . , 2010 ) . Females were mated to SP-null; ProtB-eGFP or SP+; ProtB-eGFP ( control ) males . Some of the mated females were frozen at 4d ASM ( or 5d ASM ) for sperm counts . The remaining mates of SP-null; ProtB-eGFP males were remated to spermless males at 1d ASFM . These flies were frozen at 4d ASSM . Subsequently , seminal receptacles of females singly-mated to SP-null; ProtB-eGFP and SP+; ProtB-eGFP , or doubly-mated to SP-null; ProtB-eGFP and spermless males , were dissected and eGFP sperm were counted ( at a total magnification of 200X , with FITC filter on an Echo-Revolve microscope ) . Mature sperm in the seminal receptacles of mated females were counted twice and groups were blindfolded to ensure reproducibility and avoid bias . The percent repeatability was 90–94% . Assays were repeated more than two times , with two technical replicates . Every group contained a minimum sample size of 15–25 . Differences in the sperm counts between groups were analyzed statistically through one-way ANOVA followed by Tukey’s multiple comparison tests . Control ( CS ) males were subjected to brood matings ( Misra et al . , 2014; Gilchrist and Partridge , 1995 ) to deplete SFPs , as their levels are known to become exhausted at a higher rate than sperm numbers ( Hihara , 1981 ) . Briefly , 3-day-old control males were mated to CS females in two broods ( each consisting of three virgin females ) over 2 days . The first mating of both broods was observed . On the third day , previously mated females were removed and the male was provided with an additional virgin female ( 7th mate ) , matings were observed and depleted CS males were removed . Half of the 7th mated females were frozen at 4d ASM , while the others were subsequently remated to control ( ProtB-dsRed ) males at 4d ASFM , and then frozen at 2 hr ASSM . Sperm stored in the seminal receptacle of the frozen flies were dissected and immunostained for SP . Immunofluorescence was performed to detect SP-sperm binding ( Ravi Ram and Wolfner , 2009; Peng et al . , 2005; Singh et al . , 2018 ) . Sperm dissected from seminal receptacles of experimental or control females were attached to poly-L-Lysine ( Sigma ) coated slides . Sample processing was carried out according to the protocol of Ravi Ram and Wolfner , 2009 with minor modifications . Samples were blocked with 5% bovine serum albumin ( BSA ) in 1X PBS for 30 min . Subsequently , samples were incubated overnight in rabbit anti-SP ( 1:200 ) , CG1656 ( 1:100 ) , CG1652 ( 1:50 ) , CG9997 ( 1:50 ) ( Singh et al . , 2018 ) , in 0 . 1% BSA at 4°C overnight . Samples were then washed in PBS and incubated at room temperature for 2 hr in goat anti-rabbit IgG coupled to alexa fluor 488 ( green ) or 594 ( red; Invitrogen ) at a concentration of 1:300 in 1x PBS at room temperature in the dark . Samples were then washed in PBS , incubated in 0 . 01% DAPI for 3 min at room temperature in the dark , rewashed and mounted using antifade ( CitiFluor mountant solution; EMS ) . The fluorescence was visualized under an Echo-Revolve fluorescence microscope at a magnification of 200X . A minimum of three independent immunostaining batches , with a minimum sample size of 10 , were analyzed for each group . To further examine transfer , persistence or binding of SP to sperm stored in singly-mated or doubly-mated females , the lower reproductive tract ( RT ) or sperm stored ( SS ) in seminal receptacles of mated female were dissected . The dissected tissues ( lower RT , n = 5–10 or sperm , n = 20–30 ) were suspended in 5 µl of homogenization buffer ( 5% 1M Tris; pH 6 . 8 , 2% 0 . 5M EDTA ) and processed further according to the protocol of Ravi Ram and Wolfner , 2009 . Proteins from stored sperm or lower female reproductive tract were then resolved on 12% polyacrylamide SDS gel and processed further for western blotting . Affinity purified rabbit antibodies against SP ( 1:2000 ) , CG1656 ( 1:1000 ) , CG1652 ( 1:500 ) , antares ( 1:500 ) , CG9997 ( 1:1000 ) , CG17575 ( 1:1000 ) , seminase ( 1:1000 ) ( Ravi Ram and Wolfner , 2009; Singh et al . , 2018; LaFlamme et al . , 2012 ) and mouse antibody against actin ( as a loading control; Millipore Corp . , cat no . #MAB1501 at 1:3000 ) were used as primary antibodies . HRP conjugated secondary anti-rabbit and anti-mouse antibodies ( Jackson Research ) were used for detection of SFPs at a concentration of 1:2000 .	When fruit flies and other animals reproduce , a compatible male and a female mate , allowing sperm from the male to swim to and fuse with the female’s egg cells . The males also produce proteins known as seminal proteins that travel with the sperm . These proteins increase the likelihood of sperm meeting an egg and induce changes in the female that increase the number , or quality , of offspring produced . Some seminal proteins help a male to compete against its rivals by decreasing their chances to fertilize eggs . However , since many of the changes seminal proteins induce in females are long-lasting , it is possible that a subsequent male may actually benefit indirectly from the effects of a prior male’s seminal proteins . It remains unclear whether the seminal proteins of one male are also able to directly interact with and help the sperm of another male . Male fruit flies make a seminal protein known as sex peptide . Normally , a sex peptide binds to the sperm it accompanies into the female , increasing the female’s fertility and preventing her from mating again with a different male . To test whether the sex peptide from one male can bind to and help a rival male’s sperm , Misra and Wolfner mated female fruit flies with different combinations of males that did , or did not , produce the sex peptide . The experiments found that female flies that only mated with mutant males lacking the sex peptide produced fewer offspring than if they had mated with a ‘normal’ male . However , in females that mated with a mutant male followed by another male who provided the sex peptide , the second male’s sex peptide was able to bind to the mutant male’s sperm ( as well as to his own ) . This in turn allowed the mutant male’s sperm to be efficiently used to sire offspring , at levels comparable to a normal male providing the sex peptide . These findings demonstrate that the ways individual male fruit flies interact during reproduction are more complex than just simple rivalry . Since humans and other animals also produce seminal proteins comparable to those of fruit flies , this work may aid future advances in human fertility treatments and strategies to control the fertility of livestock and pests , including mosquitoes that transmit diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology", "genetics", "and", "genomics" ]	2020	Drosophila seminal sex peptide associates with rival as well as own sperm, providing SP function in polyandrous females
Development and function of highly polarized cells such as neurons depend on microtubule-associated intracellular transport , but little is known about contributions of specific molecular motors to the establishment of synaptic connections . In this study , we investigated the function of the Kinesin I heavy chain Kif5aa during retinotectal circuit formation in zebrafish . Targeted disruption of Kif5aa does not affect retinal ganglion cell differentiation , and retinal axons reach their topographically correct targets in the tectum , albeit with a delay . In vivo dynamic imaging showed that anterograde transport of mitochondria is impaired , as is synaptic transmission . Strikingly , disruption of presynaptic activity elicits upregulation of Neurotrophin-3 ( Ntf3 ) in postsynaptic tectal cells . This in turn promotes exuberant branching of retinal axons by signaling through the TrkC receptor ( Ntrk3 ) . Thus , our study has uncovered an activity-dependent , retrograde signaling pathway that homeostatically controls axonal branching . Intracellular transport is an essential process in cell growth , maintenance , and inter- and intracellular signaling . This is especially apparent in highly polarized cells like neurons that are composed of complex dendrites and a long axon responsible for impulse propagation . Most of the proteins , mRNAs , and organelles required for cellular growth and function are produced in the cell body and must , therefore , be moved down the axon to the synaptic terminals . Microtubules serve as main longitudinal cytoskeletal tracks in axons , and it is well established that microtubule stabilization is a landmark of early axonal development that is sufficient to induce axon formation in vivo . Besides , microtubule stabilization alone can even lead to the transformation of mature dendrites into axons in differentiated neurons ( Gomis-Ruth et al . , 2008; Witte and Bradke , 2008; Witte et al . , 2008 ) . Along the axonal microtubule cytoskeleton molecular motors of the kinesin and dynein superfamily act as main transport molecules ( Hirokawa , 1998; Karki and Holzbaur , 1999; Vale , 2003 ) . Although it is well established that anterograde molecular motors are essential for synapse generation and function ( Okada et al . , 1995 ) , little is known about their exact role in neural circuit establishment in vivo . Of the kinesin superfamily , which comprises 45 members in mammals ( Miki et al . , 2001 ) and many more in zebrafish , the kinesin I subclass plays an especially prominent role in neuronal function . The kinesin motors mediate the plus end directed transport of cargo proteins along microtubules and are composed of two identical heavy chains and two identical light chains ( Hirokawa et al . , 2010 ) . In the mammalian genome , three kinesin I heavy chain genes are present: Kif5A , Kif5B , and Kif5C . While Kif5B is ubiquitously expressed , Kif5A and Kif5C are neuron-specific ( Xia et al . , 1998 ) . Their cargoes include voltage-gated potassium channels , AMPA receptor GluR2 , GABAA receptors , sodium channels , neurofilaments , and mitochondria ( Rivera et al . , 2007; Uchida et al . , 2009; Twelvetrees et al . , 2010; Karle et al . , 2012; Su et al . , 2013; Barry et al . , 2014 ) . In humans , Kif5A mutations have been implicated in a heterogenous group of neurodegenerative disorders , including a form of Hereditary Spastic Paraplegia characterized by slowly progressive lower limb paralysis ( Goldstein , 2001 ) and Charcot Marie Tooth Type 2 , a peripheral axonal neuropathy ( Crimella et al . , 2012 ) . In this study , we have taken a combined genetic , molecular biological , and in vivo imaging approach in developing zebrafish larvae to investigate the role of anterograde intracellular transport in the development of connections between retina and tectum . By TALEN-mediated gene disruption , we generated a kif5aa loss-of-function allele and could show that kif5aa mutant fish display a de-synchronisation of retinal axon and tectal growth . A delay in tectal innervation by mutant retinal ganglion cell ( RGC ) axons is followed by a period of exuberant branching , resulting in enlarged axonal arbors . GCaMP imaging revealed that kif5aa mutant RGCs do not transmit signals from the retina to their postsynaptic partner cells . Utilizing two additional zebrafish mutant lines with defects in RGC formation or function , lakritz ( Kay et al . , 2001 ) and blumenkohl ( Smear et al . , 2007 ) , or specific silencing of RGCs by expression of botulinum toxin light chain B ( BoTxLCB ) , we show that in all four cases the reduction of presynaptic activity leads to increased expression of neurotrophic factor 3 ( Ntf3 ) . The overabundance of this neurotrophin causes excessive branching by RGC axons . Thus , a defect in Kif5aa-mediated axonal transport has unmasked a homeostatic mechanism that adjusts axon arbor growth to levels of synaptic activity and depends on Ntf3 signaling . To investigate the role of axonal transport in visual-system development , we generated a series of insertion and deletion ( indel ) mutations in the open reading frame ( ORF ) of the zebrafish anterograde motor protein Kif5aa by targeted gene disruption using transcription activator like nucleases ( TALENs ) . Subsequently , we isolated two alleles with a 10 base pair ( bp ) and 13 bp deletion , respectively ( Figure 1A ) , both resulting in a frameshift in the ORF from amino acid ( aa ) 122 onwards . This frameshift leads to a premature stop codon at position 162 of 1033aa ( kif5aa162 ) of the wild-type full-length protein ( Figure 1A ) . In situ hybridization showed a strong expression of kif5aa in RGCs . In contrast , kif5aa mutant mRNA is dramatically down regulated in embryos starting from 24 hours post fertilization ( hpf ) onwards , probably by nonsense-mediated decay ( Figure 1B ) . By quantitative reverse transcription PCR ( qRT-PCR ) ( Figure 1C ) , we could see a 47% decrease of kif5aa transcript levels in mutant embryos compared to controls . Both alleles were not complementing each other , and the mutant phenotype co-segregated with the TALEN-induced mutation over three consecutive generations . This indicates that the genomic targeting was specific and argues for the absence of off-target effects of the TALEN pair used . 10 . 7554/eLife . 05061 . 003Figure 1 . Generation of loss-of-function alleles of the anterograde motor protein Kif5aa . ( A ) Employing TALENs targeting exon4 of the kif5aa open reading frame , we generated two loss-of-function alleles with a 10 bp and 13 bp deletion , respectively . These result in a frameshift at amino acid 122 and a premature stop codon after 162 of 1033aas within the motor domain of Kif5aa . L = linker region , N = neck region . ( B ) In situ hybridization shows a substantial downregulation of kif5aa mRNA in 24 hpf and 72 hpf old embryos . Scale bars ( from left to right ) = 150 μm , 100 μm , 50 μm . RGCL = Retinal Ganglion Cell layer . ( C ) Quantitative reverse transcription PCR confirms that only 47% of wild-type kif5aa mRNA expression levels are reached in homozygote mutant embryos at 4 dpf ( p < 0 . 01 ) . ( D ) Kif5aa mutant embryos show expanded melanosomes within their melanocytes and appear dark compared to wild-type embryos . Scale bars = 200 μm . ( E ) They fail to inflate their swim bladder and die 10 days post fertilization . Scale bars = 400 μm . Arrow: pointing at the respective location of the swim bladder . SB = swim bladder . DOI: http://dx . doi . org/10 . 7554/eLife . 05061 . 00310 . 7554/eLife . 05061 . 004Figure 1—figure supplement 1 . Melanosomes transport is not abolished in kif5aa mutants but they show no optokinetic response . ( A ) Phenotype of wild-type , kif5aa162 mutant , lakritz and blumenkohl embryos at 5 dpf . All three mutants show expanded melanosomes and appear dark . Application of norepinephrine ( NA ) results in aggregation of melanosomes . Scale bar = 200 μm . ( B ) Application of NA leads to aggregation of melanosomes in kif5aa mutant embryos; washing out of the drug results in their re-expansion . Scale bar = 200 μm . ( C ) Schematics illustrating the organization of the optic tectum in wild-type , lakritz and blumenkohl embryos . The tectum is subdivided in multiple sublaminae . Retinal Ganglion Cells ( RGC ) axonal arbors grow into distinct layers within the tectal neuropil where they form functional synaptic connections with dendrites of periventricular neurons ( PVNs ) . Lakritz mutant embryos lack all retinal input as they fail to specify RGCs . Blumenkohl mutants grow RGCs with increased axonal arbor sizes . The rate of synaptic transmission between RGCs and periventricular neurons is reduced . SO = stratum opticum , SFGS = stratum fibrosum et griseum superficiale , SGC = stratum griseum central , SAC = stratum album central , SPV = stratum periventriculare . ( D ) Kif5aa162 mutant embryos do not show an optokinetic response . Eye positions ( angles relative to a horizontal axis ) were plotted over time during optokinetic stimulation in one direction . During the first 60 s , no stimulus is shown followed by 60 s of motion stimulation . The OKR has a sawtooth profile in siblings , consisting of alternating quick and slow phases while kif5aa162 mutants do not show any response to the stimulus ( n = 6 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05061 . 004 In a clutch of 5 day post-fertilization ( dpf ) zebrafish larvae , derived from a cross of two heterozygous carriers of the kif5aa162 allele , 25% of the larvae exhibited a dark coloration in comparison to their wild-type siblings ( Figure 1D ) , indicating that the mutation is recessive , completely penetrant and results in a failure to adapt to a light background by melanosome re-distribution . This phenotype is frequently observed in visually defective mutants , specifically in those with RGC impairments ( Neuhauss et al . , 1999; Muto et al . , 2005 ) . For example , the lakritz ( Kay et al . , 2001 ) and the blumenkohl ( Smear et al . , 2007 ) mutations affect both vision and are darkly pigmented ( Figure 1—figure supplement 1A ) . The former represents a mutation in the basic helix-loop-helix transcription factor atonal homolog 7 ( atoh7 ) , which is required for RGC fate specification ( Kay et al . , 2001 ) . Loss of Atoh7 in zebrafish leads to a complete absence of RGCs in the retina and consequently no functional connections between the retina and other brain areas are established . Nevertheless , as Atoh7 is solely expressed in the retina , no further developmental defects are described and lakritz mutant fish develop normally apart from their complete blindness ( Kay et al . , 2001 ) . Blumenkohl mutants fail to produce a functional vesicular glutamate transporter , vglut2a , the main vesicular glutamate transporter expressed in zebrafish RGCs . The lack of functional Vglut2a leads to reduced synaptic transmission between the retina and the optic tectum . Furthermore , it was described that RGC axons consequently develop increased axonal arbors and show aberrant branching upon innervation of the optic tectum ( Figure 1—figure supplement 1A , C ) . To test visual system function in kif5aa mutant larvae , we employed the optokinetic response ( OKR ) to a moving grating as a sensitive and quantifiable indicator of visual functions in zebrafish ( Brockerhoff et al . , 1995 ) . We found that the OKR was absent in kif5aa mutants ( n = 6 ) , while it was present in all wild-type fish examined ( n = 6 ) ( Figure 1—figure supplement 1D ) . This suggests that the disruption of kif5aa causes blindness . Unlike lakritz and blumenkohl , which are viable , kif5aa mutants fail to inflate their swim bladder ( Figure 1E ) and die around 10 days post fertilization ( dpf ) . As a dynamic interaction between actin- and tubulin-based motility controls melanosome transport within melanocytes ( Evans et al . , 2014 ) , we wanted to rule out the possibility that the dark pigmentation is caused by a melanophore-autonomous defect . Kif5aa mutant embryos were treated with norepinephrin resulting in aggregation of melanosomes and consecutive re-expansion upon washing out of the drug ( [Wagle et al . , 2011] , Figure 1—figure supplement 1B ) . Similar results were obtained for lakritz and blumenkohl ( Figure 1—figure supplement 1A ) . This indicates that melanosome transport inside melanocytes in both antero- and retrograde direction is not affected in kif5aa mutant embryos . To examine if visual system defects were caused by retinal neurogenesis or patterning defects during development , we compared the expression of known marker genes between wild-type and kif5aa deficient embryos . In mutant embryos , we could not observe alterations neither in the onset of retinal neurogenesis ( characterized by sonic hedgehog expression [Shkumatava et al . , 2004] ) nor in the later specification of RGCs or other retinal cell types ( Figure 2—figure supplement 1 ) . Choroid fissure and optic stalk formation , as well as rostral-caudal patterning , were normal as revealed by the expression of pax2 . 1 and tag-1 , respectively . Further , cell type specific marker analysis revealed that all major retinal cell types were present and the layering of the retina was not affected in kif5aa mutant retinae ( Figure 2—figure supplement 1 ) . As retinal morphology and cellular composition were not altered by the loss of kifaa function , we decided to analyze the outgrowth of RGC axons from the retina and their retinotopic mapping onto the optic tectum . To this purpose , we introduced the transgenic line Tg ( pou4f3:mGFP ) ( Xiao et al . , 2005 ) , which labels a subpopulation of RGCs , into the kif5aa162 mutant background and imaged optic nerve outgrowth and optic chiasm formation at 48 hpf . No misrouting of RGCs to the ipsilateral side could be observed , and the optic chiasm was correctly established at the right developmental time ( Figure 2A ) . To confirm that pathfinding was unaffected , we used Zn5 antibody staining to bulk-label outgrowing RGC axons ( Fashena and Westerfield , 1999 ) ( Figure 2A ) . Axon tracing at later stages of development , following injection of the two lipophilic dyes DiO and DiI ( Baier et al . , 1996 ) into opposite quadrants of the contralateral eye , revealed that the retinotectal projection was correctly patterned in mutant embryos ( Figure 2B ) and no misrouting to the ipsilateral hemisphere occurred . 10 . 7554/eLife . 05061 . 005Figure 2 . Outgrowth of the optic nerve and retinotopic mapping is normal in kif5aa mutants . ( A ) Confocal imaging of the Tg ( pou4f3:mGFP ) transgene , labeling a subpopulation of Retinal Ganglion Cells ( RGCs ) with membrane bound GFP , at 48 hpf reveals that outgrowth of the optic nerve formed by RGC axons from the retina is not affected by the kif5aa mutation . Immunostaining against the Zn5 antigen ( DM- GRASP/neurolin present within the visual system only on RGCs [Laessing and Stuermer , 1996; Fashena and Westerfield , 1999] ) confirms that optic chiasm formation is normal ( marked with an asterisk ) . No pathfinding errors occur at this level of axonal growth . Scale bars = 200 μm . Embryos facing upwards . R = rostral , C = caudal . ( B ) Injections of the lipophilic dyes DiI and DiO in different quadrants of the contralateral retina ( depicted in the right panel ) show that retinotopic mapping to the optic tectum is performed in the correct manner . Asterisks = pigment cells in the skin . D = dorsal , V = ventral , R = rostral , C = caudal . No misrouting of RGC axons to the ipsilateral tectum was observed ( data not shown ) . Scale bars = 150 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05061 . 00510 . 7554/eLife . 05061 . 006Figure 2—figure supplement 1 . Patterning of the mutant retina and neurogenesis is not affected in mutants . ( A ) The expression of the Tg ( shh:eGFP ) transgene marks the onset of neurogenesis ( white arrow ) in the developing retina ( Shkumatava et al . , 2004 ) and its expression is not altered in kif5aa mutant retinae . ( B ) RGC differentiation marked by Tg ( pou4f3:mGFP ) ( Xiao et al . , 2005 ) expression is not affected . ( C ) The glycosylphosphatidyl inositol ( GPI ) -anchored protein of the immunoglobulin ( Ig ) superfamily Tag-1 I is expressed in nasal RGCs ( Lang et al . , 2001 ) . In situ staining shows that nasal patterning is identical in wild-type and kif5aa mutant retinae . ( D ) Pax2 . 1 , whose expression is restricted to the optic stalk and retinal cells around the choroid fissure and that mediates optic stalk and chiasm formation in fish ( Krauss et al . , 1991; Puschel et al . , 1992 ) shows a normal expression pattern in kif5aa mutants . ( E ) Cryosection of Tg ( Isl2:Gal4 , UAS:eGFP ) transgenic wild-type and mutant retinae and immunostaining against eGFP ( RGCs ) ( Ben Fredj et al . , 2010 ) , Parvalbumin ( marking amacrine cells ) ( Godinho et al . , 2005 ) and protein kinase C ( marking bipolar cells ) ( Godinho et al . , 2005 ) shows that all cell types are present in the right retinal layer . Blue = DAPI . Scale bar = 10 μm . RGCL = RGC layer , AC = amacrine cells , BPC = bipolar cells . DOI: http://dx . doi . org/10 . 7554/eLife . 05061 . 006 To gain deeper insights into the phenotype of single RGCs , we made use of the Tg ( BGUG ) transgenic line ( Xiao and Baier , 2007 ) . The BGUG ( Brn3C [also known as pou4f3]:Gal4; UAS:mGFP ) transgene marks RGCs projecting to the SO ( stratum opticum ) and SFGS ( stratum fibrosum et griseum superficiale ) layers of the optic tectum . Probably due to position-effect variegation of the transgene , a stochastic subset of one to ten RGCs per retina is labeled with membrane-bound GFP , which allows the imaging of RGC trajectories in the living or fixed fish brain . By in vivo imaging , we thus followed the growth behavior of single RGC axons over consecutive days . To guarantee comparability between RGC axons , we analyzed only RGC axons growing into the SFGC layer of the optic tectum at a central position on the rostral-caudal axis . While wild-type axons reach the tectal neuropil at 72 hpf and start forming a complex axonal arbor , axons of kif5aa mutant RGCs showed a delayed ingrowth into the target tissue ( Figure 3A ) . This phenotype was confirmed by analyzing a larger RGC population , which was labeled by DiI injections into quadrants of the contralateral retina ( Figure 3A ) . 10 . 7554/eLife . 05061 . 007Figure 3 . RGC axons in kif5aa mutants show a delayed ingrowth into the optic tectum and grow larger arbors at later stages . ( A ) Single membrane-GFP expressing RGC axons from the Tg ( BGUG ) transgene ( left panel ) and DiI injections into the contralateral retina of 3 dpf old wild-type and kif5aa mutant embryos ( right panel ) illustrate the delay of tectal innervation in mutants . Scale bars = 20 μm . The schematic in the lower left panel illustrates the perspective chosen for image acquisition ( indicated by an arrow ) . D = dorsal , V = ventral , R = rostral , C = caudal . ( B ) Upper panel , left: Axonal arbor of a single wild-type RGC at 3 , 5 , and 7 dpf . Lower panel , left: Tracings of an axonal arbor at time point zero . In red: Overlay of filopodia formed and retracted within 10 min ( 1 frame/2 min ) . Scale bars = 20 μm . Upper panel , right: Axonal arbor of a single kif5aa mutant RGC axonal arbor at 3 , 5 , and 7 dpf . Lower panel , right: Tracing of an axonal arbor at time point zero . In red: Overlay of filopodia formed and retracted within 10 min ( 1 frame/2 min ) . Scale bars = 20 μm . ( C ) Kif5aa mutant RGC axons grow significantly more branches , cover a larger area of the optic tectum with their arbors ( in μm2 ) and grow longer arbors ( in μm ) than wild-type cells at 7 dpf ( n = 10 , p < 0 . 01 ) . Scale bars = 20 μm . For quantification , only branches stable within 10 min of image acquisition were selected . ( D ) Quantification of filopodia numbers formed and retracted within 10 min per cell at 3 , 5 , and 7 dpf . While wild-type RGC arbors form most filopodia at 3 dpf and reduce this rate constantly until 7 dpf , kif5aa mutant RGC axons grow almost three times more filopodia at 5 dpf ( n = 4 , p < 0 . 01 ) . At 7 dpf , the rate is still more as double as high as for their wild-type counterparts ( n = 4 , p < 0 . 01 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05061 . 00710 . 7554/eLife . 05061 . 008Figure 3—figure supplement 1 . Mapping of the vertigos1614 locus by genetic linkage analysis . In 1800 meioses , the vertigos1614 allele was co-segregating with the two polymorphic markers fj61a10 and tsub1g3 located on Contig 963 of linkage group 9 ( LG9 ) . Four genes encoding for Poly ( rc ) -binding protein 2 , Kif5aa , Arp and Cdk3 are found in this genomic region . In complementation crosses the kif5aa162 allele was not complementing with the vertigos1614 allele . This strongly suggests that vertigo represents another loss-of-function allele of kif5aa . DOI: http://dx . doi . org/10 . 7554/eLife . 05061 . 008 A strikingly similar phenotype was reported in a previously published forward genetic screen for defects in the visual system in the vertigos1614 mutant without further in-depth characterization of this mutant line nor identification of the causal gene ( Xiao et al . , 2005 ) . Our genetic linkage analysis confirmed that the ethylnitrosourea-induced mutation of the vertigos1614 allele is located at linkage group 9 of the zebrafish genome ( Figure 3—figure supplement 1 ) . Utilizing the newly generated kif5aa162 allele we could confirm by complementation crosses that vertigos1614 represents a loss-of-function allele of kif5aa . Although sequencing of the genome has not identified a telltale mutation of kif5aa coding sequence or its flanking regulatory regions in vertigo mutants , judging by its penetrance and expressivity , we expect the kif5aas1614 mutation to be a strong hypomorph or null allele . Furthermore , our analysis did not reveal any phenotypic difference between the different alleles and for our subsequent analysis we used kif5aa162 . Wild-type axons continually expand their arbors during lifelong growth of the tectum . After 5 dpf , however , branching activity noticeably subsides , and axons maintain their complexity over the following days ( Meyer and Smith , 2006 ) ( Figure 3B ) . This phase of relative stability coincides with the consolidation of synaptic connections with tectal dendrites in the neuropil region ( Nevin et al . , 2010 ) . In kif5aa mutant axons , the observed delay of ingrowth into the tectum is followed by a period of highly active growth of the axonal arbor after 5 dpf , at a stage when wild-type axons are comparatively stable . We investigated the underlying branching dynamics by multi-day single axon imaging . Between 5 and 7 dpf , mutant axons added branches at about double the wild-type rate ( Figure 3B , C ) . Mutant axons also showed markedly increased numbers of active filopodia either being retracted or newly formed at any time point analyzed , as observed in 10 min timelapse movies ( Figure 3B , D ) . Together , these findings suggest that , perhaps counterintuitively , absence of the motor protein Kif5aa stimulates growth of the axon arbor and maintains high filopodia activity . The loss of optokinetic response of kif5aa mutants raises the question on which level of the visual pathway the visual information processing is perturbed . To understand this better , we used genetically encoded Ca2+ sensors that were differentially expressed in two different neuronal types of the visual system , namely in RGCs and tectal periventricular neurons ( PVNs ) ( Del Bene et al . , 2010; Nikolaou et al . , 2012; Hunter et al . , 2013 ) . First , we probed the overall activity of the larva's visual system from 5 to 7 dpf in response to defined visual stimuli by using the Tg ( HuC:GCaMP5G ) transgenic line , in which among other neuron types , all RGCs and PVNs are labeled ( Ahrens et al . , 2013 ) ( Figure 4B , left ) . We did observe clear stimulus-evoked Ca2+ transients in the wild-type PVN layer and the tectal neuropil ( the later contains both PVNs dendritic and RGC axonal arbors ) in response to a bar ( Nikolaou et al . , 2012 ) moving in a caudal-to-rostral direction ( Figure 4C , Video 1 ) across the larva's visual field . This response was almost completely absent in kif5aa mutants at 5 dpf ( Figure 4C , Figure 4—figure supplement 1 , Video 2 ) . In addition , also no response was detected in older larvae ( 7 dpf ) , arguing against a delayed onset of activity in mutants , as could have been speculated based on the observed developmental delay of RGC ingrowth ( Figure 3 ) . Besides bars running caudal-to-rostrally , we also tested bar stimuli running in the opposite direction , bars moving in different orientations of 45° steps across the visual field as well as looming stimuli . In neither of these , kif5aa mutants showed a response comparable to their wild-type siblings at any developmental stage between 5 dpf and 7 dpf ( data not shown ) . We next investigated whether this loss of Ca2+ responses in the tectum was due to a presynaptic defect in RGC axons . For this , the same stimulation paradigms were employed in compound transgenic fish carrying Tg ( Isl2b:Gal4 ) and Tg ( UAS:GCAMP3 ) , which express the Ca2+ sensor GCaMP3 in all or nearly all RGCs ( Ben Fredj et al . , 2010; Warp et al . , 2012 ) . Mutant RGC axons in the tectal neuropil remained unresponsive to visual stimulation , whereas wild-type axons were robustly activated ( Figure 4D ) . These data provide an explanation for the behavioral blindness of kif5aa mutant larvae and indicate that the Kif5aa motor is required for proper synaptic transmission from RGC axon terminals to tectal dendrites . 10 . 7554/eLife . 05061 . 009Figure 4 . Kif5aa mutant larvae show no activity in RGCs and no synaptic transmission to tectal cells . ( A ) 5–7 dpf larvae were visually stimulated by bars on an LED screen running in caudal-to-rostral direction across the larva's visual field . Wild-type and kif5aa mutant larvae expressing genetically encoded calcium indicators ( GCaMPs ) in different subsets of neurons of the visual system were confocally imaged in the tectum contralaterally to the stimulated eye ( dashed box inset ) . RGCs = Retinal Ganglion Cells , PVNs = periventricular neurons , NP = neuropil . ( B ) The activity of visual system neurons in response to visual stimuli is shown as normalized GCaMP fluorescence intensity changes ( deltaF/F0 ) over time . GCaMP intensity was averaged over manually determined regions of interest ( ROIs ) that corresponded to well-distinguishable anatomical regions in the larval tectum , the neuropil ( NP ) and the periventricular cell bodies area ( PVNs ) . In Tg ( HuC:GCaMP5 ) fish ( left ) , GCaMP5 is expressed pan-neuronaly , that is , in both neuropil and PVNs , whereas in Tg ( Isl2b:Gal4 ) × Tg ( UAS:GCaMP3 ) fish ( right ) , it is expressed in all RGCs and their processes . Scale bars = 20 μm . Asterisk = pigment cell in the skin . ( C ) Averaged deltaF/F0 ratio over time in response to a moving bar visual stimulation ( black vertical line denotes the time point of the stimulus onset ) in fish with pan-neuronal GCaMP5G expression . Four sequential rounds of stimulus presentation and the time-courses of Ca2+-transients in the neuropil ROIs ( left panel ) and periventricular cell bodies ROIs ( right panel ) of wild-type/heterozygous ( red curve ) and kif5aa mutant larvae ( blue curve ) ( n = 7 each ) are shown . Activity in the visual system of kif5aa mutants was almost absent between 5 and 7 dpf compared to wild-type larvae . Light red and blue zones indicate the 95% confidence intervals around the averaged deltaF/F0 curves for each region of interest ( NP and PVN ) , respectively . ( D ) Averaged deltaF/F0 ratio over time in response to a moving bar visual stimulation ( black vertical line denotes stimulus onset ) in fish with GCaMP3 expression in RGC axons . Five sequential rounds of stimulus presentation and time-courses of Ca2+-transients in the neuropil of wild-type/heterozygous ( red curve ) and kif5aa mutant larvae ( blue ) ( n = 9 each ) . RGC arbor activity in kif5aa mutants was strongly diminished . Light red and blue zones indicate the 95% confidence intervals around the averaged deltaF/F0 curves for the neuropil region of interest ( NP ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05061 . 00910 . 7554/eLife . 05061 . 010Figure 4—figure supplement 1 . Regression-based analysis of wild-type/heterozygous vs kif5aa−/− Tg ( HuC:GCaMP5G ) mutants to a visual stimulus . Kif5aa sibling and mutant Tg ( HuC:GCaMP5G ) larvae were stimulated with a visual bar running from caudal to rostral in the visual field . Subsequently , regression analysis of the measured Calcium time-series and a time-series representing an expected calcium response to the stimulus was performed ( Miri et al . , 2011 ) . In wild-type/heterozygous a large number of correlated pixels could be identified spanning the neuropil and PVN layers ( A ) , whereas in kif5aa mutants no pixels were activity correlated with the stimulus ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05061 . 01010 . 7554/eLife . 05061 . 011Video 1 . In vivo timelapse imaging of the optic tectum in a Tg ( HuC:GCaMP5G ) transgenic 5 dpf wild-type fish . A wild-type larva was stimulated with a visual bar running from caudal to rostral through the visual field of the contralateral eye to the imaged optic tectum . Stimulus onset is indicated by a white circle in the top right corner of the image sequence . The movie is accelerated to a framerate of 40 frames/s . DOI: http://dx . doi . org/10 . 7554/eLife . 05061 . 01110 . 7554/eLife . 05061 . 012Video 2 . In vivo timelapse imaging of the optic tectum in Tg ( HuC:GCaMP5G ) transgenic 5 dpf kif5aa mutant embryos . A kif5aa mutant larva was stimulated with a visual bar running from caudal to rostral through the visual field of the contralateral eye to the imaged optic tectum . Stimulus onset is indicated by a white circle in the top right corner of the image sequence . The movie is accelerated to a framerate of 40 frames/s . DOI: http://dx . doi . org/10 . 7554/eLife . 05061 . 012 As loss of synaptic activity in RGCs might be caused by a failure in synapse formation , we monitored synapse distribution and transport of synaptic vesicles in RGC axons in vivo . For this experiment , we made use of a Synaptophysin-GFP ( Syp-GFP ) fusion construct , a marker for stable presynaptic sites as well as for motile Synaptophysin-containing clusters ( Meyer and Smith , 2006 ) . When co-expressed with a membrane-localized red fluorescent protein ( RFPCaax ) in single RGCs in wild-type or kif5aa mutant embryos ( Figure 5A ) , we observed no difference in distribution of stable presynaptic clusters at 5 and 7 dpf by in vivo imaging ( Figure 5B ) . This indicates that the formation of stable presynaptic clusters is not affected by the loss of Kif5aa . Quantification of Synaptophysin-containing vesicle movements in axonal segments ( Figure 5—figure supplement 1 , Video 3 ) did not show a difference in either anterograde or retrograde transport upon loss of kif5aa at 4 , 5 , and 7 dpf . This is consistent with previous studies showing that Synaptophysin containing vesicles are not transported by Kinesin I ( Karle et al . , 2012 ) and furthermore argues against a general loss of vesicle movement in kif5aa mutant RGCs . To distinguish between different sized clusters , we divided vesicles into small ( <0 . 4 μm ) and middle-sized to large clusters ( >0 . 4 μm ) as defined previously for RGC axons in zebrafish ( Meyer and Smith , 2006 ) . Using these different categories , we detected a higher fraction of small , motile clusters in kif5aa mutant cells at 4 and 5 dpf , but not at 7 dpf , most probably reflecting their highly active growth behavior at that stage of development ( Figure 5—figure supplement 1C ) . 10 . 7554/eLife . 05061 . 013Figure 5 . Kif5aa mutant RGC arbors show the same density of presynaptic sites but are depleted of mitochondria . ( A ) In vivo imaging shows the distribution of presynaptic sites marked by Synaptophysin-GFP ( SypGFP ) in single kif5aa mutant and wild-type RGC arbors expressing membrane localized RFP ( RFPCaax ) . Upper panel: wild-type cell arbor , lower panel: kif5aa mutant cell arbor . Scale bars = 20 μm . D = dorsal , V = ventral , R = rostral , C = caudal . ( B ) Upper panel: Zoom in to an axonal segment indicated in the right panel of ( A ) . Stable presynaptic clusters of SypGFP larger than 0 . 4 μm were defined as synapses ( white arrows in the upper panel ) and synapse density in axonal segments of wild-type and mutant cell arbors does not show a significant difference at 5 and 7 dpf ( lower panel ) . ( C ) Distribution of mitochondria ( labeled by mitoGFP ) in single mutant and wild-type RGC arbors expressing membrane localized RFP ( RFPCaax ) in vivo . Upper panel: wild-type cell arbor , lower panel: kif5aa mutant cell arbor . Mutants RGC cell arbors show substantially less mitochondria . Scale bars = 20 μm . ( D ) Transmission electron micrograph of a transvers section of the neuropil containing RGC axonal arbors . Upper panel: wild-type neuropil , lower panel: kif5aa mutant neuropil . In yellow circles: mitochondria . Left panel: Zoom in into a single axonal segment containing a mitochondrion . The kif5aa mutant neuropil contains less mitochondria . Scale bar = 500 nm . ( E ) Quantification of mitochondria area per neuropil area comparing wild-type and mutant tecta at 6 dpf . Mutant cells contain significantly less mitochondria than wild-type cells ( p < 0 . 01 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05061 . 01310 . 7554/eLife . 05061 . 014Figure 5—figure supplement 1 . Analysis of transport dynamics of synaptic vesicles in wild-type and kif5aa mutant cells during visual system development . ( A ) Synaptic vesicles were visualized by Synaptophysin-GFP ( SypGFP ) expression and imaged in axonal segments of wild-type and kif5aa mutant cell RGC arbors . Depending on the size of vesicles we defined two different categories ( Meyer and Smith , 2006 ) : small vesicles ( <0 . 4 μm ) ( shown in red ) and medium sized and large vesicles ( >0 . 4 μm ) ( shown in yellow and green ) . ( B ) Transport dynamics of synaptic vesicles were analyzed in kymographs ( see Video 3 ) . White cross: stable vesicles . White arrowhead: vesicle transported in anterograde direction . Black star: vesicle transported in retrograde direction . Scale bar = 3 μm . ( C ) While the number of stable presynaptic sites was not altered in kif5aa mutant cells ( see Figure 5 ) , they show an increased number of small , motile vesicles at 4 and 5 dpf reflecting their highly dynamic growth state ( p < 0 . 05 ) . This difference disappears at 7 dpf . ( D ) Quantification of anterograde and retrograde transport rates at 4 , 5 , and 7 dpf does not show a difference between wild-type and kif5aa mutant RGC arbors . DOI: http://dx . doi . org/10 . 7554/eLife . 05061 . 01410 . 7554/eLife . 05061 . 015Figure 5—figure supplement 2 . Analysis of mitochondria localization and transport dynamics in wildtype and kif5aa mutant RGC arbors . ( A ) As previously shown in other experimental systems ( Obashi and Okabe , 2013 ) mitochondria are often localized in close proximity to synapses . In the upper panel , an axonal segment of a single RGC axon with labeled synapses ( SypGFP ) and mitochondria ( mitoRFP ) is shown . Scale bar = 5 μm . The middle panel and the lower panel show an axonal segment of a trigeminal Ganglion Cell ( TGC ) with labeled synapses and mitochondria in wild-type and kif5aa mutant embryos , respectively . Scale bar = 5 μm . ( B ) In wildtype and blumenkohl mutant embryos about 40% of synapses in RGCs possess an associated mitochondrion ( blu RGCs: 41/91 , wt RGCs: 68/153 ) . This percentage is also observed in wild-type TGCs while in kif5aa mutant TGCs only about 20% of synapses contain a colocalized mitochondrion ( wt TGCs: 117/235 , kif5aa TGCs: 99/452 ) . ( C ) Kymograph of labeled mitochondria within an RGC axonal segment ( see Video 4 ) . White crosses = stable mitochondria . Black asterisk = Single mitochondrion moving in anterograde direction . ( D ) Comparison of mitochondria dynamics between wild-type and kif5aa mutant RGC axons at 5 dpf . In both , most mitochondria are stable while only about 15% of mitochondria are motile . ( E ) Of these , in mutant cells mitochondria move significantly more often in retrograde direction ( p < 0 . 05 ) . Wild-type cells show a higher proportion of anterograde transport . DOI: http://dx . doi . org/10 . 7554/eLife . 05061 . 01510 . 7554/eLife . 05061 . 016Figure 5—figure supplement 3 . Retinal Ganglion Cells show a normal mitochondria distribution in blumenkohl mutants . ( A ) Distribution of mitochondria ( labeled by mitoGFP ) in single blumenkohl mutant and wild-type RGC arbors expressing membrane localized RFP ( RFPCaax ) in vivo . Upper two panels: blumenkohl mutant cell arbors at 5 and 7 dpf , lower two panels: wild-type cell arbors at 5 and 7 dpf . D = dorsal , V = ventral , R = rostral , C = caudal . Scale bars = 20 μm . ( B ) Transmission electron micrograph of a transvers section of the blumenkohl mutant neuropil containing RGC axonal arbors . In yellow circles: mitochondria . Scale bar = 500 nm . ( C ) Quantification of mitochondria area per neuropil area comparing wildtype , blumenkohl and kif5aa mutant tecta at 6 dpf . No difference is observed between wild-type and blumenkohl mutant cells while kif5aa mutants contain significantly less mitochondria than wild-type cells ( p < 0 . 01 ) . The graphs for kif5aa and wild-type cells were taken from Figure 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 05061 . 01610 . 7554/eLife . 05061 . 017Video 3 . In vivo timelapse imaging of Synaptophysin-GFP containing clusters in RGC axonal segments . Representative RGC axonal segment in a 5 dpf old wild-type larva . SypGFP labels synaptic clusters of different sizes that were grouped in small and middle-sized plus large vesicles . Compare Figure 5—figure supplements 1A , B for grouping and kymogram analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 05061 . 017 In vivo imaging of labeled mitochondria ( mitoGFP ) and membranes ( RFPCaax ) in single RGCs in contrast showed a reduction of mitochondria within the axon of mutant RGCs compared to wild-type cells ( Figure 5C ) . This is consistent with previous reports showing that mitochondria are transported by Kif5a in other experimental systems ( Macaskill et al . , 2009; Karle et al . , 2012; Chen and Sheng , 2013 ) . We confirmed this result by quantification of the area covered by mitochondria per neuropil area in electron micrographs of transverse section of the tectal neuropil ( Figure 5D ) . At 6 dpf , mutants show a significant depletion of mitochondria from their branched axons ( Figure 5E ) . To test if this mitochondria depletion from the distal axon of RGCs is caused by transport defects of this known cargo of Kif5a , we quantified the transport dynamics of mitochondria within RGC axonal segments ( Figure 5—figure supplement 2 , Video 4 ) . We did not detect a difference in overall amount of mobile vs stable mitochondria . In mutant cells , though , mitochondria were transported significantly more often in a retrograde direction than in an anterograde direction ( Figure 5—figure supplement 2D ) . This bias explains the depletion of mitochondria from the tips of kif5aa mutant RGCs . Mitochondria are preferentially localized at active synapses ( Obashi and Okabe , 2013 ) and are found in close proximity to stable Synaptophysin-containing clusters in RGC arbors ( Figure 5—figure supplement 2A ) . By co-labeling mitochondria and presynaptic clusters in the same cells in vivo we showed that approximately 40% stable presynaptic sites are associated with mitochondria in wild-type and blu−/− RGC axons ( Figure 5—figure supplement 2B ) . We extended and confirmed these data observing mitochondria distribution and presynaptic densities in trigeminal ganglion cell axons ( TGCs ) ( Figure 5—figure supplement 2A , B ) . In contrast to wild-type cells , in TGCs in kif5aa−/− we observed a marked decrease of presynaptic densities located in close proximity of mitochondria . This observation is consistent with a recent study in zebrafish reporting a reduced number of mitochondria in peripheral cutaneous axon arbors in kif5aa mutant zebrafish embryos ( Campbell et al . , 2014 ) without affecting the distribution of presynaptic densities . 10 . 7554/eLife . 05061 . 018Video 4 . In vivo timelapse imaging of mitochondria in RGC axonal segments . Representative RGC axonal segment in a 5 dpf old wild-type larva . Compare Figure 5—figure supplement 2B for kymogram analysis of mitochondria movements . DOI: http://dx . doi . org/10 . 7554/eLife . 05061 . 018 Taken together , these experiments show that kif5aa mutant RGCs form presynaptic sites at the same density as wild-type cells and transport Synaptophysin-containing clusters at the same rate and direction as wild-type . A detectable reduction in the relative anterograde transport of mitochondria , however , results in a depletion of these organelles from synaptic terminals . We aimed to identify the signal responsible for the observed increased growth of RGC arbors in kif5aa mutant tecta . As it was previously shown in the optic tectum of X . laevis that Brain-Derived Neurotrophic Factor ( BDNF ) can promote axonal arborization ( Cohen-Cory and Fraser , 1995 ) , we decided to measure the expression levels of this neurotrophic factor in kif5aa mutant embryos . In parallel , we performed quantitative reverse transcription PCR of other known members of the neurotrophic factor family , namely neurotrophin 3 ( ntf3 ) , neurotrophin 4 ( ntf4 ) , neurotrophin 7 ( ntf7 ) , and nerve growth factor ( ngf ) present in the zebrafish genome ( Heinrich and Lum , 2000 ) . Comparing 4 dpf old homozygous kif5aa mutants to their siblings , we detected a significant upregulation of ntf3 in mutants to up to 160% of its normal expression level in wild-type embryos , while the levels of bdnf , ngf , ntf4 , and ntf7 were not significantly altered ( Figure 6A ) . By in situ hybridization with an ntf3 antisense probe , we could furthermore see an increased staining intensity in mutant tecta compared to their siblings ( Figure 6B ) . 10 . 7554/eLife . 05061 . 019Figure 6 . Expression of the neurotrophic factor neurotrophin 3 in visually impaired mutants . ( A ) Relative expression levels of bdnf , ntf3 , ntf4 , ntf7 , and ngf in 4 dpf old wild-type and kif5aa mutant embryos . ntf3 is upregulated to 160% of wild-type expression levels ( p < 0 . 05 ) while all other neurotrophic factors show the same expression levels between wild-type and mutant embryos . ( B ) Transversal sections through the tectum after in situ hybridization with an ntf3 specific antisense probe detect higher expression levels of ntf3 in tecta of kif5aa . Scale bars = 25 μm . PC = pigment cell , NP = neuropil , PVN = periventricular neurons . Asterisk = strong ntf3 signal in the otic vesicle . ( C ) Confirmation of Ntf3 overexpression by Western blotting in 4 dpf old embryos . To show the specificity of the antibody , we generated a Ntf3 overexpression construct ( UAS:ntf3-E2A-RFP ) . The two visually impaired mutant lines lakritz and blumenkohl also show a substantial upregulation of ntf3 expression levels . ( D ) Quantification of Ntf3 protein expression levels based on Western blotting data in wild-type and visually impaired mutant embryos . All three mutant lines show a substantial upregulation of Ntf3 protein levels . DOI: http://dx . doi . org/10 . 7554/eLife . 05061 . 01910 . 7554/eLife . 05061 . 020Figure 6—figure supplement 1 . Silencing of all Retinal Ganglion Cells by BoTx expression leads to Ntf3 upregulation . ( A ) Expression of botulinum toxin light chain B ( Brunger et al . , 2008; Nevin et al . , 2008 ) in all RGCs leads to expansion of melanosomes similar to blind mutant fish as shown in Figure 1—figure supplement 1 . Scale bar = 200 μm . ( B ) Relative expression levels of ntf3 in 4-dpf old wildtype , kif5aa mutant , and Tg ( Isl2b:Gal4 , UAS:BoTxLCB-GFP ) transgenic embryos . ntf3 is upregulated to a similar extent in Tg ( Isl2b:Gal4 , UAS:BoTxLCB-GFP ) transgenic embryos as in kif5aa mutants . ( C ) Western blotting confirms the overexpression of Ntf3 protein in embryos without presynaptic activity . DOI: http://dx . doi . org/10 . 7554/eLife . 05061 . 020 For further validation of these results , we performed Western blotting analysis using an antibody targeting the human orthologue of Ntf3 . This antibody recognizes the zebrafish Ntf3 protein . We could demonstrate cross-reactivity by overexpressing a construct carrying the zebrafish ntf3 cDNA followed by an E2A sequence allowing multicistronic expression ( Szymczak et al . , 2004 ) of ntf3 and a RFP reporter gene from the same cDNA ( Figure 6C ) . This experiment showed that kif5aa mutant embryos produced significantly more Ntf3 protein than their siblings . To see if Ntf3 upregulation was a common feature in mutants with defective retinotectal synaptic transmission , we investigated lakritz mutants , which lack all RGCs and blumenkohl , in which RGCs show impaired glutamate secretion into the synaptic cleft . Ntf3 protein was increased in the tecta of all three mutant lines in which presynaptic input to the tectum was abolished or highly reduced ( Figure 6D ) . To directly test the causality link between the lack of presynaptic input and upregulation of Ntf3 in the optic tectum , we silenced the neuronal activity of RGCs by expressing a UAS:BoTxLCB-GFP construct in Islet2b:Gal4 zebrafish larvae . BoTx has been shown to specifically block synaptic vesicle release ( Brunger et al . , 2008 ) and its injection has been used successfully in zebrafish embryos to silence neuronal activity ( Nevin et al . , 2008 ) . Similar to what previously observed in kif5aa , lak , and blu mutants , silencing of RGCs via BoTx expression leads to melanosomes expansion and failure to adapt to a light background ( Figure 6—figure supplement 1A ) . Both , via qRT-PCR and Western blotting analysis we detected an upregulation of Ntf3 in these transgenic animals showing that , like in the previously described mutants , lack of RGCs presynaptic activity per se is sufficient to cause Ntf3 overexpression ( Figure 6—figure supplement 1B , C ) . These results strongly suggest that lack of presynaptic activity and subsequent overexpression of Ntf3 in the tectum trigger the increased size of axonal branches of RGCs in kif5aa , blumenkohl , and lakritz mutants ( see Figure 2 , [Smear et al . , 2007; Gosse et al . , 2008] ) . To further test this hypothesis , we designed an experiment to interfere with Ntf3 signaling in RGCs . We generated a kinase-dead , GFP-tagged , dominant-negative version of the zebrafish ntrk3a gene orthologous to the gene encoding the TrkC receptor in mammals ( ntrk3adN-GFP ) ( Parada et al . , 1992 ) . A similar approach was previously shown to efficiently block Ntf3 function in mammalian cells ( Tsoulfas et al . , 1996; Lin et al . , 2000 ) . In zebrafish two TrkC paralogues , ntrk3a and ntrk3b exist , which are both expressed in RGCs ( Figure 7A ) . We decided to generate our dominant-negative construct based on the ntrk3a coding sequence , which shows a higher degree of conservation ( based on amino acid identity and sequence similarity ) to the rat TrkC receptor ( Martin et al . , 1998 ) . Both receptors are predicted to bind to Ntf3 based on binding motif analysis ( Martin et al . , 1998 ) . Upon overexpression of the truncated ntrk3adN-GFP construct in single wild-type RGCs by mosaic DNA expression , we observed a substantial reduction of axon branch length and number of branches at 5 dpf ( Figure 7B , E ) , consistent with a role of Ntrk3 as a branch-promoting receptor . 10 . 7554/eLife . 05061 . 021Figure 7 . Neurotrophin 3 signaling alters axonal branch size in RGCs . ( A ) In situ hybridization with ntrk3a and ntrk3b specific antisense probes shows expression of both paralogues in broad parts of the nervous system in 5 dpf larvae . Both receptors are strongly expressed in RGCs . RGCL = Retinal Ganglion Cell Layer . Scale bars ( from left to right ) = 50 μm , 50 μm , 150 μm . ( B ) Representative pictures of single RGC axons at 5 ( upper ) and 7 dpf ( lower panel ) . Control RGC cells express a membrane bound eGFP ( control; left panel ) . To render cells unresponsive to the TrkC pathway , single RGCs express a dominant negative , kinase dead and eGFP-tagged form of the neurotrophic factor receptor ntrk3a ( ntrk3adN-GFP ) ( dominant negative , right panel ) . Consequently , RGCs grow smaller arbors with less branches . ( C ) To investigate the effect of Ntf3 on RGC axonal growth , we monitored single eGFP positive RGCs while growing into a tectum overexpressing Ntf3 ( overexpression ) at 5 ( upper left panel ) and 7 dpf ( lower left panel ) . Overexpression of Ntf3 was driven by an UAS:ntf3-E2A-RFP construct in the Tg ( gSA2AzGFF49A ) ( Muto et al . , 2013 ) transgenic line in tectal glial cells and periventricular neurons from 2 dpf onwards . By employing a 2A sequence between the ntf3 and the RFP open reading frame , both proteins were produced from the same construct . Thereby Ntf3 overexpressing cells were marked by RFP expression ( right upper panel ) and we analyzed the arbors of single eGFPCaax positive RGCs at 5 ( upper left panel ) and 7 dpf ( lower left panel ) growing in RFP expressing optic tecta . RGCs grow more complex arbors with more branches when invading into the Ntf3 overexpressing tectal environment compared to control RGCs ( A , left panel ) . Lower left panel = merge of ntf3-E2A-RFP expressing tectal cells and an eGFPCaax expressing RGC axon . Scale bars = 20 μm . D = dorsal , V = ventral , R = rostral , C = caudal . ( D ) Schematics illustrating the approach for analysis of single RGC arbors . While in control and dominant negative expression experiments , single RGCs were labeled ( upper panel ) , in the Ntf3 overexpression situation , single membrane bound eGFP ( eGFPCaax ) labeled RGCs were growing into a tectum overexpressing Ntf3 ( labeled by RFP expression , shown in magenta , lower panel ) . ( E ) Quantification of total branch length and number of branches at 5 dpf in single RGC arbors upon overexpression of the dominant negative ntrk3adN-GFP construct in single RGCs or overexpression of Ntf3 in tectal cells . Ntrk3adN-GFP expressing wild-type cells are significantly smaller and grow fewer branches at 5 dpf ( p < 0 . 001 ) . In both , kif5aa and blumenkohl mutant embryos , expression of ntrk3adN-GFP in single RGCs inhibits the overgrowth of the axonal arbor that is normally observed . The branch length is not different to the length in wild-type cells . Ntf3 overexpression in the tectum leads to increased axonal branch length and increased branch number in wild-type RGCs ( p < 0 . 05 ) ( n = 8 , 23 , 7 , 8 , 26 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05061 . 02110 . 7554/eLife . 05061 . 022Figure 7—figure supplement 1 . Blumenkohl mutant RGC arbors and RGC arbors growing into a Ntf3 overexpressing tectum do not show increased filopodia dynamics . ( A ) Upper panel: Axonal arbor of a single blumenkohl mutant RGC at 5 and 7 dpf . ( indicated by an arrow ) . D = dorsal , V = ventral , R = rostral , C = caudal . Lower panel: Tracings of an axonal arbor at time point zero . In red: Overlay of filopodia formed and retracted within 10 min ( 1 frame/2 min ) . Scale bars = 20 μm . ( B ) Upper panel: Axonal arbor of a single RGC axonal arbor growing into a Ntf3 overexpressing tectum ( compare Figure 7 ) at 5 and 7 dpf . Lower panel: Tracing of an axonal arbor at time point zero . In red: Overlay of filopodia formed and retracted within 10 min ( 1 frame/2 min ) . Scale bars = 20 μm . ( C ) Quantification of filopodia numbers formed and retracted within 10 min per cell at 5 and 7 dpf . No increased rate of filopodia formation can be observed in blumenkohl mutant RGC arbors and arbors of RGCs growing into a Ntf3 overexpressing tectum while kif5aa mutant axonal arbors form and retract significantly more filopodias . Graphs for wild-type and kif5aa mutant RGC arbors are identical to Figure 7 . DOI: http://dx . doi . org/10 . 7554/eLife . 05061 . 022 In addition , ntrk3adN-GFP overexpression in single kif5aa−/− and blu−/− RGCs could abolish the axonal overgrowth normally observed in these mutants as measured by total branch length ( Smear et al . , 2007 ) and even reduce the number of branches compared to wild-type cells ( Figure 7E ) . To analyze the consequence of Ntf3 overexpression , we injected an UAS:Ntf3-E2A-RFP construct into the Tg ( gSA2AzGFF49A ) ( Muto et al . , 2013 ) transgenic line , driving the expression of the transgene in tectal neurons and glia cells from 2 dpf onwards . Thereby , we generated larvae specifically overexpressing Ntf3 in the optic tectum just before innervation by RGC axons . Analyzing the morphology of single RGCs marked by membrane-bound eGFP and growing under these conditions , we confirmed that these cells formed larger axonal arbors than wild-type cells ( Figure 7C , E ) at 5 and 7 dpf . Taken together , these results strongly support the hypothesis that Ntf3 signaling is a signal promoting RGC arbor growth and branching and that Ntf3 upregulation is responsible for axonal arbor overgrowth when RGC presynaptic activity is impaired . We next decided to test if the observed increased formation and retraction of filopodia in the axons of kif5aa mutant RGCs was directly caused by Ntf3 upregulation . Therefore , we analyzed filopodia dynamics via time-lapse imaging both in blu−/− RGCs and in axons growing when Ntf3 was overexpressed in the tectum via our transgenic construct . In both experimental conditions we did not observe any increase in the filopodia dynamics or RGC axons ( Figure 7—figure supplement 1A–C ) , excluding the possibility that Ntf3 overexpression has a direct effect on this process . In addition , we observed that the distribution of mitochondria was not significantly altered in blu−/− RGC axons nor was the association with stable presynaptic sites ( Figure 5—figure supplements 2B , 3 ) . Together these data suggest that Ntf3 upregulation does not per se affect mitochondria localization . For the dissection of cell-autonomous vs non-cell autonomous effects of the loss of kif5aa , we generated mutant/wild-type chimeric embryos by blastomere transplantations at the 1000-cell stage ( Gosse et al . , 2008 ) . Donor RGCs were derived from a Tg ( Pou3f4:Gal4 ) × Tg ( UAS:RFP ) cross and visualized by the expression of a fluorescent membrane-targeted RFP . When cells were transplanted in low numbers from a transgenic donor into a host embryo , we could image single donor derived and fluorescently labeled RGCs , within a host environment . First , we could observe that the delayed ingrowth of kif5aa mutant RGC axons in the tectal neuropil is a cell-autonomous effect ( Figure 8—figure supplement 1 ) . When growing in a wild-type environment , kif5aa mutant RGC axons invade the tectal neuropil 1 day later ( 4 dpf ) than wild-type cells . Second , in contrast to what we observed in mutant larvae , arbor size was not increased at 5 or 7 dpf in mutant RGC axons that grow into a wild-type tectum ( Figure 8A , B ) . Mutant RGC arbors are significantly smaller , similar to those that overexpress ntrk3adN-eGFP ( see Figure 7B ) . Wild-type cells growing in a kif5aa mutant background show the opposite behavior . No stalling at 3 dpf was detected ( Figure 8—figure supplement 1 ) and , at 5 and 7 dpf , they established a significantly increased axonal arbor ( Figure 8A , B ) . To further demonstrate that the axonal arbor overgrowth observed both in kif5aa−/− and blu−/− was due to a common molecular mechanism , we transplanted single kif5aa−/− RFP labeled RGC into a blu mutant host . In these conditions , kif5aa mutant RGC axonal arbors were significantly larger than when transplanted into a wild-type host ( Figure 8—figure supplement 2 ) . 10 . 7554/eLife . 05061 . 023Figure 8 . Transplantations confirm the growth promoting-effect in kif5aa mutant tecta . ( A ) Representative pictures of single in vivo imaged RGC axons after blastula stage transplantions from wild-type donors into a wild-type tectum ( left panel ) , from kif5aa mutants into a wild-type tectum ( middle panel ) or from a wild-type donor into a kif5aa mutant tectum ( right panel ) . The same cell was analyzed at 5 dpf ( upper panel ) and 7 dpf ( middle panel ) . Scale bars = 20 μm . Schematics of RGC arbor complexity and size in the lower panel . In orange: Ntf3 overexpressing kif5aa mutant tectum . D = dorsal , V = ventral , R = rostral , C = caudal . ( B ) Quantification of total branch length of transplanted RGC axons at 5 and 7 dpf . Kif5aa mutant cell arbors are significantly smaller than wild-type cell arbors when growing into a wild-type tectum ( p < 0 . 01 ) . Wild-type cells built larger arbors when growing into a kif5aa mutant tectum ( p < 0 . 05 ) ( 5 dpf: n = 14 , 35 , 6; 7 dpf: n = 14 , 19 , 5 ) . ( C ) Schematic illustrating growth behavior of RGC axons in wildtype ( upper panel ) and kif5aa mutant tecta ( lower panel ) and upon loss of TrkC signaling ( middle panel ) . Wild-type RGCs start to grow into the wild-type neuropil at 3 dpf . They grow highly active filopodial protrusions and start to form complex axonal arbors . At 5 dpf they reach their final size and maintain their branch shape at 7 dpf . When TrkC signaling is blocked by overexpression of a dominant negative receptor ( ntrk3adN-GFP ) , wild-type cells show a substantially reduced arbor complexity ( middle panel ) . Kif5aa mutant RGC arbors show a delay of ingrowth into the tectal neuropil . This is followed by a period of highly active growth with abundant filopodia formation . This results in highly complex arbors at 7 dpf . The delay of RGC growth is cell autonomous ( Figure 8—figure supplement 1 ) . The lack of retinal input leads to an upregulation of ntf3 expression by tectal cells and constitutes a growth-promoting environment . DOI: http://dx . doi . org/10 . 7554/eLife . 05061 . 02310 . 7554/eLife . 05061 . 024Figure 8—figure supplement 1 . Phenotype of transplanted RGC arbors at early stages of development . ( A ) Kif5aa mutant RGC arbors show a delay of ingrowth into the optic tectum compared to wild-type cells ( right column ) . This delay is cell autonomous as mutant RGC axons also fail to invade their target tissue before 4 dpf when transplanted into a wild-type background ( upper panel , middle , and left column ) . In contrary , wild-type cells when transplanted into a kif5aa mutant tectum grow into the neuropil from 3 dpf onwards ( lower panel , middle , and left column ) . Scale bars = 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05061 . 02410 . 7554/eLife . 05061 . 025Figure 8—figure supplement 2 . Transplantation of kif5aa mutant RGCs into a blumenkohl mutant acceptor leads to an increased growth compared to transplantation into a wild-type acceptor . ( A ) Representative pictures of single in vivo imaged RGC axons after blastula stage transplantions from kif5aa mutant donors into a wild-type tectum ( left panel ) and from kif5aa mutants into a blumenkohl mutant tectum ( right panel ) . The same cell was analyzed at 5 dpf ( upper panel ) and 7 dpf ( middle panel ) . Scale bars = 20 μm . Schematics of RGC arbor complexity and size in the lower panel . In orange: Ntf3 overexpressing blumenkohl mutant tectum . D = dorsal , V = ventral , R = rostral , C = caudal . Pictures in the left panel are identical to Figure 8 . ( B ) Quantification of total branch length of transplanted RGC axons at 5 and 7 dpf . The reduced size of kif5aa mutant axonal arbors when growing into a wild-type tectum is partially rescued when transplanted into a blumenkohl mutant environment ( p < 0 . 05 ) ( 5 dpf: n = 35 , 8; 7 dpf: n = 19 , 8 ) . The graph for kif5aa mutant cells into a wild-type host is identical to Figure 8 . DOI: http://dx . doi . org/10 . 7554/eLife . 05061 . 025 Taken together , these experiments support a homeostatic mechanism by which tectum-secreted Ntf3 directly promotes the growth of innervating RGC axons . Lack of retinal synaptic activity results in upregulation of Ntf3 and , consequently , in an enlargement of RGC axonal arbors . Here , we report for the first time the role of the anterograde transport motor Kif5aa in the larval development of the zebrafish visual system . The loss-of-function kif5aa allele that we generated disrupts the open reading frame after 122 of 1033aa within the motor domain of the protein and results in mRNA degradation likely by nonsense-mediated decay . We therefore expect that no functional Kif5aa protein is produced . Disruption of Kif5aa created a complex and dynamically changing retinotectal phenotype . At 3 to 4 dpf , RGC axons devoid of Kif5aa grew more slowly and reached their targets in the tectum with a delay of about 24 hr . Mutant retinotectal synapses did not transmit signals to tectal cells , but were apparently silent , resulting in a complete loss of visual responses . Both the stalled growth and the synaptic transmission defects were likely a direct consequence of the absence of Kif5aa . Some of the known cargoes of mammalian Kif5a are required for normal axon outgrowth ( Karle et al . , 2012; Chen and Sheng , 2013; Schwarz , 2013; Sheng , 2014 ) . In the zebrafish kif5aa mutant , mitochondria are significantly depleted from distal RGC axon terminals , as shown by in vivo imaging and transmission electron microscopy , suggesting that these organelles are Kif5a cargoes as in mammals . Furthermore , a recent report showed that Kif5aa has a similar role in the posterior lateral line nerve and peripheral cutaneous axonal arbors ( Campbell et al . , 2014 ) . Mitochondrial ATP production is required for synapse assembly ( Lee and Peng , 2008 ) , the generation of action potentials ( Attwell and Laughlin , 2001 ) and synaptic transmission ( Verstreken et al . , 2005 ) . In addition , synaptic mitochondria maintain and regulate neurotransmission by buffering Ca2+ ( Medler and Gleason , 2002; David and Barrett , 2003 ) . This deficit can therefore explain the impairment in transmitter release at the presynaptic terminals without , however , excluding the possibility that the impaired transport of other cargoes is also involved . Interestingly , outgrowth of axons from the retina was not affected by loss of Kif5aa . Only after crossing the optic chiasm , at the entrance to the tectum , did RGC axons stall . Reduced axonal growth is caused by loss of kinesin-1 in other experimental systems or other axonal transport motors like Dynein/Dynactin ( Ferreira et al . , 1992; Ahmad et al . , 2006; Abe et al . , 2008; Karle et al . , 2012; Prokop , 2013 ) , but often the deficit is evident from the start . Early functions of kinesin I heavy chains are probably carried out by other members of this gene family in zebrafish , such as kif5ab , kif5b or kif5c . All isoforms homodimerize and may carry distinct sets of cargoes ( DeBoer et al . , 2008 ) . In mammalian neurons , KIF5C likely contributes to axon specification ( Jacobson et al . , 2006 ) . Similarly , the Drosophila KIF5 homolog kinesin heavy chain ( KHC ) drives axon initiation and transiently maintains axonal growth ( Lu et al . , 2013 ) . The zebrafish Kif5c orthologue is therefore the prime candidate to carry out this early kinesin motor function for RGCs . Synapse assembly ( as assayed by transport and distribution of Synaptophysin-containing clusters ) was not detectably affected by the kif5aa mutation , in agreement with previous experiments in mouse KO cells ( Xia et al . , 2003 ) . The observed higher percentage of mobile vesicles likely reflects the highly active growth at days 4 and 5 , as it was previously shown in RGCs that synaptic puncta stability increases with axon maturation ( Meyer and Smith , 2006 ) . Filopodial activity was previously described as a sign of immature , silent axons before the onset of presynaptic activity ( Ben Fredj et al . , 2010 ) . This is in line with the observed failure of kif5aa mutant axons to transmit neuronal signals . As a secondary , non-cell autonomous consequence of kif5aa disruption , we observed that Ntf3 was upregulated by the tectum . Overexpression of Ntf3 in wild-type tectum and blockade of its receptor Ntrk3 ( TrkC ) in RGCs demonstrated that this neurotrophin is both sufficient and necessary to alter branch dynamics in RGC axon arbors . Classical work in the optic tectum of Xenopus laevis showed that overexposure to BDNF leads to enlarged and more complex axonal arbors ( Marshak et al . , 2007 ) . Here , we identified the related Ntf3 as the intrinsic , growth-promoting signal for RGC axons in zebrafish . Ntf3 upregulation was observed not only in kif5aa mutant larvae but also in two other previously characterized mutant lines , blumenkohl and lakritz , and when directly silencing RGCs via BoTx expression . In all cases , this upregulation was correlated with RGC absence or dysfunction . In lakritz zebrafish larvae , RGCs are absent; the functional connection between the retina and the tectum thus is eliminated ( Kay et al . , 2001 ) . In blumenkohl , transmitter release is diminished ( Smear et al . , 2007 ) . Similar to our results with kif5aa , blumenkohl RGC axons show an increased arbor size and complexity ( Smear et al . , 2007 ) . The same retrograde signal may underlie the enlargement of retinal arbors following treatment with MK-801 , a blocker of glutamate receptors of the NMDA-subtype ( Schmidt et al . , 2004 ) , or in macho mutants , in which RGCs fail to generate action potentials ( Gnuegge et al . , 2001 ) . In the case of lakritz , when single wild-type RGCs were transplanted into a lakritz host , their solitary axons formed larger and more complex arbors ( Gosse et al . , 2008 ) . This result is reminiscent of the phenotype of wild-type RGCs growing in a kif5aa mutant larva , that is , when they are surrounded by inactive axons . Interestingly , Ntf3 upregulation alone as observed in blu mutants or in overexpression experiments had no significant effect on mitochondria distribution and short-lived filopodia dynamic . This suggests that these phenotypes are specific to kif5aa−/− RGCs and probably caused by direct axonal trafficking defects , and that they are not due to impaired synaptic activity . Together , these data suggest a model in which a deficit in presynaptic activity enhances the production and release of Ntf3 by tectal neurons . Tectum-derived Ntf3 , in turn , retrogradely stimulates axonal branching and , thus , the addition of presynaptic terminals ( Figure 7C ) . Such a signal could be the core motif of a compensatory pathway that is triggered when synaptic drive deviates from some homeostatic setpoint ( Davis and Bezprozvanny , 2001; Burrone and Murthy , 2003 ) . Our analysis of a mutation in a motor protein has thus unmasked a potentially general structural plasticity mechanism that together with the well-known competition-based and Hebbian mechanisms shapes the retinotectal projection and determines the final axonal arbor size ( Ruthazer et al . , 2003; Ruthazer and Cline , 2004; Hua et al . , 2005; Uesaka et al . , 2006; Schwartz et al . , 2009 , 2011; Ben Fredj et al . , 2010; Munz et al . , 2014 ) . All fish are housed in the fish facility of our laboratory , which was built according to the local animal welfare standards . All animal procedures were performed in accordance with French and European Union animal welfare guidelines . The following transgenic fish lines were used or generated: Tg ( UAS:RFP , cry:eGFP ) ( Auer et al . , 2014 ) , Tg ( UAS:SypGFP ) ( Meyer and Smith , 2006 ) , Tg ( HuC:GCaMP5 ) ( Ahrens et al . , 2013 ) , Tg ( BGUG ) ( Xiao and Baier , 2007 ) , Tg ( Pou3f4:Gal4 ) ( Xiao and Baier , 2007 ) , Tg ( gSA2AzGFF49A ) ( Muto et al . , 2013 ) , Tg ( Shh:eGFP ) ( Neumann and Nuesslein-Volhard , 2000 ) , Tg ( pou4f3:mGFP ) ( Xiao et al . , 2005 ) , Tg ( Isl2b:Gal4 , cmlc2:eGFP ) , Tg ( UAS:BoTxLCB-GFP ) ( see ‘Materials and methods’ ) . TALENs used to generate the kif5aa loss-of-function alleles were described previously ( Auer et al . , 2014 ) . All mutant and transgenic lines used in this study are described in Supplementary file 1 . For genotyping the following primers were used ( 5′ to 3′ ) : kif5aa_geno_fwd: GTTCACAGATTGTGATGTCTGTG , kif5aa_geno_rev: TGGAGGATGGAGAAATGATGACA . After PCR amplification from genomic DNA the 400 bp long amplicon was digested with NcoI . The wild-type allele is digested into two fragments of 240 bp and 160 bp length , respectively . The mutant alleles are not digested and show a band at 387 bp or 390 bp . The pIsl2b:Gal4 , cmlc2:eGFP construct was generated by a Gateway reaction ( MultiSite Gateway Three-Fragment Vector Construction Kit , ThermoFisher Scientific , Waltham , MA ) using the p5E-Isl2b ( Ben Fredj et al . , 2010 ) , pME-Gal4 , p3E-pA and the pDest-Tol2;cmlc2:eGFP ( Kwan et al . , 2007 ) vectors . We generated a p5E-4nrUAS vector with four non repetitive UAS sequences by digestion of the 4Xnr UAS:GFP vector ( Akitake et al . , 2011 ) and insertion of the 4nrUAS fragment into the p5E-10UAS vector ( Kwan et al . , 2007 ) after HindIII and AleI digestion . To obtain constructs with multiple UAS sequences , we generated a p5E-4nrUAS-tagRFPCaax-pA-4nrUAS vector . The tagRFPCaax sequence was amplified with primers listed in Supplementary file 2 and inserted into the pME-MCS vector ( Kwan et al . , 2007 ) after BamHI/NotI digestion resulting in pME-tagRFPCaax . After a Gateway reaction ( MultiSite Gateway Three-Fragment Vector Construction Kit ) using the p5E-4nrUAS , pME-tagRFPCaax , p3E-pA , and the pDest-Tol2; cmlc2:eGFP ( Kwan et al . , 2007 ) vectors , the p4nrUAS:tagRFPCaax-pA-Tol2;cmcl2:eGFP vector was digested with StuI and SnaBI . The 4nrUAS:tagRFPCaax-pA fragment was subsequentially inserted into the StuI digested and dephosphorylated p5E-4nrUAS vector to create a p5E-4nrUAS-tagRFPCaax-pA-4nrUAS vector . We generated a pME-SypGFP vector by digestion of 5× UAS:SypGFP ( Meyer and Smith , 2006 ) with EcoRI and NotI and insertion into the pME-MCS plasmid . To obtain the p4nrUAS:tagRFPCaax-pA-4nrUAS:SypGFP-pA-Tol2;cmcl2:eGFP vector we performed a Gateway reaction using p5E-4nrUAS-tagRFPCaax-pA-4nrUAS , pME-SypGFP , p3E-pA and pDest-Tol2;cmlc2:eGFP . We generated a pME-PhbGFP and a pME-Phbmcherry vector by digestion of pClontecN1-PhbGFP and pClontecN1-Phbmcherry ( a kind gift from Christian Wunder ) ( Rajalingam et al . , 2005 ) with EcoRI and NotI and insertion into pME-MCS . To obtain the p4nrUAS:tagRFPCaax-pA-4nrUAS:PhbGFP-pA-Tol2;cmcl2:eGFP vector we performed a Gateway reaction using p5E-4nrUAS-tagRFPCaax-pA-4nrUAS , pME-PhbGFP , p3E-pA and pDest-Tol2;cmlc2:eGFP . We generated a p5E-4nrUAS-SypGFP-pA-4nrUAS vector by performing a Gateway reaction using p5E-4nrUAS , pME-SypGFP , p3E-pA and pDest-Tol2; cmlc2:eGFP . The resulting p4nrUAS:SypGFP-pA-Tol2;cmcl2:eGFP vector was digested with StuI and SnaBI to create a p5E-4nrUAS-SypGFP-pA-4nrUAS vector after insertion into the StuI digested and dephosphorylated p5E-4nrUAS vector fragment . To create a p4nrUAS:SypGFP-pA-4nrUAS:PhBmcherry-pA-Tol2;cmcl2:eGFP plasmid we performed a Gateway reaction using p5E-4nrUAS-SypGFP-pA-4nrUAS , pME-Phbmcherry , p3E-pA and pDest-Tol2;cmlc2:eGFP . To create a ntf3_E2A_tagRFP expression construct we amplified ntf3_E2A from wild-type zebrafish cDNA ( 3 dpf ) and fused it to a tagRFP fragment . Primers used are listed in Supplementary file 2 . After digestion and insertion into the pME-MCS vector with HindIII and NotI , we performed a Gateway reaction using p5E-10UAS ( Kwan et al . , 2007 ) , pME-ntf3_E2A_tagRFP , p3E-pA and pDest-Tol2; cmlc2:eGFP to generate p10UAS-ntf3_E2A_tagRFP-pA-tol2 , cmcl2 :eGFP . To create a dominant negative ntrk3A expression construct , we amplified a truncated fragment of ntrk3A from wild-type zebrafish cDNA ( 3 dpf ) and fused it to the eGFP open reading frame . Primers used are listed in Supplementary file 2 . After digestion and insertion into the pME-MCS vector with HindIII and NotI we performed a Gateway reaction using p5E-10UAS , pME-ntrk3AdNeGFP , p3E-pA and pDest-Tol2; cmlc2:eGFP to generate p10UAS:ntrk3adNeGFP-pA-Tol2;cmcl2:eGFP . To create a pIsl2b:eGFPCaax construct , we performed a Gateway reaction using p5E-Isl2b , pME-eGFPCaax ( Kwan et al . , 2007 ) , p3E-pA and pDest-Tol2; cmlc2:eGFP . To generate a UAS:BoTxBLC-GFP construct , a codon-optimized cDNA encoding botulinum toxin light chain B serotype ( Kurazono et al . , 1992; Whelan et al . , 1992 ) was fused in frame with GFP and cloned downstream of the 5× UAS sequence using gateway recombination ( Asakawa and Kawakami , 2008 ) . Microinjection of the pT2UAS:BoTxBLC-GFP plasmid ( 50 ng/μl ) was based on standard protocols with Tol2 mRNA ( 25 ng/μl ) . Over 50 founders were screened for the presence of a functional transgene using a combination of behavioral assays ( touch-evoked swimming , escape response ) and the level of expression of the BoTxBLC-GFP fusion protein ( Suster et al . , in preparation ) . The genomic locus of the vertigos1614 allele ( vrt ) was determined using a PCR based simple sequence length polymorphisms ( SSLPs ) marker strategy . Vrt carriers in the TL genomic background ( in which the mutagenesis was carried out ) were crossed to the WIK genomic background to generate mapping crosses . From 1800 meioses the two SSLP markers , fj61a10 and tsub1g3 , located 0 . 1 cM apart were identified to flank the vrt locus . Both markers are placed on Contig 963 of the Sanger center BAC sequencing project built by two overlapped BACs with sequencing information and four genes , one of which is kif5aa , are predicted between the two mapping markers . Retinal sections of 5 dpf old embryos and whole mount embryos were stained using standard protocols ( Kay et al . , 2001 ) . The full list of primary and secondary antibodies is given below . Whole-mount in situ hybridization was performed according to standard protocols ( Di Donato et al . , 2013 ) . Kif5aa , ntrk3a , ntrk3b , and ntf3 specific sense and antisense probes were amplified by PCR from cDNA and cloned into the pCRII-topo vector ( ThermoFisher Scientific ) . All primers used are reported in Supplementary file 2 . The tag-1 probe was synthesized from a 3 . 1 kb tag1 cDNA clone ( Warren et al . , 1999 ) and the pax2 . 1 probe was synthesized using the complete pax2 . 1 cDNA ( Krauss et al . , 1991 ) . Probes were hydrolyzed to 200 bp fragments prior to use . The following primary antibodies were used in the course of this study: anti-Parvalbumine ( EMD Millipore , Billerica , MA , MAB1572 , 1:500 ) , anti-eGFP ( Genetex , GXT13970 , 1:500 ) , anti-PKC ( Santa Cruz Biotechnology , Santa Cruz , CA , sc-209 , 1:500 ) , anti-humanNT3 ( ThermoFisher Scientific , PA1-18385 , 1:500 ) , anti-alpha-tubulin ( Genetex , Irvine , CA , GTX11304 , 1:5000 ) . The following secondary antibodies were used in the course of this study: anti-mouse-Alexa635 ( ThermoFisher Scientific , A31574 , 1:250 ) , anti-rabbit-Alexa546 ( ThermoFisher Scientific , A11081 , 1:250 ) , anti-chicken-Alexa488 ( ThermoFisher Scientific , A11039 , 1:250 ) , anti-Rabbit IgG , HRP conjugated ( Promega , W4011 , 1:2000 ) , anti-mouse IgG , HRP conjugated ( Promega , Madison , WI , W4021 , 1:2000 ) . The behavioral test for the optokinetic response was performed as described previously ( Muto et al . , 2005 ) . The morphology of single RGCs was analyzed using the Tg ( BGUG ) , kif5aa162+/− transgenic line and single cells were imaged over consecutive days . To quantify filopodia dynamics , imaging was performed for 10 min at a rate of 1 frame/2 min . All branches not extending within this imaging period were assigned as stable branches and used for quantification of branch number and length . All branches extended or retracted within this imaging period were defined as filopodia . Single cell labeling to analyze synapse and mitochondria distribution was achieved by injection of 1 nl of naked plasmid DNA ( 25 ng/μl ) into 1 cell stage embryos of the Tg ( Isl2b:Gal4 , cmlc2:eGFP ) , kif5aa162+/− transgenic line . The following constructs were used: p4nrUAS:tagRFPCaax-pA-4nrUAS:PhBGFP-pA-Tol2; p4nrUAS:tagRFPCaax-pA-4nrUAS:SypGFP-pA-Tol2;cmcl2:eGFP; p4nrUAS:SypGFP-pA-4nrUAS:PhBmcherry-pA-Tol2;cmcl2:eGFP . To generate single RGCs expressing the dominant negative ntrk3a receptor , we injected 1 nl of naked p10UAS:ntrk3aDN-eGFP-pA-Tol2;cmcl2:eGFP plasmid DNA into 1 cell stage embryos of the Tg ( Pou3f4:Gal4 ) , nacre+/− transgenic line . To generate single eGFP expressing RGCs growing into a ntf3 overexpressing tectum , we performed injections of 0 . 1 ng/μl pIsl2b:eGFPCaax plasmid DNA , 15 ng/μl p10UAS:ntf3-E2A-tagRFP-pA-tol2 , cmcl2:eGFP plasmid DNA and 50 ng/μl Tol2 transposase mRNA into 1 cell stage embryos of the Tg ( gSA2AzGFF49A ) ( Muto et al . , 2013 ) transgenic line . In control injections , no p10UAS:ntf3-E2A-tagRFP-pA-tol2 , cmcl2:eGFP plasmid DNA was injected . Imaging was performed on a Roper confocal spinning disk head mounted on a Zeiss upright microscope , and acquisitions were done with a CoolSNAP HQ2 CDD camera ( Photometrics , USA ) through the MetaMorph software ( Molecular Devices , Sunnyvale , CA ) . Embryos were anaesthetized using 0 . 02% tricaine ( MS-222 , Sigma-Aldrich , Saint Louis , MO ) diluted in egg water and embedded in 1% low melting-point agarose in glass-bottom cell tissue culture dish ( Fluorodish , World Precision Instruments , Sarasota , FL ) . Acquisitions were done using water immersion long working distance lenses , at 40× magnification ( W DIC PL APO VIS-IR; 421462-9900 ) for z-stack images of the whole tectum and at 63× magnification ( W PL APO VIS-IR; 421480-9900 ) for single plane time-lapse imaging of linear axonal segments . Images were assembled and analyzed in ImageJ ( NIH ) . Z-stack images were manual edited to remove skin autofluorescence . Time-lapse parameters were determined similar to previous studies ( Moughamian et al . , 2013; Niwa et al . , 2013 ) based on the speed of transport in the tectum and set at 5 s intervals for 15 min ( SypGFP ) and 20 min ( mitoGFP ) total duration . Time-lapse images were assembled and analyzed in ImageJ to determine the percentage of moving vs stable particles , as well as distribution/density and size of the organelles . Kymograms were extracted for each linear segment using the kymogram tool ( Montpellier RIO Imaging , CNRS , France ) . Extraction of small structures of the kymograms was done using the rotational watershed algorithm of the KymoMaker program ( Chiba et al . , 2014 ) and trace detection was done manually in ImageJ . 5- to 7-day-old nacre ( mitfa−/− ) or TL larvae were taken from a cross of kif5aa162+/− Tg ( HuC:GCaMP5G ) or Tg ( Isl2b:Gal4 ) , Tg ( UAS:GCaMP3 ) transgenic fish . They were immobilized in 2% low melting point agarose and mounted with the dorsal side up on a plexiglas platform . The platform was then placed in a custom-made chamber and immersed in E3 solution without methylene blue . The agarose around the eyes was cut away with a scalpel to allow for an unhindered view for the larvae . The larvae faced with one eye towards a glass cover slip in the chamber wall at a distance between 8 and 10 mm . Directly behind this glass cover slip and outside the chamber , a monochromatic OLED array ( 800 × 600 px , 13 × 9 mm , eMagin ) for visual stimulus presentation covering approximately 70° by 50° of the larva's visual field was positioned . A colored filter ( Kodak Wratten No . 32 ) was placed between glass cover slip and OLED to block green light emitted from the OLED thus allowing for simultaneous imaging and visual stimulation . Visual stimuli were synchronized to the acquisition and consisted of single black bars on a white background ( or the inverse ) running along the caudal rostral axis . The long axis of the bar was orthogonal to the direction of motion . Each bar was approximately 7° in width and moved at 16°/s . Visual stimuli were generated and controlled by custom scripts written in Matlab ( MathWorks , for details see Source code 1 ) using the Psychophysics Toolbox extensions ( Brainard , 1997; Pelli , 1997 ) . Confocal imaging of visually-evoked calcium responses in the tectum contralateral to the eye receiving the visual stimulus was performed using an upright microscope ( Roper/Zeiss , Germany ) equipped with a Spinning Disk head ( CSU-X1 , Yokogawa , Japan ) and a 40×/1 . 0 NA water-immersion objective ( Zeiss ) . Time-series streams of 5 min duration were acquired at 4 Hz with 0 . 323 × 0 . 323 μm spatial resolution ( 620 × 520 pixels ) . GCaMP3 or GCaMP5G were excited by a 491 nm laser and emitted light was bandpass-filtered ( HQ 525/50 ) . Images were taken with a CCD camera ( CoolSnap HQ2 Photometrics , Tucson , AZ ) . We did not observe any differences in larvae from 5 dpf to 7 dpf . Therefore , the data sets were combined for analysis . Occasionally , a visual response at the onset of the laser illumination or the power-on of the OLED was observed that quickly returned to baseline , probably due to habituation of the fish . Therefore , we excluded the first few seconds from acquisition analysis . Each stimulus epoch was presented for 4 . 4 s to every animal with an inter-epoch interval of 5 . 6 or 10 . 6 s to allow for the GCaMP signal to return to baseline values . After the experiment , fish were genotyped . Confocal time-series were pre-processed by correcting for motion with a translation algorithm ( Fiji [Schindelin et al . , 2012] ) . For each acquisition , ROIs for Neuropil and PVNs , respectively , were determined manually . Then the averaged ROI based time-series were smoothened by a low pass filter and the baseline signal was calculated by calculating the minimum in a time interval of 10 s before stimulus onset . The smoothened fluorescence signal and the baseline fluorescence were then used to calculate normalized signal intensity changes ( % ΔF/F0 ) . To identify pixels that were response locked to the stimulus , we performed linear regression on the Calcium evoked time-series ( Miri et al . , 2011 ) . For this , we convolved the stimulus time-series with an exponentially decaying kernel with half-decay times for GCaMP5G ( 667 ms ) ( Akerboom et al . , 2012 ) or GCaMP3 ( 597 ms ) ( Tian et al . , 2009 ) . This predicted fluorescence trace for the bar stimulus was then compared with the measured calcium traces for each pixel using Pearson correlation . 6 dpf larvae were anaesthesized in 0 . 004% Tricaine in E3 solution and then fixed in 2% glutaraldehyde in 0 . 1 M phosphate buffer , pH 7 . 2 . Tails were severed to increase permeability of fixation . Four drops of 4% OsO4 were added to 1 ml of glutaraldehyde fixative , and samples were soaked for 15 min . After three washes in 0 . 1 M phosphate buffer , samples were stored in 2% glutaraldehyde solution . Prior to embedding , samples were washed three times with 0 . 1 M phosphate buffer for 2 min and dehydrated with graded acetone: 35% acetone , 50% acetone , 75% acetone , 80% acetone , 95% acetone , and 100% acetone ( three times ) for 10 min per solution while shaking . Samples were infiltrated with Epon resin/acetone mixtures: Epon resin: acetone ( 1:3 ) , Epon resin: acetone ( 1:1 ) and Epon resin: acetone ( 3:1 ) followed by pure Epon resin containing the accelerant BDMA ( three washes ) . Finally , embedded samples were cured in a vacuum oven and sectioned with a RMC MT6000 Microtome to 70 nm slices . Images were acquired with an FEI Tecnai 12 Transmission electron microscope . Total RNA was prepared from 4 or 5 dpf embryos with TRIzol reagent ( ThermoFisher Scientific ) and TURBO DNA-free reagents ( ThermoFisher Scientific ) . RNA ( 1 μg ) was retrotranscribed using random primers and the SuperScript III First-Strand Synthesis system ( ThermoFisher Scientific ) . For q-RT-PCR , the SYBR Green PCR Master Mix ( ThermoFisher Scientific ) was used according to the manufacturers protocol and the PCR reaction was performed on an ABI PRISM 7900HT instrument . Ef1a and RPL13a were used as reference genes as reported previously ( Tang et al . , 2007 ) . All assays were performed in triplicate using 11 . 25 ng of cDNA per reaction . The mean values of triplicate experiments were calculated according to the deltaCT quantification method . Western blot analysis of embryo extracts was performed using standard techniques . Briefly , about 25 5 dpf larvae were homogenized in lysis buffer ( 20 μl/embryo ) containing: 10 mM HEPES , 300 mM KCl , 5 mM MgCl2 , 0 . 45% Triton , 0 . 05% Tween , Protease inhibitor-EDTA ( Mini Complete , Roche , Switzerland ) . Protein extracts ( about 20 μg/lane ) were separated by SDS-PAGE and subsequently blotted onto a PVDF membrane . Secondary antibody couples with Horseradish peroxidase ( 1:2000 ) were used to detect the anti-NTF3 ( 1:500 ) and anti-alpha-Tubulin ( 1:5000 ) primary antibodies ( see Antibody section above for details ) , and reveled using ECL Western Blotting Detection Reagents ( GE Healthcare Life Sciences , Pittsburgh , PA ) . Western blot quantification was performed using a cheminoluminescence digital imaging system ( ImageQuant Las-4000 Mini , GE Healthcare Life Sciences ) and analyzed using ImageJ software .	Different regions of a neuron have distinct structures and roles . For example , each neuron has a cable-like structure called the axon that extends out of the body of the cell and carries electrical signals away from the cell body . To pass these messages on to neighboring cells , branches on the axon form connections called synapses with other neurons . The axon lacks most of the cellular machinery needed to make proteins and other molecules that the cell needs to work correctly . Therefore , neurons must transport these materials from the cell body—where they are produced—down to the end of the axon . Specialized proteins called molecular motors carry this cargo down the axon along ‘tracks’ composed of filaments called microtubules . Auer , Xiao et al . have now used genetic techniques to disrupt the gene that encodes an important molecular motor , called Kif5A , in developing zebrafish larvae . The effects of this manipulation on the development of the zebrafish's visual system were then examined . When zebrafish are a few days old , neurons in the retina—the structure at the back of the eye that responds to light—extend axons into a region of the brain called the tectum . The formation of synapses between cells in the retina and the tectum provides a pathway that enables information to travel from the eye to the brain . Auer , Xiao et al . found that in larvae that lack Kif5A , axons from the retina enter the brain about a day later than they do in normal larvae . However , when these mutant axons arrive , they produce large numbers of branches , each with the potential to form multiple synapses with cells in the tectum . However , none of the resulting synapses appear to respond to visual stimuli , which is consistent with the fact that Kif5A mutant larvae are blind . Experiments to identify what triggers the excessive branching of retinal axons revealed that the mutant fish had elevated levels of a growth-promoting protein called neurotrophin-3 in cells in the tectum . This increased production of neurotrophin-3 was also observed when neuronal activity was blocked , for example by toxins . The lack of neuronal activity in retinal axons therefore seems to increase the production of neurotrophin-3 , which in turn stimulates axonal branching . Future experiments could investigate the molecular signal that drives this increased production of neurotrophin-3 , and how this is regulated during normal neuronal development .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2015	Deletion of a kinesin I motor unmasks a mechanism of homeostatic branching control by neurotrophin-3
Several neurodegenerative diseases are driven by the toxic gain-of-function of specific proteins within the brain . Elevated levels of alpha-synuclein ( α-Syn ) appear to drive neurotoxicity in Parkinson's disease ( PD ) ; neuronal accumulation of tau is a hallmark of Alzheimer's disease ( AD ) ; and their increased levels cause neurodegeneration in humans and model organisms . Despite the clinical differences between AD and PD , several lines of evidence suggest that α-Syn and tau overlap pathologically . The connections between α-Syn and tau led us to ask whether these proteins might be regulated through a shared pathway . We therefore screened for genes that affect post-translational levels of α-Syn and tau . We found that TRIM28 regulates α-Syn and tau levels and that its reduction rescues toxicity in animal models of tau- and α-Syn-mediated degeneration . TRIM28 stabilizes and promotes the nuclear accumulation and toxicity of both proteins . Intersecting screens across comorbid proteinopathies thus reveal shared mechanisms and therapeutic entry points . A number of neurodegenerative diseases are caused by the gradual accumulation of specific proteins within neurons . As research on these ‘proteinopathies’ has progressed , certain unexpected commonalities have arisen . For example , alpha-synuclein ( α-Syn , SNCA ) is now thought to mediate neurotoxicity not only in Parkinson’s disease ( PD ) , but also in Parkinson’s Disease Dementia ( PDD ) and Lewy Body Dementia ( LBD ) where it accumulates in Lewy bodies ( Spillantini et al . , 1997 ) . Similarly , the microtubule associated protein tau ( MAPT ) accumulates not only in neurons affected by Alzheimer’s disease ( AD ) , but in PD and PDD , while mutations that increase the stability of tau result in frontotemporal dementia with parkinsonism ( FTDP-17 ( or FTD ) [Spillantini and Goedert , 2013] ) . Coding region mutations in SNCA are by no means necessary for PD to develop given that duplication or triplication of the SNCA locus is sufficient to result in forms of PD whose onset and severity correlate with gene dosage ( Chartier-Harlin et al . , 2004; Ibanez et al . , 2004; Singleton , 2003 ) . Supporting these clinical findings are studies that show that overexpression of wild-type forms of either SNCA or MAPT elicit neurodegeneration in model organisms , whereas suppressing their levels appears to be neuroprotective ( Dauer et al . , 2002; Ishihara et al . , 1999; Jackson et al . , 2002; Rapoport et al . , 2002; Rockenstein et al . , 2002; Wittmann , 2001 ) . Given the brain’s sensitivity to the dosage of either of these proteins , one would expect elevated levels of multiple proteins to be even more problematic , and this proves to be the case ( Moussaud et al . , 2014 ) . Pathogenic proteins involved in different proteionopathies often interact with each other and cause cellular toxicity , either due to their additive effects on downstream activities or further compromise of protein homeostasis ( Clinton et al . , 2010 ) . Abnormally aggregated α-Syn and tau are often found together in postmortem cases of PD and LBD ( Arima et al . , 1999; Colom-Cadena et al . , 2013; Iseki et al . , 2002; Ishizawa et al . , 2003 ) , and genetic interaction studies in Drosophila demonstrate that α-Syn and tau synergize in promoting toxicity ( Roy and Jackson , 2014 ) . Biochemical evidence even suggests that α-Syn may act as an amyloidogenic ‘seed’ for the accumulation of tau , and vice versa ( Guo et al . , 2013; Lasagna-Reeves et al . , 2010; Sengupta et al . , 2015 ) . Even more , several genome-wide association studies have reported genetic interaction between tau and alpha-synuclein in PD pathogenesis ( Simón-Sánchez et al . , 2009 ) . Thus , the interaction between tau and α-synuclein is gaining increased attention for its possible pathogenic role in synucleinopathies and tauopathies . The mechanisms governing the dual accumulation of α-Syn and tau remain elusive , but it seems plausible that decreasing levels of either or both of these proteins could prove an effective therapeutic strategy for this family of diseases . Inspired by the extensive overlap between α-Syn and tau pathology and their corresponding clinical phenotypes ( Galpern and Lang , 2006 ) , we reasoned that they may be regulated through shared pathways and that dysfunction of these regulatory pathways may lead to their pathogenic accumulation . Therefore , convergent screens to find common modulators for α-Syn and tau levels would yield the most insight into these disease processes and possibly open up new avenues for therapeutic intervention . Importantly , targeting the root cause of the disease – protein accumulation – in an unbiased manner makes neither assumption about the mechanism of toxicity nor which cellular process is affected . Through convergent RNAi screens targeting the steady-state levels of α-Syn and tau , we found that TRIM28 regulates their levels and toxicity through their toxic nuclear accumulation . We employed a screening strategy similar to one recently used to identify key modulators of ATXN1 stability ( Park et al . , 2013; Westbrook et al . , 2008 ) , using high-throughput flow cytometry to monitor the steady-state levels of α-Syn and tau in a fluorescent bicistronic reporter system ( Figure 1A ) . We ran parallel screens interrogating 2607 siRNAs targeting 869 potentially druggable – i . e . potentially can be targeted pharmacologically – genes to identify genes that modify the levels of both α-Syn and tau ( Figure 1—figure supplement 1A–C and Figure 1—source data 1 ) . Applying stringent criteria and validation steps to narrow down the list of putative modifiers of α-Syn and tau levels , we uncovered Tripartite motif-containing 28 ( TRIM28 ) as the most robust common modulator ( Figure 1B; Figure 1—figure supplement 2 and Figure 1—source data 1 ) . We confirmed this effect of TRIM28 knock-down on α-Syn and tau stability using three independent siRNAs on our α-Syn and tau reporter cell lines along with a negative control cell line ( DsRed-IRES-EGFP ) to ensure that TRIM28 was not affecting the stability of EGFP ( Figure 1B ) . 10 . 7554/eLife . 19809 . 003Figure 1 . TRIM28 regulates levels of α-Syn and tau . ( A ) Schematic of screen approach ( see also Figure 1—figure supplement 1 ) . The ratio of either α-Syn:EGFP/DsRed or tau:EGFP/DsRed was measured using an arrayed siRNA library covering 2607 siRNAs in biological triplicates . Venn diagram shows significant overlapping hits from both screens . ( B ) Representative traces for α-Syn:EGFP/DsRed and tau:EGFP/DsRed ratios: the black curve represents a control condition ( siScramble ) , while the red and blue curves represent siTRIM28 in the α-Syn and tau cell lines , respectively . Quantified ratiometric scores for three independent siRNAs against TRIM28 ( siTRIM28-1 , -2 and -3 ) is presented on the right . ( C ) Effects of different shRNAs targeting TRIM28 on endogenous α-Syn and tau in HEK293T cells ( left panel , shTRIM28-1 and shTRIM28-2 ) , primary mouse neurons ( middle panel , shTrim28-1 , 1 , -2 , and -3 ) and in adult mouse hippocampus ( right panel , where ‘I’ denotes the injection side [ipsilateral] and ‘C’ denotes the uninjected side [contralateral , an internal control] ) . Rightmost panel depicts the effect of the loss of one allele of Trim28 in the mouse brain . Data are presented as mean ± s . e . m . for each group . In A , p=9 . 5 × 10–4 , hypergeometric test; in B , n = 3 per cell line , *** denotes p<0 . 001 , One-Way ANOVA followed by Dunnet’s multiple comparison test; in c , n = 4–13 condition , * , , * and ns denote p<0 . 05 , p<0 . 01 , p<0 . 001 and p>0 . 05 , respectively , One-way ANOVA followed by Holm-Sidak post-hoc test in two leftmost panels and Student’s t-test in two rightmost panels . Full statistical analyses for all figures are presented as Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 19809 . 00310 . 7554/eLife . 19809 . 004Figure 1—source data 1 . List of modifier genes identified in convergent screens . Genes screened as well as the individual Z-Scores for the average of triplicate readings are presented for each siRNA/cell line in adjacent columns . Top modifiers retrieved from the primary screens meeting criteria enumerated in Online Methods . Pink column depicts screen done in DsRed-IRES-α-Syn:EGFP cells and a blue column depicts screen done in DsRed-IRES-tau:EGFP cells . Overlapping hits between both screens are indicated in grey . Validation Results from Figure 1—figure supplement 2 summarized in the last columns with legend for color-coded significant changes presented at the bottom . DOI: http://dx . doi . org/10 . 7554/eLife . 19809 . 00410 . 7554/eLife . 19809 . 005Figure 1—figure supplement 1 . Convergent screens targeting the steady state levels of α-Syn and tau identify common modifiers . ( A ) siRNA library composition as determined using Panther gene ontology analysis – Protein function ( http://www . pantherdb . org/ ) . A full list of the genes tested is presented in Figure 1—source data 1 . ( B ) Screen workflow delineating steps used to narrow down hit lists and find top shared modifiers of α-Syn and tau levels . ( C ) Raw distribution of all hits in both ratiometric screens ( % Change denotes change in EGFP:DsRed ratio compared to scramble siRNAs from each plate ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19809 . 00510 . 7554/eLife . 19809 . 006Figure 1—figure supplement 2 . Validation of shared modifiers between α-Syn and tau . 40 shared modifiers from both α-Syn- and tau-based screens were tested using three new siRNAs directed against each target ( 'si #1 , 2 , 3' ) . Changes in EGFP:DsRed levels are presented as a percentage ( % ) change in relation to Scrambled siRNAs . For each histogram , DsRed-IRES-α-Syn:EGFP , DsRed-IRES-tau:EGFP and DsRed-IRES-EGFP cell lines are represented as red , blue and white bars , respectively . % Change denotes change in EGFP:DsRed ratio compared to scramble siRNAs . Data are presented as mean ± s . e . m . Three independent biological replicates are presented for three independent siRNAs ( siRNA codes are listed in Supplementary file 1 for statistical comparison ) . * , and * denote p<0 . 05 , p<0 . 01 and p<0 . 001 , respectively , One-Way ANOVA followed by Dunnett’s multiple comparison test . Bars without asterisks denote non-significant differences . DOI: http://dx . doi . org/10 . 7554/eLife . 19809 . 00610 . 7554/eLife . 19809 . 007Figure 1—figure supplement 3 . Validation of Trim28 knockdown and lack of effect on other neurodegenerative disease-causing genes . ( A ) Confirmation of viral expression was performed in the hippocampus of 12-week-old mice , 14 days after stereotaxic delivery of lentivirus . Cryosections were stained for turbo GFP ( tGFP ) and DAPI . Note top panels denote side injected with the virus ( ‘Ipsi’ , ipsilateral ) and bottom panels , the uninjected side ( ‘Contra’ , contralateral ) . ( B ) Confirmation of Trim28 knockdown was performed as in A , but using an antibody directed against Trim28 . ( C ) The effect of Trim28 knockdown on α-Syn and tau in primary neurons can be rescued by overexpressing an shRNA-resistant cDNA of TRIM28 , ( D ) TRIM28 knockdown in human cells does not alter MAPT transcript levels and only mildly affects SNCA expression as assayed by qRT-PCR . Trim28 heterozygous ( Trim28+/- ) mice have normal levels of Snca and Mapt in their brain . ( E ) TRIM28 knockdown does not alter protein levels of several other neurodegenerative disease-causing genes . HEK293T cells were infected with virus targeting TRIM28 ( shTRIM28-1 ) or control ( non-silencing , shScram ) and levels of neurodegenerative disease-causing genes were measured by Western blot and quantified below . Data are presented as mean + s . e . m . In C , n = 3–4 replicates , * , and * denote p<0 . 05 , p<0 . 01 and p<0 . 001 , respectively , One-Way ANOVA followed Holm-Sidak post-hoc test; in D , n = 3 replicates per group , top panel: *** and ns denote p<0 . 001 and p>0 . 05 , respectively , One-Way ANOVA followed by Holm-Sidak post-hoc test , bottom panel: * and ns denote p<0 . 05 and p>0 . 05 , respectively , Student’s t-test; in e , n = 3 replicates per group , ns denotes p>0 . 05 , Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 19809 . 007 We next used species-specific shRNA lentiviral vectors to suppress TRIM28 expression in human cells , in mouse primary cerebellar neurons , and in mouse hippocampus ( Figure 1C and Figure 1—figure supplement 3A , B ) . In each case , partial reduction of TRIM28 led to a decrease of both α-Syn and tau , and virally reintroducing a human shRNA-resistant TRIM28 cDNA reversed this effect ( Figure 1—figure supplement 3C ) . In addition , we found that deleting one allele of Trim28 in mice led to a modest but significant decrease of α-Syn and tau levels in mouse brain ( Figure 1C ) . Importantly , we observed little to no changes in SNCA nor MAPT transcript levels under conditions of TRIM28 downregulation ( Figure 1—figure supplement 3D ) . Nor did reduction of TRIM28 affect other neurodegeneration-causing proteins we tested ( Figure 1—figure supplement 3E ) . TRIM28 thus appears to be a key post-translational regulator of the steady state levels of α-Syn and tau . Based on studies showing that reducing disease protein levels can rescue neurodegeneration in other proteinopathy models ( Moreno et al . , 2012; Park et al . , 2013 ) , we tested whether decreasing TRIM28 would mitigate the phenotypes resulting from overexpression of either α-Syn or tau . For a tauopathy model , we generated transgenic Drosophila that express wild-type human tau in the eye and develop a visible degenerative phenotype . Knockdown of the Drosophila TRIM28 homolog ( bonus ) significantly reduced tau levels and mitigated eye degeneration ( Figure 2A; Figure 2—figure supplement 2C ) . Similarly , expression of wild-type tau in the Drosophila nervous system led to motor deficits in the climbing assay , but two independent partial loss-of-function alleles of TRIM28 improved this behavioral phenotype without reducing the transgene expression ( Figure 2B; Figure 2—figure supplement 1; Figure 2—figure supplement 2 and Video 1 ) . To ensure that this effect was mediated through tau levels , we tested the effect of TRIM28 loss of function on transgenic tau levels and found indeed that reduction of TRIM28 could effectively decrease tau levels in this heterologous system ( Figure 2C ) . 10 . 7554/eLife . 19809 . 008Figure 2 . Loss of TRIM28 mitigates tau-mediated neurodegenerative phenotypes in Drosophila . ( A ) tau overexpression in the Drosophila eye produces a rough eye phenotype ( third panel ) compared to negative controls ( first panel ) . Decreasing the levels of Trim28 ( dsTrim28 ) ameliorates this defect ( fourth panel ) . Decreasing Trim28 alone does not result in any overt degenerative phenotypes ( second panel ) . ( B ) Expression of tau in the Drosophila nervous system ( solid blue lines ) leads to motor performance deficits that can be quantified in a climbing assay when compared with control flies ( black lines ) . This phenotype is mitigated by partial TRIM28 loss of function ( hatched blue lines ) . Two independent cohorts ( 15 animals per replicate ) are shown per genotype . ( C ) Western blot images and quantification showing decreased tau levels in the adult Drosophila retina , upon Trim28 knockdown . This reduction of tau protein levels is concordant with the suppression of tau phenotypes shown in A and B . Data are presented as mean ± s . e . m . for each group . In B , n = 15 flies per replicate and 2 replicas per genotype * denotes p<0 . 001 , Two-Way ANOVA followed by Tukey-Kramer post-hoc test; in C , n = 6 replicates per group , * denotes p<0 . 001 , Student’s t-test . Scale bars in A: 100 µm ( inset 10 µm ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19809 . 00810 . 7554/eLife . 19809 . 009Figure 2—figure supplement 1 . Reduced function of TRIM28 alone does not produce abnormal behavioral phenotypes in Drosophila but rescues tau-mediated degeneration . ( A ) expression of a Trim28 loss of function allele in the CNS ( Trim28LOF-1 , yellow line ) does not induce motor performance deficits when compared to control flies ( black line ) . ( B ) tau-mediated motor phenotypes ( solid blue lines ) are suppressed by a second loss of function TRIM28 allele ( Trim28LOF-2 , hatched blue lines ) . ( C ) Quantification of ommatidial phenotypes from Figure 2A . Data are presented as mean ± s . e . m . In A , n = 15 flies per group , , * and ns denote p<0 . 01 , p<0 . 001 and p>0 . 05 , respectively , Two-Way ANOVA followed by Tukey-Kramer post-hoc test . DOI: http://dx . doi . org/10 . 7554/eLife . 19809 . 00910 . 7554/eLife . 19809 . 010Figure 2—figure supplement 2 . TRIM28 loss does not inhibit Gal4 driver expression . Levels of Gal4 expression were assayed using qPCR from the three different driver lines: elav-Gal4 ( A ) , GMR-Gal4 ( B ) and Rh1-Gal4 ( C ) . No significant decreases in expression were observed in the driver expression that could account for the suppression of phenotypes observed in Figure 2 *** and ns denote p<0 . 001 and p>0 . 05 , respectively , Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 19809 . 01010 . 7554/eLife . 19809 . 011Video 1 . Representative video . Partial loss of TRIM28 function in fruit flies overexpressing tau rescues motor behavior in the climbing assay . Video recording shows the three groups of animals tested in Figure 2B . The number of fruit flies climbing up 9 cm ( dotted line ) in 15 s was recorded . DOI: http://dx . doi . org/10 . 7554/eLife . 19809 . 011 To test whether Trim28 mediates neurodegeneration in a synucleinopathy , we turned into a mouse model of α-Syn-overexpression-induced Parkinsonism ( Burré et al . , 2012 ) , since the degenerative phenotypes of α-Syn-overexpressing flies are generally very mild . We co-injected lentiviral vectors expressing α-Syn together with a lentiviral vector expressing an shRNA against mouse Trim28 into the Substantia Nigra pars compacta ( SNc ) of mice and evaluated dopaminergic cell integrity eight weeks afterward . Mice infected with viruses overexpressing α-Syn and a control shRNA suffered a 50% reduction in tyrosine hydroxylase ( TH ) -positive dopaminergic neurons ( Figure 3A , B-top panels and Figure 3—figure supplement 1C ) . Remarkably , mice that received the shTrim28 virus together with α-Syn overexpression maintained their dopaminergic cells in the SNc and the corresponding fibers in the striatum ( Figure 3A , B-bottom panels ) . We confirmed that Trim28 was expressed ( and effectively knocked down in the case of shTrim28 treatment ) in these nigral neurons and that this rescue was due to a decrease in α-Syn levels and not the silencing of the α-Syn expression vector ( measured by IRES-driven GFP expression , Figure 3—figure supplement 1A , B ) . Lastly , to test whether this phenotypic rescue would hold true in a non-acute paradigm , we tested whether loss of one copy of Trim28 in mice could rescue some of the pathology observed in α-Syn-overexpressing mice ( Rockenstein et al . , 2002 ) . These mice possess a transgene driving wildtype α-Syn under the control of the mThy1 promoter and demonstrate age-dependent accumulation and aggregation of α-Syn that is demarked by its S129 phosphorylated form . We found that a 50% reduction in Trim28 significantly reduced pathological , phosphorylated α-Syn accumulation in the hippocampus where it is highly expressed ( Figure 3C , D ) ( Chesselet et al . , 2012 ) . 10 . 7554/eLife . 19809 . 012Figure 3 . TRIM28 knockdown suppresses α-Syn-mediated neurodegenerative phenotypes in vivo . ( A ) Representative photomicrographs of midbrain sections stained for tyrosine hydroxylase ( TH ) at the level of the Substantia Nigra pars compacta ( SNc , top panels ) or in the striatum ( bottom panels ) on the ipsilateral side to the virus injection . ( B ) Stereological quantification of TH+ cells ( top ) in the SNc and quantification of optical density of TH+ fibers ( bottom ) is presented on the right . ( C ) Western blot analysis of 3 . 5 month old α-Syn transgenic ( TG ) mice carrying two ( no mark , Trim28+/+ ) or one ( Trim28+/- ) copies of Trim28 . ( D ) phosphorylation of α-Syn at serine 129 ( pS129 ) staining at the level of the CA1 in these mice and quantification of positive cell numbers is presented on the right . Data are presented as mean + s . e . m . for each group . In B , n = 5–11 per group , * and ** denote p<0 . 05 and p<0 . 01 , respectively , One-Way ANOVA followed by Holm-Sidak post-hoc test; in B–D , n = 3–7 mice per group , * and denote p<0 . 05 and p<0 . 01 , respectively , Student’s t-test . Scale bars in A: 400 µm ( top panels ) 150 µm ( bottom panels ) , C: 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 19809 . 01210 . 7554/eLife . 19809 . 013Figure 3—figure supplement 1 . TRIM28 is expressed in the mouse SNc , and can be effectively knocked down in vivo . ( A ) On the left , representative immunofluorescence photomicrographs depicting uniform expression of Trim28 ( Red ) in the Substantia Nigra pars compacta ( SNc , stained by TH , Green ) . Knockdown of Trim28 was confirmed on the side of the injection ( Ipsi , top panel ) compared to the non-injected side ( Contra , bottom panel ) . ( B ) Representative western blot of tissue punches from the SNc two weeks following stereotaxic injection showing that Trim28 knockdown downregulates exogenous α-Syn ( Myc tagged ) but has no effect on the transcription of the viral vector ( GFP , expressed under the same promoter as Myc-α-Syn separated by IRES ) . ( C ) Nissl staining quantification pertaining to Figure 3B . Nissl positive ( Nissl+ ) cells in the medial terminal nucleus ( MTN ) region of the SNc were quantified to assess SNc viability . Data are presented as mean ± s . e . m . In B and C , n = 4 animals per condition; in D , n = 6–7 animals per condition , and ns denote p<0 . 01 and p>0 . 05 , respectively , One-Way ANOVA followed by Holm-Sidak post-hoc test . Scale bars in B: 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 19809 . 013 Since loss of TRIM28 rescued the neurodegenerative phenotypes , we surmised that increasing TRIM28 might exacerbate pathology . We performed bilateral lentiviral overexpression of TRIM28 in pre-symptomatic mouse models of synucleinopathy ( Rockenstein et al . , 2002 ) and tauopathy ( Yoshiyama et al . , 2007 ) and found that injection with TRIM28 , but not control lentivirus , worsened each aspect of neuropathology tested in both models ( Figure 4A , B; and Figure 4—figure supplement 1A , B ) . We were particularly intrigued to find pathological , phosphorylated , forms of α-Syn ( S129 ) and tau ( S396 ) accumulating at this earlier stage . Given the efficacy of TRIM28 at tightly regulating the levels of α-Syn and tau in a post-translational manner , we reasoned that TRIM28 was likely stabilizing these proteins therefore promoting their toxic accumulation . To test this , we generated doxycycline-inducible cell lines where we could carefully assess the half-lives of α-Syn and tau without altering global protein homeostasis ( Figure 5A ) ( Meerbrey et al . , 2011 ) . Specifically , we cloned wild-type α-Syn and tau into doxycycline-inducible lentiviral vectors ( pINDUCER system ) , infected SH-SY5Y cells and selected for clones that stably express either α-Syn or tau upon induction . We found that the half-lives of α-Syn and tau ( measured following a 48 hr pulse of doxycycline ) in this system resembled those previously observed by others ( 23 . 3 hr for α-Syn and 28 . 1 hr for tau [Li , 2004; Min et al . , 2010] ) . Importantly , we found that TRIM28 lentiviral overexpression could significantly increase the half-life of each protein by approximately 50% ( LV-TRIM28 vs . LV-Ctrl , Figure 5B , C ) , without affecting their RNA stability ( Figure 5—figure supplement 1 ) . Together , these studies suggest that TRIM28 promotes the accumulation of the amyloidogenic proteins α-Syn and tau . 10 . 7554/eLife . 19809 . 014Figure 4 . TRIM28 expression worsens histopathology in mouse models of synucleinopathy and tauopathy . Transgenic mice overexpressing α-Syn ( A , mThy-Syn 'Line 61' ) or P301S tau ( B , PS19 ) were injected at a presymptomatic stage in the hippocampus with lentiviruses expressing TRIM28 . Pathological evaluation of phosphorylation of α-Syn at serine 129 ( pS129 , top panels ) ; Glial Fibrillary Acidic Protein ( GFAP , middle panels ) as well as Nissl staining ( bottom panels ) in the CA1 region of α-Syn transgenic ( TG ) mice ( solid bars ) and their wild-type littermates ( hatched bars ) was performed ( quantification on the right of each panel sets ) . Similar pathological evaluation of phosphorylation of tau at serine 396 ( pS396 tau ) ; GFAP; as well as Nissl staining in the CA1 region of P301S tau Transgenic mice . Data are represented as mean + s . e . m . In A and B , n = 3 for each genotype and treatment for each experiment , * , and * denote p<0 . 05 , p<0 . 01 and p<0 . 001 , respectively , One-Way ANOVA followed by Holm-Sidak post-hoc test . Scale bars in A and B: 25 µm ( top panels ) , 100 µm ( middle panels ) and 50 µm ( bottom panels ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19809 . 01410 . 7554/eLife . 19809 . 015Figure 4—figure supplement 1 . Effective transduction of lentivirus expressing TRIM28 in the hippocampus intensifies biochemical accumulation of α-Syn and tau in prodromal mouse models of synucleinopathy and tauopathy . ( A ) Representative photomicrographs of mCherry staining in the hippocampus of a mouse injected with a control lentivirus expressing mCherry ( left panel ) or TRIM28 fused to mCherry ( right panel ) . ( B ) Representative western blot of hippocampal punches from α-Syn Transgenic and littermate mice ( left panel ) and tau P301S transgenic and littermate mice ( right panel ) injected with lentiviruses overexpressing TRIM28 ( or control lentiviruses ) . Levels of pS129 or total α-Syn ( in the case of α-Syn TG mice ) or pS396 or total tau ( in the case of tau TG mice ) are quantified to the right of the blots . Data are represented as mean + s . e . m . In B , n = 3–6 for each genotype and treatment for each experiment , * and ** denote p<0 . 05 and p<0 . 01 , respectively . Scale bars in A: 100 µm ( top panels ) and 20 µm ( bottom panels ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19809 . 01510 . 7554/eLife . 19809 . 016Figure 5 . TRIM28 stabilizes α-Syn and tau protein levels . ( A ) Schematic of viral vector used to generate stable cell lines ( adapted from [Meerbrey et al . , 2011] ) and assay design . ( B ) Representative western blots of transgenic α-Syn ( top ) and tau ( bottom ) cell lines following different times of doxycycline ( DOX ) removal; NT denotes non-DOX treated cells; * denotes detection of endogenous tau ( not quantified ) . ( C ) Quantification of α-Syn and tau stability are presented as a % of baseline and fit to a logarithmic ( log2 ) scale . Dashed lines denote control groups whereas solid lines denote TRIM28 overexpression . Data are presented as mean ± s . e . m . for each group . In C , n = 6 per group , per time point . Protein levels follow a one-phase exponential decay with both curves being significantly different from one another ( Comparison of fits , p=0 . 0111 , α-Syn;LV-Ctrl vs . α-Syn;LV-TRIM28; p=0 . 0003 , tau; LV-Ctrl vs . tau;LV-TRIM28 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19809 . 01610 . 7554/eLife . 19809 . 017Figure 5—figure supplement 1 . TRIM28 overexpression does not affect SNCA and MAPT RNA stability . Quantification of SNCA and MAPT levels following indicated doxycycline withdrawal time points reveals expected RNA decay over time , one-phase exponential decay fit to a logarithmic ( log2 ) scale . No significant changes were observed following TRIM28 overexpression ( Comparison of fits , ns denotes p>0 . 05 , SNCA;LV-Ctrl vs . SNCA;LV-TRIM28 or MAPT;LV-Ctrl vs . MAPT;LV-TRIM28 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19809 . 017 Given that TRIM28 regulates the stability of these two proteins , we next sought the consequence of TRIM28-mediated regulation of α-Syn and tau . We asked whether TRIM28 modulates α-Syn and tau by forming a complex with them and found that they did indeed interact , albeit weakly , in human cells ( Figure 6A ) . We also used an alternative approach to confirm our findings: Bimolecular Fluorescence Complementation ( BiFC; Figure 6B ) . We generated stable cell lines expressing N-terminal YFP tagged versions of α-Syn and tau ( nYFP-α-Syn and nYFP-tau ) . By themselves , the cell lines did not generate any fluorescence ( data not shown ) but when a known interactor of each ( in this case , tau ) was co-transfected as a prey , we could elicit a strong fluorescent signal ( tested by flow cytometry ) . We found that using TRIM28 as a prey could elicit a signal roughly eight times lower than that of tau thus confirming the transiency of the interaction . Intriguingly , TRIM28:α-Syn and TRIM28:tau complexes accumulated in nuclei after some time . To confirm that this effect was not an artifact of the BiFC approach , we overexpressed a mCherry-fused TRIM28 lentiviral construct in primary cerebellar granule neurons and looked at the endogenous localization of α-Syn and tau . Again , both by immunofluorescence and by biochemical fractionation , we found that viral expression of TRIM28 dramatically promoted the nuclear accumulation of these two proteins ( Figure 6C and Figure 6—figure supplement 1 ) . Given our previous data suggesting that the effects of TRIM28 on α-Syn and tau are posttranslational , we explored whether the RING domain of TRIM28 might play a role in its modulatory effect . To this end , we focused on the known E3-ligase catalytic activity of TRIM28 . We overexpressed an inactive TRIM28 E3 ligase mutant ( [C65A/C68A] , ‘TRIM28-Mut’ ) ( Doyle et al . , 2010 ) and found that this mutant could no longer promote the toxic nuclear accumulation of these proteins ( Figure 6C and Figure 6—figure supplement 1 ) . Importantly , we found that the TRIM28 E3 ligase mutant , while less stable , could still bind α-Syn and tau ( Figure 6—figure supplement 2 ) , suggesting that its effect on α-Syn and tau nuclear accumulation is functionally , rather than structurally , driven . Taken together , TRIM28 governs the nuclear accumulation of α-Syn and tau through its E3-ligase domain , and its persistent expression aggravates the neuropathology of both proteins . 10 . 7554/eLife . 19809 . 018Figure 6 . TRIM28 binds to and drives the nuclear localization of α-Syn and tau . ( A ) Immunoprecipitation for endogenous α-Syn ( top panel ) and tau ( bottom panel ) from HEK293T cells showing the interaction between the former proteins and TRIM28 . ( B ) Bimolecular Fluorescence Complementation studies using stable cell lines expressing either α-Syn or tau fused to an n-Terminal moiety of YFP ( α-Syn:nYFP or tau:nYFP , respectively ) and TRIM28 ( solid ) , tau ( positive control , hatched ) or MSK1 ( negative control , white ) fused to a c-terminal moiety of YFP ( TRIM28:cYFP , tau:cYFP and MSK1:cYFP , respectively ) . Visualization of the epifluorescence generated through protein-protein interaction is depicted in the photomicrographs and quantified by flow cytometry on the right . Note the nuclear accumulation of the interacting proteins 72 hr following transfection . ( C ) Primary mouse neurons infected with lentiviruses harboring TRIM28 , TRIM28 E3 ligase mutant and control were stained for endogenous α-Syn ( left panel ) , and tau in ( middle panel ) and the relative nuclear fraction of each was quantified ( right panel ) . Arrows point to endogenous α-Syn and tau in the nucleus . Data are represented as mean + s . e . m . In B , n = 3 per group and *** denotes p<0 . 001 , One-Way ANOVA followed by Holm-Sidak post-hoc test; in C , n = 3 per group and * , ** and ns denote p<0 . 05 , p<0 . 01 and p>0 . 05 , respectively , One-Way ANOVA followed by Holm-Sidak post-hoc test . DOI: http://dx . doi . org/10 . 7554/eLife . 19809 . 01810 . 7554/eLife . 19809 . 019Figure 6—figure supplement 1 . TRIM28 promotes the nuclear localization and accumulation of α-Syn and tau . Primary mouse neurons infected with lentiviruses harboring TRIM28 , TRIM28-Mut or a control mCherry fluorophore were biochemically fractionated into cytoplasmic ( ‘Cyto . ’ ) and nuclear ( ‘Nuc . ’ ) components . Relative α-Syn and tau were measured both in the cytoplasm and nucleus in each condition . Notably , the TRIM28-E3 mutant construct consistently exhibited diminished levels . Data are presented as mean + s . e . m . n = 3–4 replicates per condition . * , ** and ns denote p<0 . 05 , p<0 . 01 and p>0 . 05 , respectively , One-Way ANOVA followed by Holm-Sidak post-hoc test . DOI: http://dx . doi . org/10 . 7554/eLife . 19809 . 01910 . 7554/eLife . 19809 . 020Figure 6—figure supplement 2 . TRIM28 catalytic mutant retains some binding capacity to α-Syn and tau . ( A ) α-Syn and tau were immunoprecipitated from HEK293T cells transfected with HA-tagged wild-type TRIM28 or its E3 ligase mutant ( C65A/C68A ) counterpart . * denotes IgG heavy chain immunoreactivity . ( B ) Quantification of the relative amount of HA-TRIM28 ( or its mutant ) bound to α-Syn and tau reveals a decrease in binding affinity in the catalytic mutant . * and ns denote p<0 . 05 and p>0 . 05 , respectively , Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 19809 . 020 If TRIM28 is indeed promoting accumulation of α-Syn and tau in a disease context , levels of TRIM28 could also be deregulated . We therefore obtained post-mortem tissue from individuals who had had PDD , AD , or progressive supranuclear palsy ( PSP ) , along with age-matched controls , and examined the biochemical distribution of TRIM28 . In all cases of synucleinopathy and tauopathy , a greater proportion of TRIM28 accumulated in an insoluble form ( Formic Acid fraction ) than the more soluble form ( RIPA fraction ) ( Figure 7—figure supplement 1A–D ) . Subcellular fractionation on these post-mortem tissues revealed that α-Syn and tau accumulated in the nucleus in most cases of synucleinopathy and tauopathy ( Figure 7—figure supplement 2A , B ) . Then , we looked at α-Syn and tau subcellular localization in relation to TRIM28 in the context of PD and AD , respectively . We found that α-Syn and tau nuclear co-localization with TRIM28 was greater in PD and AD than in age-matched controls ( Figure 7A , B ) . Taken together , our findings suggest that TRIM28 is aberrantly increased in cases of synucleinopathy and tauopathy and is pathologically associated with the nuclear accumulation of α-Syn and tau in a diseased state . 10 . 7554/eLife . 19809 . 021Figure 7 . α-Syn and tau colocalize with TRIM28 in the nucleus in cases of synucleinopathy and tauopathy . ( A ) Representative photomicrographs of the medial frontal gyrus of cases of PDD and age-matched controls stained for α-Syn , TRIM28 and DAPI . The relative proportion of DAPI positive cells were quantified and are presented as the fraction of total nuclei counted . ( B ) Representative photomicrographs as in A but for cases of AD and respective age-matched controls , quantification on the right . Data are represented as a fraction of total . In A , n = 4 per post-mortem group , ** and ns denote p<0 . 01 and p>0 . 05 , respectively , Student’s t-test; in B , n = 5 per post-mortem group , * and ns denote p<0 . 05 and p>0 . 05 , respectively , Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 19809 . 02110 . 7554/eLife . 19809 . 022Figure 7—figure supplement 1 . TRIM28 levels are deregulated in human cases of synucleinopathy and tauopathy . ( A ) Medial frontal gyrus ( MFG ) samples from post-mortem Parkinson’s disease with dementia ( PDD ) or age-matched control cases sequentially lysed in RIPA buffer ( left panel ) followed by Formic Acid ( right panel ) . ( B ) Pons samples from Progressive supranuclear palsy ( PSP ) and control cases prepared as in A . ( C ) MFG samples from Alzheimer’s disease ( AD ) and Control cases prepared as in A . ( D ) Quantification of the relative amount of insoluble – FA enriched – TRIM28 relative to RIPA-soluble TRIM28 . Left , middle and right panels denote quantification of PDD , AD and PSP cases , respectively . Data are presented as mean ± s . e . m . In D , n = 5–10 cases per group , * and ** denote p<0 . 05 and p<0 . 01 , respectively , Student’s t-test with Welch’s correction . DOI: http://dx . doi . org/10 . 7554/eLife . 19809 . 02210 . 7554/eLife . 19809 . 023Figure 7—figure supplement 2 . α-Syn and tau accumulate in the nucleus in cases of synucleinopathy and tauopathy . ( A ) Quantitative assessment of nuclear and cytoplasmic localization of α-Syn and tau in the pons of post-mortem cases of PDD ( top panel ) , PSP ( middle panel ) and AD ( bottom panel ) compared to age-matched/tissue-matched controls . ( B ) Quantification of relative nuclear enrichment of α-Syn ( top ) and tau ( bottom ) . ( C ) Model: In normal conditions , TRIM28 resides primarily in the nucleus , whereas α-Syn and tau perform their respective physiological functions in the cytoplasm and their levels are tightly controlled by cytosolic quality control mechanisms; in disease conditions the levels of α-Syn and tau as well as TRIM28 are abnormally high . TRIM28 mediates the nuclear accumulation , potentially via SUMOylation ( denoted by the ‘S’ ) of α-Syn and tau where they become toxic . Data are presented as mean ± s . e . m . In B , n = 10 cases per condition . * and ns denote p<0 . 05 and p>0 . 05 , respectively , Student’s t-test with Welch’s correction . DOI: http://dx . doi . org/10 . 7554/eLife . 19809 . 023 The mechanism ( s ) underlying the neurotoxicity of α-Syn and tau have been difficult to pin down ( Cookson and van der Brug , 2008; Ward et al . , 2012 ) . Although accumulating evidence suggests that α-Syn and tau pathology propagate from cell to cell , the exact mechanism driving their initial accumulation and toxicity remains unclear . Given that both α-Syn and tau accumulate in Parkinson Disease Dementia and Lewy Body Dementia ( Moussaud et al . , 2014; Sengupta et al . , 2015 ) , we rationalized that there must be some proteins that regulate both of them . Thus , when screening for posttranslational regulators that modulate the levels of each protein , we were most interested in regulators that can modify the levels of both α-Syn and tau . Through such convergent screens , we found that TRIM28 regulates the levels of both disease-driving proteins . Our finding that TRIM28 regulates α-Syn and tau by promoting their nuclear localization and accumulation dovetails nicely with studies indicating that altered cellular distribution of α-Syn and tau contributes to neurodegeneration ( Fares et al . , 2014; Fernández-Nogales et al . , 2014; Kontopoulos et al . , 2006 ) . Indeed , several lines of evidence suggest that mislocalization or missorting of α-Syn and tau as a result of mutations , post-translational modification or overexpression , rather than their respective aggregation in the form of Lewy bodies and NFTs , are what promotes neurodegeneration ( Frandemiche et al . , 2014; Wilson et al . , 2004 ) . For instance , α-Syn is much more toxic in a fly model of synucleinopathy when it is tagged with a nuclear localization sequence ( NLS ) ( Kontopoulos et al . , 2006 ) . Further , PD-causing mutations in α-Syn increase its nuclear accumulation ( Fares et al . , 2014; Kontopoulos et al . , 2006 ) . Moreover , postmortem studies of tissue from AD and Huntington disease patients indicate that misfolded tau accumulates in the nucleus of neurons in the form of rod-like deposits ( Fernández-Nogales et al . , 2014; Vuono et al . , 2015 ) . Our finding that TRIM28 drives the nuclear accumulation of α-Syn and tau provides new mechanistic insight into the steps that lead to the pathogenicity of these proteins and highlights a shared pathway for therapeutic targeting . The molecular mechanism through which TRIM28 stabilizes and mediates the nuclear localization of α-Syn and tau is not elucidated at this point . Indeed , it is possible that TRIM28 mediates its effects on α-Syn and tau indirectly through a yet undiscovered mediator , through epigenetic remodeling or indirect binding and/or modification of an intermediate ( Cheng , 2014 ) . However , the evidence we present suggests that this less likely of a mechanism . For instance , we show that TRIM28 does not affect the RNA of α-Syn and tau but rather their protein stability . Moreover , we show that TRIM28 can form a complex with the two proteins and that its effect on nuclear accumulation is mediated via its E3-ligase domain . This would therefore suggest that TRIM28 may mediate this effect through SUMOylation or ubiquitination . Since SUMOylation has previously been suggested to mediate the nuclear localization of certain proteins , we hypothesize that TRIM28 acting as a SUMO ligase for α-Syn and tau . Future studies will dissect such a mechanism through the use of genetically engineered mice that permit monitoring of either modification in vivo . In the present study , we found that TRIM28 has important post-transcriptional functions in the brain . While the Trim28 null mice are embryonic lethal ( Cammas et al . , 2000 ) , we found that the Trim28+/- mice looked and behaved normally and that 50% loss of Trim28 was sufficient to reduce α-Syn and tau and ameliorate toxicity . This finding is particularly encouraging when it comes to therapeutics as it suggests that only partially inhibiting TRIM28 function may be effective at blocking neurotoxicity . Moreover , having discovered that mutating the E3 ligase catalytic activity of TRIM28 effectively is sufficient to regulate α-Syn and tau levels and for this specific effect behaves like a null allele , narrows down a target domain within TRIM28 for inhibition . This finding sets the stage for future studies to dissect the native function of the E3 ligase domain of TRIM28 versus its other essential domains and provides a potential opportunity for targeting TRIM28 as a source of therapy for these debilitating disorders . The unusual accumulation of TRIM28 in cases of synucleinopathy and tauopathy was a curious finding , though it remains solely an observation of correlative nature at this point . While we found that both α-Syn and tau co-localized with TRIM28 in the nucleus in cases of PD and AD , respectively , it was also interesting to find that TRIM28 accumulated in the insoluble biochemical fractions from post-mortem tissue . Moreover , though TRIM28 accumulation occurs in both diseases , irrespective of the driving pathological protein , α-Syn and tau appeared to more selectively accumulate in the nucleus in the context of their respective disease . Teasing out the precise mechanisms that affect TRIM28 accumulation as well as α-Syn and tau nuclear localization in human tissue will provide additional context for the understanding of disease pathogenesis . We posit two potential mechanisms through which deregulated TRIM28 activity could promote neurodegeneration . By driving α-Syn and tau to the nucleus , TRIM28 could: ( 1 ) prevent their degradation by native quality control mechanisms , thereby enhancing their overall bioavailability and toxicity ( a passive mechanism ) ; or ( 2 ) allow for a toxic gain of function in the nucleus ( an active mechanism ) ( model , Figure 7—figure supplement 2 ) . Though some studies have suggested various roles for α-Syn and tau in the nucleus ( Fernández-Nogales et al . , 2014; Kontopoulos et al . , 2006; Sultan et al . , 2011 ) , careful studies testing the kinetics and the outcome of this nuclear build-up will surely shed light on disease pathogenesis . As the sensitivity of the brain to protein levels becomes clearer , and more neurodegenerative diseases are found to involve elevated levels of more than one protein ( Moussaud et al . , 2014; Ramanan and Saykin , 2013 ) , it will become more important to identify shared modifiers and regulatory mechanisms of the steady-state levels of disease proteins , both to understand pathogenesis and to find the best candidate targets for therapeutic interventions . α-Syn , alpha-synuclein; AD , Alzheimer disease; BCA , bicinchoninic acid; BiFC , bimolecular fluorescence complementation; CA1 , Cornu Ammonis 1; ‘C’ or CONTRA , contralateral; cDNA , complementary DNA; DsRed , Discosoma red fluorescent protein; DMEM , dulbecco’s modified eagle medium; DOX , Doxycycline; EGFP , enhanced green fluorescent protein; FACS , fluorescent activated cell sorting; FBS , fetal bovine serum; FTD , Frontal Temporal Dementia; GFAP , Glial Fibrillary Acidic Protein; HEK , human embryonic kidney; HTS , high throughput sampler; ‘I’ or IPSI , ipsilateral; IP , immunoprecipitation; IRES , internal ribosomal entry site; LBD , Lewy Body Demetia; LV , Lentivirus; Mut , mutant; PD , Parkinson disease; PDD , Parkinson disease with dementia; PFA , paraformaldehyde; RNAi , ribonucleic acid interference; PSP , Progressive Supranuclear Palsy; RIPA , Radioimmunoprecipitation assay ( buffer ) ; RRE , Rev response element; siRNA , small interfering ribonucleic acid; shRNA , short hairpin ribonucleic acid; SNc , substantia nigra pars compacta; SUMO , Small Ubiquitin-Like Modifier; TH , tyrosine hydroxylase; TU , Transducing units; TRIM28 , Tripartite Motif Containing 28; UBC9 , Ubiquitin carrier protein 9; and VSVG , vesicular stomatitis virus-G . The collection of antibodies ( and their concentrations ) , shRNAs ( and their sequences ) , Plasmids ( and their construction ) as well as qRT-PCR primers used are presented in Supplementary file 2 . Lentiviral vectors ( pHAGE , L302 , LV-mCherry , pGIPz or pZip ) and their respective packaging vectors ( VSVG , Rev and RRE for L302; psPAX2 and pMD2G for LV-mCherry , pGIPz and pZip ) were cotransfected into HEK293T cells in a 1:1:1:1 or 4:3:1 molar ratio , respectively . Media was changed 16 hr following transfection to low volume media ( 5 mL for a 10 cm dish ) . Media was collected at 48 hr following transfection , replaced with fresh media ( 5 mL ) and collected again at 72 hr . Viral supernatant was cleared from cell debris via centrifugation ( 10 min at 4000 rpm ) as well as filtration through a 0 . 45 µM polyethersulfone membrane ( VWR ) . Cleared supernatants were concentrated using Lenti-X concentrator ( Clontec ) to 1/50–1/100 of the original volume . Viruses were titered using the Open Biosystems ( Thermoscientific , Waltham , MA ) method ( counting either fluorescence-positive colonies or puromycin-resistant colonies ) . DsRed-IRES-α-Syn:EGFP , DsRed-IRES-tau:EGFP , DsRed-IRES-EGFP , α-Syn:nYFP and tau:nYFP cell lines were generated as previously described ( Park et al . , 2013 ) . Briefly , each construct was cloned into a pHAGE vector , packaged into lentiviruses and infected into Daoy cells at low multiplicity of infection ( 0 . 3 ) to promote single copy integration . Cells were then selected by puromycin ( 1 µg/mL ) and single clonal populations were obtained using fluorescence activated cell sorting ( FACS ) using an Aria II cell sorter ( BD Biosciences ) and sorting a single double-positive ( DsRed and EGFP ) cell per well . Clones were expanded and the ones with the best qualities for screening ( i . e . , low transgene expression , low variation , homogenous population ) were expanded for screening purposes . Flow cytometry was carried out on a LSRII Fortessa coupled with an HTS module ( BD Biosciences ) . Gates for analysis were set using single-positive and double-positive cells as well as negative cells . For the RNAi Screens , DsRed-IRES-α-Syn:GFP or DsRed-IRES-tau:EGFP cells were plated onto 96-well round bottom plates . For each RNAi plate , cells were plated in triplicate to assay for technical variance . The following day , individual siRNAs were transfected as previously reported ( Park et al . , 2013 ) and incubated for 72 hr for downstream flow cytometry analysis . For each plate tested , Z-Scores were calculated by comparing each sample to the plate mean . A cut-off Z-Score of 1 . 5 was set in either direction for a first pass of hits . Next , individual t-tests were performed on hits by comparing them to the average of three scrambled siRNA sequences ( with Low- , Medium- , and High GC content ) to control for variability between triplicates . Hits meeting a p-value cut-off of <0 . 05 were kept for downstream analysis . Top hits from each screen met the Z-score and p-value cut-off and had more than two siRNAs with a significant effect ( regardless of the directionality of said effect ) ( Birmingham et al . , 2009 ) . Validation of top hits was performed by ordering new siRNAs against most robust hits and revalidating in the same platform but this time including a DsRed-IRES-GFP cell line as a negative control ( see Figure 1—figure supplement 2 and Figure 1—source data 1 ) . Hits that significantly altered the DsRed-IRES-GFP cell line in the same direction were excluded . Further downstream analysis consisted of generating viral shRNAs against the top hits ( using the human pGIPz collection , ThermoScientific ) and testing for their effect on endogenous α-Syn and tau levels in HEK293T cells as shown in Figure 1C . Daoy , HeLa or HEK293T cells ( obtained and certified from ATCC ) were cultured in DMEM ( Invitrogen ) containing 10% FBS and antibiotics ( Penicillin/Streptomycin ) . These lines are not misidentified as per ICLAC and are free from mycoplasma contamination . siRNAs were transfected using DharmaFECT ( Dharmacon ) and incubated for 72 hr prior to analysis by flow cytometry or western blot . shRNA viruses were generated as described below , infected into cells , and left for 9 days prior to analysis . Plasmids were transfected using Lipofectamine 2000 ( Invitrogen ) and left to express from 30 to 72 hr , depending on the downstream application . Unless otherwise mentioned , cells were lysed with RIPA buffer ( 50 mM Tris-HCl [pH 8 . 0] , 150 mM NaCl , 1 % NP-40 , 0 . 5% sodium deoxycholate , 0 . 1% SDS ) supplemented with protease inhibitors ( Roche ) on ice for 20 min with vortexing . Lysates were cleared by centrifugation ( 20 min , 15 , 000 r . p . m , 4°C ) followed by protein quantification via BCA assay ( Pierce ) . Sample buffer with reducing agent was added to each lysate followed by a 10 min incubation at 95°C . Samples were spun down and run on a 4–12% Bis-Tris gel , transferred to a nitrocellulose membrane and blocked for one hour with 5% non-fat milk prior to primary antibody incubation . Primary cerebellar granule precursors were generated as previously described ( Aleyasin et al . , 2007 ) with slight modifications . Briefly , cerebella from P5-P9 mice were dissected and dissociated with trypsin . Granule neurons were specifically isolated using a Percoll gradient following which single cell suspensions were plated onto 12-well plates previously coated with poly-L-lysine . Cells were left to recover for 24 hr prior to infection with 4 µL of concentrated virus ( ~1E7 Transducing units [TU] / µL ) . Cerebellar granule neurons were used as opposed to cortical and hippocampal cultures due to their ease in infectivity ( >80% infected , data not shown ) and their homogeneity thus allowing for clean biochemical experiments . Stereotaxic injections were performed as previously described ( Lasagna-Reeves et al . , 2015b ) . Briefly , 8–12 week-old mice were deeply anesthetized with a ketamine/xylazine cocktail; their heads were shaven and cleaned using aseptic solution . A midline incision was performed to reveal Bregma and a burr hole was drilled at the appropriate coordinates . Mice were injected with concentrated lentiviral solution ( 1E7 TU / µL; 1 µL unilateral , left hemisphere for SNc injections , 3 µL of 2 . 5E7 TU / µL bilaterally for hippocampal injections ) at a flow rate of 0 . 125 µL/min . For the hippocampal injections , coordinates relative to Bregma were: 2 . 1 mm posterior , 1 . 3 mm lateral and 1 . 5 mm ventral . For SNc injections , coordinates relative to Bregma were 3 . 2 mm posterior , 1 mm lateral and 4 . 2 mm ventral . Incisions were glued together using VetBond ( 3 M ) . Mice were followed for proper recovery following injection and received daily fluids ( 0 . 5 mL subcutaneous saline ) and analgesics ( 100 µL IP Ketoprofen ) for 72 hr following the injection . Free floating or slide-mounted sections were stained as before ( Lasagna-Reeves et al . , 2015a; Rousseaux et al . , 2012 ) . Briefly , sections were permeabilized and blocked in PBS containing 0 . 3% Triton X-100 and 5% normal goat serum . Samples were incubated with primary antibody overnight in blocking buffer and were washed three times the next day prior to adding fluorescent-conjugated secondary antibody ( either Alexa 488 or Alexa 568 , 1:700 concentration ) and incubating an additional hour at room temperature . Tissue was then washed three more times during and DAPI was added during one of the washes to permit nuclear visualization . Slides were coverslipped using fluoromount and imaged using a Zeiss LSM710 confocal microscope . mThy1-α-Syn ( Line 61 [Rockenstein et al . , 2002] ) mice were a gift from Marie-Francoise Chesselet and were bred and maintained on a C57Bl/6J background . Note that in our hands , these mice exhibited stronger pathological phenotypes than those originally reported; this may be due to their background ( data not shown ) . tau transgenic mice , overexpressing human tau with a P301S mutation ( PS19 [Yoshiyama et al . , 2007] ) , were obtained from the Jackson Laboratory were maintained on a C57Bl/6J background . Trim28+/- mice were generated by crossing Trim28flox/+ ( Cammas et al . , 2000 ) ( B6 . 129S2[SJL]-Trim28tm1 . 1Ipc/J , Jackson Laboratory ) mice with CMV-Cre mice ( Schwenk et al . , 1995 ) ( B6 . C-Tg[CMV-cre]1Cgn/J , Jackson Laboratory ) . Genotyping for Trim28+/- mice was performed using standard methods and the following primers to detect the recombined allele: 5’-TTGTTTATTTGGGAATGGTTGTTC-3’ and 5’-GCGAGCACGAATCAAGGTC-3’ . Mapt-/- mice were a generous gift from J . Noebels ( BCM ) and Snca-/- mice were obtained from the Jackson Laboratory ( B6;129 X 1-Sncatm1Rosl/J ) . C57Bl/6J mice used for stereotaxic surgery in this study were aged 8–12 weeks . For all studies , mice of both sexes were used , unless specified . Up to five mice were housed per cage and kept on a 12 hr light; 12 hr dark cycle and were given water and standard rodent chow ad libitum . All procedures carried out in mice were approved by the Institutional Animal Care and Use Committee for Baylor College of Medicine and Affiliates . SH-SY5Y cell lines stably expressing α-Syn or tau were generated using the pINDUCER system ( Meerbrey et al . , 2011 ) . Briefly , myc-tagged α-Syn or tau was inserted into the pINDUCER20 ( ORF-UN ) cassette . Lentivirus was generated as above and naïve SH-SY5Y cells were infected and selected for more than a week in geneticin-containing medium ( G418 , 150 µg/mL ) . Cells were then split into a 24 well plate and treated with doxycycline ( DOX , 100 ng/mL ) for 48 hr . DOX-containing media was replaced with regular media ( containing no tetracycline ) at indicated time points . All cells were harvested at the same time as described above for downstream western-blot testing . Half-lives were calculated from normalized densitometric values obtained by Image J and solving x when y = 50% for the following equation: y = f ( e−kx ) as previously described ( Li , 2004 ) . Alternatively , RNA was isolated from the indicated time points qPCR was performed using the primers presented in Supplementary file 2 . HEK293T cells were infected with lentiviral constructs harboring shRNAs to TRIM28 or scrambled controls . Eight days following infection and puromycin selection , cells were spun down and RNA was extracted using the miRNeasy kit ( Qiagen ) according to the manufacturer’s instructions . For the in vivo quantification of genes expression RNA was extracted from whole mouse brain tissue . RNA was quantified using the NanoDrop 1000 ( Thermo Fisher ) and quality assessed by gel electrophoresis . cDNA was synthesized using a Quantitect Reverse Transcription kit ( Qiagen ) starting from 1 μg of RNA . Quantitative RT-polymerase chain reaction ( qRT-PCR ) experiments were performed using the CFX96 Touch Real-Time PCR Detection System ( Bio-Rad Laboratories ) with PerfeCTa SYBR Green FastMix , ROX ( Quanta Biosciences ) . Real-time PCR results were analyzed using the comparative Ct method and normalized against the housekeeping gene Hs-GAPDH or mm-Hprt . The range of expression levels was determined by calculating the standard deviation of the ΔCt ( Pfaffl , 2001 ) . Primers used to amplify specific exons of the target genes have been designed across introns to distinguish spliced cDNA from genomic contamination . Primers sequences are presented in Supplementary file 2 . Drosophila lines were obtained from the Bloomington Stock Center ( Indiana ) and VDRC ( Vienna ) . Human tau transgenic flies were generated by cloning human wild-type 4 repeat tau under the control of a UAS promoter , and injected in w- embryos following standard transgenesis procedures . For eye assays , expression of wild-type 4 repeat human tau was driven to the eye using GMR-Gal4 and fruit flies , cultured at 28°C , were processed for scanning electron microscopy analysis as previously described ( Park et al . , 2013 ) . tau levels were analyzed in Drosophila expressing tau in the adult eye under the control of Rh1-Gal4 at 28 . 5°C . Animals were aged for ten days , heads were collected , homogenized in LDS Buffer ( Invitrogen ) , 10% β-mercaptoethanol and loaded in 4–12% Bis-Tris gel , transferred to a nitrocellulose membrane , blocked for one hour with 5% non-fat milk and incubated overnight with primary antibody . For motor performance analysis , expression of tau was driven specifically in neurons with elav-Gal4 , and crosses were performed at 26°C . Motor performance was assessed in a climbing assay as previously described ( Park et al . , 2013 ) . Briefly , % motor performance was calculated as the percentage of flies that managed to climb up 9 cm in 15 s ( averaged over 10 trials per time point ) . For a list of all Drosophila strains used in this study , please see Supplementary file 2 , tab 'Fly lines' . Trim28LOF-1 corresponds to the allele P{EPg}bonHP32434 , which is a mobile activating P element inserted in the 5’UTR of bonus ( the Drosophila homologue of Trim28 ) . These P elements can cause loss of function when inserted in the orientation of the negative strand of the inserted gene . Trim28LOF-2 carries a loss of function allele [21B] , previously characterized as missing all but 12 bp of exon 1 and the first 324 bp of intron 1 . We confirmed that Trim28LOF-1 is a loss of function allele of bonus because it failed to complement Trim28LOF-2 , which has already been described as a bonus LOF allele ( Beckstead et al . , 2001 ) . Eight weeks after virus injection , mice were deeply anesthetized and transcardially perfused with 0 . 9% saline followed by 4% PFA . Brains were removed , placed in 4% PFA overnight for post-fixation and dehydrated using 10% Sucrose in PBS ( twice a day for 3 days ) . Brains were frozen in O . C . T . compound ( Tissue-Tek ) and 30 µm free-floating sections were cut from the striatum and Substantia Nigra pars compacta ( SNc ) . Tyrosine hydroxylase and Nissl staining in both the SNc and Striatum was performed as previously described ( Rousseaux et al . , 2012 ) . Unbiased stereology using the optical fractionation method was used to estimate total dopaminergic cell counts in the SNc of each treatment group ( StereoInvestigator 11 , MBF Bioscience ) . Samples missing multiple sections or with staining artifacts were excluded from stereological analysis prior to unblinding . For striatal TH optical density quantification , images were captured using a 10x objective ( both Ipsi- and Contralateral sides ) . Pictures were processes using an in house automated Adobe Photoshop CS5 workflow ( Automator , Mac OSX ) to convert the image to gray scale , invert ( so that the intensity of staining is a positive value ) and saved as a new file . Five regions were sampled per picture in the TH-positive striatum and these were normalized to negative stained background ( corpus callosum ) . Normalized values across three independent sections were used to compare the relative TH optic density between groups . An experimenter blind to the treatment or mouse genotype injected 6–8 week old α-Syn and tau transgenic mice ( mThy1-α-Syn and P301S tau ) and their respective wild-type littermate controls bilaterally with virus harboring TRIM28-mCherry or mCherry as a control into the CA1 region of the hippocampus ( Dhungel et al . , 2015 ) . One month following injection , mice were euthanized by isofluorane inhalation followed by cervical dislocation and decapitation . Brains were dissected , split down the midline and either post-fixed for 48 hr in 4% PFA or collected for downstream biochemical analysis . Post-fixed brains were dehydrated in 10% sucrose ( as above ) and sectioned directly on slides at a thickness of 20 µm . Alternatively , sections were paraffin embedded and sectioned at 5 µm for IHC analysis . Slides were then stained for pS129 α-Syn , pS396 tau and GFAP and visualization was performed using the ABC detection method ( Vectastain , Vector Labs ) as previously described ( Lasagna-Reeves et al . , 2015b ) . Nissl staining was performed as described previously ( Rousseaux et al . , 2012 ) . Quantification of each pathological parameter was performed at an area surrounding the injection site ( though not directly beside the needle tract to avoid staining artifacts ) on three independent sections per animal with at least three animals for each tested condition . Counts were performed in a defined area ( i . e . 1000 × 1000 pixels ) using ImageJ 1 . 47v . Immunoprecipitation of protein complexes was performed as previously described ( Burré et al . , 2010 ) . Briefly , cell lysis was performed on ice for 20 min with brief vortexing using lysis buffer ( 1% Triton X-100 , 150 mM NaCl , 10 mM Tris pH 8 . 0 , 10% glycerol , 20 mM N-ethyl maleimide and protease inhibitors [Roche] ) . Cell debris were removed by centrifugation ( 20 min at 15 , 000 r . p . m , 4°C ) and pre-cleared with un-conjugated beads . In parallel , 2 µg of antibody was conjugated to Dynabeads for one hour at 4°C with rocking . Lysate was then added to the conjugated beads for 2 hr . Beads were then washed 5 × 500 µL of lysis buffer before being eluted in Laemmli buffer at 85°C for ten minutes . Bimolecular fluorescence complementation ( Split-YFP assay ) was performed as previously described with some modifications ( Lee et al . , 2011 ) . N-terminal or C-terminally tagged α-Syn and tau pBABE retroviral constructs were generated containing either the N-terminal or C-terminal portion of YFP . Viruses were generated from these constructs and stable cell lines ( using HeLa cells , infected with constructs and selected with G418 for over a week ) were created for both α-Syn and tau using the N-terminal portion of YFP , termed ‘Bait’ . Once generated , these cell lines ( 5 × 104 cells ) were transfected with 250 ng of ‘Prey’ plasmid ( tau , MSK1 and TRIM28 containing the C-terminal portion of YFP ) . Cells were monitored for fluorescence for 48–72 hr ( fluorescence begins around 24–30 hr ) and cells were either collected for flow cytometry analysis ( at 48 hr , using a BD LSR Fortessa with HTS [high throughput sampler] module ) or fixed for fluorescence microscopy . Cells visualized for fluorescence microscopy were permeabilized and stained for DAPI to visualize nuclei and epifluorescence from the complementation was monitored . The relative amount of nuclear versus cytoplasmic stained protein ( α-Syn or tau ) was performed in primary granule neurons previously infected ( for 7 days ) with lentiviral constructs overexpressing mCherry , TRIM28:mCherry or TRIM28-Mut:mCherry on coverslips . The nucleus was outlined using DAPI and the cytoplasmic outline was determined by DIC ( not shown ) . For each biological replicate , at least 15 cells were sampled and the nuclear fraction of each cell was calculated as the intensity of the nuclear localized fluorophore divided by the intensity of the fluorophore spanning the cytoplasm and the nucleus ( total intensity ) and multiplied by 100 . Three independent coverslips per condition were tested and statistics were based on these three biological replicates . Alternatively , postmortem medial frontal gyrus tissue was stained for α-Syn or tau together with TRIM28 and nuclei were visualized using DAPI . Positive nuclear co-localization was set as a threshold of staining co-localizing with DAPI . At least 150 cells over the span of 5 fields ( at 40x magnification ) were counted per human case . Nuclear/Cytoplasmic fractionation was carried out in freshly spun down cells 7 days following infection ( or alternatively in frozen pre-weighed post-mortem samples ) . Briefly , the NE-PER Nuclear and Cytoplasmic Extraction Kit ( Thermo Scientific , Product #78835 ) was used to isolate nuclear and cytoplasmic fractions of primary neurons or tissue . Of note , the protocol of the NE-PER kit was modified to incorporate an additional PBS-wash of the nuclear pellet following the cytosolic extraction as we found it increased purity of the preparation . Probing with antibodies specific to each compartment ( i . e . Lamin A and histone H3 for the nuclear fraction and Vinculin and GAPDH for the cytoplasmic fraction ) was used as a measure of extraction purity and for normalization . Tissue from patients with PD , AD , PSP and control subjects were obtained from the Neuropathology Core at the Johns Hopkins Udall Centre . Tissue was obtained from consenting donors and use conformed to JHMI Institutional Review Board approved protocols . Neuropathological assessment conformed to the National Institute on Aging-Reagan consensus criteria and similar post-mortem intervals were used between samples . For this study , tissues from the frontal cortex ( for PD , AD and their respective controls ) or pons ( for PSP and their respective controls ) was utilized . Please see Supplementary file 3 for full demographics and pathological analysis . Frozen tissue was either fractionated as per the protocol above or was alternatively processed for biochemical fractionation . In the latter case , each sample was homogenized in RIPA buffer containing protease inhibitors using a dilution of brain: RIPA of 1:10 ( w/v ) . Samples were then centrifuged at 10 , 000 rpm for 20 min at 4°C . The supernatants were portioned into aliquots , snap-frozen , and stored at –80°C . The pellets were then resuspended in 88% formic acid to solubilize aggregate/insoluble protein for one hour at room temperature . Samples were then diluted to 22% in PBS and lyophilized overnight using a Savant automatic environmental speedvac system ( Aes1010 ) . Pellets were then solubilized in the same volume as the original RIPA volume and subsequently sonicated at output 2 for 30 cycles on a Branson sonicator . Independent blocks of frozen tissue were cut from PD , AD and control subjects for immunofluorescence analysis ( 10 µm sections ) . Experimental analysis and data collection were performed in a blinded fashion whenever possible . The sample size was chosen based on previous studies using the models described in the study in order to ensure adequate statistical power . Pre-determined exclusion criteria were used in the nuclear/cytoplasmic biochemical fractionation by determining purity of the prep . If the fractionation was not adequate ( as per Lamin A , Histone H3 , GAPDH and Vinculin levels ) , sample was re-fractionated or omitted from the study . Tissue sections with high background staining/staining artifacts were removed and/or re-stained for the analysis . Randomization together with blinding was used for histological quantification as well as human case studies . Randomization was done using Microsoft Excel and blinding was performed by assigning new non-descript codes to cases . Unblinding was only done following analysis . Outliers for histology were removed if there was a lack of detectable viral expression in the animal and if the sample met the Grubb’s outlier test p-value . p values were determined using the appropriate statistical method via GraphPad Prism , as described throughout the manuscript . For simple comparisons , two-tailed Student’s t-test was used whereas for multiple comparisons , ANOVA followed by the appropriate post hoc analysis were utilized . The summary of all statistical analysis is presented in Supplementary file 1 . All data presented are of mean + ( or ± ) s . e . m . * , and * denote p<0 . 05 , p<0 . 01 and p<0 . 001 , respectively . ns denotes p>0 . 05 .	Behind many neurodegenerative diseases are specific proteins that abnormally accumulate inside neurons and damage the cells . In Parkinson’s disease , the protein alpha-synuclein accumulates; in Alzheimer’s disease , the protein tau is one of the toxic culprits; and in other neurodegenerative diseases , alpha-synuclein and tau both accumulate . Genetic studies suggest that accumulation of the two proteins may be linked , but little is known about the factors that regulate the levels of these proteins inside neurons . Rousseaux et al . set out to identify how these proteins are regulated in the hope of finding new ways of targeting them and reducing their toxicity . Screening a subset of human genes led to one that encodes a protein called TRIM28 , which regulates the levels of both alpha-synuclein and tau . When the TRIM28 protein was depleted in human and mouse cells , the levels of alpha-synuclein and tau also went down . This effect was specific because the levels of other proteins with the potential to cause neurodegeneration remained unaffected . Models of neurodegenerative disease in fruit flies and mice were then used to explore how TRIM28 affects the levels of tau and alpha-synuclein in animals . In each case , the proteins’ levels dropped when TRIM28 was suppressed and this in turn protected the neurons from damage . Rousseaux et al . went on to show that TRIM28 affected how alpha-synuclein and tau were cleared in cells . Overexpressing TRIM28 revealed that it could encourage both alpha-synuclein and tau to accumulate in the nucleus of cells over time . Finally , Rousseaux et al . compared post-mortem brain tissue from people who had neurodegenerative conditions that are driven by or associated with tau and alpha-synuclein with tissue from those who did not . The cell nuclei in the diseased tissue had much more TRIM28 associated with alpha-synuclein and tau than those in the healthy tissues . Overall , the findings show that TRIM28 promotes the accumulation and damaging effects of both alpha-synuclein and tau . The next steps will be to understand how TRIM28 does this . It will also be important to determine if this effect can be targeted , whilst leaving others roles of TRIM28 intact , in order to explore it as a potential target to treat or prevent neurodegenerative diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	TRIM28 regulates the nuclear accumulation and toxicity of both alpha-synuclein and tau
During development internal models of the sensory world must be acquired which have to be continuously adapted later . We used event-related potentials ( ERP ) to test the hypothesis that infants extract crossmodal statistics implicitly while adults learn them when task relevant . Participants were passively exposed to frequent standard audio-visual combinations ( A1V1 , A2V2 , p=0 . 35 each ) , rare recombinations of these standard stimuli ( A1V2 , A2V1 , p=0 . 10 each ) , and a rare audio-visual deviant with infrequent auditory and visual elements ( A3V3 , p=0 . 10 ) . While both six-month-old infants and adults differentiated between rare deviants and standards involving early neural processing stages only infants were sensitive to crossmodal statistics as indicated by a late ERP difference between standard and recombined stimuli . A second experiment revealed that adults differentiated recombined and standard combinations when crossmodal combinations were task relevant . These results demonstrate a heightened sensitivity for crossmodal statistics in infants and a change in learning mode from infancy to adulthood . After birth infants are immediately exposed to a sensory world comprising input of multiple sensory modalities . The developing brain must adapt to the statistical properties of the sensory environment ( Fiser et al . , 2010 ) since genetically defined neural circuits are usually crude . Indeed a high sensitivity of infants to statistical regularities within single sensory systems has often been demonstrated ( Fantz , 1964; Saffran et al . , 1996; Fiser and Aslin , 2002; Bulf et al . , 2011 ) . The seminal study of Saffran et al . ( 1996 ) reported that eight-month-old infants quickly learn transitional probabilities between syllables by pure exposure to an artificial language . This ability was interpreted as a basic mechanism allowing infants to segment a language . Similar results were found for non-linguistic auditory sequences and for visual patterns ( Saffran et al . , 1999; Fiser and Aslin , 2002 ) , demonstrating a modality independent sensitivity of infants to statistical patterns in their sensory environment which moreover is not unique to linguistic material . For example , in the visual domain , there is strong evidence that infants are able to implicitly learn subtle statistical relationships among visual objects ( Fiser and Aslin , 2002; Bulf et al . , 2011; Kirkham et al . , 2002 ) . Nine-month-old infants who were exposed to multi element visual scenes , showed greater interest in element pairs which co-occurred more frequently than in pairs which co-occurred less frequently . Moreover , the infants were sensitive to the predictability between elements of the pairs as manifested by the conditional probability relations between these elements ( Fiser and Aslin , 2002 ) . The ability to extract statistical patterns of visual stimuli was found even in younger age groups ( Kirkham et al . , 2002 ) ; two- , five- , and eight-month-old infants were habituated to sequences of discrete visual stimuli whose ordering followed a statistical predictable pattern . Subsequently the infants were shown the previously encountered pattern alternating with a novel pattern of identical stimulus components . Infants of all age groups looked longer at the novel sequences providing evidence for the detection of visual statistical regularities at an early developmental stage . These results suggest that infants own powerful mechanisms for extracting the statistical properties of their sensory input without any instructions , explicit feedback , or intentional awareness ( Lany and Saffran , 2013; Krogh et al . , 2012 ) . The ability of infants to detect crossmodal statistical regularities within their sensory environment is less well understood , but some basic multisensory abilities , such as multisensory temporal synchrony detection seem to exist within the first month of life ( Lewkowicz , 1992 ) . In the next months the capability to perceive higher-level and more complex multisensory relations starts to develop . For example , at the age of six months infants were shown to perceive duration-based ( Lewkowicz , 1992 ) and spatio-temporal based crossmodal relations ( Scheier et al . , 2003 ) . Furthermore , there is evidence that similar to adults , infants take advantage of crossmodal events in terms of a better discrimination and a faster responsiveness to bimodal compared to unimodal information ( Bahrick et al . , 2004; Lewkowicz and Kraebel , 2004 ) . First evidence for multisensory facilitation was found in eight-month-old infants as indicated by faster eye movements to spatially aligned auditory and visual cues compared to eye movements to each of these stimuli alone ( Neil et al . , 2006 ) . Moreover , other studies revealed multisensory benefits for perceptual learning in infants ( Bahrick and Lickliter , 2000; Frank et al . , 2009 ) . Five-month-old infants were habituated to either an audio-visual rhythm or the same rhythm presented unimodally . In the crossmodal condition , infants were able to discriminate between the familiar and a novel rhythm , whereas no discrimination was observed for the unimodal stimuli ( Bahrick and Lickliter , 2000 ) . Corresponding results were found for the learning of an abstract rule in five-month-old infants: they were able to learn the sequence if defined by redundant visual shapes and speech sounds but not if only one sensory modality was involved ( Frank et al . , 2009 ) . These results suggest that infants are able to learn and use associations between auditory and visual stimuli . However , it must be taken into account that the multisensory effects in infants were not tested against statistical facilitation ( probability summation , see Miller , 1982 ) . Several studies on crossmodal association learning have reported that infants at the age of three months , but not younger , were able to learn specific voice-face combinations; infants were habituated to different unfamiliar voice-face pairings . In the post-familiarization test the infants showed higher attention to the learned voice-face pairs as compared to the novel combinations . The latter category comprised a voice and a face they had heard and seen previously , but the combination of the voice and face was new ( Brookes et al . , 2001; Bahrick et al . , 2005 ) . More recently , near-infrared spectroscopy ( NIRS ) and event-related potentials ( ERPs ) were used to test whether infants are able to learn crossmodal associations between arbitrary auditory and visual stimuli . Emberson et al . ( 2015 ) used an audio-visual omission paradigm with six-month-old infants and found similar visual cortex activation as a response to an auditory stimulus alone , which had been previously combined with a visual stimulus , as for the presentation of the same visual stimulus . The authors interpreted their findings as evidence for top-down mechanisms to be in place as early as six month of age . Kouider et al . ( 2015 ) exposed twelve-month-old infants to pictures of faces paired with one sound and pictures of flowers paired with a second sound . During the test phase the sound preceded the visual stimulus and was either congruent or incongruent with the learned combinations ( additionally no sound was used in one third of the trials ) . An enhanced early positive ERP for congruent visual stimuli as well as an enhanced late negative ERP for incongruent visual stimuli were found . Both studies demonstrate that infants are able to learn crossmodal combinations to which they were exposed . However , none of these studies used an adult control group . Thus , it remains an open question of whether developmental and adult crossmodal learning recruit similar mechanisms . In this context it is interesting to notice that Janacsek et al . ( 2012 ) demonstrated superior implicit statistical learning of visual sequences in young children ( <12 years ) compared to older children and adults; a follow-up study indicated that this advantage was lost when they became more reliant on explicit learning ( Nemeth et al . , 2013 ) . Based on non-human animal studies it has been proposed ( Keuroghlian and Knudsen , 2007 ) that developmental and adult plasticity , and thus learning , differ due to different brain states: during the sensitive phase molecular mechanisms dominate that allow for quick and extensive functional and structural synaptic plasticity ( synaptogenesis , synaptic strengthening and elimination ) allowing the emergence of a functional adaptive connectivity . By contrast , in adulthood these functionally tuned and to some degree stabilized neural circuits undergo adaptations when relevant to the system . Such age dependent changes from developmental to adult plasticity are impressively demonstrated by a study on auditory cortex plasticity in rats: while passive exposure to sounds of a specific frequency results in a permanent reorganization of auditory cortex during the sensitive phase , adult rats reorganized only those aspects of the auditory cortex which were task relevant: for example , rats were exposed to sounds which varied both in sound frequency and level . When they had to discriminate the sounds with respect to sound frequency the frequency representation of auditory cortex changed while the level representation changed when level rather than sound frequency was task relevant ( de Villers-Sidani et al . , 2007 ) . These findings suggest that adult learning depend to a larger degree on attention and context such as task relevance and reward expectations ( Keuroghlian and Knudsen , 2007; Bavelier et al . , 2010 ) . This hypothesis was supported by Riedel and Burton ( 2006 ) who investigated whether learning of auditory sequences is influenced by task demands; when using a serial reaction time task related to a feature of the auditory stimulus they found learning effects in adult participants while a passive exposure did not result in learning . Similarly , the statistical relations of concurrently presented visual streams were only learned by adults for the attended but not for the unattended streams ( Turk-Browne et al . , 2005 ) . Emberson et al . ( 2011 ) extended these findings by providing evidence in adults that attention was necessary for implicit statistical learning in both the visual and auditory modality . In the present study we investigated multisensory associative learning in infants and adults to test the hypothesis that infants as young as six months are not only able to learn arbitrary auditory-visual associations but that their sensitivity to crossmodal statistics is even higher compared to adults when crossmodal associations are passively encountered . Thus , in the first experiment we included a group of six-month-old infants ( Experiment 1a ) and a group of young adults ( Experiment 1b ) . While recording the electroencephalogram ( EEG ) , we presented two frequently occurring audio-visual standard combinations ( A1V1 , A2V2 , p=0 . 35 each , ‘Frequent standard stimuli’ ) , two rare recombinations of the ‘Frequent standard stimuli’ ( A1V2 , A2V1 , p=0 . 10 each , ‘Rare recombined stimuli’ ) and one rare audio-visual combination of an infrequent auditory and an infrequent visual stimulus ( A3V3 , p=0 . 10 , ‘Rare deviant stimuli’ ) . Recombining the auditory and visual elements of the ‘Frequent standard stimuli’ controls for the likelihood of the auditory and visual elements of the employed crossmodal stimuli . Thus , in order to detect ‘Rare recombined stimuli’ it is necessary to have learned the precise crossmodal combination . By contrast , the likelihood of both the visual and the auditory elements of ‘Rare deviant stimuli’ were lower than for all other auditory and visual elements . Therefore , the present experimental design allowed us to differentiate between the processing of the likelihood of sensory elements ( ‘Frequent standard stimuli’ vs . ‘Rare deviant stimuli’ ) and the processing of the conditional probabilities of crossmodal combinations ( ‘Frequent standard stimuli’ vs . ‘Rare recombined stimuli’ ) . We predicted ERP differences between the ‘Frequent standard stimuli’ and ‘Rare deviant stimuli’ in both infants ( Cheour et al . , 2000 ) und adults ( Schröger and Wolff , 1996; Näätänen and Alho , 1995 ) . In contrast , we hypothesized that only infants display an ERP difference for ‘Frequent standard stimuli’ vs . ‘Rare recombined stimuli’ due to a higher sensitivity to crossmodal statistics during infancy . In Experiment 1 we investigated a group of infants ( Experiment 1a ) and a group of young adults ( Experiment 1b ) with the same experimental design . Due to the age difference between the groups , a few adjustments in the procedure , data recording , and data analyses were necessary . ERP differences were found between ‘Rare deviant stimuli’ and ‘Frequent standard stimuli’ as well as ‘Rare recombined stimuli’ and ‘Frequent standard stimuli’: ‘Rare deviant stimuli’ ( A3V3 ) elicited a more negative going ERP than ‘Frequent standard stimuli’ ( A1V1 , A2V2 ) ( see Figure 1 ) . This effect ( 200–420 ms , 420–1000 ms ) was predominantly observed over the right hemisphere . Crucially , ‘Rare recombined stimuli’ ( A1V2 , A2V1 ) elicited a more negative going ERP compared to ‘Frequent standard stimuli’ ( see Figure 1 ) , predominantly over the left hemisphere ( 420–1000 ms ) . ERP differences were found only between ‘Rare deviant stimuli’ and ‘Frequent standard stimuli’ . ERPs to ‘Rare deviant stimuli’ were more negative going than ERPs to ‘Frequent standard stimuli’ during both time windows ( 180–220 ms , 250–1000 ms; see Figure 2 ) . As predicted , infants were more sensitive to crossmodal statistics than adults . Only infants displayed a significant ERP deviant effect for ‘Rare recombined stimuli’ . By contrast , both groups showed at a relatively earlier time epoch a ‘Rare deviant stimuli’ effect , suggesting that the overall experimental power had been sufficient to detect ERP deviant effects . In fact , the effect size for the ‘Rare deviant stimuli’ effects was smaller , both in infants ( d = 0 . 65 ) and adults ( d = 0 . 63 ) , than the effect size for the ERP effects comparing ‘Frequent standard stimuli’ and ‘Rare recombined stimuli’ in infants ( d = 0 . 73 ) . Thus , since smaller effects ( ‘Frequent standard stimuli’ vs . ‘Rare deviant stimuli’ ) than the missing effect ( ‘Frequent standard stimuli’ vs . ‘Rare recombined stimuli’ ) were detected in adults , it seems justified to conclude that the null effect in adults was not caused by a lack of power . Nevertheless , we ran a second Experiment ( Experiment 2a ) to replicate with a more adequate design for adults the lack of learning arbitrary crossmodal conditional probabilities when they were not related to a task . Moreover , in an additional experiment ( Experiment 2b ) we tested the requirements for adult learning of crossmodal statistics . We will discuss the results of Experiment 1a and 1b in light of the results of these additional experiments in the general Discussion . As we did not find any ERP difference between ‘Frequent standard stimuli’ and ‘Rare recombined stimuli’ in the adult group in Experiment 1b , we ran a second study in adults comprising two experiments , in which we systematically manipulated the task relevance of crossmodal combinations . Both experiments were very similar to Experiment 1 but comprised essential adaptations: ( a ) to enhance the power of the experiment , we increased the number of trials; ( b ) in Experiment 2a we included a fourth visual stimulus ( V4 ) , which had to be detected by participants ( target ) while all other stimuli remained task irrelevant: This manipulation guaranteed that participants attended the stimuli; ( c ) in Experiment 2b one of the ‘Rare recombined stimuli’ ( either A1V2 or A2V1 ) served as the target: this manipulation rendered crossmodal combinations task relevant to the participants . At the same time this design allowed us to analyze ERPs to crossmodal stimuli , including to the non-target ‘Rare recombined stimuli’ , which were , as in Experiment 2a , not followed by a manual response . We hypothesized that adults are not sensitive to crossmodal statistics ( no ERP difference between ‘Frequent standard stimuli’ and ‘Rare recombined stimuli’ ) when crossmodal combinations are task irrelevant ( Experiment 2a in replication of the findings from Experiment 1b ) but that such ERP differences would emerge in Experiment 2b , indicating learning of crossmodal statistics when they are task relevant . As seen in Table 1 , participants identified target stimuli with a high accuracy in both experiments . ERP differences were found only between ‘Rare deviant stimuli’ and ‘Frequent standard stimuli’: Compared to ‘Frequent standard stimuli’ ‘Rare deviant stimuli ‘elicited a more negative early ERP ( 80–160 ms , see Figure 3 ) . During the late time window ( 250–850 ms ) ERPs to ‘Rare deviant stimuli ‘were more negative over the anterior scalp and more positive over the posterior scalp compared to ERPs to ‘Frequent standard stimuli’ . ERPs to ‘Frequent standard stimuli’ and ‘Rare recombined stimuli’ did not significantly differ ( see Figure 3 ) . ERP differences were found between both , ‘Rare deviant stimuli’ and ‘Frequent standard stimuli’ and between ‘Rare recombined stimuli’ and ‘Frequent standard stimuli’ . Compared to ‘Frequent standard stimuli’ ‘Rare deviant stimuli ‘elicited a more negative early ERP ( 80–160 ms , see Figure 4 ) over the anterior scalp and a more positive ERP over the posterior scalp . During the late time window ( 250–850 ms , see Figure 4 ) ERPs to ‘Rare deviant stimuli ‘were more positive over the anterior scalp and more negative over the posterior scalp compared to the ‘Frequent standard stimuli’ . ERPs to ‘Rare recombined stimuli’ compared to ERPs to ‘Frequent standard stimuli’ were more positive going over the anterior scalp and more negative going over the posterior scalp ( 250–850 ms , see Figure 4 ) . Differences in ERPs between ‘Rare deviant stimuli’ and ‘Frequent standard stimuli’ were found in both experiments at early processing stages . Crucially , ERP differences between ‘Rare recombined stimuli’ and ‘Frequent standard stimuli’ were only found in Experiment 2b , indicating that the adults’ brains were able to differentiate ‘Rare recombined stimuli’ from ‘Frequent standard stimuli’ when crossmodal combinations were task relevant . The goal of the present study was to test for a higher sensitivity of infants as compared to adults to crossmodal statistics and to compare the mechanisms of crossmodal association learning in infants and adults . We conducted four ERP experiments in which infants and adults were exposed to audio-visual stimulus combinations with different probabilities . We presented ‘Frequent standard stimuli’ ( A1V1 , A2V2 , p=0 . 35 each ) , rare recombinations of the ‘Frequent standard stimuli’ ( A1V2 , A2V1 , p=0 . 10 each , ‘Rare recombined stimuli’ ) , and a rare deviant audio-visual combination with an infrequent auditory and visual element ( A3V3 , p=0 . 10 , ‘Rare deviant stimuli’ ) . While infants passively learned the crossmodal combinations , adults did not . Adults’ ERPs to ‘Rare recombined stimuli’ and to ‘Frequent recombined stimuli’ differed only when crossmodal combinations were task relevant . In contrast , all groups , irrespectively of learning context , showed a sensitivity to the probability of sensory elements , that is , for ‘Rare deviant stimuli’ . Table 2 graphically summarizes the main results of all four experiments . Studies using artificial languages or visual artificial scenes have repeatedly demonstrated that infants develop a sensitivity to the likelihood of events as well as to conditional probabilities ( Krogh et al . , 2012; Aslin , 2014 ) , partially as early as at the age of two months ( Kirkham et al . , 2002 ) . Two recent studies addressing crossmodal statistical learning found that six-month and twelve-month-old infants learned to predict a visual stimulus based on a preceding auditory stimulus ( Emberson et al . , 2015; Kouider et al . , 2015 ) . While Kouider et al . ( 2015 ) demonstrated that infants at the age of twelve months were able to learn an association between an arbitrary sound and a visual object category ( faces vs . flowers ) , they did not include an adult control group and were thus not able to demonstrate differences in learning between adults and infants , nor were they able to distinguish processes related to the detection of crossmodal combinations and processes related to the familiarity with certain sensory elements . Thus , the present study extends previous research by showing that the conditional probabilities of crossmodal combinations were extracted by infants as young as six months after a short exposure period while adults failed to learn crossmodal statistics under this condition . It is important to notice that we controlled for the likelihood of the auditory and visual elements of the employed crossmodal stimuli by infrequently recombining the auditory and visual elements of the ‘Frequent standard stimuli’ . We provide ERP evidence demonstrating that the processing of the conditional probabilities of crossmodal combinations and the processing of the likelihood of sensory elements can be dissociated: in infants , ‘Rare recombined stimuli’ elicited a left negative potential starting at about 420 ms post-stimulus while ‘Rare deviant stimuli’ elicited a right lateralized positivity starting at 200 ms post-stimulus ( Experiment 1a ) . Adults tested under identical conditions were only sensitive to ‘Rare deviant stimuli’ , which differed from ‘Frequent standard stimuli’ in the frequency of their auditory and visual elements ( Experiment 1b , ERP effect starting 180 ms post-stimulus ) but not for rare crossmodal stimuli which only differed from the ‘Frequent standard stimuli’ in the way the auditory and visual elements were combined . These results demonstrate that infants were able to learn arbitrary crossmodal associations as early as six months of age and thus much earlier than suggested by the study of Kouider et al . ( 2015 ) ( see Emberson et al . , 2015 ) . This finding is in line with behavioral studies employing natural stimuli , which showed that infants from three months onwards were able to learn specific face-voice-parings ( Brookes et al . , 2001; Bahrick et al . , 2005 ) . Our results extended these behavioral findings by providing first evidence that the learning of crossmodal statistics in infancy is particularly sensitive and superior to adults when crossmodal stimuli are not task relevant . It could be argued that the signal to noise ratio of the ERPs in adults was not sufficient in Experiment 1b to demonstrate crossmodal learning in the adult group . However , such an account is highly unlikely given that an effect of smaller size were detected in Experiment 1b and the fact , that in Experiment 2a an ERP difference between ‘Frequent standard stimuli’ and ‘Rare recombined stimuli’ was not significant either despite a much higher signal to noise ratio in comparison to Experiment 1b . Thus , our results provide evidence that crossmodal statistics are better implicitly learned in the developing than in the adult system . An enhanced sensitivity for low-level statistical patterns during development compared to adulthood has been reported by other studies as well . For example , Janacsek et al . ( 2012 ) and Nemeth et al . ( 2013 ) demonstrated that children are superior in implicit statistical learning of sequences compared to adults but later lose this advantage and become more reliant on explicit learning . A similar developmental time course was found in a study of Jost et al . , 2011 , who had investigated the neurophysiological correlates of visual statistical learning in children and adults: children showed learning related ERP effects earlier in the acquisition phase than the adult group indicating that they had quicker acquired the statistical structurer . It is , however , important to notice that not all studies investigating statistical learning during development have found enhanced learning performance in infants or children as compared to adults . For example , Saffran et al . ( 1996 ) , 1999 ) reported similar abilities in eight-month-old infants and adults in the extraction of the underlying statistical structure of auditory sequences . Other studies observed better learning for older children and young adults than in younger age groups ( Maybery et al . , 1995; Fletcher et al . , 2000; Kirkham et al . , 2007 ) . Arciuli and Simpson ( 2011 ) tested children between the age of 5 and 12 years in a visual triplet learning task and reported better learning with increasing age . At first glance , these findings seem to be at odds with the present results . However , a closer look at the employed paradigms suggests that these different outcomes might be related to the complexity of the implemented statistical patterns . For example , Arciuli and Simpson ( 2011 ) tested the incidental learning of four visual base triples which draws to a much larger extend from working memory than learning the conditional probability of two audio-visual combinations as used in the present and previous crossmodal learning studies in infants ( Emberson et al . , 2015; Kouider et al . , 2015 ) . Indeed , it is well known that working memory improves during childhood ( Zelazo et al . , 2008 ) and thus working memory demands might have been the limiting factor in some studies ( e . g . Arciuli and Simpson , 2011 ) . In addition , triples are usually embedded in a continuous stream while the crossmodal stimuli of the present study were individually presented with relatively long interstimulus intervals , thereby clearly demarking the individual events . Furthermore , studies have revealed that the ability to extract statistical patterns from sensory input during infancy improves from the simple tracking of event probabilities early in the development ( from three months onwards , see Fantz , 1964 ) to the learning of more complex and higher-level statistical patterns at a later developmental stage ( from twelve months onwards , see Gómez and Maye , 2005 ) . Thus , what most likely declines during development seems to be the sensitivity to simple conditional probabilities ( Janacsek et al . , 2012 ) . Janacsek et al . ( 2012 ) speculated that a decline in the sensitivity to ‘base probabilities’ is necessary for the acquisition of higher order representations and a switch to model-based ( in contrast to model-free ) learning . In line with this suggestion , adults did not learn crossmodal statistics when they were irrelevant for the task but became sensitive to them when a specific crossmodal combination was of behavioral relevance . Studies in non-human animals have suggested that during the sensitive phase , neural networks are elaborated in response to a pure exposure to the environment while during later development and in adulthood learning is context-specific and depends on task relevance ( e . g . reward ) and instructions ( Keuroghlian and Knudsen , 2007 ) . Currently , we can only speculate about the neural underpinnings of the age-dependent neuroplasticity as observed in the present study . As noted by Dehaene-Lambertz and Spelke ( 2015 ) feedforward connectivity seems to be to a larger degree genetically determined than feedback connectivity and the latter seems to be mostly experience dependent . The detection of ‘Rare recombined stimuli’ was associated with a relatively late ERP effect in both infants and adults . Indeed , multisensory binding has been found to rely on later processing stages in adults and the involvement of feedback connections ( Bruns and Röder , 2010; Bonath et al . , 2007 ) . Emberson et al . ( 2015 ) provided evidence that the crossmodal connectivity is at least partially in place at the age of six months . Here we speculate that this initial crossmodal connectivity might even be more extensive in the developing brain ( see Johannsen and Röder , 2014 ) and thus might be the neural underpinning of the enhanced sensitivity to simple crossmodal statistics in development which allows for quicker and a passive learning during infancy . We further assume in line with the ‘multisensory perceptual narrowing’ idea ( Lewkowicz and Ghazanfar , 2006 ) that experience narrows down the initial crossmodal connectivity by eliminating non-confirmed connections while elaborating connections which are useful for an individual ( Johannsen and Röder , 2014; Lewkowicz , 2014 ) . These functionally tuned networks ( including the experience dependent feedback connectivity ) constitute models of the sensory world ( Fiser et al . , 2010 ) . Their elaboration might go together with a switch towards model-based learning which is characterized by a larger context dependency . As some parts of the neural networks stabilize , learning must partially involve additional neural systems to guarantee that the adaptations necessary throughout life are realized without risking the loss of essential crossmodal knowledge . For example , prism wearing during the sensitive phase has been reported to change the connectivity between the central ( ICC ) and external ( ICX ) inferior colliculus of the auditory midbrain of barn owls . By contrast , crossmodal adaptation to prisms later in the critical period seems to be mediated by a reorganization of the optical tectum to which the ICX projects ( Knudsen , 2002 ) . Moreover , Bergan et al . ( 2005 ) reported that crossmodal adaptions to prims were enhanced in adult owls when they were allowed to hunt , that is , when such adaptations were particular needed . In accord with these findings in owls , we demonstrated that adult learning of crossmodal combinations depended on task relevance ( Experiment 2b ) . Thus , as suggested by Keuroghlian and Knudsen ( 2007 ) and Bavelier et al . ( 2010 ) , neuroplasticity in adults seem to require to a larger extend attention and behavioral relevance and thus the involvement of additional higher order neural systems . Task relevance or attention constitute specific top-down influences on sensory representations and are thus mediated via the feedback connections which become progressively tuned and elaborated during development ( Dehaene-Lambertz and Spelke , 2015 ) . The present study was able to dissociate the processes for the learning of probabilities of sensory elements and for the learning of conditional probabilities of the sensory elements comprising crossmodal stimuli . All groups were sensitive to ‘Rare deviant stimuli’ . To detect ‘Rare deviant stimuli’ the frequency of sensory elements rather than conditional probabilities had to be traced . Indeed , it was possible to detect ‘Rare deviant stimuli’ only based on one of the two sensory elements . The ERP effect to ‘Rare deviant stimuli’ started earlier than the ERP effects to ‘Rare recombined stimuli’ . Such early deviant effects are typical for an auditory mismatch negativity ( MMN , see Schröger and Wolff , 1996; Cheour et al . , 2000 ) . Therefore , we suggest that the observed ‘Rare deviant stimuli’ effect , similarly as has been proposed for the MMN reflects , indicates a sensory memory trace , which represents the frequency of sensory elements ( Näätänen and Alho , 1995 ) . By contrast , the detection of conditional probabilities of crossmodal stimuli cannot be based on such ( unisensory ) sensory memory traces . Thus , we speculate that the detection of ‘Rare deviant stimuli’ , is based on modality specific systems ( Frost et al . , 2015 ) . Although it has been reported that auditory mismatch responses are enhanced by redundant crossmodal ( somatosensory ) information in adults such a multisensory enhancement was only observed for later time epochs ( >200 ms; Butler et al . , 2012 ) than the first ‘Rare deviant stimulus’ effect of the present study . Since we argue that the change in learning mode during development is related to functional specialization , the strong lateralization of both ERP effects in infants seems rather surprising . The differentiation of ‘Frequent standard stimuli’ and ‘Rare recombined stimuli’ requires the detection of conditional probabilities . This ability has been postulated as a precursor of language learning ( Saffran et al . , 1996 ) . Indeed , it has been shown with structural imaging techniques that many hemispheric asymmetries , in particular those related to the language system ( Friederici , 2009 ) , exist at birth or shortly thereafter ( see Dehaene-Lambertz and Spelke , 2015 ) . Thus , we speculate that the strong left lateralized ERP difference between ‘Frequent standard stimuli’ and ‘Rare recombined stimuli’ might reflects a recruitment of similar neural circuits that have been proposed to enable the detection of word boundaries ( Saffran et al . , 1996 ) , non-adjacent transitional probabilities and possibly syntactical rules ( Friederici , 2002; Friederici et al . , 2006 ) . Thus , this neural system might , partially independently of sensory modality and domain , allow for detecting any type of statistical relations ( Kuhl , 2010; Aslin and Newport , 2014 ) . In fact , a correlation of syntactic competence and statistical learning skills in children has been reported ( Kidd and Arciuli , 2016 ) . The right lateralized ERP effect to ‘Rare deviant stimuli’ was not unique to the infant group , but was as well observed in the adults tested with the same passive design ( Experiment 1b ) . Interestingly such a lateralization was neither found for Experiment 2a nor for Experiment 2b , in which the adult participants were actively engaged in a task . We speculate that ‘Rare deviant stimuli’ elicited a reflexive and exogenous attention shift to the rare sensory features . Such reflexive spatial attention orienting has often been associated with right parietal brain regions ( Okada et al . , 2008; Mort et al . , 2003; Chica et al . , 2011 ) . In contrast , in Experiment 2a and 2b , participants had to allocate attention to a specific stimulus or stimulus combination and it was adaptive to avoid exogenous attention shifts . In the present study ERP effects in adults were of different polarity and had a shorter latency compared to the infant group . We linked ERP effects in infants and adults based on the experimental manipulations and their relative timing . Due to the immature brain ( e . g . incomplete myelination ) of infants and children it is a common finding that absolute latencies of ERP effects are longer in the developing brain . Moreover , it has repeatedly been reported that polarities of effects differ in infancy or children and adulthood ( Kouider et al . , 2015; Neville et al . , 2013; Nelson , 1997; de Haan and Halit , 2001 ) . In conclusion , our study demonstrates that six-month old infants were able to quickly learn crossmodal statistics through a mere passive exposure , whereas adults learned the same crossmodal combinations only when they were task relevant . Thus , we provide first evidence for a higher sensitivity for crossmodal statistics in infants compared to adults , indicating age-dependent mechanisms for the learning of arbitrary crossmodal combinations . We speculate that initial passive association learning allows infants to quickly form first internal models of their sensory environment . In adulthood these internal models are adjusted if this is behavioral adaptive .	On a crowded city street , we automatically attribute the sounds of cars to the cars we see driving past , and not to the motorcycles or trucks on the same road . Similarly , we assign the voices we hear to the pedestrians around us , and not to the dogs those pedestrians are walking . As adults , we cope with these everyday challenges effortlessly , but how do infants first learn to match what they see with what they hear ? When young animals are exposed to new stimuli , their brains undergo changes . Similar changes only occur in adult animals if they deliberately pay attention to the stimuli and if they are associated with rewards . Rohlf , Habets et al . therefore predicted that human infants would automatically learn to associate sights and sounds upon being passively exposed to them . Adults , on the other hand , would learn these associations only if explicitly asked to do so . To test this prediction , Rohlf , Habets et al . presented tones and colored shapes to 6-month-old infants and healthy adult volunteers while using scalp electrodes to monitor the electrical activity in their brains . Certain shapes and tones occurred frequently together , whereas other combinations of the same stimuli were rare . The 6-month-olds consistently outperformed the adults in associating the tones and shapes: the electrical activity in the infant brains reliably distinguished between common versus rare combinations . Adult brains made this distinction only when the adults were asked to pay attention to the tone-shape combinations as part of a task . This high sensitivity to combinations of sights and sounds that regularly occur together enables infants to quickly learn about the world around them . As adults will have done this previously , the most effective strategy for adults is to update their existing knowledge only when such learning enables them to achieve a goal . Further research is needed to find out what happens in the brain to cause this change in learning strategy . Understanding how learning differs in infants and adults will help identify stages of development in which the brain learns particularly easily . This may ultimately help us optimize learning strategies for individuals of different ages .	[ "Abstract", "Introduction", "Results", "Discussion" ]	[ "neuroscience" ]	2017	Infants are superior in implicit crossmodal learning and use other learning mechanisms than adults
When relatively sated , people ( and rodents ) are still easily tempted to consume calorie-dense foods , particularly those containing fat and sugar . Consumption of such foods while calorically replete likely contributes to obesity . The nucleus accumbens ( NAc ) opioid system has long been viewed as a critical substrate for this behavior , mainly via contributions to the neural control of consumption and palatability . Here , we test the hypothesis that endogenous NAc opioids also promote appetitive approach to calorie-dense food in states of relatively high satiety . We simultaneously recorded NAc neuronal firing and infused a µ-opioid receptor antagonist into the NAc while rats performed a cued approach task in which appetitive and consummatory phases were well separated . The results reveal elements of a neural mechanism by which NAc opioids promote approach to high-fat food despite the lack of caloric need , demonstrating a potential means by which the brain is biased towards overconsumption of palatable food . People often seek and consume calorie-dense food in the absence of hunger , and this behavior has profound implications for human health . Although preference for sweet and fatty foods may once have been adaptive , it now very likely contributes to epidemic rates of obesity and diabetes . Thus , understanding the neural mechanisms that guide seeking of highly palatable foods is an important step in the search for novel therapies that could combat these diseases by reducing caloric intake . One candidate neural substrate is the brain’s opioid system , particularly in the ventral striatum . A role for this circuitry is supported by observations that the ventral striatum , and in particular the nucleus accumbens ( NAc ) , is richly endowed with both opioid peptides and their respective receptors ( Mansour et al . , 1988 ) , and that activation of NAc µ-opioid receptors ( MORs ) selectively augments consumption of palatable food ( Bakshi and Kelley , 1993; Mucha and Iversen , 1986; Zhang et al . , 1998; Zhang and Kelley , 1997 ) and of preferred flavors ( Woolley et al . , 2006 ) . Moreover , activation of NAc MORs increases hedonic taste reactions to palatable food ( Peciña and Berridge , 2000 ) . Thus , opioids are thought to contribute primarily to the encoding of hedonic responses to food , which in turn reinforces the assignment of incentive salience to cues associated with palatable reward ( Berridge , 2009; Castro and Berridge , 2014 ) . However , several observations indicate that this view is incomplete . First , blockade of NAc MORs does not consistently reduce calorie-dense food consumption ( Bodnar et al . , 1995; Kelley et al . , 1996; Lardeux et al . , 2015; MacDonald et al . , 2003; Ward et al . , 2006 ) . In addition , activation of NAc MORs increases certain measures of reward-seeking behavior , including breaking point on a progressive-ratio task ( Zhang et al . , 2003 ) and lever pressing in the presence of food-predictive cues in a Pavlovian-to-instrumental transfer task ( Peciña and Berridge , 2013 ) . These studies suggest that NAc MOR activation could promote food-seeking behavior directly , instead of ( or in addition to ) doing so by enhancing the hedonic or reinforcing effects of the food . Consequently , we hypothesize that when people or animals are sated , their preferences shift toward palatable food because endogenous ligands of NAc MORs selectively promote seeking of calorie-dense foods . This idea has not yet been tested because few studies have examined the contribution of NAc MORs activated by endogenous ligands to appetitive ( food-seeking ) as opposed to consummatory behaviors . Here , we address this gap in our knowledge by using a conditioned-stimulus ( CS ) task that disambiguates appetitive from consummatory behavior . In this task , rats perform an approach response to a reward-predictive cue to obtain a highly palatable , calorie-dense liquid food ( cream ) . NAc neurons encode both cued approach and reward consumption phases of such behaviors ( Ambroggi et al . , 2011; du Hoffmann and Nicola , 2014; McGinty et al . , 2013; Morrison et al . , 2017; Nicola , 2010; Nicola et al . , 2004a; 2004b; Taha and Fields , 2005 ) , and cue-evoked excitations are necessary for the approach response ( Ambroggi et al . , 2008; du Hoffmann and Nicola , 2014; Yun et al . , 2004 ) . By simultaneously recording from NAc neurons and injecting a MOR antagonist into the NAc , we show that NAc MOR activation is required for both behavioral responding to reward-predictive cues and the neural encoding of those cues by NAc neurons . Importantly , these effects were observed in ad libitum chow-fed rats but not in those that had been food restricted . This striking dichotomy indicates that activation of NAc MORs promotes the approach to palatable food only in the absence of a homeostatic need for calories – i . e . , hunger – suggesting that these receptors contribute to a neural mechanism that drives intake of calorie-dense food specifically in the state of satiety . Free-fed rats were trained on a CS task ( Figure 1A ) in which they were presented with an unpredictable series of two auditory tones with a mean intertrial interval ( ITI ) of 30 s . The CS+ tone was reward predictive , such that rats could earn a droplet of heavy cream by making a head entry into the reward receptacle during the 5 s cue presentation . The 5 s CS- tone was not reward predictive and receptacle responses during this cue or the ITI were recorded but had no programmed consequence . CS task performance was assessed by computing a response ratio , defined as the percentage of cue presentations of a particular type ( either CS+ or CS- ) that the animal responded to . Once rats learned to discriminate between cues ( see Materials and methods ) , they were implanted bilaterally with cannulae targeting the NAc core ( see Figure 1—figure supplement 2A for histological examination of injection sites ) . Following recovery from surgery , rats underwent several retraining sessions before the start of the experiment , and a subset of rats was food restricted for at least 7 days concurrent with retraining . By the final day of retraining , food-restricted rats responded to a significantly higher proportion of cues than free-fed rats ( Figure 1B , see figure legend for statistics ) , with free-fed rats exhibiting a pronounced decline in responding over the course of the session ( Figure 1C , solid black line ) . Rats were then bilaterally injected every other day with the selective MOR antagonist CTAP ( 0 , 2 , or 4 µg/side; Figure 1C ) prior to the session . Bilateral CTAP injection significantly attenuated responding to the CS+ in free-fed rats ( Figure 1C , blue solid lines ) but , strikingly , had no effect in food-restricted rats ( Figure 1C , blue dashed lines ) , suggesting a state-dependent contribution of MOR activation to reward-seeking behavior . While food-restricted rats had higher levels of CS- responding than free-fed rats , CTAP had no effect on CS- response ratio in either group ( Figure 1—figure supplement 1 ) . Because there is evidence that NAc MOR activation selectively enhances consumption of fat in lieu of carbohydrates ( Katsuura et al . , 2011; Zhang et al . , 1998 ) , we next asked whether the CTAP effect could also be observed in free-fed rats performing the same task for 3% liquid sucrose reward . Interestingly , while the pattern of CS+ responding for sucrose was similar to responding for cream ( solid black lines in Figure 1D and C , respectively ) , CTAP had no effect on responding for sucrose ( Figure 1D ) . Taken together , these data suggest that MOR blockade preferentially affects responding to fat-predictive cues , and that this effect cannot be attributed to interference with more general motivational or arousal-related neural processes . We next sought to understand the neural mechanism by which CTAP attenuated behavioral responding to reward-predictive cues in free-fed , but not food-restricted rats . Because previous studies have demonstrated that many NAc neurons are excited by reward-predictive cues ( Ambroggi et al . , 2011; Ambroggi et al . , 2008; du Hoffmann and Nicola , 2014; McGinty et al . , 2013; Nicola et al . , 2004a; Yun et al . , 2004 ) , and further , that these cue-evoked excitations are required for behavioral responding to those cues ( du Hoffmann and Nicola , 2014; Yun et al . , 2004 ) , we hypothesized that NAc cue-evoked excitations serve as the neural effector of MOR activation . To address whether this is the case , we first recorded from NAc neurons in both free-fed and food-restricted rats during performance of the CS task . Because free-fed rats respond to a lower proportion of cues , and because cue-excited neurons fired much more in trials in which the rat responded than when it did not ( Figure 2—figure supplement 1A , B ) , we constrained this analysis to trials in which rats responded to the CS+ . Out of 83 neurons recorded in 12 free-fed rats , 45 neurons were excited by the CS+ ( 54 . 2%; Figure 2A ) , as opposed to 91 out of 122 neurons ( 74 . 5%; Figure 2B ) recorded from five food-restricted rats . Further , the magnitude of the cue-excited population response in food-restricted rats was significantly greater than the response in free-fed rats ( Figure 2C , D ) , as was the fraction of 50 ms time bins after cue onset with significant excitations ( Figure 2E , F , upper traces/dots ) . In addition to cue-evoked excitations , we also observed smaller populations of cue-inhibited neurons in both free-fed and food-restricted rats . In free-fed rats , 16 . 8% ( 14 out of 83 ) of neurons were inhibited by the CS+ , compared to 18 . 0% ( 22 out of 122 ) in food-restricted rats ( Figure 2A , B ) . Unlike with excitations , there was no difference in the fraction of significantly inhibited bins ( Figure 2F , lower dots ) . Moreover , there was no difference in the baseline firing rate between the two populations ( Figure 2—figure supplement 2 ) . Because cue-evoked excitations have been shown to encode certain spatial and behavioral elements of response vigor such as distance from receptacle at cue onset , latency to maximum speed , and average speed ( McGinty et al . , 2013; Morrison et al . , 2017 ) , we reasoned that the observed difference in the magnitude of excitation between free-fed and food-restricted populations could be explained by differences in either response vigor or the encoding of response vigor . To determine if this was the case , we first compared the behavioral metrics of restricted and free-fed rats . We found that the average speed of approach after cue onset was significantly greater in restricted rats , while latency to maximum speed and distance did not differ ( Figure 2G ) . To determine if this difference could account for the observed difference in cue-evoked excitation , we used a generalized linear model ( GLM ) to regress the post-cue spike count of each population ( restricted and free fed ) of cue-excited neurons against the three behavioral parameters ( Figure 2—figure supplement 1C–G ) . Next , we used the GLM and coefficients generated from neurons recorded in food-restricted rats to model the spike count on each trial obtained in free-fed rats by entering each trial’s behavioral parameters into the regression equation . Finally , we performed the same analysis , but instead used the GLM generated from free-fed rats to model the spike counts in restricted rats . To interpret the results , we reasoned that if the difference in the magnitude of excitation between restricted and free-fed animals were wholly accounted for by the differences in behavior , then there would not be a significant difference between the modeled spike counts ( using regression coefficients from the other population ) and the actual spike counts from that population . In fact , we observed significant differences in both analyses: when we compared the actual spike counts from free-fed rats to the spike counts predicted by the GLM obtained from food-restricted rats , the modeled spike counts were significantly higher ( Figure 2H , left panel ) . Similarly , spike counts that were modeled with the GLM obtained from free-fed rats were significantly lower than the actual spike counts from the restricted rats ( Figure 2H , right panel ) . We then employed Equation 2 ( See Generalized Linear Model ( GLM ) fitting in Materials and methods ) to test whether the GLMs obtained from each population were statistically distinct , or whether they model the same overall neural population . In brief , this test compares the pooled residual deviance from the two GLMs to the residual deviance of a GLM containing all data from both populations while accounting for the number of regressors . Consistent with the modeling analysis from Figure 2H , the two GLMs do in fact model separate populations , and not the same overall population ( F4 , 2634 = 10 . 97; p<0 . 001 ) . Taken together , these results indicate that lower cue-evoked excitation in free-fed than in restricted rats is not wholly accounted for by lower response vigor , which suggests that additional , unaccounted factors push firing rate lower in free-fed animals than their lower response vigor would predict ( or , equivalently , higher in restricted animals than their greater vigor would predict ) . Many NAc neurons were modulated during reward consumption ( Figure 3 ) . To determine whether neural activity during this epoch differed based on caloric need , we considered firing in the first 3 s following each initial rewarded receptacle entry . Although some excitations and inhibitions lasted longer than 3 s , receptacle exit almost always occurred later than this time point ( Figure 8—figure supplement 1A , B ) , assuring that our analysis window included only periods when the animal was in the receptacle . In free-fed rats , 37 . 3% ( 31 out of 83 ) of neurons were excited for at least a single 400 ms bin following entry into the receptacle on rewarded trials , compared to 60 . 7% ( 74 out 122 ) of neurons in food-restricted rats; moreover , more bins exhibited significant excitation in restricted than free-fed rats ( Figure 3D , upper dots ) . There was also a prominent population of reward-associated inhibitions in each group: 32 . 5% ( 27 out of 83 ) were inhibited for at least one bin following rewarded receptacle entry in free-fed rats , compared to 34 . 4% ( 42 out of 122 ) in food-restricted rats . Unlike excitations , there was no difference in the fraction of bins with significant inhibition between the two groups ( Figure 3D , lower dots ) . It has been demonstrated previously that the magnitude of cue-evoked excitation predicts the vigor of the subsequent cued approach response ( du Hoffmann and Nicola , 2014; McGinty et al . , 2013 ) and further , that these excitations are required for the behavior ( du Hoffmann and Nicola , 2014; Yun et al . , 2004 ) . Therefore , we hypothesized that activation of MORs facilitates cue-evoked excitations in free-fed rats , but not in food-restricted rats . This would explain why CTAP injection impaired cued approach behavior in free-fed rats but not restricted rats ( Figure 1 ) . To test this hypothesis , we injected CTAP into the NAc while simultaneously recording NAc unit activity . Rats trained on the CS task with cream reward were implanted bilaterally with circular microelectrode arrays surrounding a central microinjection guide cannula ( see Figure 1—figure supplement 2B for histological examination of injection sites ) . These arrays allow for the injection of a drug into the same brain region from which neural recordings are being obtained , thereby enabling within-session comparisons of pre- vs post-injection behavior and neural activity ( du Hoffmann et al . , 2011; du Hoffmann and Nicola , 2014 ) . Rats performed the CS task for a 33 min baseline period , after which CTAP was injected by remote activation of a syringe pump ( i . e . , without interrupting the ongoing behavior ) . The pre-injection baseline behavioral performance and neural activity was then compared to the 33 min window after drug injection . In a subset of subjects , rats’ positions in the operant chamber were tracked via two LEDs mounted on the neural recording headstage . As we previously observed ( Figure 1C ) , in bilaterally-injected free-fed rats , CTAP sharply attenuated responding to the CS+ ( Figure 4A , B blue trace and bars ) , while the drug had no effect in food-restricted rats ( Figure 4A , B red trace and bars ) . In contrast , both unilaterally-injected CTAP and saline-injected rats ( Figure 4A , gray and black traces , respectively ) exhibited a slower decline in responding over the session , consistent with the rate of decline previously observed in free-fed rats ( black traces in Figure 1B , D ) ( du Hoffmann and Nicola , 2016 ) . These slow declines in responding were accompanied by increases in latency to initiate movement , which were only slightly further increased after bilateral CTAP injection ( Figure 4—figure supplement 1A–C ) . In addition , locomotor activity during the ITI was not further reduced by bilateral CTAP injection ( Figure 4—figure supplement 1D ) . These results suggest that the CTAP-induced impairment of cued approach behavior ( Figure 4A , B ) was not due to generalized impairment of motor ability . During the behavior shown in Figure 4 , cue-evoked excitations in NAc neurons were significantly attenuated following CTAP injection in free-fed rats ( Figure 5A , B and Figure 5—figure supplement 1A ) . This was true of both the magnitude of the excitations ( Figure 5E , F ) and the fraction of significantly excited bins ( Figure 5J ) . In contrast , the magnitude of cue-evoked excitations in food-restricted animals was unchanged ( Figure 5C , D , G , H ) , as was the fraction of significantly excited bins ( Figure 5L ) . ( There were insufficient cue-evoked inhibitions in free-fed rats to assess the drug’s effects; for the six significantly inhibited neurons in food-restricted rats , a slight decrease in the fraction of inhibited bins pre- vs post-injection did not achieve statistical significance; p=0 . 06 , Wilcoxon . ) To determine whether reductions in behavioral performance could have contributed to the CTAP-induced reduction of cue-evoked excitation , we examined firing during unilateral CTAP injections , which had no discernable behavioral effects in free-fed rats ( Figure 4 ) . To control for the possibility that during some sessions rats would respond to fewer cues post-injection ( due to the gradual decline in cued approach responding in free-fed rats ) , we considered only trials in which rats responded to the CS+ . In neurons ipsilateral to the CTAP injection ( i . e . , neurons directly exposed to drug ) , we observed a significant reduction in the magnitude of cue-evoked excitations post-injection ( Figure 6A , B , E , F ) . In contrast , neurons contralateral to the injection ( i . e . , not exposed to drug; Figure 6C , D ) exhibited no significant reduction in the magnitude of cue-evoked excitations ( Figure 6G , H ) . Additionally , among neurons that were classified as cue-excited or cue-inhibited , the proportion of bins with significant excitation or inhibition was significantly decreased for neurons ipsilateral to the injection ( Figure 6I , J ) , but not for contralateral neurons ( Figure 6K , L ) . Finally , saline injection had no effect on cue-evoked neural activity ( Figure 7 ) , demonstrating that the observed change in firing or behavior following CTAP injection cannot be attributed to any physical perturbation by the injection itself . These results suggest that in free-fed ( but not restricted ) rats , activation of MORs by endogenous ligands in the NAc is required for cue-evoked excitations that , in turn , drive approach to the reward receptacle . Because MORs are classically inhibitory , we tested the possibility that CTAP increases the baseline firing rate of NAc neurons , an effect that could theoretically contribute to the impairment of cued approach behavior . However , CTAP injection had no effect on baseline firing rate in free-fed rats , as the slope of the regression line of pre- vs . post- baseline firing rate did not significantly differ from the unity line ( Figure 5—figure supplement 2A ) . Neurons from food-restricted rats demonstrated a slight reduction in baseline that may be attributable to the presence of outliers ( Figure 5—figure supplement 2B ) , and which is unlikely to have affected behavior , as CTAP did not impact cued approach in restricted animals . Because the existing literature suggests that ( 1 ) the µ-opioid system in the NAc maintains hedonic responses to food ( Bakshi and Kelley , 1993; Bodnar et al . , 1995; Kelley et al . , 1996; Mucha and Iversen , 1986; Peciña and Berridge , 2000; Zhang et al . , 1998; Zhang and Kelley , 1997 ) , and ( 2 ) a population of neurons in the NAc encodes relative palatability ( Taha and Fields , 2005 ) , we hypothesized that neuronal modulation during reward consumption might contribute to subsequent reinforcement of cued approach . Furthermore , we reasoned that CTAP might affect reinforcement and thus change the probability of cued approach behavior in free-fed animals by interfering with consumption-associated neural activity . We performed three analyses of our data to address this possibility . First , we reasoned that if firing during reward consumption contributed to reinforcement , then consumption-associated firing on a given trial should predict the probability of a behavioral response on the next trial . To test this idea , we used data from free-fed , uninjected rats to run a logistic regression to ask whether the number of spikes between 1–3 s after the rewarded entry on a given trial influenced the response probability on the subsequent trial . For both the consumption-excited and consumption-inhibited population , the spike count on a given trial did not significantly contribute to response probability ( p=0 . 18 and p=0 . 20 , respectively , Wald test ) , suggesting that reward-associated firing does not influence subsequent behavioral responding ( at least on a trial-to-trial basis; we cannot rule out the possibility that firing on a given trial may influence responding at some later point in the session ) . Second , we examined reward-associated firing in three free-fed populations: neurons ipsilateral to CTAP injection , neurons contralateral to CTAP injection , and neurons from uninjected subjects . Histograms aligned to receptacle entry show that subpopulations of neurons were excited and inhibited during reward consumption ( Figure 8 ) . These neural responses were not merely continuations of cue-evoked excitations and inhibitions as cue-evoked activity tended to end prior to receptacle entry ( Figure 8—figure supplement 2 ) . Although we found significant post-injection decreases in reward-evoked excitations and inhibitions in both ipsilateral and contralateral populations ( Figure 8A–H ) , we also observed similar changes in neurons from uninjected subjects simply by breaking up the responses into identical time epochs as the injected neurons ( Figure 8I–L ) . Therefore , within-session changes in neural modulation to reward consumption cannot be attributed to the presence of CTAP . ( Decreases in the magnitude of cue-evoked excitation across these epochs in uninjected animals were not observed; Figure 8—figure supplement 1C–F . ) Finally , unilateral CTAP injection did not significantly affect either the total time spent in the receptacle during reward consumption or the overall number of receptacle entries during the consumption epoch ( Figure 8—figure supplement 1A , B ) , indicating that the change in reward-associated firing observed during unilateral injection sessions is not a consequence of a change in consumption behavior . Taken together , these analyses indicate that in free-fed rats , declines in reward-associated firing over the course of the session are unlikely to be due to a MOR-dependent mechanism . In addition , they suggest that the CTAP-induced impairment of cued approach behavior ( Figures 1 and 4 ) is very unlikely to result from changes in reward-associated firing . In states of relatively high satiety , humans and animals greatly favor calorie-dense foods over less palatable options – a preference that likely contributes to overconsumption and obesity . Our results reveal a potential neural mechanism underlying this preference . We find that blockade of MORs in the NAc core attenuates both cue-evoked approach to high-fat food and the encoding of those cues by NAc neurons , and that these effects are observed only in rats fed ad libitum chow , and not in food-restricted ( relatively hungry ) rats . These effects could not be attributed to changes in consumption-related behavior or firing . Notably , NAc cue-evoked excitations are causal to cued approach ( du Hoffmann and Nicola , 2014 ) , suggesting that a novel and fundamentally important role of the NAc opioid system is to promote approach to highly palatable food specifically when there is no immediate homeostatic need for calorie intake . Together , these findings suggest NAc MORs as a target for development of treatments that limit overeating , consistent with the present use of drugs that block MORs as viable therapeutic options for the treatment of obesity ( Apovian , 2016; Ziauddeen et al . , 2013 ) . Although the NAc opioid system has long been implicated in the regulation of food intake ( Castro and Berridge , 2014; Kelley et al . , 2005; Nicola , 2016; Peciña et al . , 2006; Selleck and Baldo , 2017 ) , the MOR effects identified here are characterized by several features that were not necessarily predicted by previous studies . We find that activation of NAc MORs by endogenous ligands promotes appetitive behavior by increasing neural activity that drives approach to food , whereas NAc MORs appear to contribute little ( if at all ) to neural activity related to consumption . The latter conclusion appears to contrast with prior evidence that activation of these receptors by exogenous ligands increases hedonic taste reactions ( Peciña and Berridge , 2000 ) , which should be controlled by NAc neuronal activity occurring during consumption . However , because we targeted our electrodes to the NAc core , whereas hedonic taste reactions are promoted by MOR agonist injection in a very specific zone of the NAc shell ( Peciña and Berridge , 2000 ) , our results do not preclude the possibility that endogenous opioids promote taste reactivity ( and perhaps hedonia ) by influencing the consumption-related firing of NAc shell neurons . On the other hand , injection of MOR agonists into either the NAc core or shell increases consumption of palatable food ( Bakshi and Kelley , 1993; Katsuura and Taha , 2014; Mucha and Iversen , 1986; Ward et al . , 2006; Woolley et al . , 2006; Zhang et al . , 1998; Zhang and Kelley , 1997 ) , which must be due to promotion of some form of NAc neuronal activity that drives consumption . Our results suggest that , at least in the core , this form of neural activity is the pre-movement firing of NAc neurons , which can be activated by cues ( McGinty et al . , 2013; Morrison et al . , 2017 ) and which drives initiation of approach to calorie-dense food ( du Hoffmann and Nicola , 2014 ) . Further supporting this idea , activation of MORs in the NAc core by exogenous agonists increases consumption of a high-fat liquid in part by increasing the number of licking bouts ( perhaps as a result of increasing the number of approaches to the lickometer ) ( Katsuura et al . , 2011; Lardeux et al . , 2015 ) . Moreover , exogenous activation of NAc MORs promotes operant behavior for food reward ( Zhang et al . , 2003 ) , and in fact is sufficient to increase operant responses to cues predictive of high-calorie food in a Pavlovian-instrumental transfer ( PIT ) test ( Peciña and Berridge , 2013 ) . The latter observation in particular supports the hypothesis that MOR activation directly promotes approach behavior without an intermediate effect on neural activity during consumption because the PIT test is performed in extinction . Further arguing against an intermediate effect on consumption , we observed previously that injection of CTAP into the NAc does not reduce consumption of a high-fat liquid in free fed rats ( Lardeux et al . , 2015 ) . Finally , we find that unilateral infusion of CTAP into the NAc greatly reduces cue-evoked firing ( Figure 6 ) , an effect that could not have been due to reduced consumption or other performance deficit because cued approach performance was unaffected by unilateral infusions ( Figure 4 ) . Thus , we conclude that NAc core MORs act primarily to promote food seeking rather than consumption itself . Our results do not , however , rule out the possibility that MORs contribute to some aspect of the consumption- or reinforcement-related firing of NAc neurons . Indeed , excitations in a small population of NAc neurons encode the value of liquid rewards during consumption ( Taha and Fields , 2005 ) ; because we did not vary reward value , we may have missed an effect of CTAP on this form of encoding . However , more likely contributors to consumption are the large population of NAc neurons that are inhibited in proportion to the rate of licking during consumption ( Taha and Fields , 2005 ) . Together with findings that experimental silencing of NAc neurons drives consumption ( Reynolds and Berridge , 2001; Stratford and Kelley , 1997 ) whereas consumption is interrupted by brief excitation of the NAc ( Krause et al . , 2010 ) , these observations suggest that naturally occurring reductions in the firing of a population of NAc neurons ( possibly containing D1 dopamine receptors and projecting to the lateral hypothalamus; O'Connor et al . , 2015 ) drive consummatory behavior . Our observation that consumption-related inhibitions are not reduced by CTAP indicates that these inhibitions do not depend on MORs , consistent with previous findings that generalized inhibition of the NAc results in nonspecific increases in food intake whereas MOR activation results in preferential increases in consumption of calorie-dense , and especially high-fat , food ( Katsuura and Taha , 2014; Ward et al . , 2006; Woolley et al . , 2006; Zhang et al . , 1998; Zhang and Kelley , 1997 ) . This specificity in the case of MOR agonists may be due in part to increased approach to calorie-dense food , perhaps via enhanced firing of NAc core neurons that drive such approach behaviors . Strikingly , CTAP reduced cued approach when the reward was cream , but not liquid sucrose . The latter observation is consistent with previous studies indicating that activation of NAc MORs induces a preference for fat over carbohydrates ( Taha , 2010; Zhang et al . , 1998 ) ; however , similar manipulations also induce greater preference for the already-preferred flavor of two foods with equivalent nutritional content ( Woolley et al . , 2006 ) . Although we cannot rule out the possibility that relatively greater preference for ( or palatability of ) cream vs sucrose was instrumental to the much greater effects of CTAP when the cue predicted cream as opposed to sucrose , the sucrose concentration ( 3% ) was chosen such that the animals’ behavior was similar to that observed with cream reward ( Figure 1C , D ) , suggesting that it was the difference in nutrient content , not preference , that dictated the difference in dependence on NAc MORs . Although cream contains , in addition to fat , small quantities of certain nutrients that are absent from sucrose solution ( e . g . , protein , lactose ) , these minor components of cream are unlikely , on their own , to support appetitive approach and consumption in the free-fed state . Thus , our results suggest that NAc MORs specifically promote approach to high-fat foods . Even if flavor-based preference or palatability played a role , we note that preferred palatable foods tend to be calorie dense , supporting a role for NAc MORs in overconsumption that leads to obesity . Such a role is further supported by the remarkable observation that CTAP affected neither cued approach behavior nor cue-evoked neural activity in food-restricted animals , despite markedly reducing both in rats given ad libitum access to chow . To our knowledge , this is the first report of a satiety state-dependent contribution of endogenous NAc opioids to food-seeking behavior . Indeed , few studies have examined whether blockade of NAc MORs ( as opposed to activation with exogenous agonists ) impacts food-seeking . One exception is the observation that β-funaltrexamine ( β-FNA ) , a long-lasting MOR antagonist , reduces rats’ speed during runway approach to calorie-dense food ( Shin et al . , 2010 ) . However , NAc injection of β-FNA has also been shown to reduce spontaneous locomotion ( Kelley et al . , 1996 ) , suggesting that it may have had non-specific effects . Such effects could potentially also explain the reduction in calorie-dense food consumption after β-FNA injection in the NAc ( Bodnar et al . , 1995; Kelley et al . , 1996; Lenard et al . , 2010; Shin et al . , 2010 ) . Intriguingly , the MOR antagonist we used , CTAP , does not impair spontaneous locomotion when injected in the NAc ( Figure 4—figure supplement 1D ) , and is also apparently less effective than β-FNA in reducing consumption ( Katsuura et al . , 2011; Lardeux et al . , 2015 ) although a study in rabbits reports greater reductions in consumption ( Ward et al . , 2006 ) . One possibility consistent with our results is that CTAP impairs approach to food as opposed to consumption itself; differences in CTAP effects on amount of freely available food consumed could be due to differences in experimental conditions such as size of the test chamber ( and thus degree of approach required ) , species , nutrient content , and satiety state . The stark difference in effects of CTAP in free fed and restricted animals raises three important topics for further research . The first is to determine the degree of restriction that is sufficient to eliminate the dependence of cued approach on NAc MORs . Although MOR antagonists can reduce food consumption after mild ( <24 hr ) restriction ( Bodnar et al . , 1995; Kelley et al . , 1996 ) , this may not be the case for cued approach behavior . If endogenous opioids in the NAc promote cued approach to fatty foods when restriction is much less severe than the chronic restriction used here , it would imply that this neural system contributes to caloric intake when meal patterns are more natural than the extremes employed here ( freely available chow and severe restriction ) . The second question is the mechanism whereby endogenous MOR ligands promote cued approach . As observed previously ( Ambroggi et al . , 2011; du Hoffmann and Nicola , 2014; McGinty et al . , 2013; Morrison et al . , 2017; Nicola et al . , 2004a ) , we found that prominent populations of NAc neurons are excited and inhibited by cues that evoke approach behavior ( Figure 2A , B ) . Previously , we established that these changes in firing begin prior to initiation of approach movement , and that the magnitude of the firing changes predicts the latency and speed of approach ( McGinty et al . , 2013; Morrison et al . , 2017 ) . Cue-evoked excitations , but not inhibitions , are dopamine-dependent; because injection of dopamine receptor antagonists reduces both cued approach and cue-evoked excitations ( du Hoffmann and Nicola , 2014 ) , the excitations are likely causal to the subsequent approach behavior . These observations suggest that activation of MORs by endogenous opioids could increase cue-evoked excitation via a direct action on NAc neurons , on the glutamatergic terminals that likely drive the excitation , or on inhibitory interneurons or inputs that limit the magnitude of cue-evoked excitation . Because MOR effects tend to be inhibitory , the latter hypothesis is most likely . Alternatively , MOR activation by endogenous opioids could promote the release of dopamine , which could , in theory , promote greater cue-evoked firing and hence increase the probability of an approach response . This idea is consistent with previous findings that exogenous activation of MORs can increase dopamine levels ( Borg and Taylor , 1997; Hipólito et al . , 2008; Hirose et al . , 2005; Okutsu et al . , 2006; Yoshida et al . , 1999 ) , and with observations that dopamine release can be modulated at the terminal ( Cachope and Cheer , 2014; Wenzel and Cheer , 2018 ) . Moreover , dopamine neurons are strongly regulated by the state of caloric need , with greater activation in higher need states ( Meye and Adan , 2014; Nicola , 2016 ) , and in fact dopamine release evoked by food-predictive cues is greater in food-restricted rats than free-fed rats ( Aitken et al . , 2016; Cone et al . , 2014 ) – an observation that could explain the present finding that NAc cue-evoked excitations are greater in restricted than free-fed animals ( Figure 2 ) . According to this hypothesis , dopamine levels in the free-fed state are insufficient to raise cue-evoked firing above the threshold for reliably obtaining a cue-evoked approach response . However , when the subject is in an environment in which calorie dense and/or high-fat food is available , neurons that release the endogenous agonist of NAc MORs in the NAc are activated to release the opioid , and the resulting activation of MORs raises the dopamine level such that the magnitude of cue-evoked firing is sufficient to evoke a behavioral response . In contrast , in food-restricted subjects , the dopamine level is so high that either further increases are not possible , or the cue-evoked firing of NAc neurons is maximal such that further increases in dopamine are without effect . The third question is the source and nature of the opioid peptides that activate MORs to promote food-seeking . Presumably , the endogenous ligand for NAc MORs is enkephalin released by the large population of D2 receptor-expressing spiny neurons ( Gerfen et al . , 1990; Mansour et al . , 1995 ) . The peptide could be released by the extensive axon collaterals of these neurons within the NAc; alternatively , while it has not been demonstrated in the NAc , opioids can be released somatodendritically and act as a retrograde messengers ( Iremonger and Bains , 2009; Wagner et al . , 1993; Wamsteeker Cusulin et al . , 2013 ) . The conditions under which enkephalin release in the NAc is increased are unknown; however , in the dorsomedial striatum , enkephalin levels are elevated during meal onset ( DiFeliceantonio et al . , 2012 ) , suggesting that information about the availability of food drives release of the peptide . One possibility is that enkephalin release is tonically promoted when the subject is in an environment in which fatty foods are available; another is that release occurs precisely at cue onset in response to discrete cues that predict fat , but not carbohydrates . Further investigation of the hypothesis that NAc enkephalin release is regulated by fat availability , and the mechanisms by which this could occur , is clearly warranted . The mechanism we propose here – that enkephalin levels are elevated by fat availability , and these high enkephalin levels promote dopamine release that in turn increases cue-evoked excitations that drive cued approach to high-fat food – is partially speculative , but it is fully consistent with the present and previous results and provides a starting point for further exploration . Importantly , our results indicate that the neural mechanisms underlying appetitive behavior must be considered when studying the contribution of opioids ( and other neuromodulators ) in the forebrain to food intake regulation . Cued approach is only one form of appetitive behavior , but our demonstration that endogenous opioids bias this form of behavior towards fat seeking suggests that opioids may have similar effects on the neural mechanisms that control other , more complex and/or more cognitive appetitive behaviors , such as deciding among simultaneously-available food options . Although MOR antagonists are currently used to treat obesity ( Apovian , 2016; Ziauddeen et al . , 2013 ) , a more refined understanding of the impact of endogenous opioids on appetitive behaviors is required to understand how these drugs work , and to identify future targets for more specific and effective treatments that reduce preference for calorie-dense or high-fat foods . 52 male Long-Evans weighing between 275 and 300 g were obtained from Charles River Laboratories and singly housed for a week before handling . Each rat was then handled for several minutes daily for 3 days to habituate them to the experimenter . Rats were randomly allocated to their experimental groups . Those designated for experiments requiring food restriction were limited to ~15 g of rodent chow per day for at least one week prior to the start of the experiment ( to achieve 90% free-feeding weight ) , whereas free-fed animals had unlimited access to chow . All animals had unlimited access to water in their home cages . All procedures involving animals were in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals and were approved by the Institutional Animal Care and Use Committee at Albert Einstein College of Medicine . Two styles of operant chambers were used in this study . For behavioral pharmacology experiments , chambers measured 30 . 5 cm x 24 . 1 cm and were supplied by Med Associates ( St . Albans City , VT ) ; chambers reserved for electrophysiology experiments measured 40 cm x 40 cm and were custom-designed . All chambers were outfitted with a reward receptacle equipped with an infrared head entry detector ( Med Associates ) , as well as two 28 V house lights , a 65 dB white noise generator , and speakers for generating auditory cues . Reward was delivered via a syringe pump connected to the receptacle using 3/16" steel-reinforced PVC tubing to ensure consistent volume of reward delivery . All operant chamber hardware was controlled via custom-written Med-PC scripts . All rats used in this study were ad-libitum fed for the duration of training . The day before the first training session , rats were given access to heavy cream ( per 100 g: 37 g fat , 2 . 8 g carbohydrate including 0 . 1 g sugars , 2 . 1 g protein ) or 3% sucrose solution in their home cages to familiarize them with the reward . Training sessions lasted 2 hr . On the first day of training , rats were rewarded for simply entering the reward receptacle , with a 10 s timeout between rewarded entries . If they obtained >50 rewards , they advanced to the next phase of training; otherwise the current phase was repeated . In the second training phase , rats were presented with a reward-predictive CS+: either a siren tone ( frequency cycle between 4 and 8 kHz over 400 ms ) or an intermittent tone ( 6 kHz tone on for 40 ms , off for 50 ms ) was played for a maximum duration of 5 s at a fixed intertrial interval ( ITI ) of 15 s . Head entries into the receptacle during presentation of the CS+ resulted in termination of the cue and delivery of ~50 µl heavy cream or 3% sucrose solution ( although each rat received only one reward type ) . After rats obtained >50 rewards in a session , they were advanced to the full CS task , in which the CS+ or a neutral CS- ( the siren tone for rats whose CS+ was the intermittent tone , and vice versa ) were presented at ITIs randomly selected from a truncated exponential distribution ( mean = 30 s , minimum of 10 s , maximum of 150 s ) . The CS- was presented for 5 s , regardless of receptacle entry , and had no programmed consequence . Rats were considered fully trained once they responded to >40% of CS+ presentations and had a discrimination index ( defined as the number of CS+ responses divided by the total number of cue responses ) of at least 0 . 67 , indicating that rats reliably discriminated between the CS+ and CS- . Electrode arrays were custom-designed and assembled as previously described ( du Hoffmann et al . , 2011; du Hoffmann and Nicola , 2014 ) . Briefly , each array consisted of 8 Teflon-insulated tungsten microwires ( A-M Systems ) encircling a 27-ga microinjection guide cannula . Each electrode was checked to ensure its impedance fell in the range of 90–110 MΩ . Electrodes and cannulae were mounted inside a drivable casing; a hex screw enabled the entire assembly to be driven along the dorsal-ventral axis of the NAc . Each full revolution of the screw drove the array ~350 µm . Once assembled , wires were soldered onto 10-pin connectors ( Omnetics ) and impedances were re-measured to ensure connection patency . A silver ground wire was soldered to the last pin on the connector . After rats reached criterion performance on the CS task , they were implanted either with custom-built bilateral cannulated microelectrode arrays aimed at the NAc core or with bilateral 26 ga microinjection cannulae ( Plastics One , Roanoke , VA ) aimed at the NAc core as described previously ( du Hoffmann et al . , 2011; du Hoffmann and Nicola , 2014; Nicola , 2010 ) . Rats were anesthetized with isoflurane ( 1–2% ) and placed in a stereotaxic apparatus . From Bregma , cannulated arrays were implanted at AP +1 . 4 mm , ML ±1 . 5 mm , and DV −6 . 5 mm , while microinjection cannulae were implanted at AP +1 . 2 mm , ML ±2 . 0 mm , and DV −5 . 7 mm ( microinjectors were designed to extend 2 mm beyond the cannulae tips , to a target of DV −7 . 7 mm ) . Implants were secured using dental acrylic bound to six screws fixed to the surface of the skull . Steel obdurators ( Plastics One ) were inserted into the cannulae to prevent them from clogging . For electrode surgeries , ground wires were inserted into the brain at a posterior location , and connectors were fixed to the implant at the posterior aspect of the cap . Antibiotics ( Baytril ) were provided immediately before and 24 hr after surgery . Rats were allowed one week to recover from surgery before re-training commenced . After recovering from cannulation surgery , a subset of rats were food-restricted for one week . Food restriction was concomitant with re-training . All other rats continued to have ad-libitum access to food . After behavioral responding was re-established , we began the microinjection procedures . Microinjectors ( 33 ga , Plastics One ) were affixed to polyethylene tubing that was back-filled with mineral oil and connected to two 1 µl Hamilton syringes which were under the control of a microinjection pump ( KD Scientific , Holliston , MA ) . On the first day , rats received a mock injection to habituate them to the injection procedure . Rats were gently restrained while microinjectors were inserted into the guide cannulae and left in place for 1 min prior to the start of the infusion to allow the tissue to equilibrate around the injectors . D-Phe-Cys-Tyr-D-Trp-Arg-Thr-Pen-Thr-NH2 ( CTAP ) ( Sigma-Aldrich; 0 , 2 , or 4 µg/side ) , was dissolved in 0 . 9% saline and infused at a rate of 0 . 25 µl/min for 2 min for a total infusion volume of 0 . 5 µl per hemisphere . Post-infusion , injectors were left in the cannulae for 1 min post-infusion to allow the drug to diffuse into the tissue . After injection , rats were immediately placed in the operant chambers and the behavioral session was started . The order in which each rat received each drug dose was pseudo-randomized across injection days . Injection days were interleaved with non-injection days to ensure that rats’ behavior returned to baseline performance levels . Following recovery from cannulated electrode array implantation , a subset of rats were food-restricted as in the microinjection experiments . After consistent behavior was re-established , rats were tethered to a 16-channel commutator by the recording cable , which allowed for free rotational movement of the animal during neural recordings . On simultaneous recording/injection days , 33-ga microinjectors were affixed to polyethylene tubing pre-filled with mineral oil and connected to a 2-channel fluid swivel to allow for free rotational movement , terminating at a microinjection pump ( KD Scientific ) that sat atop the chamber . Drug was then loaded into the microinjector tips such that the interface between saline-dissolved drug and mineral oil was visible; the location of this interface was marked on the fluid lines . Prior to the start of the session , rats were tethered to the recording apparatus via the recording cable and either one ( for unilateral injections ) or two ( for bilateral injections ) microinjectors targeting a depth of 500 µm beyond the electrode tips were inserted into the guide cannula and taped to the recording cable such that they could not be readily removed by the rat . Once secured , fluid lines were visually inspected to ensure that the drug-oil interface remained at the marking on the fluid line , assuring that drug had not leaked out prematurely . Neural signals were then examined online to isolate active channels ( see Materials and methods section Acquisition of neural data ) and the behavioral session commenced . To obtain a neural and behavioral baseline , rats performed the task for 2000 s ( ~33 min ) , at which point the drug pump was remotely triggered , initiating the infusion of a 0 . 5 µl volume of either saline or 8 µg/side CTAP over a period of 12 min . This procedure allowed us to compare behavior and neural activity during the baseline window to behavior and neural activity during an equivalent-duration post-injection window . The higher 8 µg/side CTAP dose was chosen to mitigate the possibility that potentially partial drug effects using lower doses would mask changes in behavior and neural activity in the briefer window of examination ( ~33 min ) used in these experiments . Rats were connected to a recording cable outfitted with a 16-channel headstage . The cable was connected to a multichannel commutator that was in turn connected to a pre-amplifier , where the neural signals were amplified by 2 , 000–20 , 000X and band-pass filtered at 250 Hz and 8 . 8 kHz before being passed to a 40 kHz multi-unit acquisition processor . Prior to the start of a session , each channel was examined for putative unit activity using SortClient ( Plexon Inc , Dallas , TX ) and optimized for gain and threshold . Following acquisition , putative units were isolated manually using Offline Sorter ( Plexon ) . To be included in the analysis , units had to have an absolute amplitude ≥75 µV and ≤0 . 1% of all inter-spike intervals could be ≤2 ms . When multiple units were recorded on the same channel , cross-correlograms were used to ensure that spikes were assigned to the appropriate unit and that the units were well-isolated from one another . If these conditions were not met , then the spiking activity on these ambiguous channels was discarded . Spike timestamps were then imported into R , combined with the associated behavioral data , and analyzed using custom routines ( [Caref , 2018]; https://github . com/kcaref/neural-analysis; copy archived at https://github . com/elifesciences-publications/neural-analysis ) . Neurons were classified as cue-excited if they exceeded the 99 . 9% confidence interval of a Poisson distribution comprised of a 10 s pre-cue baseline for at least one 50 ms bin following CS+ onset and up to 500 ms post-cue onset . Neurons were classified as cue-inhibited if they fell below the 99% confidence interval for at least one 50 ms bin . A less stringent detection threshold was used for inhibitions because many NAc neurons exhibit low baseline firing rates , making it harder to detect inhibitions due to floor effects . Neurons were classified as significantly excited during reward consumption if firing in at least one 400 ms bin following the rewarded receptacle entry exceeded the 99% confidence interval of a Poisson distribution comprised of a 10 s pre-cue baseline; they were classified as consumption-inhibited if firing fell below the 99% confidence interval . For simultaneous recording/injection experiments , classification of neural responses was performed only during the pre-injection epoch so that any potential drug effects would not contribute to the neuronal classification . To construct heat maps illustrating the frequency and magnitude of cue-evoked excitations and inhibitions , for each neuron a receiver operating characteristic ( ROC ) curve was computed in 10 ms bins from 1 s prior to cue onset to 1 . 5 s after . The ROC curve used data from each trial to compare the firing rate in each bin to the 1 s baseline . The area under the ROC curve ( auROC ) for each bin was then displayed as the smoothed mean of a 200 ms sliding window . To construct heat maps for consumption-evoked activity , an auROC value was computed for each 200 ms bin from 1 s before the rewarded receptacle entry to 5 s after the rewarded entry using the pre-cue epoch as the baseline . auROC values are displayed as the smoothed mean of an 800 ms sliding window . An auROC value of 1 corresponds to very strong excitation; a value of 0 corresponds to very strong inhibition , and a value of 0 . 5 indicates no change in evoked firing relative to baseline . Because auROC values are always between 0 and 1 , these values can be used to visually compare cue-evoked firing across different neuronal populations and conditions . All statistical comparisons were performed on non-smoothed data . When possible , the rat’s position was tracked by an overhead camera at 30 fps using 2 LEDs mounted on the neural recording headstage . Video tracking was conducted using the CinePlex software suite ( Plexon ) . Tracking data were preprocessed as described previously ( McGinty et al . , 2013 ) . Briefly , the locomotor index ( LI ) was computed for each frame by taking the standard deviation of the frame-to-frame difference in x-y position for four preceding and four succeeding video frames . Thus , the LI for each frame is a smoothed spatial and temporal representation of the rat’s speed over nine frames ( ~300 ms ) . The resulting distribution of LIs for all video frames was then fitted with a double-Gaussian function; the subject was considered still when LI values were below the threshold between Gaussian peaks , and moving if the LI value was above this threshold . The LI threshold value differed from rat to rat and depended in part on the rat’s overall activity during the session . Nearest neighbor analysis was then conducted to determine the video frame corresponding to the start of behavioral and task events such as CS+ onsets . The latency to initiate locomotion following cue onset was computed by subtracting the timestamp of the cue from the timestamp of the first frame following cue onset whose locomotor index exceeded the threshold value . For all experiments , results were considered statistically significant if p<0 . 05 . For behavioral pharmacology experiments , drug effects were evaluated using two-way ANOVA . When appropriate , post-hoc tests were conducted using the Holm-Sidak p-value adjustment for multiple comparisons . For comparisons of event-evoked firing in separate neural populations , Wilcoxon rank sum tests were used; for within-session comparisons of the same neural population , Wilcoxon signed-rank tests were used . To compare pre- vs . post-injection baseline firing rates , a 95% confidence interval was constructed around the slope of the regression line resulting from plotting the pre-injection baseline against the post-injection baseline . If the confidence interval included 1 , the result was not statistically significant . For the modeling procedures employed in Figure 2 , the contribution of behavioral and spatial parameters to cue-evoked firing was examined using a GLM , ( 1 ) ln⁡ ( Y ) = β0+β1x1+β2x2+ε , where Y is the number of spikes in the window between 50 and 500 ms post-cue , β0…n are the regression coefficients for each dependent variable , x1…n are the values of the dependent variables ( i . e . , the regressors such as distance , speed , etc . ) , and ε is an error term . This form of the GLM assumes Poisson-distributed values of Y , and as such the log transformation of Y is in reference to the model fit , not the actual data . Each population of cue-excited neurons ( i . e . , neurons from free-fed rats and those from food-restricted rats ) was pooled to facilitate population-level comparisons . To both validate GLM fits and to evaluate whether two population GLMs model the same overall population , we used the following procedure , ( 2 ) F= SSt−SSp ( m+1 ) ( k−1 ) SSpDFpwhere SSt is the total residual sum of squares from a GLM fitted to the combined data , SSp is the pooled residual sum of squares from each individual GLM , m is the number of regressors in each individual GLM , k is the number of GLMs being evaluated , and DFp is the pooled residual degrees of freedom from each individual GLM ( Zar , 1999 ) . The resulting F statistic is then converted to a p-value; the numerator degrees of freedom is the m+1k-1 expression and the denominator degrees of freedom is DFp . If two GLMs model the same overall population , then the resulting p-value will be > 0 . 05 . To assess GLM fit , we employed a within-population bootstrapping approach . For each neural population , we fitted a GLM using a randomly-sampled selection of 50% of CS+ trials on which animals responded to the cue . We then fitted a second GLM using the remaining trials , and computed a p-value using Equation 2 . This procedure was performed 1000 times for each neural population , with the reasoning that if the model fits are adequate , then p will be > 0 . 05 on 95% of the bootstrapped permutations . Before sacrifice , all rats were injected with Euthasol ( 39 mg/kg pentobarbital ) to induce deep anesthesia . Rats from microinjection experiments were decapitated and their brains were removed and stored in 4% paraformaldehyde solution . Rats from electrophysiology experiments underwent intracardiac perfusion of saline followed by 4% paraformaldehyde solution . They were then decapitated , and their brains were removed and stored in 4% paraformaldehyde solution . All brains were then sectioned into 50 µm slices using a vibratome . The slices were mounted on slides and cresyl violet-stained to facilitate examination of injection sites and cannula placements . One food-restricted rat from the microinjection experiment ( Figure 1 ) died before its brain could be extracted and examined , but because the rest of its cohorts’ cannulae were placed accurately , we decided to include this data in the study .	Imagine that you have just finished Thanksgiving dinner . You are completely full , having eaten large portions of turkey , green beans and mashed potatoes . Yet , despite feeling full , you still find yourself tempted by a slice of pie for dessert , maybe even with ice cream on top . Why is it that in such a state of fullness , you desire a slice of pie but not , say , another helping of green beans ? The answer may lie in the way the brain responds to food when we do not need any more calories . At such times , your brain drives you to continue eating only those foods that are tasty and calorie-dense . This preference for fatty and sweet foods may have been helpful in the past when we could not be certain where our next meal would come from . But in modern times , the widespread availability of food makes this preference potentially harmful . For example , the drive to consume fatty and sweet foods even when not hungry may now be contributing to soaring levels of obesity and type 2 diabetes . What exactly is happening inside the brain to produce this behavior ? Previous work has implicated a structure called the nucleus accumbens . When scientists activated proteins called mu opioid receptors within the nucleus accumbens , animals ate more of the foods that they find tasty . However , they were not as interested in eating more of the foods that they are more ambivalent towards . Caref and Nicola now show that preventing opioid binding makes rats unwilling to respond to a cue to obtain cream , an appetizing , high-fat reward . It also abolishes the brain activity that drives the rats to respond the cue . Crucially , however , this effect only occurs in rats that are not hungry . It therefore appears that opioid binding in the nucleus accumbens drives animals to approach and eat high-fat foods , but only when they do not need the calories . That is , it increases fat consumption in animals that are not actually hungry . A drug that selectively blocks mu opioid receptors in the nucleus accumbens may reduce this behavior . Such a drug could potentially help to prevent obesity and the health problems associated with it .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2018	Endogenous opioids in the nucleus accumbens promote approach to high-fat food in the absence of caloric need
Understanding how climate-mediated biotic interactions shape thermal niche width is critical in an era of global change . Yet , most previous work on thermal niches has ignored detailed mechanistic information about the relationship between temperature and organismal performance , which can be described by a thermal performance curve . Here , we develop a model that predicts the width of thermal performance curves will be narrower in the presence of interspecific competitors , causing a species’ optimal breeding temperature to diverge from that of its competitor . We test this prediction in the Asian burying beetle Nicrophorus nepalensis , confirming that the divergence in actual and optimal breeding temperatures is the result of competition with their primary competitor , blowflies . However , we further show that intraspecific cooperation enables beetles to outcompete blowflies by recovering their optimal breeding temperature . Ultimately , linking abiotic factors and biotic interactions on niche width will be critical for understanding species-specific responses to climate change . Recent anthropogenic climate warming makes understanding species vulnerability to changing temperatures one of the most pressing issues in modern biology . A cornerstone for understanding the distribution and associated ecological impacts of climate change on organismal fitness is the concept of the ecological niche , which describes a hyperspace with permissive conditions and requisite resources under which an organism , population , or species has positive fitness ( Hutchinson , 1957; Chase and Leibold , 2003; Colwell and Rangel , 2009 ) . More than a half century ago , Hutchinson ( Hutchinson , 1957 ) distinguished the fundamental niche—characterized by abiotic environmental measures like temperature and precipitation—from the realized niche—characterized by biological interactions like competition and predation . Although numerous studies since then have used the relationship between a species’ distribution and the climate in which it occurs to estimate a species’ fundamental niche ( Kearney and Porter , 2009; Quintero and Wiens , 2013; Gaüzère et al . , 2015 ) , a growing number of studies have also employed theoretical , observational , or experimental approaches to evaluate the detailed mechanisms ( e . g . physiological limits ) ( Huey et al . , 2012 ) , demographic factors , and species interactions that shape a species’ fundamental and realized niches ( Monahan , 2009; Nilsson-Örtman et al . , 2013; Estay et al . , 2014 ) . In general , differentiating between how abiotic factors and biotic interactions influence niche width is critical for understanding species-specific responses to climate change . Ultimately , generating a complete understanding of a species’ ecological niche not only requires understanding how abiotic and biotic factors interact to affect organismal performance and fitness , but also identifying the detailed mechanisms that shape differences in the fundamental and realized niches . The concept of the thermal performance curve ( TPC ) was first developed to study how body temperature affects ectotherms’ physiological responses and behaviors ( Kearney and Porter , 2009; Quintero and Wiens , 2013; Gaüzère et al . , 2015 ) . Although the TPC concept has received increasing attention in studies of how warming temperatures influence organismal fitness ( Clusella-Trullas et al . , 2011; Sinclair et al . , 2016 ) , it has been developed largely independently from niche-based studies . Yet , characterizing the TPC is essentially a way to mechanistically quantify a species’ thermal niche . Since the TPC describes the detailed relationship between temperature and fitness , the concept may actually be more informative than that of the thermal niche , which is typically defined as the range of temperatures over which organisms occur in nature ( i . e . thermal niche width ) ( Huey and Stevenson , 1979; Hillaert et al . , 2015 ) . In other words , the TPC concept not only describes thermal niche width , it also quantifies an organism’s optimal temperature and how fitness varies with changes in temperature . In contrast , most TPC studies focus largely on what amounts to the fundamental TPC and do not explicitly consider biotic interactions ( Dillon et al . , 2010; Gunderson et al . , 2016; Swaddle and Ingrassia , 2017 ) . The few TPC studies that have considered biotic interactions have almost all focused on predator-prey interactions ( Ockendon et al . , 2014; Gibert et al . , 2016 ) . To the best of our knowledge , no study has differentiated between fundamental and realized TPCs , nor experimentally quantified a species’ realized TPC in the context of interspecific competition , despite the fact that ecologists have long acknowledged that interspecific competition is a key driving force shaping a species’ realized niche ( Schoener , 1983; Eurich et al . , 2018; Freeman et al . , 2019 ) . Furthermore , although a few studies have shown that intraspecific cooperation can also help social species expand their realized niche width ( Sun et al . , 2014; Lin et al . , 2019 ) , little is known about how intraspecific cooperation influences the realized TPCs of social organisms . Burying beetles ( Silphidae , Nicrophorus ) are ideal for investigating how social interactions influence both fundamental and realized TPCs because the potentially antagonistic effects of interspecific competition and intraspecific cooperation on the realized TPC can be studied simultaneously . Burying beetles rely on vertebrate carcasses for reproduction and often face intense intra- and interspecific competition for using these limiting resources ( Pukowski , 1933; Scott , 1998; Rozen et al . , 2008 ) . Competition for carcass access is temperature-dependent because the beetle’s main competitors , blowflies ( family Calliphoridae ) , are more abundant and active at higher ambient temperatures ( Sun et al . , 2014 ) . Blowfly maggots also grow faster and digest carcasses more quickly at higher temperatures ( Donovan et al . , 2006; Kotzé et al . , 2016 ) . In addition to interspecific competition , intraspecific cooperative behavior among beetles may also modulate their realized TPC . Previous studies have shown that cooperative carcass preparation and burial enables beetles to outcompete blowflies and expand their thermal niche to warmer environments ( Sun et al . , 2014; Chan et al . , 2019; Chen et al . , 2020; Liu et al . , 2020 ) . Here , we extend classic ecological niche theory by introducing the concepts of fundamental and realized TPCs . We first construct a theoretical model by using a hypothetical TPC to predict how interspecific competition influences the width and optimal temperature of realized TPCs in order to provide a general understanding of the relationship between fundamental and realized TPCs . We then describe a series of laboratory and field experiments designed to test the predicted relationship between fundamental and realized TPCs in the Asian burying beetle Nicrophorus nepalensis ( Figure 1 ) . We began our empirical work by measuring breeding performance without interspecific competitors in the laboratory and field to determine N . nepalensis’s fundamental TPC . We also measured the beetle’s locomotor ability , which is less likely to be influenced by biotic interactions , at different temperatures to determine the physiological basis of the fundamental breeding TPCs . We then quantified breeding performance in the presence of interspecific competitors , mainly blowflies ( Putman , 1978; Putman , 1983; Scott , 1994; Sun et al . , 2014 ) , to determine the beetle’s realized TPC . Finally , we used a group size manipulation in the presence of interspecific competitors in the field to explicitly examine the role of intraspecific cooperation and interspecific competition on the beetle’s realized TPC . Experimentally distinguishing between fundamental and realized TPCs will not only serve as a starting point for better understanding the relationships between abiotic and biotic drivers of organismal performance and fitness , but also for better predicting responses to climate change as the earth continues to warm . Lab experiments were conducted using N . nepalensis individuals from a laboratory-reared population . Our stable lab population was established in 2014 from 24 male and 24 female beetles caught near Meifeng , which is 2100 m above sea level on Mt . Hehuan , Taiwan ( 24°5' N , 121°10'E ) . Since establishment , we have supplemented the lab strain with new individuals from the same location every one or two years to avoid inbreeding . We used hanging pitfall traps baited with rotting pork ( mean ± SE: 100 ± 10 g ) to collect adult beetles in the field . We checked traps and collected beetles on the fourth day after traps were set . In every generation , we established at least 20 families , approximately 600 individuals in total , to maintain the population within the lab . To ensure that beetles in the lab population were unrelated to each other , we always paired beetles collected from different traps . We then put one female and one male in a 20 × 13 × 13 cm box with 10 cm of soil and a rat carcass ( 75 ± 7 . 5 g ) . Approximately two weeks after introducing adult beetles , all of the dispersing larvae that were ready to pupate from each breeding box were collected and allocated to a small , individual pupation box . After roughly 45 days , beetles that emerged from pupae were housed individually in 320 ml transparent plastic cups and fed once a week with superworms ( Zophobas morio ) . All breeding experiments were conducted in walk-in growth chambers that imitated natural conditions where the lab population was collected at 2100 m on Mt . Hehuan . Temperature was set to daily cycles between 19°C at noon and 13°C at midnight , and relative humidity was set to 83–100% . We completed all of the laboratory experiments within three generations . To investigate breeding TPCs , we conducted solitary pairing experiments in six temperature conditions—8 , 10 , 12 , 16 , 20°C and 22°C—in a common garden with no temperature variation in the lab . For each replicate , one male and one female were arbitrarily chosen from different nests to avoid inbreeding . We chose adult beetles that were sexually mature , roughly 2 to 3 weeks after their emergence . Each individual was weighed to the nearest 0 . 1 mg . We then placed the pair with a mouse carcass ( 75 ± 7 . 5 g ) under each temperature condition in a transparent plastic container ( 21 × 13 × 13 cm with 10 cm of soil depth ) for two weeks . Cases in which pairs fully buried the carcass and produced offspring were regarded as successful breeding attempts . Cases in which pairs failed to bury a carcass , or they buried it but did not produce offspring , were regarded as failed breeding attempts . The only instance ( of 118 replicates ) when beetles died during the experiment was excluded from analysis . To determine the TPC for locomotor behaviors , we conducted a series of treadmill experiments under three temperature conditions—12 , 16°C and 20°C—in a common garden with no temperature variation in the lab . We chose these temperatures because 16°C is the optimal performance temperature for reproduction , and we wanted to further test whether this optimal temperature coincided with optimal physiological function . We set 72 replicates in total ( 12°C: 25 replicates; 16°C: 25 replicates; 20°C: 22 replicates ) . We arbitrarily selected individuals from different nests for replicates at each temperature . The beetles were brought to the experimental chamber one day before data collection began . Monofilament glued to the pronotum by UV glue attached each beetle to the treadmill , where it was allowed to walk at a stable speed of 150 cm/min . We turned off the treadmill if a beetle’s abdomen began to drag or if the beetle started to fly , both behaviors that indicated that the beetle could no longer walk . An individual was tested only once per day . After each experiment , beetles were returned to the transparent container with 3 cm of soil for recovery . We measured each beetle’s pronotum and the ambient temperature during running with a thermal imaging infrared camera ( FLIR Systems , Inc , SC305; thermal sensitivity of <0 . 05°C ) at a resolution of 320*240 pixels . Pronotum temperature was measured at the center of the thorax and calculated as the average pronotum body temperature each minute until an individual dragged its abdomen or started flying . The ambient temperature was the average temperature of a 6 × 6 cm surface of the treadmill located near where the beetle was tested . The temperature difference was depicted by the difference between the beetle’s body and ambient temperatures . Since our previous study showed that blowflies are the beetle’s main interspecific competitor ( Sun et al . , 2014 ) , we conducted series of experiments in the field to investigate the breeding TPC with and without interspecific competition . In 2013 to 2016 ( May-October ) , we investigated the natural pattern of N . nepalensis reproduction and its breeding success along an elevational gradient from 673 m to 3422 m on Mt . Hehuan in central Taiwan ( 24°11’ N , 121°17’ E ) that encompasses broadleaf forests at lower elevations and mixed conifer-broadleaf forests at higher elevations . We chose 37 study sites , primarily in natural forests to avoid cultivated or open areas where temperatures are more variable ( De Frenne et al . , 2019 ) and replicated each treatment at least three times at each site . In each trial , a 75 g ( ± 7 . 5 g ) rat carcass was placed on the soil to attract beetles and covered with a 21 × 21 × 21 cm ( length x width x height ) iron cage with 2 × 2 cm mesh to prevent vertebrate scavengers from accessing the carcass . We checked each carcass daily until it began to decay due to microbial activity ( Payne , 1965 ) , was consumed by maggots or other insects , or was buried under the soil by beetles . If burying beetles completely buried the carcass , we checked the experiment after 14 days to determine if third-instar larvae appeared . Cases in which pairs produced third-instar larvae were regarded as successful breeding attempts . Cases in which pairs failed to produce larvae were regarded as failed breeding attempts . Breeding experiments without blowflies were conducted in the same experimental sites from 2014 to 2017 , and 2019 ( May-October ) . The experimental design was the same as that described above , but we used screen mesh above the pots to also keep blowflies out . To record air temperature at every site , we placed iButton devices approximately 120 cm above the ground within a T-shaped PVC pipe to prevent direct exposure to the sun but allow for air to circulate . One male and one female beetle that were reared in the lab were released into the pot to record fundamental breeding performance . After 14 days , we checked the pots to determine whether the burying beetles’ third-instar larvae appear . Cases in which pairs fully buried the carcass and produced larvae after 14 days were regarded as successful breeding attempts . Cases in which pairs failed to bury a carcass , or they buried it but did not produce larvae , were regarded as failed breeding attempts . Criteria of data exclusion were the same as the common garden experiments . Three instances ( of 178 replicates ) when beetles died during the experiment were excluded from analysis . To investigate how cooperative behavior influences TPCs , we manipulated the group size of beetles in the field at 38 sites along the elevational gradient . Our experimental device comprised a small plastic container ( 21 × 13 × 13 cm with 10 cm of soil ) placed inside a large container ( 41 × 31 × 21 . 5 cm with 11 cm of soil ) . There were several holes on the small container’s side wall that allowed beetles to move freely between the two containers . A 2 × 2 cm iron mesh was placed around the top of the large container’s wall to let flies access the carcass but to keep out larger animals that might scavenge the carcass . Small , non-cooperative groups contained one male and one female , whereas large , cooperative groups contained three males and three females ( Chen et al . , 2020; Liu et al . , 2020 ) . We captured the local beetles using the same hanging pitfall traps described above , and then conducted two group size treatments at each site . Based on our previous work exploring the natural pattern of arrival times , we released the marked beetles into the experimental device 1 , 2 and 3 days after the trials began at elevations of 1700–2000 m ( low ) , 2000–2400 m ( intermediate ) and 2400–2800 m ( high ) , respectively ( for details , see Sun et al . , 2014 ) . Each experiment was recorded by a digital video recorder ( DVR ) to determine whether N . nepalensis successfully buried the carcass . We placed the same temperature measurement device as described above at every site . Cases in which beetles buried the carcass completely and produced larvae after 14 days after were regarded as successful breeding attempts . Cases in which beetles failed to produce larvae were regarded as failed breeding attempts . We used generalized linear mixed models ( GLMMs ) with binomial error structure to compare thermal performance curves among treatments ( with/without interspecific competitors; with/without intraspecific cooperation ) in the field . The outcome of breeding success ( 1 = success , 0 = failure ) was fitted as a binomial response term to test for differences in the probability of breeding successfully . The variables of interest ( i . e . mean daily temperature , type of experimental treatment ) were fitted as fixed factors . Other environmental factors ( elevation , daily minimum air temperature ) were fitted to test the generality of the results . However , since elevation , daily minimum air temperature , and mean air temperature were highly correlated , we only included mean daily temperature in the final model . We also modeled the potential nonlinear effects of the environmental factors by fitting a quadratic regression model and compared the model fit with the linear model . Thus , the thermal performance curves ( TPC ) were determined statistically by the GLMM . To account for repeated sampling in the same plot , we set the field plot ID as a random factor ( coded as 1\|plot ID ) in the R package lme4 ( Bates et al . , 2014 ) . We also included year as a random factor to account for sampling at different time points ( See Source code 1 for further details ) . We used a general linear models ( GLMs ) to determine N . nepalensis's breeding rate and locomotor performance in the lab . The outcome ( 1 = success , 0 = failure ) was fitted as a binomial response term to test the difference in the probabilities of interest ( burial or non-burial ) under different temperatures . For locomotor performance , the outcome ( 1 = flying , 0 = not flying ) was fitted as a binomial response term to test for a difference in the probability of interest ( flying or not ) under different temperatures conditions . The relationship of temperature difference ( body temperature minus ambient temperature ) was fitted as a Gaussian response to different temperature conditions . ( See Source code 1 for further details ) . Finally , the optimal temperature ( Toptimal ) of the TPC was calculated by taking the derivative of the regression line that described the relationship between temperature and the likelihood of breeding successfully . In other words , Toptimal was estimated from the unimodal statistical model of the TPC . To estimate TPC breadth , we calculated the 95% confidence interval of the regression line . The boundaries of the TPC were the points that there was not significant difference between regression lines and zero . All statistical analyses were performed in the R v3 . 0 . 2 statistical software package ( R Development Core Team , 2018 ) . ( See Source code 1 for further details ) . We employed the following commonly-used thermal performance curve ( TPC ) function ( Deutsch et al . , 2008; Vasseur et al . , 2014 ) to describe the fundamental TPCs of two competing species with strategies similar to those in our empirical system of beetles and blowflies . A low-temperature thermal specialist species resembles the burying beetle’s breeding thermal performance ( Tsai et al . , 2020 ) , whereas a high temperature generalist species that has a similar life history to the blowfly ( Figure 1a ) : ( 1 ) P ( T ) ={exp⁡ ( − ( T−Topt ) /2σp ) 2 ) , when T≤Topt1− ( ( T−Topt ) / ( Topt−Tmax ) ) 2 , when T>Topt}where T is environmental temperature , σp is the shape parameter describing the steepness of the curve at the lower end , Topt is the optimal environmental temperature at which organisms have their highest performance , and Tmax is the upper critical temperature . We assumed that performance becomes zero when T > Tmax . Since environmental conditions also directly influence a species’ average performance , we used a Gaussian function to describe the chance of encountering a particular temperature: ( 2 ) f ( T\|Tmean ) = ( 2πV ) −0 . 5exp ( − ( Ts−Tm ) 2/2V ) where fT\|Tmean represents the probability of getting T given Tmean , Tmean represents the mean environmental temperature , and V describes the environmental temperature variability . We combined equations ( 1 ) and ( 2 ) to obtain the temperature-weighted performance function: ( 3 ) w ( T ) =∫−∞+∞[P ( T ) f ( T\|Tmean ) ]dtby integrating the product of performance and probability of the temperature across all environmental temperatures . We then used the relative performance of the two species ( i . e . wS ( T ) /wG ( T ) and wG ( T ) /wS ( T ) ) to represent the realized TPCs of specialist and generalist species , respectively . We began by addressing how interspecific competition influences the realized TPC of a focal species , finding that when a low temperature specialist species ( e . g . burying beetle ) competes with a high temperature generalist species ( e . g . blowfly ) , the optimal temperature of the realized TPC of the thermal specialist shifts towards a lower temperature and the width of the TPC decreases ( Figure 2b ) . In other words , our model predicts that the optimal temperature of the realized TPC will decrease to below that of the optimal of fundamental TPC when a low temperature specialist competes with a high temperature generalist . To make the theoretical framework complete , we also explored the scenario of a high temperature specialist competing with a low temperature generalist . We found that if a high temperature specialist species competes with the low temperature generalist species ( Figure 1—figure supplement 1a ) , the optimal temperature of the realized TPC of the thermal specialist shifts towards a higher temperature and the width of the realized TPC decreases ( Figure 1—figure supplement 1b ) , which suggests that a shift in the optimal realized TPC away from the optimal temperature of the competing species is a general result . ( See Source code 2 for further details ) . We first explored N . nepalensis’s fundamental breeding TPC in a controlled lab environment . We found that the probability N . nepalensis breeding successfully changed unimodally with ambient temperature ( Figure 3a , GLM , χ²2 = 26 . 29 , p < 0 . 001 , n = 117 ) . Accordingly , we found that the beetle’s optimal breeding temperature or fundamental TPC—defined as the mean temperature at which breeding success was highest ( calculated from our GLM ) —was 15 . 6°C . To determine the physiological basis of this optimal breeding TPC , we measured locomotion ability at different temperatures by performing a treadmill running experiment . We found that beetles had a greater likelihood of flying at 16°C while running at a stable speed on the treadmill ( Figure 3b , GLM , χ²2 = 13 . 22 , p = 0 . 001 , n = 72 ) . In other words , N . nepalensis took less energy to raise its body temperature enough to begin flying at 16°C than at other temperatures ( Figure 3c and d , GLM , χ²2 = 23 . 18 , p < 0 . 001 , n = 29 ) . Next , we investigated N . nepalensis’s realized and fundamental breeding TPCs by studying breeding performance along an elevational gradient ( 1600 to 2800 m above sea level ) . As predicted by our model , in the presence of interspecific competitors ( blowflies ) in the wild , the optimal breeding temperature ( i . e . the realized TPC ) of N . nepalensis was roughly 13 . 1°C , which is lower than the optimal temperature in the lab in the absence of blowflies ( i . e . the fundamental TPC ) ( Figure 4a , GLMM , χ²2 = 16 . 56 , p < 0 . 001 , n = 343 ) . Intriguingly , when excluding blowflies and removing the threat of intraspecific competition in the field , the optimal breeding temperature of N . nepalensis increased to approximately 14 . 6°C , such that the realized TPC began to approach the fundamental TPC in the absence of blowflies , ultimately becoming broader than when in the presence of interspecific competitors ( Figure 4b , GLMM , χ²2 = 16 . 08 , p < 0 . 001 , n = 175 ) . Thus , our experiment confirmed the causal relationship between interspecific competition and the shift in the realized TPC under natural conditions . Since our previous study found that N . nepalensis will cooperate at carcasses to compete against blowflies , particularly in warmer environments ( Sun et al . , 2014 ) , we predicted that a group of N . nepalensis in a warm environment would have a better chance of expanding its realized TPC towards the fundamental TPC than would an individual pair . To test this prediction , we performed a group size manipulation to determine the realized TPCs of cooperative groups and solitary beetles . We found that N . nepalensis in cooperative groups had an optimal breeding temperature ( i . e . realized TPC ) of 15 . 6°C , which is identical to their optimal temperature from the fundamental TPC in the lab . In contrast , the optimal breeding temperature of solitary pairs was 14 . 1°C , similar to that of the realized TPC of 14 . 6°C in the field ( Figure 5 , group size × temperature interaction , χ22 = 6 . 84 , p = 0 . 033 , n = 328; for large groups , χ22 = 9 . 40 , p = 0 . 009 , n = 162; for small groups , χ22 = 18 . 42 , p < 0 . 001 , n = 166 ) . These results suggest that beetles that cooperate are able to expand their realized TPCs such that they converge on their fundamental TPCs , whereas those do not cooperate have divergent realized and fundamental TPCs . By combining the concepts of fundamental and realized niches from ecological niche theory with the that of the thermal performance curve ( TPC ) , we found that a species’ realized thermal performance curve is likely to change in time and space in response to biotic factors such as interspecific competition ( see Figure 1—figure supplement 2 for the summary ) . Our theoretical model suggests that if thermal specialist species compete with thermal generalist species adapted to higher or lower temperatures , the optimal performance temperature of specialists will decrease or increase , respectively . Our empirical results examining competition between burying beetles ( thermal specialists ) and blowflies ( thermal generalists ) for access to carcasses support this theoretical prediction , finding that blowflies force the beetle’s optimal breeding temperature lower and the realized TPC narrower . Intriguingly , our experiment also showed that intraspecific cooperation in this facultatively social species not only enables beetles to overcome interspecific competition ( Sun et al . , 2014; Shen et al . , 2017; Lin et al . , 2019 ) , but to better match their fundamental and realized TPCs ( from 14 . 1°C to 15 . 6°C ) . Therefore , the mechanism that enables cooperative beetles to expand their range to lower elevations relative to non-cooperative beetles ( Sun et al . , 2014; Liu et al . , 2020 ) appears to be their ability to align their fundamental and realized thermal niches . The idea that interspecific competition will reduce the realized niche width of a species is well-accepted in ecology . However , our study further suggests that a more mechanistic understanding of how interspecific competitors affect the optimal temperature performance of species will be critical for understanding how climate change affects species’ vulnerability . Since it is generally assumed in studies of macroecology and climate change that thermal performance is largely influenced by physiology , a single function is often used to describe a species’ thermal performance curve ( Sinclair et al . , 2016 ) . However , if biotic interactions are key to indirectly influencing the thermal performance of a species , as we have shown here , the realized TPC of a species is likely to change in time and space and should not be described by a single function to represent the thermal performance of a species . Integrating the idea of TPCs into the ecological niche concept helps bridge two rich , but largely independent , traditions of studying thermal adaptation . By simply recognizing the concept of realized TPCs , it becomes clear that we know little about how realized and fundamental TPCs differ in most species . We show that the realized TPC provides a way to quantify how temperature mediates species interactions , which also influence organismal fitness . Thus , the realized TPC extends the realized thermal niche concept , which only considers the temperature ranges in which a species can occupy or breed successfully ( Huff et al . , 2005; Rehfeldt et al . , 2008 ) by considering how abiotic factors ( i . e . climate ) additionally indirectly affect fitness by driving biotic interactions like species competition . For example , a greater population size can facilitate intraspecific cooperation because there are more individuals to form groups quickly to outcompete interspecific competitors , as we found in burying beetles at lower elevations . We predict that Allee effects—when higher population density has a positive effect on individual fitness and population growth rate until it reaches the maximum ( Allee , 1931; Courchamp et al . , 1999; Stephens and Sutherland , 1999 ) —will likely occur in N . nepalensis in warmer environments . Therefore , conserving high population densities , especially at lower elevations , will be crucial for N . nepalensis to compete against blowflies under increased climate warming . By examining the relationship between cooperative behavior and interspecific competition , our study thus helps understand the pressing issue of how habitat destruction affects the vulnerability of social organisms to climate change ( Travis , 2003 ) . When population density influences the likelihood of intraspecific cooperation in social species , habitat destruction will not only decrease habitat availability but also weaken a species’ competitive ability against interspecific competitors , which in turn will lower the realized thermal performance of social organisms . Our study has implications beyond interspecific competition in insects . Many classic studies of TPCs investigate how changes in body temperature influence physiological or behavioral performance ( Chen et al . , 2003; Zhang and Ji , 2004; Huey et al . , 2012 ) . Body temperature is often assumed to be the same as the environmental temperature in ectotherms . However , accumulating evidence suggests that many ectotherms can at least partially regulate their own body temperature behaviorally or physiologically ( Heinrich , 1993; Tattersall et al . , 2016; Clarke , 2017 ) . Thus , to identify fundamental TPCs , it is crucial to perform both lab and field experiments in order to understand the exact relationship between body and ambient temperature , as well as how realistic environmental conditions ( e . g . temperature variation , humidity , or precipitation ) influence the fundamental TPC . For example , TPCs are often considered to be left-skewed for physiological reasons ( e . g . the response of thermoregulation proceeds faster at higher temperatures ) ( Woodin et al . , 2013; Huey and Pianka , 2018 ) , but our results and those of many other studies have shown that TPCs can be diverse in their shape ( Dell et al . , 2011; Monaco et al . , 2017 ) . Although we show that variation in the shape of TPCs can be due to interspecific competition , other types of biotic interactions , including mutualistic and host-parasites interactions ( Cohen et al . , 2017 ) , should also be carefully considered and compared when quantifying fundamental TPCs . Ultimately , separating and experimentally quantifying fundamental and realized thermal performance in the field and lab will be critical for understanding how both biotic and abiotic factors interact to influence organismal fitness , particularly in an era of rapid climate change . The concept of the TPC has received renewed interest because the earth has been warming rapidly for the past few decades . Yet , apparent gaps exist between studies of physiological function and those examining fitness consequences in changing environments . Our study shows that employing the concepts of fundamental and realized TPCs can help us predict the ecological impacts of climate change , especially because environmental change will likely reshuffle ecological communities and alter the strength of species interaction ( Alexander et al . , 2016 ) . The importance of biotic interactions in shaping species distributions and community composition is intuitively obvious , yet historically has been difficult to quantify . We believe that the concept of realized TPCs can help fill this important knowledge gap and , ultimately , deepen our understanding of the ecological impact of climate change .	Insects , reptiles and many other animals are often referred to as being ‘cold-blooded’ because , unlike mammals and birds , their body temperature fluctuates with the temperature of their surrounding environment . As a result , many cold-blooded animals are very sensitive to changes in local climate . Environmental factors , such as temperature and precipitation , as well biotic factors , such as two species competing for food or the presence of a predator , may influence how well an animal performs at different temperatures . However , few studies have examined how both environmental and biotic factors affect the range of temperatures in which a cold-blooded animal is able to survive and reproduce . When Asian burying beetles reproduce , they lay their eggs around buried animal carcasses that can provide food for their offspring . Previous studies have found that individual burying beetles can cooperate with each other to defend themselves against their main competitor , blowflies , which also lay their eggs on animal carcasses . Here , Tsai et al . used mathematical and experimental approaches to study how blowflies affect the range of temperatures in which burying beetles are able to live under different environmental conditions . The experiments showed that when blowflies were present , the range of temperatures that burying beetles were able to survive and reproduce in was smaller . Furthermore , the optimal temperature for the burying beetles to live in shifted back , away from that of their competitor . Larger groups of burying beetles were able to survive and reproduce in a greater range of temperatures than smaller groups , even when blowflies were present . This suggests that increasing the amount bury beetles cooperate with each other may make them more resilient to changes in temperature . The Earth is currently experiencing a period of climate change and therefore it is important to understand how different species of animals may respond to to changing temperatures . These findings reinforce the idea that even a small change in temperature may lead to changes in how different species interact with each other , which in turn influences the ecosystem in which they live .	[ "Abstract", "Introduction", "Materials", "and", "methods", "Results", "Discussion" ]	[ "ecology" ]	2020	Antagonistic effects of intraspecific cooperation and interspecific competition on thermal performance
Hygrosensation is an essential sensory modality that is used to find sources of moisture . Hygroreception allows animals to avoid desiccation , an existential threat that is increasing with climate change . Humidity response , however , remains poorly understood . Here we find that humidity-detecting sensilla in the Drosophila antenna express and rely on a small protein , Obp59a . Mutants lacking this protein are defective in three hygrosensory behaviors , one operating over seconds , one over minutes , and one over hours . Remarkably , loss of Obp59a and humidity response leads to an increase in desiccation resistance . Obp59a is an exceptionally well-conserved , highly localized , and abundantly expressed member of a large family of secreted proteins . Antennal Obps have long been believed to transport hydrophobic odorants , and a role in hygroreception was unexpected . The results enhance our understanding of hygroreception , Obp function , and desiccation resistance , a process that is critical to insect survival . Hygroreception is a critical sensory modality in the animal world ( Altner and Loftus , 1985; Filingeri , 2015; Okal et al . , 2013; Sayeed and Benzer , 1996; Shelford , 1918; von Arx et al . , 2012 ) . Mosquitoes , for example , use hygroreception to find humans on which to feed , and to find water sources on which to lay eggs ( Okal et al . , 2013; Takken , 1991 ) . Hygroreception helps animals avoid desiccation , a peril that is increasing due to climate change . Small insects , which have a high ratio of surface area to volume , are especially vulnerable to water loss ( Gibbs et al . , 2003 ) . The ability to sense humidity levels may allow an insect to avoid dangerously dry conditions or to initiate physiological changes that protect it against desiccation ( Stinziano et al . , 2015 ) . The antenna functions as a humidity detector in many insects ( Altner and Loftus , 1985; Tichy , 1987; Yokohari , 1978 ) . In the antenna of Drosophila , humidity detection occurs largely in a three-chambered cavity called the sacculus . The second chamber of the sacculus contains a small number of sensilla that act as hygroreceptors and thermoreceptors ( Enjin et al . , 2016; Frank et al . , 2017; Kim and Wang , 2016; Knecht et al . , 2017; Knecht et al . , 2016; Shanbhag et al . , 1995; Silbering et al . , 2011 ) . These sensilla belong to a morphological class known as coeloconic sensilla , which are small relative to other sensilla ( Shanbhag et al . , 1995 ) . The molecular basis of hygroreception remains enigmatic . A major advance was recently made through the discovery that four ionotropic receptors expressed in the sacculus ( IR93a , IR25a , IR68a , and IR40a ) are required for hygrosensation ( Enjin et al . , 2016; Frank et al . , 2017; Kim and Wang , 2016; Knecht et al . , 2017; Knecht et al . , 2016 ) . However , the precise role of these receptors in hygroreception remains unclear . Moreover , the downstream effects of hygrosensory signaling remain poorly understood . Also enigmatic has been a family of small secreted proteins called Odorant binding proteins ( Obps ) . These proteins are remarkably numerous , extremely abundant , and highly divergent in sequence ( Graham and Davies , 2002; Hekmat-Scafe et al . , 2002; Menuz et al . , 2014; Vogt et al . , 1989 ) . There are 52 Obp genes in Drosophila , of which 27 were found expressed in the antenna in a recent RNAseq analysis ( Larter et al . , 2016; Menuz et al . , 2014; Younus et al . , 2014 ) . Five of the 10 most abundantly expressed genes in the antenna are Obps . Although Obps are widely believed to carry odorants to odor receptors in olfactory sensilla ( Leal , 2013; Leal et al . , 2005 ) , there is limited in vivo evidence to support this role ( Leal , 2013; Pelosi et al . , 2006; Vogt and Riddiford , 1981 ) , and a recent genetic study found that a mutant olfactory sensillum lacking abundant Obps did not show a decreased magnitude of response to a variety of odorants ( Larter et al . , 2016 ) . One highly abundant member of the Obp family , Obp59a , is striking in two respects . First , it is exceptional in its high degree of sequence conservation among insects . Unlike nearly all other Drosophila Obps , it has clear orthologs in a variety of insect orders examined ( Vieira and Rozas , 2011; Zhou et al . , 2010 ) . Second , it is the most highly localized of the abundant antennal Obps: its expression is restricted to the sacculus ( Larter et al . , 2016 ) . Here we show that Obp59a is expressed in the same sensilla as the IRs that are essential to hygroreception . We generate Obp59a mutants and find that they are defective in three distinct hygrosensory behavioral paradigms: one operating over the course of seconds , one over minutes , and one over hours . Finally , we show that Obp59a mutants survive desiccation better than controls . The results , taken together , add a new dimension to our understanding of hygroreception , of Obp function , and of a process that is critical to insect life and will become even more critical as climate change progresses . We mapped Obp59a and examined its specificity of expression , initially by two means . First , detailed in situ hybridization analysis showed that Obp59a expression is restricted to the second chamber of the antennal sacculus ( Figure 1A ) . Second , we generated an Obp59a-GAL4 construct , and used it to drive GFP . The expression pattern in the antenna was again restricted to the second chamber of the sacculus ( Figure 1B ) . The labeling produced by the Obp59a probe and by the Obp59a-GAL4 driver coincided ( Figure 1C ) . We found no expression of Obp59a-GAL4 elsewhere in the fly head or body , or in any of the three larval instars . These results are consistent with data from the Flybase High Throughput Expression Pattern Database , which revealed no expression of Obp59a RNA in tissues or developmental stages other than the adult head , which presumably included antennae ( Gelbart and Emmert , 2013 ) . We wondered whether the specificity of Obp59a expression is conserved in other insects . We examined the antenna of the tsetse fly Glossina morsitans morsitans , which diverged from Drosophila melanogaster ~75 million years ago ( Wiegmann et al . , 2011 ) and which carries African sleeping sickness . We found that Obp59a again mapped to the sacculus ( Figure 1D , E ) . We next asked whether Obp59a maps to hygrosensitive sensilla in the second chamber of the sacculus . We carried out a double-label analysis , using an Obp59a probe and five IR-GAL4 constructs that drive expression in the sacculus , four of which label hygrosensitive sensilla in the second chamber and one of which does not . IR93a-GAL4 can be seen to label a hygrosensitive neuron in the second chamber ( Figure 2 , top panel of left column , green; the arrowhead indicates the dendrite ) . Obp59a labels cells immediately adjacent to this neuron ( Figure 2 , center and bottom panels , left column ) . Likewise , IR25a-GAL4 , IR40a-GAL4 , and IR68a-GAL4 all label hygrosensitive neurons of the second chamber , and in each case , Obp59a labels adjacent cells . We did not observe axons or dendrites in any of the cells labeled by Obp59a , consistent with its expression in non-neuronal cells of the sensilla , as expected of an Obp . As a negative control , we examined IR8a-GAL4 , which does not label hygrosensitive neurons of the second chamber . The neurons it labels are not immediately adjacent to cells labeled by Obp59a ( Figure 2—figure supplement 1 ) . Having shown that Obp59a RNA maps to hygrosensitive sensilla in the second chamber of the sacculus , we wanted next to localize Obp59a protein within these sensilla . We generated an anti-Obp59a antibody and found that the antibody labels the second chamber of the sacculus ( Figure 3A ) , consistent with that of Obp59a RNA ( Figure 3B , C ) . We further validated the antibody by generating an Obp59a mutant and asking whether immunolabeling was lost . We deleted the entire Obp59a coding region using the CRISPR-Cas9 system ( Supplementary file 1 ) . When the anti-Obp59a antibody was tested against the antenna of the Obp59a1 deletion mutant , no labeling was observed in the sacculus or anywhere else ( Figure 3D ) . These results indicate that the anti-Obp59a antibody specifically labels the Obp59a protein . We note moreover that we examined the structure of the sacculus by confocal microscopy and observed no gross morphological defects in the Obp59a mutant ( Figure 3—figure supplement 1 ) . We cannot exclude the possibility of subtle morphological defects in the sacculus or its sensilla . We then examined the antibody labeling at higher resolution . We were especially interested in whether the protein was secreted into the shaft of the sensillum ( Figure 3E ) , where dendrites reside ( Shanbhag et al . , 1999 ) . In addition to labeling antennal sections ( Figure 3F ) with the antibody ( Figure 3G ) , we co-labeled with the Obp59a RNA probe to identify the non-neuronal cells that synthesize Obp59a ( Figure 3H ) . The merged images show that Obp59a is in fact found within the shafts of the sensilla , as expected of a protein that is secreted by auxiliary cells of a sensillum into the dendritic lymph ( Figure 3I , J ) ( Pelosi and Maida , 1995 ) . We note that Obp59a contains a signal sequence , consistent with secretion of the protein into the shaft . We asked whether Obp59a is required for response to humidity in three different behavioral paradigms . Prior to testing , mutants carrying the Obp59a mutation were backcrossed to the control stock for five generations to minimize genetic background effects . First we tested hygrotaxis in a paradigm that operates on a time scale of minutes ( Ji and Zhu , 2015 ) . Flies are placed in a Petri dish that contains an inner region held at high humidity and an outer region held at low humidity ( Figure 4A ) . Humidity was controlled through the use of saturated salt solutions and was verified with a hygrometer ( Enjin et al . , 2016 ) . Initially , flies were distributed uniformly on the plate , and their distributions were then measured at 10 s intervals over the course of 5 min . A hygrotaxis index was calculated at each time point as the fraction of flies in the central region of high humidity , following established convention ( Ji and Zhu , 2015 ) . Since the area of high humidity was ~1/10 that of the total area , and the distribution of the flies was initially uniform , the initial index was ~0 . 1 . When the inner region was at 70% relative humidity ( RH ) and the outer region was at 20% RH , the control flies showed a rapid hygrotaxis behavior: control flies quickly began to move inward toward the more humid region ( Figure 4B ) . By contrast , the distribution of Obp59a mutants showed little if any change . A striking phenotype was also observed when the inner and outer RH were 96% and 20% , respectively ( Figure 4C ) , and 96% v . 70% ( Figure 4D ) . When inner and outer RH were equal ( and set to 45% ) , neither genotype showed any place preference ( Figure 4E ) . We note that the hygrotaxis responses were very rapid in these experiments , which provides an explanation for why the values at the first time points , taken ~15 s after the flies were distributed uniformly in the dish , were generally somewhat greater than 0 . 1 . We carried out a rescue experiment to determine whether the phenotype in fact mapped to the Obp59a gene . We found that when inner and outer RH were 70% and 20% , mutant flies carrying both an Obp59a-GAL4 construct and a UAS-Obp59a construct showed stronger responses than mutant flies carrying either construct alone ( Figure 4F ) . Rescue was also observed in the 70% v . 20% and 96% v . 20% cases ( Figure 4G , H ) . All genotypes showed no place preference when inner and outer RH were both set to 45% ( Figure 4I ) . Similar hygrotaxis phenotypes were observed when flies were desiccated prior to the test ( Figure 4—figure supplement 1A–D ) . We note that the desiccated Obp59a mutants appeared to gravitate toward the more humid region near the end of the test period ( e . g . Figure 4—figure supplement 1A ) . This response may arise from extreme thirst and perhaps other pathways for humidity detection ( Liu et al . , 2007 ) . We note that the behavior of control flies that were desiccated was not dramatically different from those that were not ( Figure 4 vs . Figure 4—figure supplement 1 ) . An earlier study showed using a different paradigm that hygrotaxis behavior can be altered by desiccation ( Knecht et al . , 2017 ) ; perhaps we have not observed a major alteration because of the different geometry , larger size , and shorter duration of the paradigm shown in Figure 4 , or because of differences in the desiccation procedure we used , which for example did not provide a sucrose source to flies . As further confirmation that the Obp59a gene is required for normal hygrotaxis behavior , we tested an independent allele , Obp59a2 , which was also backcrossed five times to the control strain . We again found a strong defect in all three cases: 70% v . 20% , 96% v . 20% , and 96% v . 70% RHs ( Figure 4—figure supplement 2A–D ) . As another control , we tested a mutant of a related gene , Obp28a , which was generated in the same manner as the Obp59a mutants , and found no defects in any test in this paradigm ( Figure 4—figure supplement 2E–H ) . We further tested Obp59a2 and Obp28a following desiccation and again found a phenotype for Obp59a2 but not Obp28a ( Figure 4—figure supplement 2I–P ) . These results provide additional evidence that the humidity response phenotype maps to the Obp59a gene . Next we tested Obp59a1 in a second paradigm in which we measured humidity preference over the course of hours ( Figure 5A ) ( Knecht et al . , 2016 ) . Flies were given a binary choice between higher and lower regions of humidity , and a ‘wet preference’ was calculated at one hour intervals as ( H-L ) / ( H + L ) , where H is the number of flies in the region of high humidity and L is the number in the region of low humidity . Thus , the wet preference may vary between 1 . 0 ( complete preference for high humidity ) to −1 . 0 ( complete preference for low humidity ) . We first gave flies a choice between 70% and 20% RH and found that by the first time point ( 5 min ) , control flies showed a strong preference for high humidity; this preference continued for 24 hr ( Figure 5B ) . The response of Obp59a1 was much lower at this first time point , and remained lower over the course of many hours . Response of the mutant was also lower when given a choice between 70% and 50% RH , and between 96% and 70% RH ( Figure 5C , D ) . The Obp59a1 mutant retains some limited ability to respond to humidity , as evidenced most clearly by the responses in the first two cases at 24 hr , when flies may be very thirsty ( Figure 5B , C ) . We note finally that although the preference we observe in control flies between 96% and 70% RH is weak , the valence is opposite that found in some other studies ( Enjin et al . , 2016; Knecht et al . , 2017; Perttunen and Salmi , 1956 ) . Knecht et al . , 2017 and Perttunen and Salmi , 1956 have shown that the valence of this preference is dependent on the hydration status of the flies , and it is possible that the different preference we have observed in our experiments reflects a difference in the conditions in which the flies are cultured . We then tested the response of Obp59a1 to humidity in a third paradigm: a modified proboscis extension response ( PER ) test ( Figure 5E ) ( Ji and Zhu , 2015 ) . We measured the response of a desiccated fly to the water vapor emanating from a moistened cotton swab . The response of Obp59a1 was lower than that of the control genotype ( Figure 5F ) . Neither genotype responded to a dry stimulus . Could Obp59a mutants be defective in these paradigms because they have lost the humidity-sensing neurons of the sacculus ? We addressed this question with three independent reagents: an anti-IR93a antibody , an anti-IR25a antibody , and an IR68a-GAL4 driver . All three reagents labeled neurons in the second chamber of the sacculus of Obp59a1 , in a pattern comparable to that observed in controls ( Figure 2—figure supplement 2 ) . Do Obp59a1 mutants show defects in these three behavioral paradigms because of a general deterioration in health or mobility ? We tested the ability of Obp59a1 flies to rapidly climb walls in the Rapid Iterative Negative Geotaxis ( RING ) assay and found that Obp59a1 showed the same robust climbing behavior as wild type ( Figure 5G ) . Obp59a1 also shows robust chemosensory responses to an attractive stimulus and several repellent stimuli in an olfactory paradigm ( Figure 5H ) . The repellent stimuli include acetic acid and propionic acid , which are detected at least in part by sensilla in the third chamber of the sacculus ( Ai et al . , 2013 ) . If Obp59a mutants do not perceive humidity normally , are there consequences for their physiology ? Specifically , we wondered whether mutants might be affected in desiccation resistance . Desiccation is a critical threat to insect survival , and a major function of the humidity detection system is likely to prevent flies from dying of desiccation . Accordingly , we placed Obp59a1 males in a chamber under desiccating conditions ( 0% RH ) and measured survival . Remarkably , Obp59a1 males survived longer than control males ( Figure 6A , p<0 . 0001 ) . Consistent with these results , we found in an independent experiment that Obp59a2 males also survived longer than control males ( Figure 6B , p<0 . 0001 ) . Females survived longer as well , in the case of both Obp59a1 and Obp59a2 ( Figure 6C , D , p<0 . 0001 in both cases ) . Survival of males and females of both alleles was normal at 70% RH ( Figure 6—figure supplement 1A–D ) . As another control , Obp28a mutants showed normal survival , in the case of both males and females and at both 0% and 70% RH ( Figure 6—figure supplement 1E–H ) . One interpretation of these results is that the loss of Obp59a leads to an abnormal pattern of humidity signaling , which triggers defensive physiological changes in the fly that protect it from desiccation . Expression of Obp59a in Drosophila is highly localized to the sacculus . The tsetse ortholog is also expressed in the sacculus , suggesting that its location has been conserved for at least 75 million years . Within the sacculus of Drosophila the protein is found in the shafts of hygrosensory sensilla , where dendrites of hygrosensory neurons are located ( Shanbhag et al . , 1999 ) . Expression levels of Obp59a are remarkably high . Obps are the most abundantly expressed genes in the antenna: in a recent RNAseq analysis , Obp19d and Obp83a were detected at 31 , 000 and 25 , 000 RPKM ( reads per kilobase per million mapped reads ) respectively , while a typical Odor receptor ( Or ) gene was expressed at ~40 RPKM ( Menuz et al . , 2014 ) . Obp59a was expressed at ~2 , 000 RPKM , and showed the most highly restricted expression of the antennal Obp genes ( Larter et al . , 2016 ) : most other Obps are expressed in many more sensilla . These results suggest that the level of Obp59a expression in an individual hygrosensitive sensillum is comparable to that of the most abundantly expressed Obps . Not only is the expression pattern of Obp59a conserved , but its sequence is conserved as well ( Stanley and Kulathinal , 2016; Vieira and Rozas , 2011; Zhou et al . , 2010 ) . Obp59a is one of only two Obps with clear orthologs across insect orders . This conservation suggests that its structure represents a good solution to a difficult problem that is common to many insects . A role for an Obp in humidity detection was unexpected . Antennal Obps are widely believed to transport hydrophobic odorants through the aqueous sensillum lymph to odor receptors in the dendritic membranes of olfactory receptor neurons ( Leal , 2013 ) . An Obp59a mutant was normal in response to both attractant and repellent odorants in behavioral tests . Obp59a seems unlikely to carry water molecules across the aqueous lymph to the dendrites of hygrosensory neurons . One proposed model for the mechanism of hygroreception is that a change in humidity alters the structure of hygrosensory sensilla , with the structural change being transduced into neuronal responses ( Altner and Loftus , 1985 ) . It is conceivable that Obp59a mutants contain a subtle defect in the structure or composition of the sensillum , and therefore do not undergo a normal structural change in response to changes in humidity . Such a role might fit well with the high abundance of Obp59a within hygrosensory sensilla . Obp59a could possibly affect sensillum structure or composition via a role in transporting hydrophobic components of the cuticular wall of the sensillum . The concept of an alternative role for an antennal Obp is consistent with a recent study showing that the classic odorant-transport model may not apply to all antennal Obps and all olfactory sensilla ( Larter et al . , 2016 ) . An Obp-to-sensillum map was constructed for all 10 of the abundant Obps , and when one particular sensillum , ab8 , was genetically depleted of its sole abundant Obp , it showed a robust electrophysiological response to odorants; in fact the peak response was increased in many cases . These results suggested that Obp28a is not required for the transport of odorants to Ors in the ab8 sensillum . Taken together , the results reveal an unexpected molecular component required for normal humidity response . Our findings add further support to the concept that antennal Obps do not have a single , unifying function , but rather play diverse roles . The results also identify a new target that could be useful in controlling insect vectors that rely on humidity to find their human hosts and oviposition sites . Deletion of a gene in Drosophila often causes a decrease in fitness . We were surprised that deletion of Obp59a caused an increase in fitness under desiccating conditions . How might greater desiccation resistance be achieved ? We suspect that a constellation of metabolic changes together produce desiccation resistance . How would these changes be triggered ? We speculate that owing to the Obp59a defect , the pattern of integrated sensory input from all humidity-sensing circuits in the fly – those activated by the neurons studied here and others ( Yao et al . , 2005 ) , including other antennal neurons expressing certain Trp channels ( Liu et al . , 2007 ) – is abnormal . This abnormality would trigger the induction of a defensive state that protects the fly against the existential threat of desiccation . Flies were reared on standard cornmeal-dextrose agar food at 25°C and 60% humidity . Flies used in behavioral experiments were backcrossed to wCS for at least five generations to minimize genetic background effects . The following IR-GAL4 lines were used: IR40a-GAL4 ( BDSC #41727 ) , IR25a-GAL4 ( BDSC #41728 ) , IR93a-GAL4 ( from Dr . Marco Gallio ) , IR68a-GAL4 ( from Dr . Paul Garrity ) , IR8a-GAL4 ( BDSC #41731 ) . Obp59a-GAL4 flies were created using 5’ and 3’ fragments cloned into pBGRY1 . Primers used for the 5’ end were: CTGCTGTTTGATGGCTTGC ( −500 to −480 ) and CTTGGGAACTGAATGGAGGA ( −1 to −20 , reverse complement ) . Primers for the 3’ end were: GATTAAACTCACCCCACTTTTTAGG ( +1 to+25 ) and AACATTTTAATCAGAAACTAAATACACAGCT ( +469 to+500 , reverse complement ) . 40 bp of yeast sequence remained between the GAL4 stop and the attB3 site , into which the 3’ end of Obp59a is cloned . Plasmids were injected into y1w67c23; P{CaryPattP2 ( BDSC #8622 ) flies . Guide chiRNAs were cloned into pU6-BbsI-chiRNA plasmids . Selected cut sites were located 1-nt upstream of the 5’ end , and 162-nt upstream of the 3’ end , removing 80 . 6% of the coding sequence of Obp59a . Homology arms extending 1 . 02 kb upstream and 1 . 02 kb downstream of the cut site were incorporated into the pHD-DsRed-attP vector ( Gratz et al . , 2014 ) . Cloning methodology was described in Larter et al . ( 2016 ) . w[1118]; Pbac{y[+mDint2]=vas-Cas9}VK00027 ( BDSC #51324 , [Gratz et al . , 2013] ) embryos were injected by Bestgene , Inc . ( Chino Hills , CA ) . Non-sibling G1 adults expressing DsRed were identified . Primer pairs extending beyond the Obp59a coding sequence and homology arms were used to verify gene deletion . Mutation strategy is provided in Supplementary file 1 . 7 day old flies were immobilized with CO2 . Four female and four male flies were placed in a collar fashioned from two inward-facing razor blades stabilized on a stack of microscope slides , as described previously ( Larter et al . , 2016 ) . Animals were covered in OCT compound ( Tissue-Tek ) and rapidly frozen on dry ice . 14 µm sections were collected on slides and stored at −80°C until use . Double-label in situ hybridization utilized Obp59a RNA probes and GAL4 lines driving expression of UAS-mCD8-GFP . RNA probes were synthesized as described in Larter et al . ( 2016 ) . The staining protocol was previously described ( Menuz et al . , 2014 ) . Glossina experiments were similarly conducted using GmObp59a RNA probe , with some modifications . A 450 bp segment of GmmObp59a was PCR-amplified from Glossina morsitans morsitans antennal cDNA using the Forward Primer: TGCCGTACAGATGATGGACC , and Reverse Primer: GGCGATGCTGTGATTCCAAG . From this GmmObp59a DNA template , an unfragmented digoxigenin ( DIG ) -labeled RNA antisense probe was synthesized using standard methods . Antennae were cryosectioned at 40 µm , RNA probes were hybridized at 55°C overnight , sheep anti-DIG-POD primary antibodies were incubated for 45 min , and Cy3 TSA was used for signal detection . Immunohistochemistry was conducted using polyclonal mouse αObp59a produced by Cocalico Biologicals , Inc . ( Stevens , PA ) from insect cell-expressed Obp59a prepared by Novoprotein ( Summit , NJ ) , and anti-IR93a and anti-IR25a antibodies provided by Richard Benton . 7 day old female flies were collected and prepared in collars as described above for expression analysis . Antennae were sectioned at 10 µm and sections mounted on 9 . 5 mm aluminum stubs ( Electron Microscopy Services #75180 ) using carbon paint ( Electron Microscopy Services #12691–30 ) . Samples were coated with 8 nm iridium with a Cressington 208 iridium sputtering tool . SEM was carried out with a Hitachi SU-70 electron microscope equipped with solid-state backscatter detector for enhanced imaging of grain boundaries .	Some insects have a sense – called hygroreception – that allows them to detect changing levels of moisture in the air . These insects use this sense to avoid becoming too dry , or to find food or places to lay their eggs . In many species , including the fruit fly Drosophila melanogaster , the antennae are important for hygroreception . Cells in the antennae produce lots of small proteins called odorant binding proteins , or Obps for short . These proteins are believed mostly to help the antennae to detect various chemical signals in the air , but it was not known if any of these proteins were also involved in hygroreception . Obp59a is an odorant binding protein that is found in the parts of the antennae that sense moisture , and Sun et al . set out to establish whether it has a role in hygroreception in the fruit fly . A closer look confirmed that Obp59a proteins were indeed found specifically in the moisture-sensitive parts of the antennae , the hygroreceptive sensilla . Further experiments showed that flies without Obp59a could not respond properly to changing humidity over periods of seconds , minutes and hours . These results indicated that Obp59a is important for insect hygroreception . Perhaps unexpectedly , these mutant flies were also more resistant to drying out . Sun et al . suggest that , because flies without Obp59a struggle with hygroreception , they may also become more cautious to avoid becoming too dry . Further experiments could now test this hypothesis . Since insects like mosquitoes use hygroreception to find their human hosts or choose where to lay their eggs , Obp59a may become a useful target for controlling insect-borne infections . Also , understanding insect hygroreception may yield new insights into how climate change will affect insect populations around the world .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2018	Humidity response depends on the small soluble protein Obp59a in Drosophila
Positional information is essential for coordinating the development of multicellular organisms . In plants , positional information provided by the hormone auxin regulates rhythmic organ production at the shoot apex , but the spatio-temporal dynamics of auxin gradients is unknown . We used quantitative imaging to demonstrate that auxin carries high-definition graded information not only in space but also in time . We show that , during organogenesis , temporal patterns of auxin arise from rhythmic centrifugal waves of high auxin travelling through the tissue faster than growth . We further demonstrate that temporal integration of auxin concentration is required to trigger the auxin-dependent transcription associated with organogenesis . This provides a mechanism to temporally differentiate sites of organ initiation and exemplifies how spatio-temporal positional information can be used to create rhythmicity . Specification of differentiation patterns in multicellular organisms is regulated by gradients of biochemical signals providing positional information to cells ( Rogers and Schier , 2011; Wolpert , 1969 ) . In plants , graded distribution of the hormone auxin is not only essential for embryogenesis , but also for post-embryonic development , where it regulates the reiterative organogenesis characteristic of plants ( Dubrovsky et al . , 2008; Vanneste and Friml , 2009; Benková et al . , 2003 ) . Plant shoots develop post-embryonically through rhythmic organ generation in the shoot apical meristem ( SAM ) , a specialized tissue with a stem cell niche in its central zone ( CZ; Figure 1A ) . In Arabidopsis thaliana , as in a majority of plants , organs are initiated sequentially in the SAM peripheral zone ( PZ surrounding the CZ ) at consecutive relative angles of close to 137° , either in a clockwise or anti-clockwise spiral ( Figure 1A; Galvan-Ampudia et al . , 2016 ) . SAM organ patterning or phyllotaxis has been extensively analyzed using mathematical models ( Douady and Couder , 1996; Mitchison , 1977; Veen and Lindenmayer , 1977 ) . A widely accepted model proposes that the time interval between organ initiations ( the plastochron ) and the spatial position of organ initiation emerge from the combined action of inhibitory fields emitted by pre-existing organs and the SAM center ( Douady and Couder , 1996 ) . Tissue growth then self-organizes organ patterning by moving organs away from the stem cells and leaving space for new ones . Auxin is the main signal for positional information in phyllotactic patterning ( Reinhardt et al . , 2003a; Reinhardt et al . , 2000 ) . Auxin , has been proposed to be transported directionally toward incipient primordia where it activates a transcriptional response leading to organ specification ( Benková et al . , 2003; Reinhardt et al . , 2003a; Heisler et al . , 2005; Vernoux et al . , 2000 ) . PIN-FORMED1 ( PIN1 ) belongs to a family of auxin efflux carriers whose polarity determines the direction of auxin fluxes ( Benková et al . , 2003; Gälweiler et al . , 1998 ) . PIN1 proteins are present throughout the SAM and regulate the spatio-temporal distribution of auxin cooperatively with other carriers ( Reinhardt et al . , 2003a; Bainbridge et al . , 2008 ) . Convergence of PIN1 carriers toward sites of organ initiation was proposed to control an accumulation of auxin that triggers organ initiation . This spatial organization of PIN1 polarities was also proposed to deplete auxin around organs , locally blocking initiation and thus establishing auxin-based inhibitory fields ( Reinhardt et al . , 2003a; Heisler et al . , 2005; Vernoux et al . , 2011; de Reuille et al . , 2006; Stoma et al . , 2008; Jonsson et al . , 2006; Smith et al . , 2006a ) . In addition , a reduced responsivity of the CZ to auxin has been demonstrated , providing an auxin-dependent mechanism for the inhibition of organogenesis in the CZ ( Vernoux et al . , 2011; de Reuille et al . , 2006 ) . Several models converge to suggest that together , these auxin-dependent regional cues determine new organ locations in the growing SAM . The genetically-encoded biosensor DII-VENUS , a synthetic protein degraded directly upon sensing of auxin , recently allowed an unprecedented qualitative visualization of spatial auxin gradients in the SAM ( Vernoux et al . , 2011; Brunoud et al . , 2012 ) . However , quantification of the spatio-temporal dynamics of auxin is required to fully evaluate both experimental and theoretical understanding of the action of auxin in SAM patterning . This is all the more important given that the continuous helicoidal reorganization of auxin distribution in the growing SAM , suggests that auxin might convey complex positional information . Here , we used a quantitative imaging approach to question the nature of the auxin-dependent positional information . We further investigate how efflux and biosynthesis regulate the 4D dynamics of auxin , and explore how this information is processed in the SAM to generate rhythmic patterning . In the SAM , DII-VENUS fluorescence reports auxin concentration with cellular resolution ( Vernoux et al . , 2011; Brunoud et al . , 2012 ) . To extract quantitative information about auxin distribution , we generated a DII-VENUS ratiometric variant , hereafter named qDII ( quantitative DII-VENUS ) . qDII differs from previously used tools ( Liao et al . , 2015 ) in producing DII-VENUS and a non-degradable TagBFP reference stoichiometrically from a single RPS5A promoter ( Wend et al . , 2013; Goedhart et al . , 2011; Figure 1—figure supplement 1A–H ) . By introducing a stem cell-specific pCLV3:mCherry nuclear transcriptional reporter into plants expressing qDII ( Pfeiffer et al . , 2016 ) we generated a functional and robust geometrical reference for the SAM center ( Figure 1B , C and Figure 1—figure supplement 1I–M ) . All analyzed meristems ( 21 individual SAM ) showed qDII patterns similar to those obtained with DII-VENUS , with locations of auxin maxima following the phyllotactic pattern ( Vernoux et al . , 2011; Figure 1B–E ) . Despite the fact that SAMs were imaged independently and not synchronized , qDII patterns appeared highly stereotypical with easily identifiable fluorescence maxima and minima . This was confirmed by image alignment using SAM rotations ( applying prior mirror symmetry if necessary; Figure 1D and Figure 1—figure supplement 2A–C ) . All images could be superimposed preserving the spatial distribution of auxin maxima and minima ( Figure 1—figure supplement 2B ) . Our analysis shows that auxin distribution follows the same synchronous pattern across a population of SAMs , with low angular and rhythmic variability ( Figure 1—figure supplement 2D–E , Appendix 2 ) , with apparent stationarity up to a 137° rotation ( Figure 1H ) . To further quantify auxin distribution , we developed a mostly automated computational pipeline to measure SAM fluorescence ( Appendix 3 ) ( Cerutti et al . , 2020 ) . We used the spatial distribution of 1-DII-VENUS/TagBFP as a proxy for auxin distribution , hereafter named ‘auxin’ ( Figure 1C ) and focused on the epidermal cell layer ( L1 ) where organ initiation takes place ( Jonsson et al . , 2006; Kierzkowski et al . , 2013; Smith et al . , 2006b; Reinhardt et al . , 2003b ) . The location of the absolute auxin maximum value was defined as Primordium 0 ( P0 ) . Other local maxima with lower auxin values were called Pn ( Appendix 1 ) , with n corresponding to their rank in the phyllotactic spiral ( Figure 1C and Figure 1—figure supplement 2B ) . Note that the dynamic range of qDII allows measuring an auxin value for the vast majority of cells in the PZ and only a few cells at P0 had undetectable values of DII-VENUS , leading to an auxin value of 1 . The pipeline then permits the quantification of nuclear signals and aligns all the SAMs onto a common clockwise reference frame with standardized x , y , z-orientation and with the P0 maximum to the right . This automatic registration confirmed that auxin maxima follow a phyllotactic pattern with a divergence angle close to 137 . 5° ( Figure 1—figure supplement 2F ) . It also demonstrated that maxima are positioned with a precision close to the size of a cell both in distance from the SAM center and in azimuth ( angular distance ) with a maximal standard deviation of 8 . 4 µm or 1 . 5 cell diameters ( Figure 1G ) . We then considered the temporal changes in auxin distribution by using time-lapse images over one plastochron , which corresponds to the period of this rhythmic system . P0 and successive auxin maxima moved radially ( Figure 1—figure supplement 2D ) . Remarkably , while the average radial distance from each local maximum Pn to the SAM center progresses ( Figure 1—figure supplement 2G ) , the spatial deviation of this distance does not change significantly over time , reflecting the synchronized movement of local maxima , with limited meristem to meristem variation . After 10 hr , every Pn local maximum has almost reached the starting position of the next local maximum , Pn+1 , but after 14 hr they have passed this position ( Figure 1—figure supplement 2G ) . This suggests that a rotation of 137 . 5° , which replaces Pn by Pn+1 , corresponds to a temporal progression of 10 to 14 hr ( Figure 1H ) . This was supported by dissimilarity measurements obtained using different rotation angles between maps ( Figure 1—figure supplement 2H ) , allowing us to confirm that plastochron last 12h ± 2h . We could thus derive a continuum of primordium development by placing Pn+1 time series one plastochron ( 12 hr ) after Pn time series on a common developmental time axis ( Figure 1H ) . Together with the observed developmental stationarity , this permitted the reconstruction of auxin dynamics over several plastochrons from observations spanning only one . The resulting quantitative temporal map of auxin distribution in the SAM reveals the dynamic genesis of auxin maxima in the PZ first as finger-like protrusions ( visible at P-2 , P-1 and P0 ) from a permanent high auxin zone at the center of the SAM ( Figure 1—figure supplement 2I–L and Video 1 ) , as previously predicted ( de Reuille et al . , 2006 ) . At later stages , auxin maxima become confined to fewer cells while auxin minima are progressively established precisely in between auxin maxima and the CZ ( Figure 1—figure supplement 3 ) . We next wondered whether the motion of auxin maxima and minima could result purely from cellular growth , an hypothesis used in several theoretical models ( Douady and Couder , 1996; Jonsson et al . , 2006; Smith et al . , 2006b; Heisler and Jönsson , 2006 ) . By following a P1 maximum , we observed that cells within the auxin maximum zone closest to the CZ at time 0 hr gradually transfer to the depletion zone at time 10 hr ( Figure 2A–C; nuclei circled in white ) . At the same time , cells on the distal edge of the maximum zone show a progressive increase in their auxin level ( Figure 2A–C; nuclei circled in red ) , suggesting a spatial shift of the auxin maximum relatively to the cellular canvas . To explore further this phenomenon , we used nuclear motion to estimate cell motion vectors and compare them with the motion of the center of auxin maximum zones , we further found that the average radial speed of auxin maxima between stages P1 and P4 can surpass the average displacement of individual nuclei , with a peak velocity of more than 1 µm/h at the P2 stage ( Figure 2D–E ) . These results show that auxin maxima are not attached to specific cells; instead they travel through the tissue , resulting in an apparent centrifugal wave of auxin accumulation . Consequently , the SAM cellular network provides a dynamic medium in which auxin maximum zones can move radially with their own apparent velocity relative to the growing tissue ( Figure 2D–E ) . Analysis on time-courses of up to 14 hr revealed significant auxin variations in certain cells over one plastochron while auxin levels remained unchanged in others ( Figure 2F–G ) . However , neighboring cells always showed limited differences in their temporal auxin profiles ( Figure 2F–G ) . We concluded from these observations that there is a high definition spatio-temporal distribution of auxin , with auxin apparent movement occurring faster than growth within the tissue and providing cells with graded positional information in space and time ( Figure 2H ) . The creation of auxin maxima first as protrusions of a high auxin zone in the CZ contrasts with the current vision of organogenesis being triggered by local auxin accumulation at the periphery of the CZ with concomitant auxin depletion around auxin maxima ( Reinhardt et al . , 2003a; de Reuille et al . , 2006; Stoma et al . , 2008; Jonsson et al . , 2006; Smith et al . , 2006b ) . This , in addition to the partial uncoupling of auxin distribution dynamics and growth , led us to reevaluate the spatio-temporal patterns of PIN1 localization , given their central role in controlling auxin distribution ( Reinhardt et al . , 2003a; de Reuille et al . , 2006; Jonsson et al . , 2006; Smith et al . , 2006b ) . Co-visualization of a functional PIN1-GFP ( Benková et al . , 2003 ) and qDII/CLV3 fluorescence over time showed that PIN1 concentration increases from P0 and reaches a maximum at P2 before decreasing ( Figure 3A , H and Figure 3—figure supplement 2 ) , consistent with previous observations ( Heisler et al . , 2005; Bhatia et al . , 2016; Caggiano et al . , 2017 ) . To quantify PIN1 cell polarities , we used confocal images after cell wall staining with the fluorescent dye propidium iodide ( PI ) as a reference to position the PIN1-GFP signal relative to the L1 anticlinal cell walls at each cell-cell interface ( Shi et al . , 2017; Figure 3B and Appendix 4 ) . This allowed us to compute PIN1-GFP polarity for each cell-cell interface of the SAM by extracting the 3D distribution of fluorescence for PI and GFP and quantifying the difference of intensity on membranes on both sides of the cell wall ( Figure 1C and Appendix 4 ) . These cell interface polarities measure in which direction each cell interface locally contributes to orient the flow of auxin transport . Using super-resolution radial fluctuation ( SRRF ) microscopy ( Gustafsson et al . , 2016 ) on the same samples , we could show that this method recovers cell interface PIN1 polarities with an error below 10% ( 8 out of 94 interfaces analyzed ) . When calculating cellular PIN1 polarity vectors by integrating the cell interface polarity information for each cell , we could further show that more than 80% of the cellular polarities deviate by less than 30° between the two approaches . This quantitative evaluation ( Figure 3D–G , Figure 3—figure supplement 1 and Appendix 4 ) validates the robustness of our method , showing that , in spite of a coarse image resolution , a vast majority of cellular polarity directions are consistent with super resolution imaging techniques . Our approach is therefore particularly suitable for monitoring global trends at the scale of a tissue . Local averaging of the cellular vectors obtained from confocal images was then used to calculate continuous PIN1 polarity vector maps in order to identify the dominant trends in auxin flux directions in the SAM ( Figure 3I , and Appendix 4 ) . At the tissue scale , the vector maps demonstrate a strong convergence of PIN1 toward the center of the SAM ( Figure 3I–J and Figure 3—figure supplement 2 ) . In addition , PIN1 polarities deviate locally toward the radial axes followed by auxin maxima when they protrude from the CZ . We detected the previously observed inversion of PIN1 polarities at organ boundaries ( Heisler et al . , 2005 ) and our quantifications show that this occurs only from P7 ( Figure 3—figure supplement 2C ) , thus isolating the flower from the rest of the SAM from this late stage . P3 to P5 show a general flux toward the SAM that is locally deflected around the zones of auxin minima before converging back toward the SAM center ( Figure 3I and Figure 3—figure supplement 2 ) . Over the course of one plastochron , only limited changes in the PIN1 polarities are observed ( Figure 3—figure supplement 2 ) , suggesting that changes in auxin distribution at this time resolution do not require major adjustments in the direction of auxin efflux at the tissue scale . We next asked where auxin could be produced in the SAM . YUCCAs ( YUCs ) have been shown to be limiting enzymes for auxin biosynthesis ( Cheng et al . , 2006; Liu et al . , 2016 ) . We thus mapped expression of the eleven YUC encoding genes in the SAM , using GFP reporter lines with a promoter fragment size shown to be functional for YUC1 , 2 and 6 ( Figure 3—figure supplement 3A–N; Liu et al . , 2016; Robert et al . , 2013 ) . Only YUC1 , 4 , 6 were expressed ( Figure 3K , Figure 3—figure supplement 3A–F ) . While YUC6 showed a very weak expression in the CZ , both YUC1 and YUC4 are expressed in the L1 layer on the lateral sides of the SAM/flower boundary from P3 for YUC4 ( Figure 3K ) and P4 for YUC1 ( Figure 3—figure supplement 3A and D; Cheng et al . , 2006 ) . From P4 , YUC4 expression extends over the entire epidermis of flower primordia . This is coherent with genetic and other expression data ( Supplementary file 1; Cheng et al . , 2006; Armezzani et al . , 2018 ) . In addition , yuc1yuc4 loss-of-function mutants show severe defects in SAM organ positioning and size ( Shi et al . , 2018; Pinon et al . , 2013; Figure 3L and Figure 3—figure supplement 3O–U ) . Taken with the organization of PIN1 polarities , these results suggest that P3-P5 are auxin production centers for the SAM that regulate phyllotaxis and that PIN1 polarity organization allows for pumping auxin away from these production centers and towards the meristem . In conclusion , our results suggest a scenario in which auxin distribution depends on high concentrations of auxin at the center of the SAM , and also at P-1 and P0 , acting as flux attractors and on auxin production primarily in P3-P5 ( Figure 3M ) . To assess quantitatively whether and how the spatio-temporal distribution of auxin is interpreted in the SAM , we next introduced the synthetic auxin-induced transcriptional reporter DR5 ( Friml et al . , 2003; Sabatini et al . , 1999; Ulmasov , 1997 ) driving mTurquoise2 into the qDII/CLV3 reporter line ( Figure 4A–D ) . Cells expressing DR5 closest to the CZ were robustly positioned at an average distance of 32 µM ± 7 ( SD ) from the center . This corresponds to a distance at which the intensity of CLV3 reporter expression is less than 5% of its maximal value ( Figure 4—figure supplement 1A ) . The distance from the center at which transcription can be activated by auxin is thus defined with a near-cellular precision . To obtain a global vision of how auxin-controlled transcription is related to auxin concentration , we performed a Principal Component Analysis ( PCA ) using quantified levels of DR5 , auxin and CLV3 in each nucleus of the PZ during a 10 hr time series , together with their distance from the center ( Figure 4E ) . With the first two axes accounting for around 75% of the observed variability , we unexpectedly observed orthogonality between auxin input and DR5 output , clearly marking the absence of a general correlation in the SAM ( Figure 4E , inset ) . This unexpected finding was confirmed by the low Pearson correlation coefficients between DR5 and auxin values at the cell-level ( Figure 4—figure supplement 1B ) . We refined our analysis by focusing on the different primordia regions . We assembled all the observed couples of values ( auxin , DR5 ) , averaged over each primordium region , on a single graph ( Figure 4F ) . This demonstrated that , spatially , a given auxin value does not in general determine a specific DR5 value . However , values corresponding to primordia at consecutive stages follow loop-like counter-clockwise trajectories in the auxin x DR5 space ( indicated by the arrow in Figure 4F ) . Such trajectories are symptomatic of hysteresis reflecting the dependence of a system on its history . In other words , it appears that the relationship between auxin level and DR5 expression is not direct , but is affected by another factor depending on the previous developmental trajectory of each cell ( determined by parameters such as genetic activity , protein content , signal exposure , chromatin state ) . We then tried to identify what in this developmental history can explain the observed differences in DR5 response to auxin . We first used our reconstructed continuum of primordium development to study the joint temporal variations of DR5 and auxin within a group of cells during primordium initiation ( Appendix 5 ) . This showed that the start of auxin-induced transcription follows the build-up of auxin concentration with a delay of nearly one plastochron ( Figure 4G–H ) . The duration of the observed phenomenon suggests the existence of an additional process , over and above fluorescent protein maturation ( Vernoux et al . , 2011; Balleza et al . , 2018 ) , that creates a significant auxin response delay in primordium cells during development . Due to this delay , DR5 is not a direct readout of auxin concentration , explaining the absence of correlation between DR5 expression and auxin levels in these cells . We next wondered what could explain a time-dependent acquisition of cell competence to respond to auxin . A first possible scenario is that cells exiting the CZ proceed through different stages of activation of an auxin-independent developmental program enabling them to sense auxin only after a temporal delay . A second possibility is that auxin controls this developmental program through a time integration process . In this scenario , cells exiting the CZ would need to be exposed to high auxin concentrations for a given time to build up an auxin transcriptional response . To test these scenarios , we treated SAMs with auxin for different periods using physiologically relevant concentrations ( Reinhardt et al . , 2000; Figure 5A–I ) . All treatments , even the shorter ones , equally degraded DII-VENUS throughout the PZ ( Figure 5—figure supplement 1A–I ) . This suggests that auxin uptake was similar throughout the PZ , although we cannot totally discard that some differences exist . In the shorter auxin treatments ( 30’ and 120’ ) , the auxin transcriptional response was mainly enhanced at P-1 and P-2 and to a lesser extent at the position of the predicted P-3that is where cells are already being exposed to auxin ( Figure 5I ) . The longer auxin treatments ( 300' ) lead to an activation of signaling in most cells in the PZ and organs , with the strongest activation being observed again at P-1 and P-2 but also at the predicted azimuth for P-3 , P-4 and P-5 ( Figure 5H–I ) . We could further show that a 300’ treatment with a lower auxin concentration ( 200 nM ) activated signaling similarly ( at P-1 ) or more strongly ( at P-2 , P-3 , P-4 and P-5 ) than a 120’ 1 mM auxin treatment . Conversely , a 120’ treatment with higher auxin concentration ( 5 mM ) lead to an activation of signaling almost as strongly as a 300’ 1 mM treatment at P-1 , although the activation was lower at P-2 ( Figure 5I ) . In all treatments , no significant effect was detected at P0 , consistent with the fact that DR5 activation is already maximal at this stage of development ( Figure 4 ) . We next treated pinoid ( pid ) mutant SAMs with exogenous auxin . pid mutants are strongly affected in polar auxin transport and in aerial organ production ( Reinhardt et al . , 2003a; Friml et al . , 2004; Christensen et al . , 2000 ) . DR5 expression was low and radially uniform in pid SAMs , suggesting a uniform auxin distribution ( Figure 5J; Friml et al . , 2004 ) . When treated with 1 mM auxin , DR5 could be activated in all cells of the periphery of the SAM ( suggesting an uptake throughout the PZ as in the wild-type ) only with a 300’ treatment , while a 120’ treatment had only a weak effect ( Figure 5J–M and Figure 5—figure supplement 2A–C ) . This indicates that , even with the reduced complexity in PZ patterning of the pid mutant ( Friml et al . , 2004 ) , activation of auxin signaling is still dependent on the time of exposure to auxin in all cells surrounding the CZ . Taken together , our observations support the second scenario , with the activation of signaling being a function of both time of exposure to auxin and auxin concentration . Conversely , our results are incompatible with the first scenario , where the capacity of the cells to respond to auxin is intrinsic and is not dependent upon auxin exposure time . Notably , the results with pid SAMs suggest that all cells at the SAM periphery show no intrinsic differences in their capacity to respond to auxin , in agreement with published data ( Reinhardt et al . , 2003a; Heisler et al . , 2005; Smith et al . , 2006a ) . Our results thus support the hypothesis that temporal integration of auxin concentration is required for downstream transcriptional activation in the SAM . The Auxin Response Factor ( ARF ) ETTIN ( ETT/ARF3 ) plays an important role in promoting organogenesis in the SAM ( Wu et al . , 2015; Chung et al . , 2019 ) . Despite the fact that ETT is a non-canonical ARF , genetic data indicate that it acts together with ARF4 and MONOPTEROS/ARF5 to promote organogenesis at the SAM . We found that in a loss-of function ett3 mutant the expression of DR5 was restricted to only 2–3 cells at sites of organogenesis , an observation consistent with a role for ETT in promoting organogenesis . In addition , a 300’ 1 mM auxin treatment did not induce DR5 in the SAM ( Figure 5N–Q and Figure 5—figure supplement 2D–E ) . Auxin signaling and ARF3 in particular have been shown to act by modifying acetylation of histones ( Wu et al . , 2015; Chung et al . , 2019; Long et al . , 2006 ) . Pharmacological inhibition of histone deacetylases ( HDACs ) alone was able to trigger concomitant activation of DR5 at P0 and P-1 sites in the SAM ( Figure 5—figure supplement 1Q–S ) . Taken together , these results suggest that auxin signal integration likely depends on a functional ARF-dependent auxin nuclear pathway . Phyllotaxis is perturbed in ett mutant SAMs ( Figure 5—figure supplement 1M–P; Simonini et al . , 2017 ) . Our results thus suggest that a perturbation of the temporal reading of auxin information can result in phyllotaxis defects . Supporting this idea , we also found that daily exogenous auxin treatments at the SAM affected phyllotaxis and that the efficiency of the treatment increased with both auxin concentration and treatment length . This was particularly evident for 30’ and 120’ treatments ( Figure 5—figure supplement 1J–L ) . 300’ treatments were less efficient at higher auxin concentrations , possibly due to compensation mechanisms . These results suggest that temporal integration of auxin information at the SAM is essential for phyllotaxis . In a recent modeling study , a stochastic induction of organ initiation based on temporal integration of morphogenetic information was proposed ( Refahi et al . , 2016 ) . Here we provide evidence that organ initiation in the SAM is indeed dependent on temporal integration of the auxin signal . Our quantitative analysis of the dynamics of auxin distribution and response supports a scenario in which rhythmic organ initiation at the SAM is driven by the combination of high-precision spatio-temporal graded distributions of auxin with the use of the duration of cell exposure to auxin , to temporally differentiate sites of organ initiation ( Figure 6 ) . Importantly our results suggest that a time integration mechanism is essential for rhythmic organ patterning in the SAM since auxin-based spatial information pre-specifies several sites of organ initiation and is thus unlikely to provide sufficient information ( Video 1 ) . Whether temporal integration of auxin information exists in other tissues remains to be established . We provide evidence that temporal integration of the auxin signal likely requires the effectors of the auxin signaling pathway . Activation of transcription downstream of auxin by ARFs relies on chromatin remodeling , increasing the accessibility of ARF targets and possibly allowing for the recruitment of histone acetyltransferases ( Wu et al . , 2015 ) , together with the release of histone deacetylases ( HDACs ) from target loci through degradation of Aux/IAA repressors ( Long et al . , 2006 ) . Chromatin state change is one mechanism that allows the temporal integration of signals in eukaryotes , including plants ( Angel et al . , 2011; Coda et al . , 2017; Nahmad and Lander , 2011; Sun et al . , 2009 ) . It is thus plausible that time integration of the auxin signal in cells leaving the CZ is set by progressive acetylation of histones triggered by ARFs at their target loci . As chromatin deacetylation also represses auxin signaling in the CZ ( Ma et al . , 2019 ) , balancing the acetylation status of ARF target loci could provide a mechanism to tightly link stem cell maintenance to differentiation by precisely positioning organ initiation at the boundary of the stem cell niche , while at the same time allowing sequential organ initiation . Temporal integration might as well rely on mechanisms that fine-tune the intracellular distribution of auxin , such as auxin metabolism but also intracellular transport ( Sauer and Kleine-Vehn , 2019 ) . Determining how different mechanisms might act in parallel to provide a capacity to activate target genes as a function of auxin concentrations over time will require further analyses . It will notably be important to determine whether other ARF than ARF3 act in the temporal integration of auxin . The existence of high definition spatio-temporal auxin gradients suggests that as for several morphogens in animals ( Nahmad and Lander , 2011; Dessaud et al . , 2007; Scherz et al . , 2007; Maden , 2002 ) the robustness of SAM patterning results from highly reproducible spatio-temporal positional information . Our results indicate that auxin maxima could first emerge from the CZ at the confluence of centripetal auxin fluxes . Confluences creating auxin maxima would at the same time divert fluxes away from areas where auxin minima appear ( Figure 3M ) . Our analysis raises the question of how auxin transport could generate this high definition signal distribution and whether the different models that have been proposed can explain this distribution ( Bainbridge et al . , 2008; Stoma et al . , 2008; Jonsson et al . , 2006; Smith et al . , 2006a; van Berkel et al . , 2013 ) . Further analysis of the spatio-temporal control of auxin distribution needs also to consider that early developing flowers act as auxin production centers . These flowers could not only provide a memory of the developmental pattern through lateral inhibition but also contribute positively to a self-sustained auxin distribution pattern by providing auxin to the system ( Figure 3M ) . Finally , our work indicates that the stem cell niche could act as a system-wide organizer of auxin transport , consistent with previous work ( de Reuille et al . , 2006 ) . This could provide another layer of regulation tightly coordinating differentiation with the presence of a largely auxin-insensitive stem cell niche ( Vernoux et al . , 2011; Ma et al . , 2019 ) . Seeds were directly sown in soil , vernalized at 4 °C , and grown for 24 days at 21 °C under long day condition ( 16 hrs light , LED 150µmol/m²/s ) . Shoot apical meristems from inflorescence stems with a length between 0 . 5 and 1 . 5 cm were dissected and cultured in vitro as described in Prunet et al . ( 2016 ) for 16 hrs . When required , meristems were stained with 100 µM propidium iodide ( PI; Merck ) for 5 min . Auxin treatments were performed by immersing meristems in solutions containing indicated concentrations of indole-acetic acid ( IAA ) and 10 mM MES-hydrate ( buffer ) for indicated periods of time . Trichostatin A ( TSA – Invivogen ) was added to the culture medium to a final concentration of 5 µM . Meristems were cultured in TSA for 16 hrs prior to auxin treatment . For time lapses , the first image acquisition ( T=0 ) corresponds to 2 hrs after the end of the dark period . In planta treatments were carried out on 24 day-old Col-0 plants by dropping 10 µL of IAA solution ( IAA at different concentrations , 10 mM MES-hydrate and 0 . 01% v/v Tween-20 ) onto the SAM , followed by incubation for indicated lengths of time . Meristems were then washed with 100 µL of 10 mM MES buffer with 0 . 01% v/v Tween-20 . Treatments were carried out on 5 consecutive days and perturbations in organ positioning were recorded 7 days after the end of the treatments . Previously published transgenic lines used in this study are PIN1-GFP ( Benková et al . , 2003 ) , pCLV3:mCherry-NLS ( Pfeiffer et al . , 2016 ) , pYUC1-11:GFP and yuc1 yuc4/+ pDR5rev::GFP ( Liu et al . , 2016; Robert et al . , 2013 ) , ett-22 ( Pekker et al . , 2005 ) , pid-14 ( Huang et al . , 2010 ) . pRPS5a:DII-VENUS-N7-p2A-TagBFP-SV40 ( qDII ) and pDR5rev:2x-mTurquoise2-SV40 constructs were cloned cloned using Gateway technology ( Life Sciences ) , and transformed in Arabidopsis thaliana ( Col-0 ) . Stable qDII homozygous lines were then crossed with pCLV3:mCherry-NLS , pDR5rev:2x-mTurquoise2-SV40 and PIN1-GFP reporter lines . All confocal laser scanning microscopy was carried out with a Zeiss LSM 710 spectral microscope or a Zeiss LSM700 microscope . Multitrack sequential acquisitions were always performed using the same settings ( PMT voltage , laser power and detection wavelengths ) as follows: VENUS , excitation wavelength ( ex ) : 514 nm , emission wavelength ( em ) : 520–558 nm; mTurquoise2 , ex: 458 nm , em: 470–510 nm; EGFP , ex: 488 nm , em: 510–558 nm; TagBFP , ex:405 nm , em: 430–460 nm; mCherry , ex: 561 nm , em: 580–640 nm; propidium iodide , ex: 488 , em: 605–650 nm . Scanning electron microscopy of meristems were carried out using a HIROX SH-3000 microscope . Time lapses for Super Resolution Radial Fluctuation ( SRRF ) imaging were performed on an inverted Zeiss microscope ( AxioObserver Z1 , Carl Zeiss Group , http://www . zeiss . com/ ) equipped with a spinning disk module ( CSU-W1-T3 , Yokogawa , www . yokogawa . com ) and a Prime95B SCMOS camera ( https://www . photometrics . com ) using a 63x Plan-Apochromat objective ( numerical aperture 1 . 4 , oil immersion ) , pixel size 175 nm or a 100x Plan-Apochromat objective ( numerical aperture 1 . 46 , oil immersion ) , pixel size 110 nm . GFP was excited with a 488 nm laser ( 150 mW ) and fluorescence emission was filtered using a 525/50 nm BrightLine single-band bandpass filter ( Semrock , http://www . semrock . com/ ) . PI was excited with a 561 nm laser ( 80 mW ) and fluorescence emission was filtered using a 609/54 nm BrightLine single-band bandpass filter ( Semrock , http://www . semrock . com/ ) . To obtain high resolution images , 200 frames were acquired with 50% laser power and 70 ms exposure time using Stream Acquisition mode . The green and red channels were acquired sequentially . For drift correction , 200 nm TetraSpeck beads ( Life Technologies ) were added to samples . Images were processed using the NanoJ-SRRF plugin ( Gustafsson et al . , 2016 ) with the following parameters: Ring Radius 0 . 5 , Radiality Magnification 5 , Axes in ring 6 , Temporal Analysis TRPPM . SRRF time-lapses were produced by running SRRF analysis on groups of 50 frames . If aberrant PSF of Tetraspeck beads were observed , datasets were discarded . All confocal images were pre-processed using the ImageJ software ( http://rsbweb . nih . gov/ij/ ) for the delimitation of the region of interest . Then the CZI image files were processed using a computational pipeline relying on the numpy , scipy , pandas , czi_file Python libraries , as well as other custom libraries . Extensive details about the computational methods and algorithms are given in Appendix 3 , 4 and 5 . Given the non- linear positive DR5 response , the raw values were logarithmically transformed in order to obtain a symmetric distribution of the noise . Nadaraya-Watson estimates and confidence intervals were then calculated with a confidence level of 95% in the R environment ( RStudio Team , 2015 ) . The boxplots displayed in the article were obtained by computing the median ( central line ) , first and third quartiles ( lower and upper bound of the box ) and first and ninth deciles ( lower and upper whiskers ) using the R environment or numpy percentile function and rendered using the matplotlib Python library . Linear regressions were performed using the polyfit and polyval numpy functions . P-values were obtained using the scipy anova implementation in the f_oneway function . Principal component analysis was performed using the PCA implementation from the scikit-learn Python library . All data were generated with at least three independent sets of plants . All experimental data and quantified data that support the findings of this study are available from the corresponding authors upon request . Generic quantitative image and geometry analysis algorithms are provided in Python libraries timagetk , cellcomplex , tissue_nukem_3d and tissue_paredes ( https://gitlab . inria . fr/mosaic/ ) made publicly available under the CECILL-C license . Specific SAM sequence alignment and visualization algorithms are provided in a separate project providing Python scripts to perform the complete analysis pipelines ( Cerutti , 2020; copy archived at https://github . com/elifesciences-publications/sam_spaghetti ) .	Plants , like animals and many other multicellular organisms , control their body architecture by creating organized patterns of cells . These patterns are generally defined by signal molecules whose levels differ across the tissue and change over time . This tells the cells where they are located in the tissue and therefore helps them know what tasks to perform . A plant hormone called auxin is one such signal molecule and it controls when and where plants produce new leaves and flowers . Over time , this process gives rise to the dashing arrangements of spiraling organs exhibited by many plant species . The leaves and flowers form from a relatively small group of cells at the tip of a growing stem known as the shoot apical meristem . Auxin accumulates at precise locations within the shoot apical meristem before cells activate the genes required to make a new leaf or flower . However , the precise role of auxin in forming these new organs remained unclear because the tools to observe the process in enough detail were lacking . Galvan-Ampudia , Cerutti et al . have now developed new microscopy and computational approaches to observe auxin in a small plant known as Arabidopsis thaliana . This showed that dozens of shoot apical meristems exhibited very similar patterns of auxin . Images taken over a period of several hours showed that the locations where auxin accumulated were not fixed on a group of cells but instead shifted away from the center of the shoot apical meristems faster than the tissue grew . This suggested the cells experience rapidly changing levels of auxin . Further experiments revealed that the cells needed to be exposed to a high level of auxin over time to activate genes required to form an organ . This mechanism sheds a new light on how auxin regulates when and where plants make new leaves and flowers . The tools developed by Galvan-Ampudia , Cerutti et al . could be used to study the role of auxin in other plant tissues , and to investigate how plants regulate the response to other plant hormones .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "plant", "biology", "developmental", "biology" ]	2020	Temporal integration of auxin information for the regulation of patterning
Most research on neurodegenerative diseases has focused on neurons , yet glia help form and maintain the synapses whose loss is so prominent in these conditions . To investigate the contributions of glia to Huntington's disease ( HD ) , we profiled the gene expression alterations of Drosophila expressing human mutant Huntingtin ( mHTT ) in either glia or neurons and compared these changes to what is observed in HD human and HD mice striata . A large portion of conserved genes are concordantly dysregulated across the three species; we tested these genes in a high-throughput behavioral assay and found that downregulation of genes involved in synapse assembly mitigated pathogenesis and behavioral deficits . To our surprise , reducing dNRXN3 function in glia was sufficient to improve the phenotype of flies expressing mHTT in neurons , suggesting that mHTT's toxic effects in glia ramify throughout the brain . This supports a model in which dampening synaptic function is protective because it attenuates the excitotoxicity that characterizes HD . Neurodegenerative conditions involve a complex cascade of events that takes many years to unfold . Even in the case of inherited disorders due to mutation in a single gene , such as Huntington’s disease ( HD ) , the downstream ramifications at the molecular level are astonishingly broad . Caused by a CAG repeat expansion in Huntingtin ( HTT ) ( The Huntington’s Disease Collaborative Research Group , 1993 ) , HD pathology is prominent in the striatum and cortex , yet transcriptomic studies consistently reveal thousands of changes in gene expression across the brain and different neuronal cell types , involving pathways ranging from autophagy to vesicular trafficking ( Saudou and Humbert , 2016 ) . To disentangle changes that are pathogenic from those that represent the brain’s effort to compensate for the disease , we recently integrated transcriptomics with in silico analysis and high-throughput in vivo screening using a Drosophila model of HD ( Al-Ramahi et al . , 2018 ) . This study demonstrated that HD pathogenesis is driven by upregulation of genes involved in the actin cytoskeleton and inflammation , but that neurons compensate by downregulating the expression of genes involved in synaptic biology and calcium signaling . The finding that synaptic changes were protective caught our attention because HTT itself is necessary for normal synaptogenesis and maintenance within the cortico-striatal circuit ( McKinstry et al . , 2014 ) , largely through its role in retrograde axonal trafficking of neurotrophic factors ( Saudou and Humbert , 2016 ) . But synapses involve more than just neurons: glial cells also contribute to synapse formation , function , and elimination ( Filipello et al . , 2018; McKinstry et al . , 2014; Octeau et al . , 2018; Stogsdill et al . , 2017 ) . There is , in fact , emerging evidence that various glial subtypes affect outcomes in HD . The accumulation of mutant Huntingtin ( mHTT ) in astrocytes and oligodendrocytes hinders their development and function and contributes to disease pathophysiology ( Benraiss et al . , 2016; Ferrari Bardile et al . , 2019; Osipovitch et al . , 2019; Wood et al . , 2018 ) . Conversely , healthy glia can improve the disease phenotype in HD mice ( Benraiss et al . , 2016 ) . Recent studies using single-cell sequencing in astrocytes isolated from post-mortem tissue from HD patients and mouse models of HD ( Al-Dalahmah et al . , 2020; Diaz-Castro et al . , 2019 ) developed molecular profiles that distinguish HD-affected astrocytes from astrocytes found in healthy brain tissue , but the physiological consequences of the gene expression changes were unclear . Whether mHTT affects glial participation in synapse formation or maintenance remains unknown , but then , we are only just now beginning to understand the range of glial types and their functions ( Bayraktar et al . , 2020; Darmanis et al . , 2015 ) . The combination of synaptic degeneration in HD and the fact that both HTT and glia contribute to synaptic formation and maintenance led us to further investigate the influence of mHTT in glia . Because Drosophila have been used to elucidate glial biology ( Freeman and Doherty , 2006; Olsen and Feany , 2019; Pearce et al . , 2015; Ziegenfuss et al . , 2012 ) and are a tractable model system for studying HD and other neurodegenerative diseases ( Al-Ramahi et al . , 2018; Bondar et al . , 2018; Donnelly et al . , 2020; Fernandez-Funez et al . , 2000; Filimonenko et al . , 2010; Goodman et al . , 2019; Ochaba et al . , 2014; Olsen and Feany , 2019; O'Rourke et al . , 2013; Rousseaux et al . , 2018; Yuva-Aydemir et al . , 2018 ) , we decided to generate flies that express mHTT solely in glia so that we could compare their transcriptomic signature with that of flies expressing mHTT in neurons . We took an unbiased approach , first establishing the repertoire of evolutionarily conserved genes that show concordant expression changes across HD human and mouse striata and HD fly brains . We then integrated this comparative transcriptomic data with high-throughput in vivo behavioral screening to acquire insight into glial contributions to HD pathogenesis and identify disease-modifying targets that mitigate the HD phenotype . To study the contributions of neurons and glia to HD pathogenesis , we first needed to define a transcriptomic signature that would enable us to move across species ( human , mouse , and fly ) ( Figure 1A ) . We began with human tissue . Since the striatum is the brain region most prominently affected in HD , we compared the gene expression profiles of human post-mortem striatal samples from healthy individuals and patients with HD , from different stages of the disease ( i . e . , Vonsattel Grade 0–4 ) ( Hodges et al . , 2006; Vonsattel et al . , 1985 ) . We identified 1852 downregulated and 1941 upregulated differentially expressed genes ( DEGs ) in patients with HD compared to healthy individuals ( Figure 1B ) . We then reanalyzed published RNA-seq data from mouse striata using an allelic series of knock-in mouse models with varying CAG repeat lengths at 6 months of age ( Langfelder et al . , 2016 ) . Because it is unclear which CAG tract length in mice most faithfully recapitulates HD pathogenesis , the triplet repeat length was treated as a continuous trait , and we narrowed our analysis to DEGs that correlate with increasing CAG repeat length . Comparing the striata of wildtype mice to the knock-in HD mouse models , there were 3575 downregulated and 3634 upregulated DEGs ( Figure 1B ) . ( The greater genome coverage provided by RNA-seq [Miller et al . , 2014] yielded larger datasets for mouse and , below , for Drosophila than for humans . ) We performed RNA-seq leveraging Drosophila HD models ( Kaltenbach et al . , 2007; Romero et al . , 2008 ) ( see Materials and methods ) to compare the effect of expressing mHTT in either neurons or glia . The binary GAL4-UAS system was used to drive the expression of human mHTT either in neurons ( elav >GAL4 ) or glia ( repo >GAL4 ) . Both full-length ( HTTFLQ200 ) and N-terminal ( HTTNT231Q128 ) models were used in this set of experiments since both the full protein and N-terminal HD fragments accumulate in the human brain as a result of proteolysis and mis-splicing ( Kim et al . , 2001; Neueder et al . , 2017; Sathasivam et al . , 2013; Wellington et al . , 2002 ) . Principal component analysis ( PCA ) showed that the greatest differences between samples are attributable to the cell-specific drivers , and not to the use of N-terminal versus full-length protein ( Figure 1—figure supplement 1 ) . Expressing mHTT in neurons resulted in 3058 downregulated and 2979 upregulated DEGs , while expressing mHTT in glia resulted in 3127 downregulated and 3159 upregulated DEGs . There were also DEGs common to both neurons and glia expressing mHTT: 1293 downregulated and 1181 upregulated ( Figure 1B ) . With these transcriptomic signatures in hand , we were able to compare gene expression profiles across the three species . We focused on genes with significantly altered expression ( using a false discovery rate [FDR] < 0 . 05; see Materials and methods ) in the same direction ( i . e . , upregulated or downregulated ) in response to mHTT expression across these three species , including both Drosophila HD models . We call genes that meet this criterion concordantly altered DEGs ( Supplementary file 1 ) . We compared DEGs using a graph-based approach ( see Materials and methods ) that allows for evolutionary divergence and convergence , instead of imposing one-to-one relationships . 815 upregulated DEGs observed in HD patient-derived striatal tissue had an orthologous gene in the HD mouse model and at least one Drosophila model of HD that was concordantly upregulated . Similarly , 791 DEGs identified in HD patients had an orthologous gene in mouse and Drosophila models that was concordantly downregulated ( Figure 1C ) . About 40% of the alterations in gene expression in patient striatal samples are concordant with orthologous genes in both Drosophila and mice models of HD . To determine whether this result could be an artifact of overlapping a large number of DEGs in each model , we randomly selected and overlapped 815 and 791 orthologous genes across the three species 20 , 000 times . Based on the resulting distribution , we concluded that the overlap of concordant , orthologous DEGs across the various HD models was not random ( p=6 . 37×10−158 and p=1 . 66×10−165 , probability distribution test ) . To compare the consequence of expressing mHTT in glia versus neurons , we recalculated the overlaps between the three species , distinguishing DEGs from the neuron-only and glia-only HTT-expressing Drosophila . There were 425 concordantly upregulated and 545 concordantly downregulated DEGs in glia . We also found 522 upregulated DEGs and 453 downregulated specific to neurons . Out of these groups of DEGs , 310 were upregulated and 320 were downregulated in both neurons and glia . To acknowledge the proportion of transcriptional alterations we excluded by specifying concordant expression with the HD Drosophila models , we also calculated the overlap between concordant DEGs observed only in striata from HD patients and mice . We found that 83 . 7% of upregulated DEGs and 77 . 7% of downregulated DEGs that were altered concordantly in human and mouse HD striata were also concordantly altered in the brains of the neuronal and/or glial HD Drosophila models ( Figure 1D ) . Of the genes that showed concordantly altered expression only in human and mouse striata , 64 ( 40% ) of the upregulated and 68 ( 30% ) of the downregulated DEGs did not have an ortholog in Drosophila . To investigate the cellular pathophysiology represented by DEGs in neurons and glia , we constructed protein-protein interaction ( PPI ) networks using the STRING-db database ( Szklarczyk et al . , 2015 ) . The upregulated and downregulated networks of DEGs responding to mHTT expression in neurons or glia had a significant PPI enrichment compared to networks constructed from an equivalent number of random genes selected from a whole-proteome background ( Supplementary file 2 ) . To control for potential artifacts that could arise from using the whole proteome background , we performed a more stringent analysis using only proteins that are found in the striatum ( Al-Ramahi et al . , 2018 ) . Using average node degree and betweenness as proxies for connectivity , we found that the glial and neuronal networks show higher network connectivity than expected by random chance among proteins present in the striatum ( Supplementary file 2 ) . This high connectivity suggested that the networks are enriched in specific biological processes and/or pathways . We therefore clustered the glial mHTT response and neuronal mHTT response networks using the InfoMap random walks algorithm ( iGraph Package for R and Python ) ( Rosvall and Bergstrom , 2007 ) . Clusters that had fewer than four nodes were filtered out of subsequent analysis . The glial networks formed 23 and 24 clusters for upregulated and downregulated DEGs , respectively . Both the upregulated and downregulated neuronal networks formed 29 clusters . We applied this clustering method to the networks of randomly selected striatal proteins in order to determine the expected number of clusters for networks of a similar size . Both the glial and neuronal networks formed significantly more clusters than would be expected from random selection ( Supplementary file 2 ) . To gain insight into biological processes represented by each cluster , we queried the five most significantly enriched terms ( FDR < 0 . 05 ) using the GO Biological Process and Kyoto Encyclopedia of Genes and Genomes ( KEGG ) terms within each cluster ( Supplementary file 3 ) . A synthesis of these terms was used to identify clusters in both the glial and neuronal networks ( Supplementary file 3 , Figure 2—figure supplement 1B , C ) . We compared the membership within clusters across the glial and neuronal networks using a pairwise hypergeometric test and identified 14 clusters of upregulated DEGs common to both glial and neuronal networks . Similarly , there were 15 clusters of downregulated DEGs common to the both networks ( Figure 2—figure supplement 1A ) . Given the aims of our study , the clusters of DEGs specific to glia ( represented by nodes in Figure 2 ) were of particular interest to us . Six clusters were specifically upregulated in response to mHTT expression , enriched in genes involved in transcription and chromatin remodeling , amino acid metabolism , cell proliferation , cytokine signaling/innate immunity , arachidonic acid metabolism , and steroid synthesis ( Figure 2A ) . Six clusters were downregulated in response to glial mHTT expression , containing genes involved in synapse assembly , calcium ion transport , immune system regulation , phagocytosis , mRNA processing , and fatty acid degradation ( Figure 2B ) . We applied the same network analysis to genes that had concordantly altered expression in HD patient striata and HD mouse model striata but not in HD Drosophila models ( Figure 2—figure supplement 2A ) . We observed that clusters comprising DEGs specific to the HD patients and the mouse models were functionally related to DEGs in both the glial and neuronal networks ( Figure 2—figure supplement 2B ) . Gene expression data from bulk tissue does not provide the resolution required to define cell-autonomous gene expression alterations resulting from mHTT toxicity . Therefore , we compared DEGs ( false discovery rate [FDR] < 0 . 1 ) in human embryonic stem cells from individuals with HD ( carrying 40–48 CAG repeats ) with healthy embryonic stem cells that have been differentiated into either CD140+ oligodendrocyte progenitor cells ( OPCs ) or CD44+ astrocyte progenitor cells ( APCs ) ( Osipovitch et al . , 2019 ) . We compared the resulting list of DEGs identified in the HD OPCs ( 1439 genes ) and HD APCs ( 193 genes ) to the list of conserved HD DEGs from flies expressing mHTT in glia . We identified 46 upregulated and 91 downregulated DEGs in common ( Figure 3A ) . APCs had 4 upregulated and 12 downregulated genes in common . We next asked whether any clusters in the fly glial networks were enriched in genes dysregulated in HD OPCs or APCs . The Synapse Assembly cluster ( Figure 2B ) was significantly enriched in genes with reduced expression in HD OPCs ( Fisher’s exact test , p<0 . 001 ) , including SYT13 , LRRTM1 , GRM1 , EPB41L2 , DLGAP3 , and AGAP2; the only gene of this cluster that was upregulated in HD OPCs was NRXN3 ( Figure 3B , C ) . In sum , by using a comparative , network-based analysis of the HD transcriptome , we associated dysregulation of several biological processes with the expression of mHTT in glia . Layering the gene expression profile of homogenous glial populations affected by mHTT onto these networks , we were able to extract from the bulk-tissue analysis a cluster of genes related to synaptic assembly that are altered in response to glial mHTT toxicity . The next question we sought to answer is whether changes in expression of synaptic assembly genes are compensatory or pathogenic . We reasoned that if lowering the expression of a downregulated HD DEG aggravated mHTT-induced toxicity , then the downregulation of that gene is pathogenic . Conversely , if reducing the expression of a DEG led to an improvement in HD-related phenotypes , we considered that reduction to be compensatory . We previously used this approach , which takes advantage of the genetic tractability of Drosophila and the availability of high-throughput behavioral screening as a proxy for neurological function , to discover modifier genes that reduce HTT protein levels in HD patient cells ( Al-Ramahi et al . , 2018 ) . Here we assessed the effect of various genetic changes in the same group of animals over time , following the expression of mHTT in either glia , neurons , or both cell types . We used a custom , robotic assay system that video-records flies climbing upwards to the top of a vial after being knocked to the bottom ( negative geotaxis ) to track the behavior of individual Drosophila in real time and measure several motor metrics including speed ( see Materials and methods ) . Healthy flies reliably climb to the top at a steady rate until the effects of aging gradually reduce their speed . In contrast , animals expressing mHTT specifically in glia or neurons show much more rapid , if still age-dependent , loss of climbing speed compared to animals expressing a non-targeting hairpin RNA ( hpRNA ) . While we only focus on the effect of these genetic perturbations on speed , we also observe impairments in coordination , balance , and direction ( output as number of turns and stumbles ) in Drosophila expressing mHTT ( data not shown ) . The expression of SYT13 , LRRTM1 , GRM1 , EPB41L2 , DLGAP3 , and AGAP2 is reduced in HD OPCs derived from human embryonic stem cells ( Osipovitch et al . , 2019 ) , which is consistent with the expression patterns we observed in patient-derived striatal tissue , knock-in mouse model striatal tissue , and in neuronal tissue from Drosophila expressing mHTT in glia ( Figure 3C ) . We performed genetic perturbation analysis on the Drosophila orthologs of these genes to assess whether their downregulation was pathogenic or compensatory in glia . Diminishing expression of the Drosophila orthologs of these six genes mitigated the behavioral deficits induced by mHTT expression in glia ( Figure 3D , additional controls in Figure 3—figure supplement 1B ) . We concluded that reduced expression of these genes is a compensatory response to mHTT expression in glia . There were additional protein interactors in Synapse Assembly whose expression was not altered in the HD-affected OPCs or APCs compared with controls but that were nonetheless downregulated across all three HD models . In our behavioral assay , reducing expression of these interactors , including NLGN3 , NLGN4X , HOMER1 , and SLITRK5 , was also protective against glial mHTT toxicity ( Figure 3—figure supplement 1A , Supplementary file 4; additional controls in Figure 3—figure supplement 1B ) . In sum , comparative transcriptomic analysis indicated that genes within the Synapse Assembly cluster are associated with the glial response to HD , and the high-throughput behavioral assay further defined this response as compensatory . NRXN3 was identified as a DEG in both our cross-species comparative transcriptomic analysis and in the gene expression profile of the HD glial progenitor population . NRXN3 expression was lower in the bulk HD transcriptome across species compared to their respective controls , but it was more highly expressed in the HD OPCs than in controls . This discordance between the bulk and single-cell-type gene expression profiles might be a result of time-dependent changes in gene expression as neurons age , but it prevented us from classifying the NRXN3 expression changes as being compensatory or pathogenic . We were particularly interested in neurexins , including NRXN3 , because they mediate contact between pre- and post-synaptic neurons ( Ushkaryov et al . , 1992; Zeng et al . , 2007 ) . We therefore asked whether downregulation of Drosophila NRXN3 ( dNRXN3 , also known as nrx-1 ) is damaging or protective when both neurons and glia express mHTT . In the Drosophila behavioral assay , heterozygous loss of dNRXN3 function in animals expressing mHTT in both neurons and glia mitigated mHTT toxicity and improved behavior ( Figure 4A , left panel ) . Reproducing this experiment with flies expressing mHTT only in glia yielded the same benefit ( Figure 4A , middle panel ) . The obvious next question , given its canonical role in neuron-neuron contact , was whether dNRXN3 heterozygosity would protect against mHTT pathogenesis in neurons . Interestingly , the answer was no ( Figure 4A , right panel ) . Consistent with this , glia-specific knockdown of dNRXN3 ( using the repo-GAL4 driver ) mitigated mHTT toxicity in glia ( Figure 4B , left panel ) , but neuron-specific knockdown ( using the elav-GAL4 driver ) of dNRXN3 did not mitigate mHTT toxicity in neurons ( Figure 4B , right panel ) . In sum , reducing dNRXN3 in both neurons and glia protects against glial pathogenesis—and the combination of neuronal and glial pathogenesis—but not neuronal pathogenesis . This implies that mHTT disrupts some aspect of glial-neuronal interaction that is driven by the glia since lowering expression of dNRXN3 in glia is necessary and sufficient to mitigate behavioral impairments caused by mHTT . To investigate whether Nrxn3 is expressed in astrocytes in the striatum of HD mice , we performed in situ hybridization ( ISH ) in coronal sections of striatal tissue taken from a mouse model of HD ( HdhzQ175/+ ) to probe Nrxn3 mRNA . Nrxn3 was expressed in striatal astrocytes ( Figure 4C , D , Figure 4—figure supplement 1 ) . In conclusion , modulating the expression genes other than mHTT in glia could be an effective strategy for ameliorating HD-induced central nervous system ( CNS ) dysfunction . We were curious to identify modifiers that concordantly affect mHTT-induced pathogenesis in both neurons and glia as these might be particularly attractive therapeutic targets for HD . We were particularly interested to discover whether any such shared modifiers exert their effect by reducing mHTT levels , which is considered a promising approach to therapy ( Al-Ramahi et al . , 2018; Barker et al . , 2020; Caron et al . , 2020; Li et al . , 2019; Tabrizi et al . , 2019; Wang et al . , 2014; Wood et al . , 2018; Yamamoto et al . , 2000; Yao et al . , 2015 ) . We therefore again integrated network analysis with high-throughput experimentation . Genes were sampled from both the neuronal and glial mHTT response networks by prioritizing those candidates with high centrality ( calculated as a cumulative rank-score of node betweenness and node degree ) within each cluster . When available , we used alleles that perturb the expression or activity of the Drosophila orthologs in the same direction as the gene expression change in the HD patient population ( Figure 5A ) . We screened 411 alleles , representing 248 Drosophila genes homologous to 211 human genes , for perturbations that improve the age-dependent behavior of Drosophila expressing mHTT in neurons or glia ( Supplementary file 5 ) . Alleles that ameliorated neuronal or glial function were verified in a subsequent trial in animals expressing mHTT across the CNS ( in both neurons and glia ) . In all , we identified 25 genes with altered expression in HD that suppressed mHTT-induced behavioral deficits in neurons , glia , or both ( Figure 5B , C , Figure 5—figure supplement 1 , Supplementary file 6 ) . Many of the modifiers common to neuronal and glial mHTT-induced dysfunction are involved in the regulation of the actin cytoskeleton ( RHOC , TIAM1 , ENAH , and CFL2 ) , vesicular trafficking ( SNAP23 , SNX9 , and SNX18 ) , and inflammation ( JUN , GTF3A , and ATF3 ) . Multiple reports have implicated components of these pathways in the pathogenesis of not only HD , but in other neurodegenerative disorders as well ( Al-Ramahi et al . , 2018; Bardai et al . , 2018; Bondar et al . , 2018 ) . We previously established an axis of genes with altered expression that regulate actin cytoskeleton and inflammation pathways driving forward HD pathogenesis ( Al-Ramahi et al . , 2018 ) . Our current results would indicate that these pathways are not only critical to disease progression in neurons , but also in glia . We previously observed that reducing the activity of RAC GTPase , a regulator of the actin cytoskeleton , and inflammation mediating nuclear factor kappa-light-chain-enhancer of activated B cells ( NF Kappa-B ) ameliorated pathogenesis by lowering mHTT protein levels through the activation of autophagy ( Al-Ramahi et al . , 2018 ) . Thus , in a secondary screen we tested whether these disease modifiers common to both neurons and glia exerted their beneficial effects by lowering levels of the mutant HTT protein . We collected protein lysates from Drosophila expressing mHTT across the CNS that also bore alleles that suppressed mHTT-induced behavioral deficits in both neurons and glia . We assessed the quantity of mHTT protein in these lysates by western blot , comparing experimental ( candidate modifiers ) and control animals ( carrying a non-targeting hpRNA ) . This secondary screen identified Spn42De as a modifier whose knockdown lowered mHTT levels . Spn42De is one of the four Drosophila homologues of human SERPINA1 ( which encodes alpha-1-antitrypsin , a member of a large group of protease inhibitors ) . Spn42De , human SERPINA1 , and mouse Serpina1 are all upregulated in HD , and they are part of the Wound Healing and Inflammation cluster in both the neuronal and glial mHTT response networks ( Figure 2—figure supplement 1C ) . Knockdown of Spn42De ( henceforth dSERPINA1 ) in Drosophila expressing mHTT in both neurons and glia mitigated behavioral impairments ( Figure 6A ) . In independent immunoblots , dSERPINA1 knockdown consistently reduced mHTT protein levels in lysates extracted from the heads of Drosophila expressing mHTT in both neurons and glia ( Figure 6B , C ) . As a control , we performed immunoblot analysis of lysates from a green fluorescent protein ( GFP ) reporter line to ensure that this allele of dSERPINA1 did not reduce the function of the GAL4-UAS system ( Figure 6—figure supplement 1 ) . To validate this observation across model systems , we performed homogenous time-resolved fluorescence ( HTRF ) on HdhQ111/Q7 mouse striatal cell lysates that were treated with either a pool of non-targeting scramble small interfering RNAs ( siRNAs ) , a pool of siRNAs against Htt , or a pool of siRNAs against Serpina1a ( the murine ortholog of SERPINA1 ) . Serpina1a knockdown significantly reduced mHTT signal ( Figure 6D ) . Knockdown of SERPINA1 thus protected against mHTT toxicity in neurons and glia by reducing levels of mutant HTT . Verifying this effect in multiple model organisms increases confidence in this observation and suggests that SERPINA1 could potentially prove useful as a target for treating HD . Interestingly , SERPINA1 expression is low in the healthy brain but it is upregulated in several disease conditions , consistent with a potential role in neuroinflammation ( Abu-Rumeileh et al . , 2020; Cabezas-Llobet et al . , 2018; Gollin et al . , 1992; Peng et al . , 2015 ) . We found increased Serpina1a protein staining in the striatum of HdhzQ175/+ compared to wildtype mice at 8 . 5 months ( Figure 6—figure supplement 2 ) , confirming its upregulation from the transcriptomic data . Previously we had shown that other genes in the subnetwork implicated in neuroinflammation can be manipulated to lower mHTT protein levels ( Al-Ramahi et al . , 2018 ) . SERPINA1 may thus warrant investigation as a target for other neurological disorders as well . We found a high degree of overlap of DEGs across tissues from human HD brains , brains of HD mice , and flies that express mHTT in glia . This may seem unexpected given obvious differences between vertebrate and Drosophila glia , such as a lack of documented microglia or distinct morphology of endothelial/glial cells forming the blood-brain barrier in Drosophila ( Freeman and Doherty , 2006 ) . Our observations are however consistent with previous evidence that Drosophila glia perform many of the same functions as mammalian astrocytes , oligodendrocytes , endothelial cells , and microglia including phagocytosis ( Chung et al . , 2020; Freeman , 2015; Freeman and Doherty , 2006; Ziegenfuss et al . , 2012 ) . In fact , the overlap of concordant DEGs between mammalian and Drosophila glia may be underestimated in our analysis because it was limited to CD44+ and CD140+ cells from human embryonic stem cell-derived glial progenitors and therefore we may have missed DEG overlaps from other glial types , or from more mature state of oligodendrocytes or astrocytes . Several studies have also shown that wildtype glial cells ameliorate disease when transplanted into HD mice , and mHTT exerts a deleterious effect on glial development and function , which in turn influences HD pathogenesis ( Benraiss et al . , 2016; Bradford et al . , 2009; Garcia et al . , 2019; Huang et al . , 2015; Osipovitch et al . , 2019 ) . More recently , it was discovered that transcription factors involved in glial differentiation and myelin synthesis are downregulated in glial progenitor cells ( Osipovitch et al . , 2019 ) . Yet despite this progress , the overall contributions of glial genes to synaptic impairments and other key neurodegenerative pathologies remain poorly understood . The genetic malleability of Drosophila enabled us to thoroughly examine the neuron-glia interface from both the glial and the neuronal directions . Synaptic dysfunction is a common theme among many neurodegenerative disorders ( McInnes et al . , 2018; Phan et al . , 2017; Prots et al . , 2018 ) . While it is clear that the dysfunction of the glia-synapse interface is central to the pathophysiology of neurodegeneration ( Filipello et al . , 2018; Garcia et al . , 2019; Lian et al . , 2015; Litvinchuk et al . , 2018 ) , the underlying mechanisms remain underexplored relative to the interactions between pre- and post-synaptic neurons . Our results support the observation that the expression of mHTT in glia is sufficient to drive synaptic dysfunction ( Wood et al . , 2018 ) . In HD , pre-synaptic neurons release elevated levels of glutamate into the synapse , driving medium spiny neurons ( MSNs ) into excitotoxicity ( Estrada Sánchez et al . , 2008; Hong et al . , 2016 ) . Hyperactivity of receptors at the post-synaptic densities sensitizes MSNs to excitotoxicity , further contributing to neurodegeneration ( Estrada Sánchez et al . , 2008 ) . Astrocytic mHTT expression may contribute to neuronal excitotoxicity by elevating levels of glutamate , potassium , and calcium at the synapse ( Garcia et al . , 2019; Jiang et al . , 2016; Tong et al . , 2014 ) . Modifiers of mHTT-induced pathogenesis identified in our study , such as metabotropic glutamate receptors and the scaffold protein HOMER1 , regulate calcium and glutamate signaling in astrocytes ( Buscemi et al . , 2017; Spampinato et al . , 2018 ) . Reducing the expression of these genes could prevent excess calcium and glutamate from accumulating at the synapse . Indeed , we previously found that HD neurons downregulate the expression of genes involved in calcium signaling in an effort to compensate for HD pathogenesis ( Al-Ramahi et al . , 2018 ) . Glial calcium signaling can also influence neuronal activity , however , at the neuronal soma ( Weiss et al . , 2019 ) . In Drosophila , cortical glia modulate neuronal activity through potassium buffering , a process that is regulated by calcium-mediated endocytosis of potassium channels ( Weiss et al . , 2019 ) . Glia can also physically disrupt synapses in disease states: Förster resonance energy transmission in vivo revealed that , in HD , the distances between astrocytes and pre-synaptic neurons are increased at the cortico-striatal circuit ( Octeau et al . , 2018 ) . Thus , knocking down the genes in the Synapse Assembly cluster could reduce physical interaction between glia and synapses , promoting normal synaptic function . If in HD synapses grow more fragile and fewer in number as the disease progresses , why would downregulating the expression of glial genes required for synapse formation and function be protective ? We postulate it is for the same reason that downregulating calcium-signaling genes is compensatory ( Al-Ramahi et al . , 2018 ) : the brain is attempting to protect against the excitotoxicity described above . Mutant HTT disrupts neuronal development ( Ring et al . , 2015 ) and skews embryonic neurogenesis toward producing more neurons ( Barnat et al . , 2020 ) ; by the time HD mutation carriers reach the age of 6 years , they have greatly enlarged striata and functional hyperconnectivity to the cerebellum ( Tereshchenko et al . , 2020 ) . The more hyperconnected , the more abrupt the loss of these connections , and the more rapid the striatal atrophy that follows Tereshchenko et al . , 2020 . The hyperfunction of a given brain region puts considerable strain on the circuit , and it seems that over the course of a lifetime , the brain keeps trying to compensate for the abnormalities that arise at different stages of HD . The recent observation that deletion of astrocytic neurexin-1α attenuates synaptic transmission but not synapse number supports this hypothesis ( Trotter et al . , 2020 ) . We do not think that the protection provided by modifiers in this cluster is limited to modulating neurotransmission . In astrocytes , calcium signaling also controls the activity of reactive astrocytes ( Buscemi et al . , 2017 ) . Astrogliosis , or the proliferation of immune active astrocytes , is typically observed at later stages of HD ( Al-Dalahmah et al . , 2020; Buscemi et al . , 2017 ) . These immune-activated glia not only eliminate synapses ( Liddelow et al . , 2017; Sofroniew , 2009 ) but can also transmit mHTT aggregates through the synapse ( Donnelly et al . , 2020 ) . In Drosophila , knockdown of draper prevents astrocytic phagocytosis and stops the spread of mHTT protein aggregates from pre-synaptic neurons to the post-synaptic compartment ( Donnelly et al . , 2020; Pearce et al . , 2015 ) . mHTT protein can also enter the synaptic space by endosomal/lysosomal secretion mediated by Syt7 ( Trajkovic et al . , 2017 ) . In this study , we observed that knockdown of synaptotagmins in Drosophila ameliorates glial mHTT-induced dysfunction . Thus , knocking down genes in the Synapse Assembly cluster could also benefit the circuit by reducing the transmission of aggregated mHTT protein from pre- to post-synaptic neurons . Intriguingly , loss-of-function variants in NRXN1-3 , NLGN1 , NLGN3 , DLGAP3 , and LRRTM1 have been associated with various disorders of synaptic dysfunction , including autism spectrum disorder ( ASD ) , schizophrenia , and obsessive compulsive disorder ( OCD ) ( Nakanishi et al . , 2017; Jamain et al . , 2003; Südhof , 2008; Vaags et al . , 2012; Wang et al . , 2018; Windrem et al . , 2017 ) . We speculate that the consequences of loss of function of these genes depend on both dosage and context: modest reductions of gene expression can be protective in the context of HD pathogenesis , whereas a more severe loss of function results in ASD and OCD . It is interesting that many HD patients develop schizophrenia-like psychosis , suggesting that the compensatory mechanism at place in HD may eventually lead to schizophrenia-like symptoms ( Connors et al . , 2020; Tsuang et al . , 2018 ) . Future studies should investigate whether these loss-of-function variants associated with neurodevelopmental and psychiatric disorders alter the age of disease onset in patients with HD . It could be of particular interest to assess if these neurodevelopmental and psychiatric-associated variants ameliorate neurodevelopmental changes observed early in HD or blunt synaptic hyperactivity later in disease . Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact , Juan Botas ( jbotas@bcm . edu ) . We began with Drosophila models expressing either N-terminal human HTT ( HTTNT231Q128 ) or full-length HTT ( HTTFLQ200 ) ( Kaltenbach et al . , 2007; Romero et al . , 2008 ) . The mHTT was expressed using either a pan-neuronal ( elav ) or a pan-glial driver ( repo ) . Mutant strains for screening were obtained from Bloomington Drosophila Stock Center , GenetiVision , and the Vienna Drosophila Resource Center . All strains were maintained at 18°C in standard molasses , yeast extract , and agar media until their experimental use . For RNA-sequencing , the full-length models were raised at 29°C and the N-terminal models were raised at 28°C . All behavioral experiments were performed on females raised at 28°C . In Figure 3D , we used the following mutants to assess the effect of reduced expression of synaptic genes in mHTT animals on behavior: UAS-non-targetinghpRNA ( Vienna Drosophila Resource Center , ID:13974 ) , CenG1ALOF or y1w*;Mi{MIC}CenG1AMI06024 ( Bloomington Drosophila Stock Center , ID: 44301 ) , vlcLOF or y1w67c23;P{w+mc = lacW}vlck01109/CyO ( Bloomington Drosophila Stock Center , ID: 10366 ) , trnLOF or y1w67c23;P{w+mc = lacW}trnS064117/TM3 , Sb1 Ser1 ( Bloomington Drosophila Stock Center , ID: 4550 ) , coraLOF or P{ryt7 . 2=neoFRT}43D cora14/CyO ( Bloomington Drosophila Stock Center , ID: 9099 ) , UAS-SytbetahpRN A ( Vienna Drosophila Resource Center , ID:106559 ) , and UAS-mGluRRNAi ( National Institute of Genetics , Japan , ID: 11144-R3 ) . To generate Drosophila that expressed siRNA that knocked down human HTT ( UAS-siHTT ) , we cloned a 378 bp inverted EcoRI , XbaI fragment of N-terminal Htt into the pMF3 vector ( Drosophila Genome Resource Center ) . This fragment maps to base pairs 406–783 of the human mRNA Huntingtin , which we cloned using the following primers: We first digested the PCR product with EcoRI and ligated it with itself to obtain inverted repeats . We then digested the inverted repeat with XbaI and pasted the fragment into the pMF3 vector ( also cut with XbaI ) ; the resulting plasmid was injected into Drosophila embryos using standard methods ( Dietzl et al . , 2007 ) . We validated that this line lowers mHTT levels . Immortalized mouse striatal cells heterozygous for mHTT ( STHdhQ111/Q7 ) were obtained from Coriell Cell Repositories ( Camden , NJ ) and cultured in DMEM ( Life Technologies , cat . no . 11965 ) supplemented with 10% fetal bovine serum ( Life Technologies , cat . no . 10082–147 ) . The cells were tested every two months by a TransDetect PCR Mycoplasma Detection Kit ( Transgen Biotech , cat . no . FM311-01 ) to ensure that they are mycoplasma free . The identity has not been authenticated by STR profiling , but has been validated by western blot , morphology , and phenotypic experiments . We performed RNA-seq on head tissue collected from Drosophila expressing N-terminal ( UAS-HTTNT231Q128 ) or full-length ( UAS-HTTFLQ200 ) human mHTT in neurons ( elav-GAL4 ) or glia ( repo-GAL4 ) . For each combination of HD model and driver , RNA-seq was performed at three timepoints to capture the early , middle , and late phases of disease pathogenesis , corresponding to behavioral deficits caused by mHTT-induced neuronal or glial dysfunction . At each timepoint , samples for HD and age-matched controls were collected in triplicate . Drosophila expressing the N-terminal construct and corresponding controls were obtained at 7 , 9 , and 11 days post-eclosion for the neuronal driver , and at 5 , 7 , and 8 days post-eclosion for the glial driver . Drosophila expressing the full-length construct , samples were obtained at 18 , 20 , and 22 days post-eclosion for both the neuronal and glial driver . For RNA-seq , the neuronal N-terminal , glial N-terminal , and glial full-length model Drosophila were raised at 28°C . The neuronal full-length model Drosophila were raised at 29°C . For each genotype at each timepoint , we collected an equivalent number of control animals ( elav-GAL4 or repo-GAL4 ) that were raised in the same conditions . Three replicates of 50 virgin females were collected for each genotype and timepoint . Animals were aged in the appropriate temperature and were transferred to fresh food daily until tissue was harvested . At the selected ages , animals were transferred to 1 . 5 mL tubes , flash frozen in liquid nitrogen , vigorously shaken , and then sieved to collect 50 heads/genotype/replica ( ~5 mg tissue/replica ) . Total RNA was extracted using the miRNeasy Mini Kit ( Qiagen cat . no . 210074 ) . RNA-seq profiling and preprocessing was performed by Q2 Solutions ( Morrisville , NC ) . Samples were converted into cDNA libraries using the Illumina TruSeq Stranded mRNA sample preparation kit ( Illumina cat . no . 20020595 ) and were sequenced using HISeq-Sequencing-2 × 50 bp-PE . Initial analysis was performed using Q2 Solution in-house mRNAv7 pipeline with a median of 49 million actual reads . After adapter sequences were removed , the reads were aligned to the Drosophila melanogaster transcriptome using Bowtie version 0 . 12 . 9 ( Langmead and Salzberg , 2012 ) . Expression was quantified using RSEM version 1 . 1 . 19 , resulting in a median of 11 , 214 genes and 18 , 604 isoforms detected ( Li and Dewey , 2011 ) . Three homology maps were constructed to define conserved genes that were concordantly dysregulated in response to mHTT toxicity: a Drosophila-human map , a Drosophila-mouse map , and a mouse-human map . The Drosophila-human map and Drosophila-mouse map were both obtained from DIOPT version 6 . 0 . 2 ( Hu et al . , 2011 ) . To capture homology that results from evolutionary convergence and divergence , we included lower DIOPT scores between Drosophila and mammals instead of fitting one-to-one mappings between these species . The mouse-human homology mapping was obtained from the Mouse Genome Informatics ( MGI ) database hosted by Jackson Laboratories ( Blake et al . , 2017 ) . We integrated these three homology maps by representing each map as an undirected bipartite graph , where nodes are genes of one species and edges represent homology between two genes across species . All components were then merged to form an undirected graph where each node represents a gene name and corresponding species . We applied this integrated homology map consisting of nodes representing the Drosophila , mouse , and human dysregulated genes , and all edges induced by the corresponding nodes , to obtain a subgraph consisting of multiple connected components . If any individual connected component contained nodes that belong to all three species , we characterized all genes within the connected component as concordant . To examine how the upregulated and downregulated core genes interact functionally , we used STRING v10 . 5 ( Szklarczyk et al . , 2015 ) . Only high-confidence interactions ( edge weight >0 . 7 ) were considered . Each node is converted from an ENSEMBL ID to human Entrez ID via the provided mapping file ( v10 , 04-28-2015 ) . Four subgraphs of STRING were then induced on each core gene set separately . Nodes were further clustered with the InfoMap community detection algorithm ( Rosvall and Bergstrom , 2008 ) , implemented in the Python iGraph package , with the default settings ( trials = 10 ) ( Csardi and Nepusz , 2000 ) . We crossed female virgins that carried the mHTT transgene under the control of either the neuronal or glial driver , or the cell-specific driver alone , to males carrying the experimental allele . We introduced a heat-shock-induced lethality mutation on the Y chromosome ( YP{hs-hid} ) to the disease and cell-specific driver stocks to increase the efficiency of virgin collection ( Starz-Gaiano et al . , 2001 ) . For crosses involving alleles that were lethal or sterile mutations on the X chromosome , this mating strategy was reversed . For behavioral assays , elav >HTTNTNT231Q128 and repo >HTTNTNT231Q128 animals were raised and maintained at 28 . 5°C . elav , repo >HTTNTNT231Q128 animals were raised and maintained at 25°C . Individual Drosophila in each genotype were randomly grouped into replicates of 10 . The negative geotaxis climbing assay was performed using a custom robotic system ( SRI International , available in the Automated Behavioral Core at the Dan and Jan Duncan Neurological Research Institute ) . The robotic instrumentation elicited negative geotaxis by ‘tapping’ Drosophila housed in 96-vial arrays . After three taps , video cameras recorded and tracked the movement of animals at a rate of 30 frames per second for 7 . 5 s . For each genotype , we collected 4–6 replicates of 10 animals to be tested in parallel ( biological replicates ) . Each trial was repeated three times ( technical replicates ) . The automated , high-throughput system is capable of assaying 16 arrays ( 1536 total vials ) in ~3 . 5 hr . To transform video recordings into quantifiable data , individual Drosophila were treated as an ellipse , and the software deconvoluted the movement of individuals by calculating the angle and distance that each ellipse moves between frames . Replicates were randomly assigned to positions throughout the plate and were blinded to users throughout the duration of experiments . The results of this analysis were used to compute more than two dozen individual and population metrics , including distance , speed , and stumbles . Software required to run and configure the automation and image/track the videos include Adept desktop , Video Savant , MatLab with Image Processing Toolkit and Statistics Toolkit , RSLogix ( Rockwell Automation ) , and Ultraware ( Rockwell Automation ) . Additional custom-designed software include Assay Control – SRI graphical user interface for controlling the assay machine; Analysis software bundles: FastPhenoTrack ( Vision Processing Software ) , TrackingServer ( Data Management Software ) , ScoringServer ( Behavior Scoring Software ) , and Trackviewer ( Visual Tracking Viewing Software ) . mRNA ISH and immunofluorescence were performed on 25-µm-thick coronal brain sections cut from fresh-frozen brain harvested from a 6-month-old HdhzQ175/+ mouse . We generated digoxigenin ( DIG ) -labeled mRNA antisense probes against Nrxn3 using reverse-transcribed mouse cDNA as a template and an RNA DIG-labeling kit from Roche ( Sigma ) . Primer and probe sequences for the Nrxn3 probe are available in Allen Brain Atlas ( http://www . brain-map . org ) . ISH was performed by the RNA In Situ Hybridization Core at Baylor College of Medicine using an automated robotic platform as previously described ( Yaylaoglu et al . , 2005 ) with modifications of the protocol for fluorescent ISH . In brief: after the described washes and blocking steps , the DIG-labeled probe was visualized using a tyramide-Cy3 Plus kit ( 1:50 dilution , 15 min incubation , Perkin Elmer ) . Following washes in phosphate buffered saline ( PBS ) , the slides were stained with 1:500 anti-GFAP rabbit polyclonal antibody ( DAKO , Z0334 ) diluted in 1% blocking reagent in Tris buffered saline ( Roche Applied Science , 11096176001 ) overnight at 4°C . After washing , slides were treated with 1:500 anti-rabbit IgG Alexa 488 secondary antibody for 30 min at room temperature ( Invitrogen , A-11008 ) . The slides were stained with DAPI and cover slipped using ProLong Diamond ( Invitrogen , P36970 ) . Images were taken at ×63 magnification using a Leica SP8 confocal microscope . For visualizing Serpina1 , 8 . 5-month-old male zQ175 mice ( four wildtype and three knock-in ) were deeply anesthetized and transcardially perfused with 1× PBS . The tissues were then treated with 70% ethanol for 24 hr , 95% ethanol overnight , 100% ethanol for 4 hr , and chloroform overnight . Next tissues were treated with paraffin at room temperature overnight and again with paraffin at 65°C for 2 hr . Paraffin-embedded tissue blocks were coronally sectioned at the thickness of 8 μm , starting from Bregma 0 . 98 mm . Immunofluorescence for Serpina1 was conducted with rabbit anti-Serpina1a primary antibody ( Invitrogen , PA5-16661 ) , followed by the biotin labeled secondary antibody and detected by Alexa Fluor 488 conjugated streptavidin . Fluorescent imaging of the striatal region was performed on a Leica Sp8 confocal microscope . For all immunoblot experiments , Drosophila were raised and maintained at 25°C . Female F1 progeny were collected and flash-frozen 24 hr after eclosion . Heads were separated by genotype and divided into eight individuals per replicate . Drosophila heads were lysed and homogenized in 30 µL of lysis buffer ( 1× NuPage LDS Sample Buffer , 10% beta-mercaptoethanol ) and boiled at 100°C for 10 min . Lysates were loaded on a 4–12% gradient Bis-Tri NuPage ( Invitrogen ) gel and run at a constant voltage of 80 V for an hour and then 120 V for 30 min . For mHTT levels , a 20% methanol transfer buffer was used to transfer proteins at 4°C overnight using a 200 mA current . For mCD8::GFP , proteins were transferred using a 10% methanol buffer for 2 hr at 4°C using a 200 mA current . Prior to antibody treatment , all membranes were treated with blocking solution ( 5% non-fat milk in 1× TBST ) . For primary antibody treatment , all antibodies were diluted in blocking solution . To assess mHTT levels , membranes were then treated with a 1:500 mouse anti-HTT solution ( mAb5490 , EMD Millipore ) overnight . For a loading control , membranes were subsequently treated with a 1:1000 alpha-tubulin antibody ( Abcam EP1332Y ) . 1:1000 Rabbit anti-GFP ( ThermoFisher A-11122 ) was used to assess levels of mCD8::GFP , and 1:1000 anti-lamin C ( Hybridoma Bank LC28 . 26 ) was used as a loading control . All blots were treated with 1:5000 Goat anti-Mouse ( IRDye 800CW Goat anti-Mouse IgG ) and Goat anti-Rabbit ( RDye 680RD Goat anti-Rabbit IgG ) secondary antibodies diluted in blocking solution for 1 hr and imaged using the Odyssey CLx imager ( LI-COR Biosciences ) . STHdhQ111/Q7 cells were reverse transfected with pooled siRNAs using Lipofectamine 2000 ( Life Technologies , cat . no . 11668 ) . Cells were treated with a pool of four small siRNAs per gene with the following sequences ( Qiagen 1027280 ) : Following siRNA treatment , cell lysis buffer ( 1× PBS with 1% TrintonX-100% and 1% EDTA-free protease inhibitor; Calbiochem , #539134 ) was added to each well and the plate was put on ice for 30 min . After incubation , cells were homogenized and lysates were extracted . Separately , HTRF assay buffer was prepared using 50 mM NaH2PO4 ( pH 7 . 4 ) , 400 mM KF , 0 . 1% bovine serum albumin , 0 . 05% Tween-20 , and Quant-ITTM PicoGreen ( 1:1500 ) . The donor antibody , 2B7 conjugated to terbium , was diluted in HTRF assay buffer to a concentration of 0 . 023 µg/mL , and the acceptor antibody , mAb2166 ( SigmaAldrich ) conjugated to fluorescent dye D2 , was diluted to a final concentration of 1 . 4 μg/mL . 5 µL of the HTRF buffer was added to 5 µL of cell lysates ( 5 µL ) in each well of a 384-well plate . Lysates were then incubated at 4°C overnight . HTRF was performed in a Perkin Elmer EnVision multilabel plate reader ( model #2104 ) , measuring the 615 nM and 665 nM , as well as the PicoGreen signal at 485 nM . Each sample was measured following 30 cycles of the excitation at an interval of 16 . 6 ms . Differential expression analysis used the DESeq2 R package on a total of 12 comparisons ( two HD models , two cell-specific drivers , and three timepoints ) ( Love et al . , 2014 ) . Outlier detection was performed using PCA on normalized gene expression data , resulting in one sample being removed . To establish a list of upregulated and downregulated DEGs in Drosophila , we examined the FDR at every timepoint in both genetic models . If the FDR was <0 . 05 at any data point in the HD models compared to control , we established that that gene was dysregulated due to the presence of mHTT in either neurons or glia . We did not take the magnitude of fold-change into account , only the direction ( upregulated or downregulated ) ( Langfelder et al . , 2016 ) . The identification of DEGs from humans was based on microarray data from brain tissue collected post-mortem in patients with HD and age-matched , healthy individuals . For consistency with the reported results , we examined the summary statistics of the caudate probe on the Affymetrix U133 A and B microarrays . We computed the FDR by applying the Benjamini–Hochberg procedure to the p-values reported in Hodges et al . , 2006 . A probe was said to be dysregulated if the absolute value of its fold-change was >1 . 2 ( or log2FC > 0 . 263 ) and the FDR was <0 . 05 . Since multiple Affymetrix probes can match to the same Entrez ID , we specified that an Entrez-identified human gene was dysregulated , if there exists a matching probe that is also dysregulated . We established the lists of upregulated and downregulated DEGs in mice from RNA-seq data presented in Langfelder et al . , 2016 , where the authors profiled mRNA of an allelic series in a HD knock-in mouse model . We reanalyzed data from the striatum at 6 months , identifying gene expression alterations that were significant ( FDR < 0 . 05 ) in the continuous-Q case , a summary regression variable derived from DESeq that tests the association of the expression profile with Q-length as a numeric variable ( Love et al . , 2014 ) . We randomly sampled 471 proteins ( equivalent to the average number of input proteins in the mHTT Responding networks ) 1000 times from 15 , 884 proteins that are expressed in the striatum . Implementing the same parameters that were used for the mHTT responding networks , we constructed clustered PPI networks with the random striatal protein lists as inputs . We calculated the average node degree and average node betweenness within each network of random genes and compiled a distribution using these results . A Z-score was calculated using the distribution compiled from the random striatal networks . These Z-scores were then used to calculate the p-values that are reported in Supplementary file 2 . All simulations and statistical calculations were performed in R—this script can be found as Source code 2 . We assessed behavior in Drosophila as the speed at which individual animals within one vial moved as a function of age and genotype using a nonlinear random mixed effects model regression . Specifically , we looked at differences in regression between genotypes with time ( additive effect , represented by a shift in the curve ) or the interaction of genotype and time ( interactive effect , represented by a change in the slope of the curve ) . We estimated the expected statistical power to detect differences by each of our models using a stringent threshold for statistical significance ( alpha = 0 . 001 ) . We reported p-values representative of the pairwise post-hoc tests for testing whether all possible pairs of genotype curves are different in both models . We considered differences between positive controls and experimental perturbations of p<0 . 001 to be significant . p-values were adjusted for multiplicity using Holm’s procedure . Code for this analysis is available upon request from the Botas Laboratory . All graphing and statistical analyses were performed in R . Images of western blots were analyzed using the Image Studio Lite software . We used an equivalent area to measure signal intensity across all replicates . We present proteins of interest as a ratio of the target protein to loading control ( n = 5 immunoblots ) . Experimental replicates were compared to controls using a one-sided Student’s t-test . For HTRF , levels of mHTT were calculated by taking the ratio of the fluorescence signals ( 665 nM/615 nM ) and normalizing to the PicoGreen signal in experimental groups after subtracting the signal from wells containing only sample buffer and HTRF buffer , without protein lysates . Results are presented as the average and standard error of the mean of the ΔF ( % ) ( ΔF ( % ) = ( Sample ratio − blank ratio ) /blank ratio ×100 ) . Each treatment group consisted of nine replicates ( n = 9 ) . p-values were calculated using Fisher’s LSD test . The mean intensity for images of Serpina1a stained brain slices was measured using ImageJ . For each sample , five images were measured and the mean was calculated The control group consisted of four samples ( n = 4 ) , and the HD group consisted of three samples ( n = 3 ) . Groups were compared using a two-tailed t-test assuming unequal variances .	When a neuron dies , through injury or disease , the body loses all communication that passes through it . The brain compensates by rerouting the flow of information through other neurons in the network . Eventually , if the loss of neurons becomes too great , compensation becomes impossible . This process happens in Alzheimer's , Parkinson's , and Huntington's disease . In the case of Huntington's disease , the cause is mutation to a single gene known as huntingtin . The mutation is present in every cell in the body but causes particular damage to parts of the brain involved in mood , thinking and movement . Neurons and other cells respond to mutations in the huntingtin gene by turning the activities of other genes up or down , but it is not clear whether all of these changes contribute to the damage seen in Huntington's disease . In fact , it is possible that some of the changes are a result of the brain trying to protect itself . So far , most research on this subject has focused on neurons because the huntingtin gene plays a role in maintaining healthy neuronal connections . But , given that all cells carry the mutated gene , it is likely that other cells are also involved . The glia are a diverse group of cells that support the brain , providing care and sustenance to neurons . These cells have a known role in maintaining the connections between neurons and may also have play a role in either causing or correcting the damage seen in Huntington's disease . The aim of Onur et al . was to find out which genes are affected by having a mutant huntingtin gene in neurons or glia , and whether severity of Huntington’s disease improved or worsened when the activity of these genes changed . First , Onur et al . identified genes affected by mutant huntingtin by comparing healthy human brains to the brains of people with Huntington's disease . Repeating the same comparison in mice and fruit flies identified genes affected in the same way across all three species , revealing that , in Huntington's disease , the brain dials down glial cell genes involved in maintaining neuronal connections . To find out how these changes in gene activity affect disease severity and progression , Onur et al . manipulated the activity of each of the genes they had identified in fruit flies that carried mutant versions of huntingtin either in neurons , in glial cells or in both cell types . They then filmed the flies to see the effects of the manipulation on movement behaviors , which are affected by Huntington’s disease . This revealed that purposely lowering the activity of the glial genes involved in maintaining connections between neurons improved the symptoms of the disease , but only in flies who had mutant huntingtin in their glial cells . This indicates that the drop in activity of these genes observed in Huntington’s disease is the brain trying to protect itself . This work suggests that it is important to include glial cells in studies of neurological disorders . It also highlights the fact that changes in gene expression as a result of a disease are not always bad . Many alterations are compensatory , and try to either make up for or protect cells affected by the disease . Therefore , it may be important to consider whether drugs designed to treat a condition by changing levels of gene activity might undo some of the body's natural protection . Working out which changes drive disease and which changes are protective will be essential for designing effective treatments .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "genetics", "and", "genomics" ]	2021	Downregulation of glial genes involved in synaptic function mitigates Huntington's disease pathogenesis
Arrhythmogenesis from aberrant electrical remodeling is a primary cause of death among patients with heart disease . Amongst a multitude of remodeling events , reduced expression of the ion channel subunit KChIP2 is consistently observed in numerous cardiac pathologies . However , it remains unknown if KChIP2 loss is merely a symptom or involved in disease development . Using rat and human derived cardiomyocytes , we identify a previously unobserved transcriptional capacity for cardiac KChIP2 critical in maintaining electrical stability . Through interaction with genetic elements , KChIP2 transcriptionally repressed the miRNAs miR-34b and miR-34c , which subsequently targeted key depolarizing ( INa ) and repolarizing ( Ito ) currents altered in cardiac disease . Genetically maintaining KChIP2 expression or inhibiting miR-34 under pathologic conditions restored channel function and moreover , prevented the incidence of reentrant arrhythmias . This identifies the KChIP2/miR-34 axis as a central regulator in developing electrical dysfunction and reveals miR-34 as a therapeutic target for treating arrhythmogenesis in heart disease . Cardiac excitability is controlled by a combination of depolarizing and repolarizing currents , whose dysregulation during heart failure ( HF ) or myocardial infarction ( MI ) play a significant role in clinically relevant arrhythmias ( Tomaselli and Marbán , 1999; Wang and Hill , 2010 ) . Aberrant remodeling culminates in altered Ca2+ current ( ICa ) ( Houser et al . , 2000; Wang et al . , 2008 ) , Na+ current ( INa ) ( Pu and Boyden , 1997; Maltsev et al . , 2002 ) , and a host of outward K+ currents ( IK ) ( Näbauer and Kääb , 1998 ) , creating impaired cardiac excitability and performance , accounting for high rates of mortality in HF patients ( Nattel et al . , 2007; Tomaselli and Zipes , 2004 ) . However , large variability and breadth of electrical changes present challenges in determining which mechanisms are critical in driving arrhythmias and disease progression . Intriguingly , loss of the Potassium Channel Interacting Protein 2 ( KChIP2 ) has proven to be a consistent event following cardiac stress , sparking interest into understanding its contribution in disease remodeling ( Näbauer and Kääb , 1998; Nass et al . , 2008; Jin et al . , 2010 ) . It is well described that KChIP2 associates with and modulates the Kv4 family of potassium channels , which together define the fast transient outward potassium current ( Ito , f ) , maintaining early cardiac repolarization ( An et al . , 2000; Niwa and Nerbonne , 2010 ) . However , emerging evidence suggests KChIP2 may not be limited to this role ( Thomsen et al . , 2009; Li et al . , 2005; Deschênes et al . , 2008 ) . Investigations following KChIP2 knock-down show reduced transcript expression for the cardiac sodium channel gene , SCN5A , and its accessory subunit SCN1B , in addition to Kv4 . 3 protein , prompting the loss of both Ito , f and INa ( Deschênes et al . , 2008 ) . Considerably , these changes reflect conditions observed in the diseased heart , but more importantly implicate potential transcriptional significance for KChIP2 at the center of that remodeling . Indeed , other members of the KChIP family not expressed in the myocardium behave as transcriptional repressors , while also maintaining the ability to interact with Kv4 channels ( An et al . , 2000; Carrión et al . , 1999; Savignac et al . , 2005; Gomez-Villafuertes et al . , 2005; Ronkainen et al . , 2011 ) . Therefore , we sought to identify the existence of cardiac KChIP2 transcriptional activity and its significance in electrical remodeling and arrhythmia susceptibility . Here , we find KChIP2 transcriptionally represses a set of miRNAs known as miR-34b and-34c . Through KChIP2 loss , miR-34b/c are elevated , subsequently targeting other ion channel genes defining INa and Ito densities . Either restoring KChIP2 expression or blocking miR-34b/c activity during cardiac stress reverses this remodeling and completely negates the occurrence of re-entrant arrhythmias . Together , this work unveils a novel , transcriptional mechanism for KChIP2 , and defines it as a central mediator of cardiac electrical activity . This study was approached with the knowledge that acute KChIP2 loss affected the SCN5A ( Nav1 . 5 ) , SCN1B ( Navβ1 ) , and KCND3 ( Kv4 . 3 ) genes in a manner suggesting miRNA activity ( Deschênes et al . , 2008 ) . We therefore performed a miRNA microarray following KChIP2 silencing in neonatal rat ventricular myocytes ( NRVMs ) , resulting in the induction of a number of miRNAs ( Figure 1A ) . We evaluated the miRNAs that achieved at least a two fold increase ( Figure 1B ) using TargetScan 7 . 1 ( Lewis et al . , 2005 ) to identify potential targeting to the mRNAs SCN5A , SCN1B , and KCND3 . Ultimately , we identified miR-34b and −34c as the only miRNAs predicted to target not just one of these ion channel genes , but notably target all three collectively ( Figure 1C ) . Notably , we also observed 14 miRNAs decreased greater than two fold ( Figure 1B ) . However , a loss in miRNA expression is not consistent with the role of KChIP2 as a transcriptional repressor , and also would not lead to a decrease in ion channel mRNA expression . Real-time qPCR was used to confirm the array results , showing elevation in the mature transcripts for miR-34b and −34c ( Figure 1D ) . Importantly , we also performed overexpression of three different cardiac KChIP2 isoforms which reduced the expression of miRs-34b/c ( Figure 1D ) . Together , these changes are consistent with the novel idea that KChIP2 behaves as a transcriptional repressor . 10 . 7554/eLife . 17304 . 003Figure 1 . miR-34 regulation linked to changes in KChIP2 expression . ( A ) Results of miRNA microarray showing the log2 of the fold changes in miR expression following 72 hr of KChIP2 siRNA treatment . Arrow identifies miR-34b and −34c amongst the panel of altered miRNAs . Analysis of miRNAs for mRNA targets using TargetScan 7 . 1 was restricted to those above two fold induction ( dashed line ) ( B ) Tables showing the list of those miRNAs showing at least a two fold increase or decrease following KChIP2 silencing . ( C ) Alignment of the 3’-UTR of SCN5A , SCN1B , and KCND3 genes with miRs-34b/c from rat , showing hybridization of the seed region . Grayed letters indicate variation in sequence between miR-34b and −34c . A single site of interaction is indicated for SCN5A and SCN1B while two sites exist for KCND3 . ( D ) Real-time qPCR analysis showing percent change of miR-34b/c expression from control cells in NRVM transfected with KChIP2 . 3 ( n = 5 ) , KChIP2 . 6 ( n = 6 ) , KChIP2 . 4 ( n = 4 ) , or KChIP2 siRNA ( n = 4–5 ) . ( E ) Cytosolic , membrane , and nuclear fractions of native adult rat heart tissue . KChIP2 nuclear localization was assessed by using lactate dehydrogenase ( LDH ) , Serca2a , and Lamin-B as cytoplasmic , membrane , and nuclear markers respectively . ( F ) Representative z-stack images of adult rat ventricular myocyte . Nuclear stained regions ( DAPI , blue ) show the absence of cytosolic protein LDH ( green ) , while KChIP2 ( red ) staining reveals significant colocalization . Data presented as mean ± SEM . p<0 . 05; p<0 . 01 , compared to control . DOI: http://dx . doi . org/10 . 7554/eLife . 17304 . 003 Because KChIP2 is dominantly known as a modulator of Kv4 channels , with cytoplasmic localization ( Takimoto et al . , 2002 ) , we addressed whether it could also localize to the nucleus where it could act as a transcriptional regulator . Indeed , fractionation of adult rat cardiomyocytes into nuclear fractions reveals endogenous KChIP2 nuclear expression in the absence of contaminating cytosolic ( lactate dehydrogenase ) and membrane associated proteins ( Serca2a ) ( Figure 1E ) . This is reinforced in the localization patterns of adult cardiomyocytes , showing marked endogenous KChIP2 colocalization in the nucleus in the absence of the cytosolic marker lactate dehydrogenase ( Figure 1F ) . To assess whether the transcriptional changes seen in miR-34b/c were the consequence of KChIP2 activity on the promoter , a luciferase assay was conducted containing the cloned minimal miR-34b/c promoter in the presence of KChIP2 . Notably , both miR-34b and −34c are transcribed in tandem under the regulation of a shared , intergenic promoter ( Toyota et al . , 2008 ) . To identify potential DNA binding locations for KChIP2 , we borrowed from what is known about the putative nucleotide binding sequence for the transcriptional repressor DREAM ( KChIP3 ) . This member of the KChIP family shares a high degree of homology with KChIP2 , but more importantly has known transcriptional activity occurring through interaction with a nucleotide sequence known as the DRE motif ( Carrión et al . , 1999 ) . MatInspector software ( Cartharius et al . , 2005 ) was used to evaluate the miR-34b/c promoter for occurrences of this motif , revealing a potential site beginning 254 bp upstream of the miR-34b stem-loop ( Figure 2A ) . A region of the promoter 500 bp to 191 bp upstream of the miR-34b stem-loop was cloned into the pGL4 . 10 luciferase vector and co-transfected with several KChIP2 isoforms into cos-7 cells . When compared to a GFP transfected control without KChIP2 , we observed significant repression in the presence of KChIP2 . 3 , 2 . 4 , and 2 . 6 ( Figure 2A ) , showing that KChIP2 can directly act on the miR-34b/c promoter to impart repressive action . To determine if physical KChIP2 interaction with the promoter mediates the repressive state , native adult rat cardiomyocytes were used to perform chromatin immunoprecipitation , followed by qPCR with a primer set flanking the identified DRE site . KChIP2 pull-down resulted in significant enrichment of the DRE containing PCR fragment when compared to an IgG control ( Figure 2B ) . 10 . 7554/eLife . 17304 . 004Figure 2 . KChIP2 represses miR-34b/c expression by direct interaction with a putative DRE motif in promoter . ( A ) A region from −500 to −191 of the miR-34b/c promoter was cloned into the promoterless luciferase construct , pGL4 . 10 . This construct was co-transfected into COS-7 cells in the presence of KChIP2 . 3 ( n = 3 ) , KChIP2 . 6 ( n = 8 ) , or KChIP2 . 3 ( n = 3 ) and compared to GFP alone . Renillin ( pGL4 . 74 ) was used as a normalization control . Results are depicted as a % change in activity compared to GFP alone . ( B ) IgG and KChIP2 ChIP-PCR conducted on native adult rat cardiomyocytes . The target primer site residing within the cloned promoter was evaluated for enrichment following pull down ( n = 3 ) , showing significant enrichment of the target region . ( C ) Luciferase assay conducted in COS-7 cells to evaluate the outcome of deleting the putative DRE site in the miR-34b/c promoter . COS-7 cells were transfected with the same WT reporter construct inserted into the pGL4 . 10 vector or with the DRE motif deleted , both in the presence of KChIP2 . 6 . Activity was normalized to renillin ( pGL4 . 74 ) . Deletion of a putative KChIP2 interaction site ( DRE motif ) partially abolished the repressive effect KChIP2 . 6 had over the miR-34b/c promoter ( n = 4 ) compared to WT ( n = 9 ) . ( D ) COS-7 cells transfected with KChIP2 . 6 and the pGL4 . 10 containing the WT miR-34b/c promoter were treated with or without 10 mM caffeine for 6 hr , leading to promoter activation ( n = 4 ) . Results were normalized to renillin activity . Data presented as mean ± SEM . p<0 . 05; *p<0 . 01 , as indicated or compared to control . DOI: http://dx . doi . org/10 . 7554/eLife . 17304 . 004 To identify if the DRE site within the promoter fragment is responsible for the repression caused by KChIP2 , the core nucleotide sequence was deleted from the promoter ( Figure 2C ) . This attenuated the repressive action of KChIP2 , implying that KChIP2 is capable of recognizing the same putative DNA binding motifs as DREAM and uses it to induce repressive action . Additionally , it is known that transcriptional derepression of DREAM is regulated through Ca2+ binding to EF-hand motifs ( Carrión et al . , 1999 ) . Therefore , to further characterize KChIP2 activity , the reporter assay was conducted following incubation with 10 mM caffeine to induce global elevations in Ca2+ . This led to significant activation of the promoter ( Figure 2D ) , reinforcing the transcriptionally repressive nature of KChIP2 and its conserved mechanisms with DREAM . Together , this data demonstrates that KChIP2 behaves as a transcriptional repressor on the promoter of miR-34b/c by direct binding to the putative DRE motif . Previous studies identified reduction in Nav1 . 5 , Navβ1 , and Kv4 . 3 following KChIP2 silencing ( Deschênes et al . , 2008 ) . Having observed that KChIP2 knock-down elevates miR-34b/c , we next sought to determine whether miR-34b/c targets these ion channel transcripts to mediate their loss in expression . Precursor miRNAs for miRs-34b/c were transfected into NRVMs to directly elevate their expression . Assessment of the resulting transcripts showed reduced mRNA for SCN5A and SCN1B following miR-34 expression , compared to a non-targeting control miR ( Figure 3A ) . While KCND3 levels remained unchanged ( Figure 3A ) , Kv4 . 3 protein experienced significant reduction that reinforces the miRNA mode of translational inhibition without mRNA degradation previously noted ( Deschênes et al . , 2008 ) ( Figure 3B and C ) . 10 . 7554/eLife . 17304 . 005Figure 3 . Cardiac ion channel directly regulated by miR-34a/b/c through interaction with their 3’-UTR . ( A ) NRVM over-expressing precursors for miR-34b/c were collected for mRNA transcript levels . Results ( normalized to non-targeting miR ) show down-regulation of SCN5A , and SCN1B , but unchanged levels for KCND3 ( n = 7–8 ) . ( B ) Protein levels from NRVM with over-expressed miR-34b showing reduced protein expression for Kv4 . 3 ( KCND3 ) . Multiple bands for Kv4 . 3 represent different glycosylation states of the protein . ( C ) Summary data of the immunblot for Kv4 . 3 ( n = 4 ) . ( D ) Alignment of the 3’-UTR of SCN5A , SCN1B , and KCND3 genes with miRs-34b/c , with mutations made to the seed regions ( highlighted in red ) to disrupt interaction at the seed region . ( E ) Reporter assay with the 3’-UTR cloned into pmiRGlo reporter construct . Luciferase activity in HEK cells transfected with WT or mutant 3’-UTRs . Results are presented as a percent change from a non-targeting miR precursor ( n = 5 ) normalized to renillin activity . ( F ) I/V curves for INa measured in NRVM over-expressing precursors for control ( n = 24 ) , miR-34b ( n = 24 ) , or miR-34c ( n = 18 ) . ( G ) I/V curves for Ito , total measured in NRVM over-expressing precursors for control ( n = 15 ) , miR-34b ( n = 12 ) , or miR-34c ( n = 11 ) . ( H ) Ito , f was also assessed in NRVM through kinetic subtraction of Ito , s . Resulting I/V curves now reveal a significant reduction in current density in miR-34b ( n = 14 ) and miR-34c ( n = 15 ) precursor treated cells compared to control ( n = 16 ) . ( I ) The same experiments conducted in human derived cardiomyocytes ( iCells ) expressing either control or miR-34b/c precursor together , measuring INa ( control , n = 24; miR-34b/c , n = 21 ) and ( J ) Ito , total ( control , n = 24; miR-34b/c , n = 25 ) . Data presented as mean ± SEM . p<0 . 05; *p<0 . 01 , as indicated or compared to control . See also Figure 3—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 17304 . 00510 . 7554/eLife . 17304 . 006Figure 3—figure supplement 1 . Kv4 . 2 ( kcnd2 ) expression in NRVM following expression of miR-34b/c precursor . RT-qPCR detection of Kv4 . 2 ( kcnd2 ) following over-expression of miR-34b/c precursors expressed in NRVM . Results reflect fold changes relative to a control miR-precursor ( n = 7 ) . While the elevation in kcnd2 following miR-34b/c over-expression is not significant , a strongly trended elevation in suggests compensatory upregulation of Kv4 . 2 , contributing to the minimal loss of Ito in NRVM , despite significant Kv4 . 3 loss . Data presented as mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 17304 . 006 To determine if the changes in channel expression was the consequence of miR-34 targeting to the 3’-UTR of these genes , and not the result of an indirect pathway , fragments of the 3’ region containing the seed sequence were fused to the end of a luciferase reporter construct . This construct was co-expressed with the miR-34b/c precursors in HEK293 cells , resulting in reduced activity in all three constructs when compared to a control miR-precursor ( Figure 3E ) . Subsequently , mutations were made within the seed region where miR-34 targeting is predicted to bind ( Figure 3D ) , which significantly attenuated the repressive action ( Figure 3E ) . This suggests that miR-34b/c are indeed targeting the predicted seed region in the SCN5A , SCN1B , and KCND3 genes and directly influencing their expression . Functional assessment of changes to INa and Ito were determined through patch clamp recordings in NRVM . Reflecting the changes in mRNA and protein , expression of miR-34b/c precursor produced a significant decline in INa ( Figure 3F ) . Ito , however , while having trended reductions , did not produce significant loss despite the loss in Kv4 . 3 protein levels ( Figure 3G ) . This can be attributed to a number of reasons . The current evaluation was conducted in rodent myocytes , where Ito is comprised of the shared alpha subunits Kv4 . 2 and Kv4 . 3 , which comprise a fast component of Ito referred to as Ito , f . Additionally , there are the contributions of Kv1 . 4 , another potassium channel subunit , which encodes a slow component , referred to as Ito , s . These descriptions are attributed to the respective rates of recovery from inactivation for each of these channels ( Niwa and Nerbonne , 2010 ) . Notably , our patch protocol in Figure 3G took into account the contributions of all three subunits , or Ito , total . Therefore , despite reductions in Kv4 . 3 protein expression the change in current resists as it is not the predominant channel contributing to Ito . Importantly , Kv4 . 2 and Kv1 . 4 do not contain a miR-34 seed region . In fact , in response to miR-34b/c expression , mRNA levels for Kv4 . 2 actually experienced a trended elevation ( Figure 3—figure supplement 1 ) which could also contribute to the lack of reduction in Ito . We therefore modified our patch protocol to probe just Ito , f and remove the contribution of Kv1 . 4 , which now revealed a significant reduction in the Ito , f density ( Figure 3H ) . To further identify if the presence of Kv4 . 2 , Kv4 . 3 , and Kv1 . 4 in rats could explain the resisted change in Ito , cardiomyocytes derived from human induced pluripotent stem cells ( iCells ) were used , since in the human background , Kv4 . 3 is the dominant contributor to Ito ( Niwa and Nerbonne , 2010 ) . Expression of miR-34b/c precursors now produced a significant loss in Ito density ( Figure 3J ) , while also maintaining reductions in INa ( Figure 3I ) . Importantly , this not only satisfies why Ito loss was resistant in the NRVMs , but identifies conservation of miR-34 activity across species , implicating the importance of miR-34s in human cardiac ion channel regulation . To begin to understand the pathogenic importance of this pathway , NRVMs were cultured in 100 μM phenylephrine ( PE ) for 48 hr to mimic neuro-hormonal overload in a stressed myocardium . PE stimulation resulted in a dramatic decline of KCNIP2 ( KChIP2 ) , while also yielding significant elevation in miR-34b/c ( Figure 4A and B ) . These conditions resulted in reductions in expression for SCN5A , SCN1B , and KCND3 transcripts ( Figure 4C ) . Critically , maintaining KChIP2 levels through use of adenovirus encoding KChIP2 ( Ad . KChIP2 ) normalized the expression of miRs-34b/c while reversing the loss in SCN5A and SCN1B; however , KCND3 levels remained suppressed ( Figure 4B and C ) . Functional evaluation on both INa and Ito , f shows significant loss in density following PE treatment ( Figure 4D and E ) , reflecting the changes we see in transcript expression and mimicking ion channel remodeling observed in HF . However , Ad . KChIP2 treatment restored the current density for both currents , despite KCND3 transcript expression being unaffected by KChIP2 expression . These observations strongly implicate a role for KChIP2 in maintaining proper electrical expression during pathological remodeling in the stressed heart . Moreover , we were able to observe significant reduction of KChIP2 and elevation of miR-34b/c within failing human heart tissue compared to non-failing ( Figure 5A ) . Reinforcing this conservation was the identification of a predicted DRE motif proximal to the transcriptional start site , in the human miR-34b/c promoter as evaluated by MatInspector ( Figure 5B ) . At the same time , significant loss of SCN5A and KCND3 transcripts in failing tissue ( Figure 5C ) also show conservation of miR-34b/c targeting within their 3’-UTRs ( Figure 5D ) . Interestingly , SCN1B does not preserve its target site in humans , however , we also observed no significant reduction in transcript expression from failing heart tissue ( Figure 5C ) . Together , this reinforces the concept of KChIP2 as a core transcriptional regulator of electrical activity under normal and pathologic conditions . 10 . 7554/eLife . 17304 . 007Figure 4 . In vitro cardiac disease signaling links KChIP2 loss with miR-34 elevation . ( A ) Real-time qPCR evaluation of relative kcnip2 following treatment with 100 μM PE for 48 hr in NRVM ( n = 6 ) . Results normalized to ribosomal protein RPL27 . ( B ) Evaluation of miR-34b ( n = 8 ) and miR-34c ( n = 7 ) relative expression in NRVM under control ( no PE with Ad . GFP ) , 100 μM PE with Ad . GFP , or 100 μM PE with Ad . KChIP2 to maintain KChIP2 expression during the 48 hr treatment . Expression levels were normalized to small nucleolar RNA , U87 . ( C ) The same treatment conditions in ( B ) , evaluating relative mRNA expression for SCN5A ( n = 10 ) , SCN1B ( n = 10 ) , and KCND3 ( n = 7 ) . ( D ) Functional current-voltage measurements of INa from NRVM under control ( n = 29 ) , PE+Ad . GFP ( n = 27 ) , and PE+Ad . KChIP2 ( n = 30 ) . ( E ) I/V curve for Ito , f recordings in control ( n = 7 ) , PE+Ad . GFP ( n = 9 ) and PE+Ad . KChIP2 ( n = 9 ) . Data presented as mean ± SEM . p<0 . 05 , *p<0 . 01 , as indicated or compared to control , #p<0 . 05 , compared to PE+Ad . GFP . DOI: http://dx . doi . org/10 . 7554/eLife . 17304 . 00710 . 7554/eLife . 17304 . 008Figure 5 . Preservation of the KChIP2/miR-34b/c axis in human heart failure . ( A ) Human tissue taken from the left ventricle of non-failing ( NF ) ( n = 8 ) and failing patients ( n = 20 ) evaluating KChIP2 and miR-34b/c RNA expression . KChIP2 levels were normalized to GAPDH and miR expression to small nucleolar RNA U6 . ( B ) Evaluation of the human miR-34b/c reveals a conserved DRE motif in proximity of the miR-34b stem loop ( −242 bp ) , as predicted by MatInspector , suggesting conservation of KChIP2 activity in the regulation of miR-34b/c expression . ( C ) Human heart failure tissue evaluating RNA levels for SCN5A , SCN1B , and KCND3 . Significant reductions in heart failure samples ( n = 20 ) were observed for SCN5A and KCND3 , but not for SCN1B , compared to non-failing tissue ( n = 8 ) . ( D ) Alignment of the 3’-UTR of SCN5A , SCN1B , and KCND3 genes with miRs-34b/c from human . Grayed letters indicate variation in sequence between miR-34b and −34c . A single site of interaction is indicated for SCN5A , matching observations in the rat , while KCND3 has three potential sites , compared to two observed in the rat . Notably , SCN1B miR-34b/c targeting is not conserved in human shown by imperfect hybridization in the seed region . Data presented as mean ± SEM . p<0 . 05; *p<0 . 01 , as indicated or compared to control . #p<0 . 05 , ##PP<0 . 01 compared to PE+Ad . GFP . DOI: http://dx . doi . org/10 . 7554/eLife . 17304 . 008 To address the specific activity of miR-34b/c in mediating these changes in ion channel expression , NRVMs and iCells were transfected with miR-34b/c antimir molecules during the duration of PE treatment . Much like KChIP2 delivery which reduced miR-34b/c expression , directly blocking miR-34b/c activity maintained INa in both rat and iCells ( Figure 6A and D ) , further implicating miR-34b/c in the direct regulation of these ion channel transcripts . However , Ito , total density in the NRVMs did not observe the same rescue ( Figure 6B ) . We believe this is once again explained by the contributions of Kv1 . 4 and Kv4 . 2 , in addition to Kv4 . 3 in defining rodent Ito . In fact , by probing just Ito , f , we revealed a significant , but incomplete restoration following miR-34b/c block ( Figure 6C ) . Notably , the same experiment conducted in iCells where Kv4 . 3 is the dominant subunit , resulted in the full restoration of Ito ( Figure 6E ) . To be more certain the restoration of current density was specific to miR-34b/c targeting the underlying subunits encoding INa and Ito , rather than a general rescue in the molecular state of the cell , the repolarizing current IKr was assessed in iCells . PE successfully reduced this current , which is known to be reduced by cardiac stressors , however , it was unable to be rescued by miR-34b/c block ( Figure 6—figure supplement 1 ) . Critically , this shows that KChIP2 regulation of INa and Ito is enacted through specific targeting of miR-34b/c activity , while the use of iCells displays mechanistic conservation in human derived cells . 10 . 7554/eLife . 17304 . 009Figure 6 . miR-34 block reverses loss of both INa and Ito in disease signaling . ( A ) INa I/V curve measured in NRVM transfected with either non-targeting antimirs ( control , n = 26 ) , non-targeting miR + 100 μM PE ( PE+control antimir , n = 20 ) , or miR-34b/c antimirs + 100 μM PE ( PE+miR-34 antimir , n = 21 ) for 48 hr . ( B ) Ito , total I/V measurements in NRVM showing current density is lost in PE+control ( n = 16 ) and remains down in the PE+miR-34 antimir ( n = 16 ) , compared to control ( n = 17 ) cells . ( C ) Ito , f I/V measurements in NRVM . Cells treated with PE+control antimir ( n = 22 ) have reduced current density , that is now partially restored in the PE+miR-34 antimir ( n = 23 ) cells compared to control ( n = 27 ) . ( D ) I/V curve for INa taken in iCells showing that miR-34 antimirs ( n = 6 ) can rescue current density back toward control ( n = 6 ) , when compared to PE+control ( n = 6 ) . ( D ) I/V curve for Ito , total measurements in iCells showing miR-34b/c antimir in the presence of PE ( n = 15 ) can rescue current density towards control ( n = 15 ) while PE+control ( n = 15 ) remains reduced . Data presented as mean ± SEM . p<0 . 05 versus control , *p<0 . 01 , as indicated or compared to control antimir , #p<0 . 05 , ##p<0 . 01 compared to PE+control antimir . See also Figure 6—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 17304 . 00910 . 7554/eLife . 17304 . 010Figure 6—figure supplement 1 . IKr is insensitive to miR-34 block following PE stimulation . ( A ) IKr I/V curve measured in iCells transfected with either non-targeting antimirs ( control , n = 6 ) , non-targeting miR + 100 μM PE ( PE+control antimir , n = 6 ) , or miR-34b/c antimirs + 100 μM PE ( PE+miR-34 antimir , n = 6 ) for 48 hr . Lack of restoration suggests specificity of miR-34b/c targeting to specific ion channel transcripts . Data presented as mean ± SEM . p<0 . 05 , *p<0 . 01 , versus control . DOI: http://dx . doi . org/10 . 7554/eLife . 17304 . 010 Dysregulation of INa and Ito have been previously associated with arrhythmogenesis ( Starmer et al . , 2003; Kuo et al . , 2001 ) . Therefore , in order to test the consequence of INa and Ito loss and the involvement of miR-34b/c in regulating susceptibility to arrhythmic events , optical mapping was performed in NRVM monolayers . As before , cells were exposed to 100 μM PE for 48 hr following treatment with either a control or miR-34b/c antimir . Using point stimulation , we submitted the monolayers to baseline pacing ( S1 ) followed by a single premature stimulus ( S2 ) over a range of S1-S2 coupling intervals . Immediately following S2 capture , the occurrence of rapid , non-paced activity ( arrhythmia ) was assessed . Figure 7A shows representative activation maps during S1 ( top ) and S2 ( bottom ) pacing . In all conditions , activation during S1 pacing shows uniform wavefront propagation , with evidence of conduction slowing following PE + control antimirs , consistent with reduced INa density . Compared to S1 pacing , propagation during S2 pacing was slower in all conditions; however , in PE + control antimirs significant impulse slowing ( isochrone crowding ) and block ( solid line ) were observed . Critically , this block was sufficient to cause sustained reentrant excitation in 5 of 7 monolayers ( Figure 7B and C ) . Remarkably , PE + miR-34b/c inhibition prevented conduction block and mitigated conduction slowing , protecting all monolayers from sustained re-entry ( Figure 7C ) . 10 . 7554/eLife . 17304 . 011Figure 7 . miR-34 block retains excitability in NRVM monolayers following prolonged PE treatment . ( A ) Isochronal conduction maps of monolayers submitted to PE ( 100 μM ) with either a non-targeting control or miR-34b/c antimir . Conduction maps on the top row represent the final S1 ( 750 ms ) preceding the S2 , showing no pre-existing abnormalities in propagation . The square function represents the site of pacing . The second row shows the first incidence of capture of the premature stimulus ( S2 ) . PE + control antimir results in significant conduction block around the pacing site ( solid line ) . Conduction block was minimal in control and PE + miR-34b/c antimir groups . ( B ) Conduction map showing an example of sustained reentry for the PE + control antimir treated group shown in ( A ) . ( C ) Summary data for the occurrence of sustained reentry following S1S2 pacing . ( D ) Restitution curve of APD80 in paced NRVM monolayers treated with either control antimir ( n = 6–8 ) , PE + control antimir ( n = 6–11 ) , or PE + miR-34b/c antimir ( n = 7–12 ) . ( E ) Conduction velocity restitution curve in paced NRVM monolayers treated with either control antimir ( n = 6–8 ) , PE + control antimir ( n = 6–11 ) , or PE + miR-34b/c antimir ( n = 17–12 ) . ( F ) Measurement of the effected refractory interval evaluated by identifying the shortest premature stimulus that would elicit capture or arrhythmia induction , under control ( n = 6 ) , PE+control antimir ( n = 13 ) , and PE+miR-34b/c antimir ( n = 12 ) . Data presented as mean ± SEM . p<0 . 05 , **p<0 . 01 , as indicated or compared to control antimir . DOI: http://dx . doi . org/10 . 7554/eLife . 17304 . 011 To determine the electrophysiological substrate responsible for the reentrant activity observed , monolayers were evaluated for changes in APD and conduction velocity . Reflecting the changes in INa and Ito expression , exposure to PE for 48 hr significantly prolonged APD and slowed conduction velocity compared to control dishes across multiple pacing cycle lengths ( Figure 7D and E ) . APD prolongation from PE was unresponsive to miR-34b/c inhibition; however , this was anticipated as we previously determined Ito is not restored by miR-34b/c block in NRVM due to the additional Kv1 . 4 and Kv4 . 2 mediated current . However , treatment with miR-34b/c antimir , which maintained INa density in isolated myocytes ( Figure 6A ) , produced a trend towards restoration of conduction velocity , suggesting other mechanisms of conduction slowing following PE treatment that are uninfluenced by miR-34b/c activity . Therefore , to more precisely assess changes in cellular excitability , we determined the effective refractory period ( ERP ) under each condition . Reflecting the prolonged APD and reduced INa , PE treated cells displayed a significantly longer ERP ( Figure 7F ) than control cells . However , treatment with the miR-34 antimir significantly shortened ERP towards control . Notably , this recovery occurred in the absence of a shortened APD , suggesting a significant recovery of INa excitability . Thus , even without being able to rescue Ito , we were still able to restore cellular excitability through miR-34b/c inhibition and limit the occurrence of conduction block and reentry . Overall , the observation KChIP2 can normalize electrical remodeling in a setting of myocardial stress highlights a much expanded and multimodal role in establishing the cardiac electrical state . This study established a novel transcriptional role for cardiac KChIP2 , whereby it maintains a repressive influence over the miR-34b/c promoter . KChIP2 loss either by direct silencing or pathologic means , removes repression over miR-34b/c expression . Consequentially , reductions in transcript and protein expression for Nav1 . 5 , Navβ1 , and Kv4 . 3 are observed as an outcome of miR-34b/c targeting to seed regions present in the 3’-UTR of these genes , allowing KChIP2 to manipulate functional expression of a host of critical cardiac ion channel genes , ultimately acting as a key regulator of cardiac excitability and arrhythmia susceptibility . While we evaluated a discrete pathway targeted by KChIP2 transcriptional activity , there are doubtless many other gene targets . To begin to address this discussion , a gene expression array was performed on NRVM following 48 hr of KChIP2 silencing . Evaluation of genes that experienced at least a two fold change revealed an increase in expression for 293 genes and a decrease in expression for 407 genes in response to KChIP2 silencing ( Supplementary file 1 ) . Notably , of the genes experiencing increased expression , 192 of them ( 65 . 5% ) were predicted to contain a DRE motif within promoter elements , implicating the potential of KChIP2 transcriptional activity directly mediating these changes . Additionally , of the genes responding with reduced expression , 71 of them ( 17 . 4% ) were predicted to contain a miR-34b/c target site within their 3’-UTR . Importantly , we see significant reduction to SCN5A and SCN1B , consistent with previous data and the mechanisms investigated here . Considerably , whether these changes are the direct consequence of KChIP2 transcriptional repression , or more indirect KChIP2 dependent mechanisms , including prolonged APD from loss in Ito density contributing to altered Ca2+ handling , the results are still relevant to cardiac remodeling , particularly given the associated loss of KChIP2 in cardiac disease states . Notably , a diverse range of gene pathways were implicated in response to KChIP2 loss , including cardiovascular signaling , regulation to G-protein coupled receptor pathways , relaxation and contraction , TGFβ signaling , apoptotic , and NFκB dependent signaling mechanisms , all of which have a relevance in disease remodeling ( Supplementary file 1 ) . Importantly , these data are supportive of our conclusion that KChIP2 is a key regulator of cardiac pathology . While our study focused on the transcriptional pathway by which KChIP2 could exert a concerted regulation of INa and Ito , further investigations pursuing some of the targets highlighted from this gene expression array will likely reveal a broader role for KChIP2 as a transcriptional regulator of cardiac physiology much like is seen for KChIP3 ( DREAM ) in the brain . Indeed , the role of KChIP2 as a multimodal regulator of cardiac ion channels has been an emerging topic . Recent work has identified KChIP2 regulation of Cav1 . 2 through direct interaction with an inhibitory N-terminal domain on the channel , effectively reducing ICa , L in the absence of KChIP2 ( Thomsen et al . , 2009; Foeger et al . , 2013 ) . Our previous work has also suggested that KChIP2 is part of a larger macromolecular complex that includes subunits for both Nav1 . 5 and Kv4 channels that lead to functional increases in both currents when coexpressed with KChIP2 ( DeschenesDeschênes et al . , 2008 ) . In the same study we also identified transcriptional changes in SCN5A and SCN1B following acute knockdown of KChIP2 in NRVM , which provided the motivational basis for the work presented here . Taken together , these observations suggest a highly promiscuous nature for KChIP2 . Notably , another member of the KChIP family , KChIP3 , has also been discovered to interact with multiple membrane proteins , including regulation of Kv4 channels ( Mellström et al . , 2008; Buxbaum et al . , 1998 ) , while also displaying Ca2+ regulated transcriptional repression ( Carrión et al . , 1999 ) . Even the role of KChIP3 as a transcriptional repressor is multimodal , including direct binding to DRE motifs , in addition to interacting with and suppressing the activity of the cAMP response element-binding protein ( CREB ) , an established transcriptional activator ( Ledo et al . , 2002 ) . Both of these processes are Ca2+ regulated , due to three functional high affinity EF-hand motifs residing in the protein ( Carrión et al . , 1999 ) . Occupancy of these sites upon increases in intracellular Ca2+ lead to conformational changes that cause DNA binding release ( Carrión et al . , 1999 ) or dissociation from CREB ( Ledo et al . , 2002 ) , causing de-repression of downstream gene targets . Given that the entirety of the KChIP gene family displays strong conservation around these EF-hand residues , suggests conservation of these Ca2+ regulated responses . Indeed , we observed that caffeine stimulation produced increased activity of the miR-34b/c promoter ( Figure 2D ) in the presence of KChIP2 . Additionally , when we deleted the DRE element in the miR-34b/c promoter , we observed an incomplete removal of suppression ( Figure 2C ) . However , as KChIP3 represses gene expression through alternative CREB dependent regulation , the partially retained repressive activity may be attributed to this secondary function . Further analysis of the promoter by MatInpsector revealed several potential sites of predicted CREB binding that may have allowed for partially maintained KChIP2 suppression , even in the absence of the DRE site . Given that KChIP2 and KChIP3 share a high degree of homology only reinforces the observation of multiple activities for KChIP2 as well . The physiologic implications of KChIP2 targeting miR-34b/c expression is one of tremendous significance for many cardiac pathologic states . Rapid depletion of KChIP2 protein is a widespread event that underlies remodeling in many cardiac diseases , including chronic HF , MI , and atrial fibrillation ( Nattel et al . , 2007 ) . Considerably , these diseases also present with reductions in Ito and INa . The relationship between KChIP2 and Ito has been heavily studied , frequently identifying that KChIP2 loss induces the destabilization of Kv4 . 2/4 . 3 channels and mediates the decline in current density ( Foeger et al . , 2013 ) . However , the work presented here offers the unique alternative that translation block through miRNA interaction mediates the decline in Kv4 . 3 . Given that Kv4 . 2 does not contain a miR-34 target region in its 3’-UTR , but still experiences degradation following KChIP2 loss , it is likely that both mechanisms contribute to the resulting loss in Ito , f . However , it is also observed that reduced KChIP2 expression stimulated by phenylephrine + propranolol in in vitro cultures of NRVM experienced increased Kv4 . 2 protein while KChIP2 and Kv4 . 3 levels were reduced ( Panama et al . , 2011 ) , supporting the opportunity for miRNA dependent translational block targeting Kv4 . 3 , rather than just destabilization of all Kv4 channels . In the same settings of cardiac disease where KChIP2 is down , there are also observations of INa depletion ( Valdivia et al . , 2005; Zicha et al . , 2004 ) . Our data of miR-34 targeting Nav1 . 5 provides a means for describing this loss in activity . Notably , others have shown a loss in the full length transcript for Nav1 . 5 mRNA and a corresponding increase in a truncated isoform without the miR-34b/c target region present ( Shang et al . , 2007 ) , reinforcing the observations for miR-34b/c mediating the decline of SCN5A . Overall , the consequential loss of both INa , and Ito , suggests KChIP2 loss during cardiac stress may be a nodal event in a cascade of gene regulation defining electrical remodeling in the stressed myocardium . Indeed , earlier work was done that sought to determine the significance of KChIP2 in the development of hypertrophic remodeling . In a rat TAC banding model , it was observed that maintaining KChIP2 expression attenuated hypertrophy and pathogenic remodeling that otherwise lead to a worsening myocardium during pressure overload ( Jin et al . , 2010 ) . This reverse in remodeling was attributed to changes in intracellular Ca2+ signaling brought on by restoration of an abbreviated APD . Yet , we were able to observe that inhibition of miR-34b/c could also attenuate adverse remodeling without influencing APD ( Figure 7 ) implicating multiple pathways of KChIP2 intervention . Indeed , the miR-34 family has recently been implicated in the development and progression of hypertrophy and heart failure , in rodent models of both MI and pressure overload ( Bernardo et al . , 2012 ) . Critically , these studies , combined with our own data , show that blockade of the miR-34 family can attenuate pathologic remodeling , expanding the significance of KChIP2 and miR-34 in cardiac pathogenesis . There are still some challenges in understanding the role of KChIP2 in the progression of hypertrophy and heart failure . Investigations conducted in KChIP2 null mice have shown that when submitted to TAC banding , there is no worsened phenotype when compared to wild type mice ( Speerschneider et al . , 2013 ) . In fact , arrhythmia susceptibility was lowered in the KChIP2 null mice during heart failure , believed to be the result of reduced dispersion of repolarization . At the same time , there were no observed changes to INa . While our current understanding is unable to account for this disparity , it may be that compensatory regulation exists in these mice as a consequence of constitutive KChIP2 absence during development , fundamentally changing its regulatory significance . Evidence for this is observed when restoring KChIP2 expression in myocytes isolated from KChIP2 null mice , which resulted in no rescue of Kv4 . 2 protein expression or recovery of Ito , f ( Foeger et al . , 2013 ) . However , restoration of KChIP2 following acute loss from pathologic consequences in a rat model was able to rescue Ito , f ( Jin et al . , 2010 ) , consistent with what we see in our own maintenance of KChIP2 following prolonged PE exposure ( Figure 4E ) . The significance of this begins to suggest deviations in KChIP2 regulatory impact depending on acute versus constitutive loss . Ultimately , our endpoint was to determine whether electrical dysregulation brought on by KChIP2 loss was able to influence arrhythmia susceptibility through the activity of miR-34b/c . Despite only rescuing INa and not Ito in the NRVMs , as evidenced by the shortened ERP with sustained APD prolongation ( Figure 7D and F ) , we found this was sufficient to rescue arrhythmia induction following PE treatment ( Figure 7C ) . Indeed , previous studies have revealed the relationship between changes in Na+ channel density and arrhythmia induction . As INa becomes compromised , it begins to resolve an expanding interval of premature stimuli declared the vulnerability period . Within this interval , reentry is more likely to occur as a result of non-uniform conduction block surrounding the point of excitation ( Starmer et al . , 2003 ) . Both theoretical ( Starmer et al . , 1991 , 1993 ) and experimental ( Nesterenko et al . , 1992; Starmer et al . , 1992 ) studies show that when Na+ channel availability is reduced , the vulnerable period increases . Therefore , by restoring Na+ channel through miR-34b/c inhibition , we are effectively minimizing the vulnerable period and making unidirectional conduction block less likely to occur . Care must still be taken before translating these mechanisms to the clinical setting . Our investigated pathway was developed using cultured rodent myocytes , differing from human electrophysiology in its APD and the impact of underlying currents . We must also understand the electrical impact of miR-34 inhibition in vivo . However , we know from this investigation that miR-34b/c are elevated in native human HF tissue ( Figure 5A ) , and that functionally , the inhibition of miR-34b/c in human derived cardiomyocytes following stress can achieve restoration of both INa and Ito ( Figure 6C and D ) , reinforcing species dependent conservation . At the same time , conduction block due to compromised cellular excitability has long been understood to be important for clinically relevant arrhythmias ( Shah et al . , 2005 ) . These observations together suggest strong therapeutic potential for targeting miR-34 in the treatment of electrical instabilities . Currently , the use of locked nucleic acids and related technologies have been used to successfully target miRNA activity in vivo ( Olson , 2014 ) . While miR-34b/c is also expressed outside the heart , it is unclear what long-term consequences its inhibition will have as a therapeutic . However , these outcomes will have to be weighed against the potential therapeutic advantage it will have in alleviating cardiac events . Overall , this newly identified KChIP2/miR-34 pathway reflects electrical remodeling observed within multiple cardiac pathologies . Moreover , the events brought on by KChIP2 loss are critical in initiating electrical instabilities and arrhythmias implicated in sudden cardiac death . The identification of KChIP2 transcriptional capacity significantly reshapes its role in cardiac biology as a core mediator of cardiac electrical activity and reveals KChIP2 and miR-34 as therapeutic targets for managing arrhythmogenesis in heart disease . Rat neonatal ventricular myocytes were isolated 1–2 days after birth as previously described ( Dennis et al . , 2011 ) . Briefly , hearts were minced in HBSS , and tissue fragments were digested overnight with trypsin at 4°C . Trypsinized fragments were treated repeatedly for short periods of time with collagenase at 37°C followed by trituration . Dissociated cells were pre-plated for 2 hr at 37°C in DMEM supplemented with 5% fetal bovine serum ( FBS ) and penicillin/streptomycin . NRVMs were collected and replated in DMEM/5% FBS/penicillin/streptomycin with 0 . 1 mM bromodeoxyuridine ( BrdU ) to suppress fibroblast growth and maintained at 37°C , 5% CO2 . These conditions were maintained for 24–36 hr , after which culture conditions deviated based on application of cells . Human-induced pluripotent stem cell ( hiPSC ) -derived cardiomyocytes ( iCell Cardiomyocytes; Cellular Dynamics International , Madison , WI ) were cultured in iCell Cardiomyocytes Maintenance Medium ( Cellular Dynamics International ) in an atmosphere of 93% humidified air and 7% CO2 at 37°C . For electrophysiological recordings , 20000–40000 cardiomyocytes were plated on glass coverslips coated with 0 . 1% gelatin as described ( Ma et al . , 2011 ) . Single ventricular myocytes were isolated from adult rat hearts . Briefly , rats were anesthetized by injection of ketamin . Hearts were quickly removed and perfused via the aorta with a physiological salt solution ( PSS ) containing ( in mmol/L ) NaCl 140 , KCl 5 . 4 , MgCl2 2 . 5 , CaCl2 1 . 5 , glucose 11 , and HEPES 5 . 5 ( pH 7 . 4 ) . After 5 min , perfusate was switched to a nominally calcium-free PSS with collagenase ( Roche , 0 . 5 mg/mL ) being added after an additional 5 min . After 15–20 min of digestion , hearts were perfused with a high K+ solution containing ( in mmol/L ) potassium glutamate 110 , KH2PO4 10 , KCl 25 , MgSO4 2 , taurine 20 , creatine 5 , EGTA 0 . 5 , glucose 20 , and HEPES 5 ( pH 7 . 4 ) . Ventricles were minced in high K+ solution , and single myocytes were obtained by filtering through a 115 μm nylon mesh . Myocytes were then plated on laminin coated coverslips for 1 . 5 hr before fixing with 4% formaldehyde in PBS to be used for immunohistochemistry . Alternatively , cells were resuspended in a 1% formaldehyde/PBS solution to be used for ChIP studies . NRVM cultures used for transfection and total RNA and protein collection were conducted on 35 mm dishes seeded with 1 . 5 × 106 cells . Following the initial 24–36 hr of plating , NRVMs were transfected with KChIP2 . 3 ( NM_173192 . 2 ) , KChIP2 . 4 ( NM_173193 . 2 ) , or KChIP2 . 6 ( NM_173195 . 2 ) for the overexpression of KChIP2 , which was inserted into the pIRES2-EGFP plasmid from Clontech as previously conducted ( Deschênes et al . , 2002 ) . The plasmid without the KChIP2 insert was used as the control . Lipofectamine 2000 reagent ( Invitrogen ) was used to deliver the constructs according to the manufacturer’s instructions . Following the transfection period , media was changed to DMEM/5% FBS/penicillin/streptomycin . Cells were cultured for 72 hr total before collection for total RNA , with a media change once after 48 hr of culture . Knockdown of KChIP2 was conducted by transfecting with siRNA for KChIP2 ( Ambion , Cat#: 4390771 , ID: s132782 ) , or a scrambled siRNA control ( Ambion , Cat#: 4390843 ) . 180 pmol of siRNA was transfected using 15 μL of Lipofectamine 2000 reagent according to the manufacturer’s instructions . Following the transfection period , media was changed to DMEM/5% FBS/penicillin/streptomycin . Cells were cultured for 72 hr total before collection for total RNA , with a media change once after 48 hr of culture . NRVM were also transfected with 180 pmol of miR-34b/c precursors ( miR-34b MC12558 , miR-34c MC11039 , Invitrogen ) or a non-targeting control ( negative control 4464058 , Invitrogen ) using 15 μl lipofectamine RNAi Max ( Invitrogen ) according to the manufacturer’s instructions . Cells were left for 48–72 hr and then collected for RNA . NRVM were also used for patch clamp recordings to measure INa and Ito . These were plated at 100 , 000 cells/dish in 35 mm dishes and the miR-precursors were modified with an attached FAM reporter to visualize transfected cells . 25 pmol of miR-34 precursor with 2 μl Lipofectamine RNAiMax was used according to the manufacturer’s instructions . Transfection of control or miR-34b/c antimirs were also used during the phenylephrine induction assays for evaluation with patch-clamp recordings in NRVM and iCells and optical mapping in NRVM only . NRVM seeded at 100 , 000 cells/35 mm dish for patch-clamping received 22 . 5 pmol of miR-34b inhibitor ( Invitrogen , MH12558 ) with 22 . 5 pmol of miR-34c inhibitor ( Invitrogen , MH11039 ) or 45 pmol of a non-targeting miR-inhibitor ( Invitrogen , 4464076 ) using 3 . 75 μl Lipofectamine RNAi Max ( Invitrogen ) . NRVM used for optical mapping were seeded at 1 . 5 × 106 cells/35 mm dish and 225 pmol each of the miR-34b and −34c inhibitor or 450 pmol of the non-targeting control miR-inhibitor were delivered using 22 . 5 μl of Lipofectamine RNAi Max . miR-inhibitors were also delivered in iCells , for cells seeded at 20 , 000–40 , 000 cells per each well of a 12-well for patch-clamp studies . 5 pmol each of the miR-34b and −34c inhibitor or 10 pmol of the non-targeting control miR-inhibitor were delivered using 1 . 0 μl of Lipofectamine RNAi Max . Total RNA was isolated from NRVM using Trizol Reagent ( Invitrogen ) according to the manufacturer’s instructions . RNA was also collected from human control and heart failure tissue samples . Tissue samples were first pulverized using liquid N2 and mortar and pestle to assist in the homogenization with Trizol . Subsequent RNA was used as a template for cDNA synthesis in reverse transcriptase reactions using the Multiscribe Reverse Transcriptase kit ( Invitrogen ) for detecting both mRNAs and miRNAs . The quantitative PCR reactions were performed with the ABI 7500 Real-Time PCR system using either SYBR green technology for coding genes or Taqman reagent for detecting mature miRNAs . mRNAs were normalized with GAPDH or ribosomal protein 27 ( RPL27 ) and miRNAs with small nucleolar RNA U87 or U6 . All miRNA primer sets were designed and provided by Invitrogen Taqman Assays . Real-time PCR reactions were conducted using TaqMan Universal Master Mix II ( Invitrogen ) . miRNA primer sets for real-time PCR detection were as follows: Rat miR-34b: Assay name , mmu-miR-34b-5p; Assay ID , 002617; Catalogue # , 4427975 Human/Rat miR-34c: Assay name , hsa-miR-34c; Assay ID , 000428; Catalogue # , 4427975 Rat U87 ( housekeeping gene ) : Assay name , U87; Assay ID , 001712; Catalogue # , 4427975 Human miR-34b: Assay name , hsa-miR-34b; Assay ID , 000427; Catalogue # , 4427975 Human U6 ( housekeeping gene ) : Assay name , U6 snRNA; Assay ID , 001973; Catalogue # , 4427975 Primer sets used in the detection of mRNA transcripts were designed in Primer 3 Plus and specificity to the intended target verified using Primer Blast ( NCBI ) . Rat Scn5a Forward primer , 5’-TCAATGACCCAGCCAATTACCT-3’ , Reverse primer , 5’-CCCGGCATCAGAGCTGTT-3’ Rat Scn1b Forward primer , 5’-ACGTGCTCATTGTGGTGTTAACC-3’ , Reverse primer , 5’-CCGTGGCAGCAGCAATC-3’ Rat Kcnd3 Forward primer , 5’-GCCTTCGAGAACCCACA-3’ , Reverse primer , 5’-GATCACCGAGACCGCAATG-3’ Rat Kcnip2 Forward primer , 5’-ACTTTGTGGCTGGTTTGTCG-3’ , Reverse primer , 5’-TGATACAGCCGTCCTTGTTGAG-3’ Rat GAPDH Forward primer , 5’-AGTTCAACGGCACAGTCAAG-3’ , Reverse primer , 5’-ACTCCACGACATACTCAGCAC-3’ Rat Rpl27 Forward primer , 5’-GCTGTCGAAATGGGCAAGTT-3’ , Reverse primer , 5’-GTCGGAGGTGCCATCATCAA-3’ Human Kcnip2 Forward primer , 5’-TGTACCGGGGCTTCAAGAAC-3’ , Reverse primer , 5’-GGCATTGAAGAGAAAAGTGGCA-3’ Human Scn5a Forward primer , 5’- CTGCGCCACTACTACTTCACCAACA-3’ , Reverse primer , 5’- TCATGAGGGCAAAGAGCAGCGT-3’ Human Scn1b Forward primer , 5’- GACCAACGCTGAGACCTTCA-3’ , Reverse primer , 5’- TCCAGCTGCAACACCTCATT-3’ Human Kcnd3 Forward primer , 5’- TCAGCACGATCCACATCCAG-3’ , Reverse primer , 5’- CTCAGTCCGTCGTCTGCTTT-3 Human GAPDH Forward primer , 5’- TCCTCTGACTTCAACAGCGA-3’ , Reverse primer , 5’- GGGTCTTACTCCTTGGAGGC-3’ . RNA collected from NRVM following KChIP2 and control siRNA treatment were submitted to miRNA microarray analysis to determine miRNAs regulated by the loss of KChIP2 . The array was performed by the Gene Expression and Genotyping Facility at Case Western Reserve Univesity using the Affymetrix GeneChip miRNA 4 . 0 array . The resulting . CEL files were used with ExpressionConsole to conduct RMA analysis to derive the relative intensities of the miRNA probe set . The raw datasets are available from the Gene Expression Omnibus ( Accession GSE75806 ) . Additionally , RNA collected from NRVM following KChIP2 silencing with an adeno-shRNA expression system with non-targeting ( control ) and KChIP2 targeting constructs were used to assess global gene changes following KChIP2 loss . A total of 1 . 5 x 106 cells were plated on 35 mm dishes . Cells were cultured in DMEM/5% FBS/penicillin/sptreptomycin with 0 . 1 mM BrdU for 24 hrs . After 24 hrs , media was replaced with fresh DMEM/5% FBS/penicillin/sptreptomycin and the corresponding control and KChIP2 shRNA virus . Cells were cultured for 48 hrs ( with a media change after 24 hrs ) and collected for total RNA and evaluated using a whole-transcriptome microarray . The array was performed by the Gene Expression and Genotyping Facility at Case Western Reserve University using the Affymetrix rat Clariom S Assay . The resulting . CEL files were used with Expression Console and the Transcriptome Analysis Console provided by Affymetrix ( available here: http://www . affymetrix . com/support/technical/software_downloads . affx ) to derive the relative changes in gene expression . The raw datasets are available from the Gene Expression Omnibus ( Accession GSE94623 ) The design of non-targeting ( control ) and KChIP2 shRNAs was conducted as described ( Campeau et al . , 2009 ) with modifications . shRNA inserts were optimally designed using a compilation of the RNAi Consortium and Invitrogen design algorithms . To begin the design , oligos were ordered that contained the shRNA sequence for control: 5’- GTTGACAGTGAGCGATCTCGCTTGGGCGAGAGTAAGTAGTGAAGCCACAGATGTACTTACTCTCGCCCAAGCGAGAGTGCCTACTGCCTC-3’ and KChIP2: 5’- GTTGACAGTGAGCGCGAGCTGGGCTTTCAACTTATATAGTGAAGCCACAGATGTATATAAGTTGAAAGCCCAGCTCATGCCTACTGCCTC-3’ . These oligos were modified in a PCR reaction to add on cloning sites for insertion into the pSM2 vector using the primer set: 5’- CAGAAGGCTCGAGAAGGTATATGCTGTTGACAGTGAGCG-3’ and 5’- CTAAAGTAGCCCCTTGAATTCCGAGGCAGTAGGCA-3’ . Underlined are the XhoI and EcoRI restriction sites used for cloning . Following the insertion of this sequence into the pSM2 vector , the insertion was cloned out again using the primer set 5’- GAGCTCGCTAGCGCTACCGGTCGCCACCATGGTGAGCAAGGGCGAGG-3’ and 5’-GATTGCCAAGCTTCTAGATAAACGCATTAGTCTTCCAATTG-3’ . Underlined are the NheI and HindII restriction sites , which were then used to clone the inserts into the adenovirus construct , Ad . CGI . During insertion , the GFP encoded by the viral vector was digested out as the shRNA system expression GFP in tandem with the shRNA . The modified Ad . GFP construct was transfected with the psi5 vector into CRE8 cells for the production and amplification of packaged viral constructs to then be used for silencing studies . Fractionation of adult rat heart tissue was performed as described ( Baghirova et al . , 2015 ) with slight modifications . Briefly , freshly isolated heart tissue was minced in ice cold PBS . Tissue was washed several times to remove residual blood from sample . Approximately 300 mg of tissue was weighed out and suspended in cytosolic lysis buffer , consisting of 150 mM NaCl , 50 mM HEPES ( pH 7 . 4 ) , 25 µg/mL Digitonin , and 10% Glycerol . Tissue pieces were homogenized then filtered through a QIAshredder homogenizer column ( Qiagen , 79656 ) . Filtered lysate was then incubated at 4°C on an end-over-end rotator for 10 min . Samples were then centrifuged at 4000 x g for 10 min at 4°C . Supernatant was collected as the cytosolic fraction . The remaining pellet was resuspended in membrane lysis buffer consisting of 150 mM NaCl , 50 mM HEPES ( pH 7 . 4 ) , 1% IGEPAL , and 10% glycerol . Sample was incubated for 30 min in end-over-end rotator at 4°C , followed by centrifugation at 6000 x g for 10 min at 4°C . The supernatant was collected as the membrane associated fraction , while the remaining cell pellet was resuspended in the nuclear lysis buffer consisting of 150 mM NaCl , 50 mM HEPES ( pH 7 . 4 ) , 0 . 5% sodium deoxycholate , 0 . 1% sodium dodecyl sulfate , and 10% glycerol . Lysate was placed on an end-over-end rotator for 10 min at 4°C , which was then followed by brief sonication . The lysate was then centrifuged at 6800 x g for 10 min at 4°C . The supernatant was collected as the nuclear fraction . Roche protease inhibitor tablets were added fresh before the addition of each lysis buffer . In order to perform western blot experiments looking at KChIP2 nuclear expression , cytosolic , membrane , and nuclear extracts were isolated as described above . 20–30 μg of protein extracts were loaded into SDS-PAGE gels , transferred to nitrocellulose membranes , and western blotting performed using lactate dehydrogenase ( Abcam Cat# ab52488 RRID:AB_2134961 , 1:1000 ) to represent the cytosolic fraction , Lamin-B1 ( Abcam Cat# ab16048 RRID:AB_443298 , 1:1000 ) representing the nuclear fraction , Serca2a ( 1:1000 , Dr . Periasamy , Ohio State University ) and KChIP2 ( UC Davis/NIH NeuroMab Facility Cat# 75–004 RRID:AB_2280942 , 1:50 ) to observe localization . Western blot performed on NRVM was conducted to assess Kv4 . 3 protein expression following miR-34 precursor treatment . NRVM were rinsed with PBS then scraped and collected . Cell pellets were re-suspended in RIPA Buffer ( 150 mM sodium chloride , 1 . 0% NP-40 or Triton X-100 , 0 . 5% sodium deoxycholate , 0 . 1% SDS ( sodium dodecyl sulphate ) , 50 mM Tris , pH 8 . 0 , plus Roche Inhibitor tablet ) and then sonicated on ice to disrupt cell membranes . 30–40 μg of whole cell extract was loaded into SDS-PAGE gels , transferred to nitrocellulose membrane , and western blotting performed using Kv4 . 3 ( UC Davis/NIH NeuroMab Facility Cat# 75–017 RRID:AB_2131966 , 1:500 ) , and actin ( Sigma-Aldrich Cat# A4700 RRID:AB_476730 , 1:1000 ) . Freshly isolated adult rat ventricular myocytes were plated on laminin coated coverslips for 1 . 5 hr to allow for attachment . Cells were quickly rinsed with room temperature PBS before being fixed by 4% formaldehyde in PBS for 15 min . Cells were permeabilized for 10 min in PBS + 0 . 03% Triton X-100 and blocked for 2 hr in a solution of PBS , 5% normal goat serum , and 1% BSA . Cells were incubated overnight with primary antibody lactate dehydrogenase ( Abcam Cat# ab52488 RRID:AB_2134961 , 1:100 ) and KChIP2 ( UC Davis/NIH NeuroMab Facility Cat# 75–004 RRID:AB_2280942 , 1:50 ) in PBS with 2% normal goat serum and 1% BSA . Cells were rinsed 3x in PBS then incubated with secondary antibody ( Alexa-568 Thermo Fisher Scientific Cat# A11036 RRID:AB_10563566 1:500 against LDH and Alexa-647 Thermo Fisher Scientific Cat# A-21236 RRID:AB_2535805 1:500 against KChIP2 ) in PBS with 2% normal goat serum and 1% BSA for 2 hr at room temperature . Coverslips were mounted onto glass slides with mounting media containing DAPI . Labeled cardiomyocytes were scanned with a Leica DMi8 confocal microscope . Chromatin Immunoprecipitation was performed as described with minor modifications ( Schmidt et al . , 2009 ) . Briefly , freshly isolated adult rat cardiomyocytes were fixed in a 1% formaldehyde solution in PBS for 14 min and quenched with 0 . 125 M glycine for 5 min . Cells were treated with a 0 . 05% trypsin/0 . 02% EDTA 1x PBS solution for 8 min at 37°C to partially digest the cells aiding in removal of cytoplasmic extract and purification of nuclear extract during cell lysis steps . Trypsin was inactivated by the addition of 10% FBS , and the cell pellet was rinsed 3x in ice cold PBS . Chromatin was extracted by the treatment with several lysis buffers . Lysis buffer 1 ( 50 mM Hepes-KOH , Ph7 . 5; 140 mM NaCl; 1 mM EDTA; 10% Glyerol; 0 . 5% Igepal; 0 . 25% Triton-X ) was added to the cells for 10 min with rocking , followed by 15–20 dounces with a glass teflon douncer on ice . This cell lysate fraction was discarded and the remaining cell pellet was resuspended in Lysis buffer 2 ( 10 mM Tris-HCl , pH 8 , 0 , 200 mM NaCl; 1 mM EDTA; 0 . 5 mM EGTA ) for 5 min with rocking . This was again followed by 15–20 dounces with a glass teflon douncer on ice . Lastly , remaining cell pellet was resuspended in Lysis buffer 3 ( 10 mM Tris-HCl , pH 8 . 0; 100 mM NaCl; 1 mM EDTA; 0 . 5 mM EGTA; 0 . 1% Na-Deoxycholate; 0 . 5% N-lauroylsarcosine ) . Cell suspension was split in half to be used for IgG or KChIP2 ChIP . Samples were then sheared on a BioRuptor ( Diagenode , total 18 cycles , hi-power , 30 s on/off ) . The sonicated chromatin was immunoprecipitated with 15 ug of antibody ( either α-KChIP2 or IgG control ) bound to Dynabeads ( Invitrogen ) followed by washing and elution . Immuoprecipitate and input chromatin samples were then reverse crosslinked followed by purification of genomic DNA . Target and nontarget regions of genomic DNA were amplified by qRT-PCR using SYBR Green . Data were analyzed by calculating the immunoprecipitated DNA enrichment normalized to a region 8 kb upstream of the target site in the KChIP2-IP compared to the IgG-IP . Antibodies used in ChIP were KChIP2 ( UC Davis NeuroMab 75–004 ) and IgG ( Millipore Cat# 12–371 RRID:AB_145840 ) ChIP-PCR primer sequences were: miR-34b target site: forward 5’- GGTCACTCGGCCAGTAGGA-3’ , reverse 5’- GGAGTCCTGCTCTCCCTCA-3’ . miR-34b 8 kb upstream: 5’- CCACCCTCTCAGTAGCTTGC-3’ , reverse 5’- CAGTGCCAGGGGATAGGAAG-3’ Phenylephrine stimulation experiments were performed to evaluate gene expression changes , functional changes in ionic current by patch-clamp technique , or conduction properties by optical mapping . RNA studies for gene expression changes were conducted in 6-well plates with 1 . 5 × 106 cells plated per well for the collection of RNA . For patch-clamp recordings , NRVMs were on coverslips coated with laminin ( Sigma , L2020 ) inside of 35 mm dish at a density of 100 , 000 cells/well . For optical mapping 1 . 5 × 10∧6 NRVMs were plated on aclar coverslips ( Electron Microscopy Sciences ) coated with fibronectin ( BD Biosciences , 356008 ) in a 35 mm dish . Following the initial 24–36 hr of plating , media was switched to 1:1 DMEM:F12 ( without serum or BrdU ) and supplemented with 1x insulin-transferrin-selenium-X ( Invitrogen ) , 1% PS , and 142 μM Na+ Ascorbate for an additional 24–36 hr . After this time , treatment media was applied , consisting of the same DMEM:F12 media with supplements and 100 μM phenylephrine . At the same time , control cells without phenylephrine were transduced with adeno . GFP , while phenylephrine treated cells received either the adeno . GFP or adeno . KChIP2 . 6 to restore KChIP2 expression during phenylephrine treatment . Alternatively , cells were transfected using Lipofectamine RNAi Max using manufacturer’s protocol to deliver a control or combination of miR-34b and −34c antimir . In the case of transfected cells , the transfection was performed prior to the initiation of phenylephrine treatment . Phenylephrine treatment was sustained for 48 hr ( fresh media was swapped after 24 hr , maintaining phenylephrine treatment , but no more virus was applied ) . Phenylephrine studies were also performed on iCells . iCell Cardiomyocyte Maintenance Medium was supplemented with 142 μM Na+ Ascorbate . iCells were only treated with the antimirs and submitted to patch-clamp recordings with the same treatment conditions applied to the NRVM . Notably , cells used for patch clamp recordings or optical mapping were washed at least three times over a minimum of 20 min in media without phenylephrine present for washout . Macroscopic INa and Ito were recorded using the whole-cell configuration of the patch clamp technique . INa was recorded in the solution containing 50 mM NaCl ( for NRVM ) and 25 mM ( for iCells ) , 80 or 105 mM N-methyl D-glucamine , 5 . 4 mM CsCl , 1 . 8 mM MgCl2 , 1 . 8 mM CaCl2 , 10 mM glucose , 10 mM HEPES , pH 7 . 3 . 1 µM of nisodipine was used to block L-type Ca currents . INa was elicited from a holding potential of −80 mV with depolarizing voltage pulses from −60 mV to 45 mV for 16 ms . To measure Ito , cells were placed in the Tyrode's solution containing ( mmol/L ) NaCl 137 , KCl 5 . 4 , CaCl2 2 . 0 , MgSO4 1 . 0 , Glucose 10 , HEPES 10 , CdCl2 0 . 3 , and TTX 100 mM , pH to 7 . 35 with NaOH . Patch pipettes were pulled from borosilicate capillary glass and lightly fire-polished to resistance 0 . 9–1 . 5 MΩ when filled with electrode solution composed of ( mmol/L ) aspartic acid 120 , KCl 20 , MgCl2 2 , and HEPES 5 , NaCl 10 , EGTA 5 , Na-GTP 0 . 3 , Phosphocreatine 14 , K-ATP 4 , Creatine phosphokinase two and brought to a pH of 7 . 3 . Ito , total amplitude was measured as the difference between peak current and steady-state current during a 400 ms voltage step ranging from –30 to +60 mV from a holding potential of –70 mV . Recording Ito , f used a modified protocol to kinetically isolate the current . A 150 ms voltage step to −80 mV from a holding potential of −20 mV was used to allow recovery of Ito , f but not Ito , s . This was followed by a 50 ms prepulse to −20 mV to eliminate INa . Ito , f amplitude was then measured as the difference between peak current and steady-state current during 500 ms voltage steps ranging from −30 to +40 mV . Ionic current density ( pA/pF ) was calculated from the ratio of current amplitude to cell capacitance . All experiments were performed at 35°C except INa ( room temperature ) . Low-resistance electrodes ( <2 MΩ ) were used , and a routine series resistance compensation was performed to values of >80% to minimize voltage clamp errors . The uncompensated Rseries was therefore <2 MΩ . Command and data acquisition were operated with an Axopatch 200B patch clamp amplifier controlled by a personal computer using a Digidata 1200 acquisition board driven by pCLAMP 7 . 0 software ( Axon Instruments , Foster City , CA ) . Current densities , cell capacitance , current-voltage relationship , and conductance , were measured as previously described ( Shinlapawittayatorn et al . , 2011 ) . Following 48 hr of PE treatment of the NRVM , cells were prepared for optical mapping studies . Prior to recordings , NRVMs were washed twice for 10 min each in DMEM:F12 treatment media without PE to wash out the PE and remove any acute effects . They were then transferred to Tyrodes solution ( 140 NaCl , 4 . 56 KCl , 0 . 73 MgCl2 , 10 HEPES , 5 . 0 dextrose , 1 . 25 CaCl2 ) containing 10 µM Di4 ( Sigma , D8064 ) for 20 min . Monolayers were then washed with normal Tyrodes solution before mounting on stage adapter to maintain cells at 34–35˚C . Di4 fluorescence 685/80 nm was measured using an upright microscope ( MVX10 , Olympus ) with a cooled CCD camera ( Princeton Instruments ) . A solid-state light source ( Sola Light Engine , Lumencore ) was used for dye excitation ( 510/80 nm ) over a 16 × 12 mm field of view . Cells were paced by point stimulation at cycle lengths of 1000 ms , 750 ms , 500 ms , 350 ms and 350 ms to obtain conduction velocity and APD restitution curves . Analysis of recordings were conducted via custom software developed in Matlab ( MathWorks ) as described previously ( PMID: 12960954 ) . Additional Matlab custom software ( Rhythm ) was also used for analysis ( Laughner et al . , 2012 ) . Arrhythmia data was collected using baseline pacing ( S1 , 750 ms ) followed by a single premature stimulus ( S2 ) with a coupling interval beginning at 150 ms and prolonged by 10 ms until either capture of a single beat or arrhythmia ensued . This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health . The protocol for tissue isolation from neonatal rat ( Protocol Number: 2013–0015 ) was approved by the Committee on the Ethics of Animal Experiments of Case Western Reserve University . Tissue from the left ventricular free wall of non-failing and failing human heart samples were acquired from the Cleveland Clinic Foundation ( CCF ) tissue repository . All protocols were approved by the CCF Institutional Review Board ( IRB# 2378 ) . Samples were received coded and no identifying metrics were documented for the study . Results are expressed as mean ± SEM and represent data from at least three independent experiments . Statistical analysis for continuous data was performed using a two-tailed Student’s t-test . When multiple comparisons were evaluated , a Bonferroni correction was performed . The null hypothesis was rejected if p<0 . 05 . Statistical testing of non-continuous data , as seen with arrhythmia susceptibility measurements , was performed using the Mann-Whitney Test . Evaluation of samples sizes were initially performed using stringent conditions for expected molecular and functional changes . Assuming as a little as a 20% change in control to treated conditions , an error rate of 10% , and a power of 0 . 8 at a threshold of 0 . 05 , provided a sample size of 4 per experimental condition . However , because of anticipated variability from the use of primary cells for many of the experiments and multiple comparisons in some datasets , larger sample sizes were used .	The heart pumps blood throughout the body to provide oxygen and nourishment . To do so , proteins in the heart create electrical signals that tell the heart muscles to contract in a coordinated manner . Heart disease can cause cells to lose control of the production or activity of these proteins , creating disorganized electrical signals called arrhythmias that interfere with the heart’s ability to pump . Sometimes these arrhythmias lead to sudden death . Researchers do not know exactly what triggers these changes in the heart’s normal electrical rhythms . This has made it difficult to develop strategies to prevent these disruptions or to fix them when they occur . By studying rat and human heart cells , Nassal et al . now show that a protein called KChIP2 stops working properly during heart disease . Most importantly , because of the decreased level of KChIP2 in heart disease , KChIP2 loses the ability to restrict the production of two microRNA molecules – a role that KChIP2 was not previously known to perform . This loss of activity sets off a cascade of signals that worsens the balance of electrical activity in the heart cells , creating arrhythmias . Treatments that restored proper levels of the fully working KChIP2 protein to the heart cells or that blocked the signals set off by a lack of KChIP2 returned the electrical activity of the cells back to normal . This also stopped the development of arrhythmias . Further studies are now needed to investigate whether these treatments have the same effects in living mammals . If effective , this could ultimately lead to new treatments for heart diseases and arrhythmias .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2017	KChIP2 is a core transcriptional regulator of cardiac excitability
Anxiety disorders affect approximately 1 in 5 ( 18% ) Americans within a given 1 year period , placing a substantial burden on the national health care system . Therefore , there is a critical need to understand the neural mechanisms mediating anxiety symptoms . We used unbiased , multimodal , data-driven , whole-brain measures of neural activity ( magnetoencephalography ) and connectivity ( fMRI ) to identify the regions of the brain that contribute most prominently to sustained anxiety . We report that a single brain region , the intraparietal sulcus ( IPS ) , shows both elevated neural activity and global brain connectivity during threat . The IPS plays a key role in attention orienting and may contribute to the hypervigilance that is a common symptom of pathological anxiety . Hyperactivation of this region during elevated state anxiety may account for the paradoxical facilitation of performance on tasks that require an external focus of attention , and impairment of performance on tasks that require an internal focus of attention . Current models of anxiety disorders suggest that pathological anxiety results from excessive or inappropriate activation of the same neural circuits that are responsible for adaptive anxiety in the face of threat ( Insel et al . , 2010; Insel , 2014 ) . Although there is a long history of translational work studying neural systems mediating the acute fear response ( Pavlov , 1927; Fanselow and Poulos , 2005; Fullana et al . , 2016 ) , much less is known about the neural systems mediating prolonged periods of elevated state anxiety . Closing this knowledge gap is critical because the occurrence of prolonged periods of elevated state anxiety is one of the primary symptoms of all anxiety disorders ( American Psychiatric Association , 2013 ) . Therefore , understanding neural mechanisms underlying prolonged periods of elevated anxiety has the potential to identify targets for the treatment of anxiety disorders , which are among the most prevalent psychiatric disorders ( Kessler et al . , 2005 ) . The gold-standard translational paradigm for studying elevated state anxiety in the laboratory is the threat of shock paradigm , during which subjects are exposed to periods when they are either safe , or at risk for receiving unpredictable aversive electrical stimulations ( Schmitz and Grillon , 2012; Grillon , 2008; Grillon and Baas , 2003 ) . It allows for the experimental manipulation of state anxiety within subjects ( Grillon and Baas , 2003 , 1998; Grillon et al . , 1991 , 2007 , 2008 , 2009 ) , which can be quantified using psychological and physiological measures ( Grillon , 2008; Grillon and Baas , 2003 ) and can be implemented in healthy subjects ( Balderston et al . , 2017a , 2017b; Cornwell et al . , 2007 , Cornwell et al . , 2008 , 2012; Lissek et al . , 2007 ) , patients ( Grillon et al . , 2009; Balderston et al . , 2017a; Vytal et al . , 2016 ) , and non-human animals ( Davis et al . , 2010 ) . A key feature of prolonged periods of threat of shock is that they induce a stable increase in anxiety that can be probed at random intervals using the acoustic startle reflex ( Grillon , 2008; Grillon and Baas , 2003 , 1998 ) , suggesting that this anxious state is mediated by a fundamental sustained change in the pattern of ongoing brain activity . This is in contrast to more phasic event-related fear responses in typical cued fear conditioning ( Schmitz and Grillon , 2012; Grillon , 2008; Davis et al . , 2010 ) . Current neuroscientific models of anxiety are based in part on translational work using Pavlovian fear conditioning ( Pavlov , 1927; Fanselow and Poulos , 2005; Fullana et al . , 2016 ) . Decades of work in non-human animals has shown that acute fear responding is dependent upon the amygdala ( Kwapis et al . , 2009; Bailey et al . , 1999; Parsons et al . , 2006 ) , and functional magnetic resonance imaging ( fMRI ) during fear conditioning in humans has been used to identify a canonical fear network that includes the amygdala , the dorsomedial prefrontal cortex , the thalamus , and the anterior insula ( Fullana et al . , 2016; Schultz et al . , 2012; Cheng et al . , 2003 , 2006 ) . However , much less is known about the network mediating extended periods of elevated state anxiety . In addition , cognitive scientific research in humans shows that attentional processing is profoundly influenced by both state ( Vytal et al . , 2013 , 2012; Patel et al . , 2016; Shackman et al . , 2006 ) and trait anxiety ( Derakshan et al . , 2009; Eysenck et al . , 2007 ) , suggesting that multiple neural systems are affected by anxiety . Although there have been some studies investigating the neural systems that mediate anxiety , these studies often depend on an a priori focus that is centered on the regions of the canonical fear network , and typically rely on a priori methods to increase statistical sensitivity in these regions such as lowered statistical thresholds ( Robinson et al . , 2013a; Mobbs et al . , 2010; Hooker et al . , 2006; Tabbert et al . , 2010 ) , region of interest analyses ( Balderston et al . , 2015 , 2014 , 2013 ) , and seed-based functional connectivity ( Schultz et al . , 2012; Vytal et al . , 2014; Gold et al . , 2015 ) . Importantly , the increased sensitivity gained by using these statistical methods comes at the cost of assessing anxiety-related changes in regions not identified a priori , thus resulting in a possible under-reporting of anxiety-related changes in other areas of the brain , such as regions important for attentional processing . Therefore , the purpose of this study was to use exploratory analytical methods to identify the most prominent activity/connectivity changes induced by the threat of shock paradigm . Toward this aim , we collected data from two complimentary imaging modalities , fMRI and magnetoencephalography ( MEG ) during a threat of shock paradigm . In both MEG and fMRI experiments , subjects underwent alternating blocks of safety and threat , and rated their anxiety continuously using a centrally located visual analog scale ( see Figure 1 ) . During the MEG experiment , subjects also received randomly timed white noise presentations , which served to probe the subject’s current anxiety level ( via the acoustic startle response ) and their ongoing brain activity ( via the preceding pattern of neural oscillations ) . 10 . 7554/eLife . 23608 . 003Figure 1 . Schematic of experimental paradigm . ( A ) Subjects underwent alternating blocks of threat and safety . ( B ) Visual display present on the screen during the experiment . During the experiment subjects saw two circles . The color of the outer circle indicated the block type . The color of the inner circle was controlled by the subject , and reflected the subject’s then-current anxiety level . DOI: http://dx . doi . org/10 . 7554/eLife . 23608 . 003 In both experiments , we used unbiased , data-driven , whole-brain approaches to identify changes in activity ( MEG ) and connectivity ( fMRI ) as a function of threat . To assess functional connectivity changes in the fMRI signal , we used the global brain connectivity ( GBC ) metric , which does not rely on a priori seed-selection for the connectivity analysis . To assess ongoing patterns of activity in the MEG study , we evaluated changes in oscillatory power during the 2 s prior to the startle probes as a function of threat . According to the translational approach , one might predict that the most prominent changes in spontaneous neural activity and connectivity would emerge in regions of the canonical fear network ( Fanselow and Poulos , 2005; Kim et al . , 2011 ) . However , given that impaired attentional control is a key feature of clinical anxiety ( Derakshan et al . , 2009; Eysenck et al . , 2007 ) , and that threat of shock has been repeatedly shown to impact performance on tasks that require attention control ( Vytal et al . , 2013 , 2012; Patel et al . , 2016; Shackman et al . , 2006 ) , one might also expect that the most prominent changes would emerge within regions of the frontoparietal attention network ( Ptak , 2012; Posner , 2012; Petersen and Posner , 2012 ) . We began by assessing the ability of our threat manipulation to induce a sustained state of anxiety . Results from the psychological questionnaires , and the subjective rating scales can be found in Table 1 . Consistent with the online ratings , subjects during both experiments reported more anxiety ( MEG: t ( 26 ) = 8 . 65; p<0 . 001; fMRI: t ( 24 ) = 13 . 98; p<0 . 001 ) and fear ( MEG: t ( 26 ) = 8 . 03; p<0 . 001; fMRI: t ( 24 ) = 9 . 15; p<0 . 001 ) during the threat blocks than during the safe blocks . In addition , two sample t-tests did not reveal any significant differences between experiments for either the psychological questionnaires , or the affective rating scales ( all ps > 0 . 05 ) . For the MEG study , we analyzed both the acoustic startle reflex and the online self-reported anxiety ratings . Because startle probes could not be presented during the MRI study , we relied only on the ratings . 10 . 7554/eLife . 23608 . 004Table 1 . Individual differences for MEG ( N = 28 ) and MRI ( N = 25 ) experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 23608 . 004MeasureMEGMRISTAI State26 . 04 ( 1 . 37 ) 23 ( 0 . 9 ) Trait27 . 12 ( 0 . 93 ) 28 . 18 ( 1 . 27 ) ASI11 . 59 ( 1 . 21 ) 8 . 64 ( 1 . 18 ) BAI1 . 37 ( 0 . 42 ) 0 . 58 ( 0 . 26 ) BDI0 . 89 ( 0 . 32 ) 0 . 42 ( 0 . 19 ) Shock Intensity ( mA ) 5 . 66 ( 0 . 66 ) 6 . 91 ( 1 . 01 ) Rating8 . 51 ( 0 . 2 ) 9 . 09 ( 0 . 19 ) Anxiety Pre2 . 04 ( 0 . 27 ) 1 . 98 ( 0 . 25 ) Safe2 . 47 ( 0 . 31 ) 1 . 76 ( 0 . 21 ) Threat5 . 41 ( 0 . 37 ) 5 . 97 ( 0 . 39 ) Fear Pre1 . 41 ( 0 . 15 ) 1 . 5 ( 0 . 23 ) Safe1 . 84 ( 0 . 27 ) 1 . 27 ( 0 . 12 ) Threat4 . 44 ( 0 . 39 ) 4 . 7 ( 0 . 42 ) Note: Numbers reflect the mean and standard deviation of the results [M ( SD ) ] . For each startle probe in the MEG study , we extracted the subject’s startle magnitude , and anxiety rating recorded just prior to the startle probe . Both the startle magnitude and anxiety ratings were normalized and converted to T-scores ( Blumenthal et al . , 2005 ) within subjects . These values were then averaged across trials and submitted to a paired-sample t-test ( Safe vs . Threat ) . Both ratings ( See Figure 2A and Figure 2—source data 1; t ( 27 ) = 10 . 03; p<0 . 001 ) and startle ( See Figure 2B and Figure 2—source data 1; t ( 27 ) = 4 . 65; p<0 . 001 ) indicated greater anxiety during the threat blocks compared to the safe blocks . Next , we created Threat > Safe difference scores for both startle ( anxiety potentiated startle; APS ) and the online ratings , and correlated the values across subjects . Ratings within the MEG study were significantly correlated with startle across subjects ( See Figure 2D and Figure 2—source data 1; r ( 26 ) = 0 . 61; p=0 . 001 ) . 10 . 7554/eLife . 23608 . 005Figure 2 . Behavioral results from both experiments . ( A ) Anxiety ratings during the MEG study . ( B ) Startle magnitude during the MEG study . ( C ) Anxiety ratings during the fMRI study . Bars represent the mean ± within-subject SEM ( Cousineau , 2005 ) . ( D ) Correlations between anxiety potentiated startle ( APS ) and differential anxiety ratings . The black squares represent the correlation between APS and ratings during the MEG session . The red dots represent the correlation between APS during the MEG study and anxiety ratings during the fMRI study in the subset of subjects who participated in both studies . DOI: http://dx . doi . org/10 . 7554/eLife . 23608 . 00510 . 7554/eLife . 23608 . 006Figure 2—source data 1 . Source data for all graphs in Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 23608 . 006 During the MRI study , we averaged the ratings across time in the threat and safe blocks , and converted these values to T-scores . As with the MEG study , these values were submitted to a paired-sample t-test ( Safe vs . Threat ) , and indicated more anxiety during the threat blocks than the safe blocks ( See Figure 2C and Figure 2—source data 1; t ( 24 ) = 23 . 06; p<0 . 001 ) . We also created Threat > Safe difference scores for these values , and correlated these difference scores with startle ( recorded during the MEG study ) in the subjects who participated in both sessions . There was a nonsignificant small positive correlation between startle and rating during the MRI session ( see Figure 2D and Figure 2—source data 1; r ( 16 ) = 0 . 24; p=0 . 344 ) . Many threat of shock studies have used seed-based functional connectivity analyses to identify changes in emotional processing centers in the brain ( Vytal et al . , 2014; Gold et al . , 2015; Satterthwaite et al . , 2011; Prater et al . , 2013; Hrybouski et al . , 2016; Birn et al . , 2014; Cha et al . , 2014; Heitmann et al . , 2016 ) ; however , seed-based functional connectivity methods suffer from bias because they require the experimenter to select a seed region ahead of time , while ignoring all other possible seed regions . To address this limitation , researchers have developed complementary data-driven functional connectivity metrics such as GBC , which do not rely on a priori seed-selection for the connectivity analysis . By assessing the connectivity between each voxel and every other voxel , this analysis allows the user to identify the most connected regions of the brain ( Cole et al . , 2010 ) , as well as the seed regions where connectivity impacts behavior across subjects ( Cole et al . , 2012; Gotts et al . , 2012 ) . By identifying regions that show the largest changes in GBC during periods of threat vs . periods of safety , it is possible to identify hubs that contribute most prominently to the sustained anxious state during threat of shock . We collected whole brain multi-echo echo-planar imaging data , and used the echo time-dependent independent components analysis to remove sources of noise unrelated to the blood oxygenation level dependent ( BOLD ) response from the timeseries ( Kundu et al . , 2012; Evans et al . , 2015 ) . Subjects were exposed to alternating 2-min blocks of safety and threat , without startle probes . Given that this design lacked external timing information ( i . e . external stimulus presentations ) , we examined changes in functional connectivity as a function of block type . We opted for a data-driven GBC approach where the connectivity of every voxel was assessed . We first computed GBC maps independently for the safe and threat conditions by correlating each voxel’s timecourse with every other voxel’s timecourse , applying the Fisher’s Z transformation , and averaging across these correlation maps ( See Figure 3A , B and Figure 3—source data 1 ) . As a first pass , we averaged across all voxels to obtain the whole brain GBC for safe and threat . We then conducted a ( Safe vs . Threat ) paired-sample t-test on these values , and found significantly more GBC for threat blocks than for safe blocks ( see Figure 3C and Figure 3—source data 1; t ( 24 ) = 2 . 13; p=0 . 044 ) . 10 . 7554/eLife . 23608 . 007Figure 3 . Overview of global brain connectivity ( GBC ) measure . ( A ) Map showing average GBC across all safe and threat TRs . ( B ) Cartoon schematic of a correlation matrix . The 43204 voxel x 43204 voxel cross correlation matrix was calculated separately for each subject and each condition . Correlations were averaged across rows for the entire grey matter mask , to create a single map reflecting the average correlation between each voxel and all other voxels in the mask . ( C ) Graph representing the mean GBC following the Fisher’s Z transformation for safe and threat averaged across the entire grey matter mask . Bars represent the mean ± within-subject SEM ( Cousineau , 2005 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23608 . 00710 . 7554/eLife . 23608 . 008Figure 3—source data 1 . Source data for graph in Figure 3C . DOI: http://dx . doi . org/10 . 7554/eLife . 23608 . 008 To follow-up the whole brain analysis , we conducted a voxelwise analysis of GBC . Using the same GBC maps created above , we conducted a voxelwise ( Safe vs . Threat ) paired-sample t-test . We used Monte Carlo simulations to estimate a null distribution for statistical testing , and used a cluster-based method based on this null distribution to correct for multiple comparisons . We found a significant increase in GBC in the threat blocks compared to the safe blocks in three clusters ( see Table 2 ) . The largest cluster was in the right angular gyrus . The two remaining clusters were found bilaterally in the IPS ( see Figure 4A and Figure 4—source data 1 ) . In all three clusters , we found significantly higher GBC for threat blocks than for safe blocks ( see Figure 4B and Figure 4—source data 1 ) . To determine whether these differences were affected by the delivery of the shock , or differences in motion across blocks , we repeated the threat vs . safe analysis at the cluster level after censoring the 10 TRs following shock delivery , and an equivalent number of safe TRs closely matched for motion . In addition , we covaried out any remaining differences in motion using an analysis of covariance . Using this approach , we still found a significant effect of threat on GBC in all three regions ( Right angular gyrus: f ( 1 , 23 ) =4 . 71 , p=0 . 04; Right IPS: f ( 1 , 23 ) =7 . 22 , p=0 . 01; Left IPS: f ( 1 , 23 ) =5 . 39 , p=0 . 03 ) , suggesting that our initial findings were not due to differences in motion , censoring , or residual neural activity evoked by the shock . 10 . 7554/eLife . 23608 . 009Table 2 . Results from voxelwise GBC analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 23608 . 009LabelVolumet-valuePeak activation ( LPI ) xyzRight Angular Gyrus1583 . 4548−5127Right Intraparietal Sulcus833 . 4221−6066Left Intraparietal Sulcus813 . 6−18−636610 . 7554/eLife . 23608 . 010Figure 4 . Results from voxelwise global brain connectivity ( GBC ) analysis . ( A ) Statistical map showing results from a threat vs . safe paired-sample t-test . ( B ) Graph representing average GBC values after applying the Fisher’s Z transformation for clusters shown in panel A . Bars represent the mean ± within-subject SEM ( Cousineau , 2005 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23608 . 01010 . 7554/eLife . 23608 . 011Figure 4—source data 1 . Source data for graph in Figure 4B . DOI: http://dx . doi . org/10 . 7554/eLife . 23608 . 011 Given that the IPS emerged as a hub differentiating global connectivity in threat compared to safe , we used this region as a seed-region to identify changes in functional connectivity during the threat vs . safe blocks . We understand this follow-up analysis could be interpreted as circular . According to this perspective , the seed-based analysis is not independent from the GBC analysis , which was used to identify the seed . However , the purpose of the follow-up analysis ( i . e . probing functional connectivity using clusters identified from the global connectivity analysis ) is to limit the interpretations of the global connectivity results to those supported by seed-based functional connectivity results , which has been done in previous group-level GBC studies ( e . g . [Gotts et al . , 2012] ) . We extracted the timecourse of activity averaged across the voxels in the bilateral IPS functional ROIs , and correlated this timecourse with the timecourse of activity across all voxels in the brain , independently for safe and threat , and applied the Fisher’s Z transformation to the resulting correlation coefficients . We conducted a voxelwise ( Safe vs . Threat ) paired-sample t-test on the resulting maps . We used Monte Carlo simulations and cluster thresholding to correct for multiple comparisons . Consistent with the GBC results , we found an increase in connectivity in threat blocks compared to safe blocks , in several regions of the frontoparietal attention network ( See Table 3 , Figure 5 , and Figure 5—source data 1 ) . 10 . 7554/eLife . 23608 . 012Table 3 . Results from voxelwise IPS connectivity analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 23608 . 012LabelVolumet-valuePeak activation ( LPI ) xyzLeft Thalamus3423 . 92-9612Right Inferior Parietal Lobule2083 . 6757−5739Left Superior Medial Gyrus1843 . 6533642Left Precuneus1793 . 593−6948Right Middle Frontal Gyrus1373 . 64331560Left Angular Gyrus1133 . 51−57−5430Left Middle Frontal Gyrus963 . 69−241560Left Middle Frontal Gyrus903 . 48−4551-310 . 7554/eLife . 23608 . 013Figure 5 . Results from bilateral IPS seed-based connectivity analysis . ( A ) Statistical map showing results from a threat vs . safe paired-sample t-test . ( B ) Graph representing average IPS connectivity values for clusters shown in panel A . Bars represent the mean ± within-subject SEM ( Cousineau , 2005 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23608 . 01310 . 7554/eLife . 23608 . 014Figure 5—source data 1 . Source data for graph in Figure 5B . DOI: http://dx . doi . org/10 . 7554/eLife . 23608 . 014 Next , we characterized the pattern of activity in the MEG study . It is well established that spontaneous neural activity at rest is dominated by oscillations in the alpha ( 8–12 Hz ) range ( de Munck et al . , 2008; Sadaghiani et al . , 2010; Mo et al . , 2013; Mayhew et al . , 2013 ) , which are most prominent when the subject is in an alert state of restful relaxation ( Doufesh et al . , 2014; Kim et al . , 2014; Khalsa et al . , 2015; Lagopoulos et al . , 2009 ) , and that alpha asymmetries can reflect differences in arousal across subjects ( Nitschke et al . , 1999 ) . Theoretical models of alpha function suggest that alpha oscillations are generated by coherent activity in local inhibitory interneurons ( Klimesch et al . , 2007 ) and that decreases in alpha power reflects increases in cortical excitability ( Klimesch et al . , 2007; Lange et al . , 2013 ) . Consistent with these theories , studies collecting simultaneous measures of electroencephalography ( EEG ) and fMRI have shown that alpha power is negatively correlated with functional connectivity ( Laufs et al . , 2003; Scheeringa et al . , 2012; Chen et al . , 2008; Chang et al . , 2013 ) . We extracted and cleaned the 2 s of data prior to each startle probe , avoiding contamination by blink artifacts . Because the timing of the startle probes was random , the pre-stimulus recording reflected random sampling across the sustained threat period . Therefore , we collapsed across the 2 s interval and examined oscillatory activity . First , we transformed the values into the frequency domain using a Fast Fourier Transform ( FFT ) with upper and lower limits of 20 Hz and 1 Hz , respectively . Then , we averaged these values across sensors , trials , and subjects , and examined the spectrogram for peaks . We detected a peak at ~10 Hz , suggesting that alpha oscillations were a prominent feature of these recordings ( Figure 6A and Figure 6—source data 1 ) . We identified the largest local maxima in each subject’s spectrogram . In all but four subjects we detected a peak in the alpha frequency band ( 8 Hz – 12 Hz ) . The power within a 2 Hz band around these individual alpha frequencies ( IAF ) s was used in all subsequent analyses ( Figure 6B and Figure 6—source data 1 ) . For subjects without a detectable IAF , power in a narrow band around IAF averaged across subjects was used . Subsequent analyses were performed in both sensor space ( Figure 6C; black dots ) and source space ( Figure 6C; green dots ) . 10 . 7554/eLife . 23608 . 015Figure 6 . Overview of MEG analyses . ( A ) Spectrogram representing power averaged across all subjects and all sensors with peak in the alpha frequency band . ( B ) Graph showing the frequency of peak alpha ( individual alpha frequency ) averaged across subjects . Bars represent the mean ± SEM . ( C ) Example of single subject alignment with sensors ( black dots ) source grid ( green dots ) and headmodel ( surface ) plotted together . DOI: http://dx . doi . org/10 . 7554/eLife . 23608 . 01510 . 7554/eLife . 23608 . 016Figure 6—source data 1 . Source data for graph in Figure 6A and B . DOI: http://dx . doi . org/10 . 7554/eLife . 23608 . 016 Given that alpha oscillations were the dominant feature in the MEG recordings across all blocks , we examined whether these oscillations differed as a function of threat . For the sensor space analysis , we averaged IAF across trials within conditions , and then performed a paired-sample t-test ( Safe vs . Threat ) on the resulting sensor space averages . We used Monte Carlo simulations and cluster thresholding to correct for multiple comparisons . Importantly , we found a large cluster of sensors over the left parietal lobe and a smaller cluster of frontal sensors showing significantly less IAF power in the threat blocks than during the safe blocks ( Figure 7A–B and Figure 7—source data 1 ) . 10 . 7554/eLife . 23608 . 017Figure 7 . Alpha results from threat vs . safe t-test . ( A ) Statistical map in sensor space showing a significant reduction in alpha power . Black symbols represent clusters of sensors showing significant threat vs . safe differences . ( B ) Graph showing average alpha power for safe and threat conditions in the largest cluster of sensors in panel A . ( C ) Statistical map in source space showing a significant reduction in alpha power . ( D ) Graph showing average alpha power for safe and threat conditions in the cluster in panel C . Bars represent the mean ± within-subject SEM ( Cousineau , 2005 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23608 . 01710 . 7554/eLife . 23608 . 018Figure 7—source data 1 . Source data for graph in Figure 7B and D . DOI: http://dx . doi . org/10 . 7554/eLife . 23608 . 018 To localize the source of these power differences , we projected these signals into source space using a dynamic imaging of coherent sources ( DICS ) beamformer . First , we created a common filter using all trials , then we projected the safe and threat trials through the filter independently , to obtain power estimates at the source level for each condition . Finally , we conducted a ( Safe vs . Threat ) paired-sample t-test on the source space IAF power estimates . As with the sensor space data , we used Monte Carlo simulations and cluster thresholding to correct for multiple comparisons . Consistent with the sensor space results , we found two clusters of voxels showing significantly less IAF power in the threat and safe conditions ( Figure 7C–D and Figure 7—source data 1 ) , the larger cluster was located in the left IPS , while the smaller cluster was located in the mid-cingulate gyrus . In addition , both the sensor and the source space results held if we analyze power across the entire alpha frequency band ( 8 Hz to 12 Hz; See below ) . Finally , when comparing the MEG results with the fMRI results , we found that the left IPS cluster in the fMRI data substantially overlaps with the alpha cluster ( 46/81 voxels; See Figure 8 ) . 10 . 7554/eLife . 23608 . 019Figure 8 . Conjunction map from voxelwise fMRI GBC analysis and MEG alpha power differences . Colors represent significant safe vs . threat differences from the fMRI analysis ( yellow ) , MEG analysis ( blue ) , and both analyses ( green ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23608 . 019 As with the MRI data , it is important to show that our denoising steps did not affect the comparisons between the safe and threat conditions . Specifically , it is important to show that the number of trials rejected either ( 1 ) did not differ across the safe and threat blocks , or ( 2 ) did not affect the safe vs . threat comparisons . In the safe condition , there were on average 58 . 21 ± 5 . 5 trials , while in the threat condition there were on average 55 . 32 ± 7 . 6 trials , which was significantly different across subjects ( t ( 27 ) = 2 . 78 , p=0 . 01 ) . Therefore , we decided to determine whether this difference in trial number impacted our results at the censor and source level . Accordingly , we included the difference in trial number across safe and threat blocks as a covariate in an ANCOVA examining the effect of threat on alpha responses . At both the sensor ( f ( 1 , 26 ) = 7 . 48 , p=0 . 01 ) and at the source ( f ( 1 , 26 ) = 17 . 797 , p<0 . 001 ) , we find that even with the difference in number of trials covaried out , the effects of threat on alpha power is still significant , suggesting that this difference did not significantly impact our results . Previous studies exploring the relationship between threat and functional connectivity focused on how threat affected connectivity within networks centered on seeds placed in emotional processing regions ( Vytal et al . , 2014; McMenamin et al . , 2014; McMenamin and Pessoa , 2015 ) . However , in this study because we were specifically interested in identifying the regions of the brain that contributed most prominently to anxiety , we chose to forego an a priori seed-based approach , and employ data-driven connectivity measures that do not rely on choosing a seed ( Cole et al . , 2010; Calhoun et al . , 2009 ) . To start , we used GBC as a method to determine whether connectivity en masse changed with threat , and to identify where in the brain the largest changes occurred ( Cole et al . , 2010 , 2012; Chu et al . , 2011 ) . We found the largest changes to occur bilaterally in the IPS , where connectivity with the rest of the brain increased as a function of threat . Then , using these regions as a seed , we found that they supported an increase in connectivity within a set of regions important for executive control ( Posner , 2012; Corbetta and Shulman , 2002 ) . These results suggest that the IPS is a critical connectivity hub in the network of regions that contribute most prominently to the sustained elevation of state anxiety induced by the threat of shock paradigm ( Cole et al . , 2010 , Cole et al . , 2014 ) . Although our GBC approach did not reveal any hubs within the amygdala , or in other regions typically associated with emotional processing ( Simmons et al . , 2006; Mechias et al . , 2010; Etkin and Wager , 2007 ) , we do see an increase in connectivity between the IPS and the thalamus and dorsomedial prefrontal cortex , two regions known to be part of the canonical fear network . In addition to the IPS , we found changes in GBC as a function of threat in the right angular gyrus . Although it is currently unclear how the angular gyrus might contribute to threat processing , there is work suggesting that this region may be a key site for multisensory integration ( Seghier , 2013 ) . One hypothesis is that enhanced GBC in this region may reflect a heightened awareness of the current context . However , this is a post hoc explanation that should be explored in future studies . An obvious parallel to the fMRI connectivity analysis would be to conduct a similar whole brain connectivity analysis of the MEG data . Not only would corroborating evidence from an independent imaging modality strengthen the fMRI connectivity results , but results from MEG specifically would allow for a frequency-specific analysis of the effects of threat on functional connectivity ( Hillebrand et al . , 2012; Brookes et al . , 2011 ) . However , the current study was not optimized to reliably detect differences in MEG connectivity . In the current study , we included the white noise probes as a way to obtain a quantitative measure of anxiety throughout the safe and threat blocks ( Grillon et al . , 1999 ) . These white noise probes trigger the acoustic startle reflex , which varies as a function of an individual’s current level of anxiety ( Grillon , 2008 ) . Unfortunately , these reflexive blinks also inject an artifact into the MEG signal , and because the magnitude of these blinks differs across conditions , the blink artifact also differs across conditions . Although adaptive beamforming can theoretically remove the artifact induced by the blink response ( Van Veen et al . , 1997 ) , the only way to ensure that the blink artifact does not differentially affect the MEG signal is to remove the contaminated time periods from the analysis , or remove the startle probes from the design at the outset . In the current study , we chose to address this limitation by extracting two-second time windows prior to each startle presentation to minimize the effect of the blink artifact on our power estimates . However , it has been shown that reliability of the MEG connectivity estimates increases as the duration of the recording increases , and durations of ~10 min or greater may be needed to maximize reliability ( Liuzzi et al . , 2016 ) . Therefore , using such short intervals did not allow for the ability to obtain reliable estimates of MEG connectivity . Future studies should be conducted to address this limitation . In addition , it will be important to use appropriate correction methods to account for signal leakage between source space regions ( Colclough et al . , 2015 ) , and verify that the resulting connectivity estimates match previously published work ( i . e . strong alpha connectivity in occipital regions and strong beta connectivity linking the parietal cortex with other frontal , temporal , and occipital regions; [Hunt et al . , 2016] ) . In addition to the increases in functional connectivity measured with fMRI , we also found decreases in alpha as a function of threat of shock in the same parietal region using MEG . Although the focus of the paper was on alpha , our initial approach ( described in the introduction ) was to examine all frequency bands independently ( other bands not shown ) . The focus on alpha emerged out of the observations that , ( 1 ) alpha was the strongest signal in the recordings , ( 2 ) alpha showed the largest power changes as a function of threat , ( 3 ) alpha was the only frequency band that showed consistent results at both the sensor and the source level , and ( 4 ) the source space results aligned nicely with the corresponding fMRI GBC data . These results are also consistent with several previous studies using simultaneous recordings of fMRI and EEG ( Sadaghiani et al . , 2010; Mo et al . , 2013; Mayhew et al . , 2013; Scheeringa et al . , 2012; Chang et al . , 2013; Wu et al . , 2010; Walz et al . , 2015; Scheeringa et al . , 2011 ) . For instance , functional connectivity at the whole brain level ( Chang et al . , 2013 ) , and within the visual system ( Scheeringa et al . , 2012 ) is negatively correlated with posterior alpha power . In addition , alpha power is negatively correlated with activity with dorsal attention network ( Sadaghiani et al . , 2010 ) , and positively correlated with activity in the default mode network ( Mo et al . , 2013 ) . Finally , intrinsic connectivity within networks is negatively correlated with alpha during eyes open vs . eyes closed resting state studies ( Wu et al . , 2010 ) , and both the phase and power of pre-stimulus alpha affects event-related BOLD responses in sensory regions ( Mayhew et al . , 2013; Walz et al . , 2015; Scheeringa et al . , 2011 ) . Together these results suggest that the increase in IPS connectivity and decrease in IPS alpha power may reflect a common process . We also show that threat reduces the power of resting state alpha oscillations . These oscillations are thought to reflect cortical inhibition , driven by inhibitory interneurons , and triggered by top-down modulatory control ( Klimesch et al . , 2007 ) . According to this view , our current findings may be driven by a release from top-down inhibition , resulting in a net increase in cortical excitability . Importantly , reductions in alpha power ( i . e . increases in excitability ) are associated with enhanced sensory and motor processing ( Cornwell et al . , 2007; Baas et al . , 2006 ) . For instance , reduced pre-stimulus alpha is associated with enhanced visual and illusory visual perception ( Lange et al . , 2013 ) , and increased pre-stimulus alpha is associated with reduced transcranial magnetic stimulation-induced motor-evoked potentials ( Klimesch et al . , 2007 ) . Accordingly , alpha reduction ( i . e . cortical excitation ) may provide a mechanistic explanation for one of the most commonly reported symptoms of elevated state anxiety , namely anxiety-potentiated startle ( Schmitz and Grillon , 2012; Grillon and Ameli , 1998; Grillon et al . , 1991 ) . That is , threat of shock potentiates the startle reflex by increasing cortical excitability . Cognitively , alpha oscillations are thought to play a key role in selective attentional processes ( Klimesch , 2012 ) , such that increases in alpha typically reflect inhibition or filtering of information that is to-be-ignored ( Bonnefond and Jensen , 2013; Kelly et al . , 2006; Händel et al . , 2011 ) . For instance , alpha oscillations have been shown to filter out noise during distracting listening conditions ( Strauß et al . , 2014 ) . Similarly , increases in pre-stimulus alpha to predictable stimuli are lateralized to to-be-ignored locations ( Horschig et al . , 2014; Rihs et al . , 2009 ) . In addition , alpha power increases during the maintenance of items in working memory ( WM ) ( Klimesch et al . , 2007; Meyer et al . , 2013 ) , and this increase in alpha serves to protect these items from distractors ( Bonnefond and Jensen , 2012; Manza et al . , 2014 ) . Taken together , these results suggest that the decreases in alpha power observed in our study may reflect greater perceptual sensitivity , consistent with the idea that threat may increase vigilance ( Eilam et al . , 2011 ) . In addition , alpha-gamma coupling is thought to be important for allocating attention to unattended salient stimuli ( Jensen et al . , 2012 ) . The reductions in alpha power observed here were lateralized to the left hemisphere , leading to a right dominant parietal asymmetry . According to prominent models , left dominant frontal alpha is associated with positive affect and/or approach behavior , while right dominant alpha is associated with negative affect and/or avoidance behavior ( Davidson , 2004; Harmon-Jones et al . , 2010 ) . Like the valence model of alpha asymmetry , one prominent model on parietal alpha asymmetry is rooted in the arousal-valence model of emotional processing ( Heller , 1993 ) . According to this model , right dominant parietal alpha is associated with increased arousal . Consistent with this theory , we find that threat of shock , which increases arousal , also reduces left parietal alpha , resulting in a right dominant profile . However , more research should be conducted to specifically test this hypothesis . Although our results only provide indirect evidence for the hypothesis that threat facilitates orienting , a bias toward hyper-orienting during periods of elevated state anxiety could explain many of the conflicting findings related to the cognition/anxiety interaction . The relationship between cognition and anxiety has been extensively studied ( For reviews , see [Eysenck et al . , 2007; Robinson et al . , 2013b] ) . Importantly , threat of shock improves performance on sustained attention tasks ( Torrisi et al . , 2016; Grillon et al . , 2016; Robinson et al . , 2011 ) , while impairing performance on working memory tasks ( Vytal et al . , 2016 , 2013; Patel et al . , 2016; Balderston et al . , 2016 ) . These tasks can be distinguished from one another based on the locus of attention required for performance . In sustained attention tasks , subjects are constantly monitoring the external environment for odd-ball stimuli , while in working memory tasks , subjects are required to maintain information internally . Our hypothesis is that threat-induced hyper-orienting improves performance on sustained attention tasks by reducing lapses in attention ( Torrisi et al . , 2016 ) . In contrast , we hypothesize that threat-induced hyper-orienting impairs performance on working memory tasks by lowering the threshold for distractors to gain access to WM resources ( Balderston et al . , 2016 ) . Consistent with these hypotheses , high pre-stimulus alpha is associated with better memory performance while low pre-stimulus alpha is associated with better perception performance ( Klimesch et al . , 2007 ) . This study had a number of strengths . First , we used the gold standard threat of shock paradigm , which has been extensively applied to manipulate state anxiety within subjects ( Grillon , 2008; Grillon and Baas , 2003 , 1998; Grillon et al . , 1991 , 2007 , 2008 , 2009; Balderston et al . , 2017a , 2017b; Cornwell et al . , 2007 , Cornwell et al . , 2008 , 2012; Lissek et al . , 2007 ) . Unlike previous studies , we also collected anxiety ratings throughout the recordings , and found that these ratings were similar across studies and correlated with anxiety-potentiated startle within session . Similarly , retrospective anxiety ratings were similar across studies , suggesting a comparable anxiety induction in both studies . We also collected data from multiple imaging modalities that both support the finding of enhanced IPS processing during periods of threat . In addition , we used the state-of-the-art multi-echo fMRI acquisition , and accompanying echo time-dependent independent components analysis to eliminate non-BOLD sources of noise from our fMRI data . Although there were a number of strengths to our study , there were also several limitations . First , as mentioned above , our MEG study was not optimally designed to reliably detect differences in connectivity , which would have been an obvious parallel to the fMRI connectivity analysis . Second , because we had a relatively small sample size , this study was not optimally designed to study how individual variability in anxiety affected IPS activity/connectivity . Third , our initial goal was to collect both the fMRI and the MEG data in all subjects , but a number of subjects could not participate in both , making it difficult to compare the results at the single subject level . Fourth , given our interest in the relationship between alpha and connectivity , it might have been better to collect simultaneous fMRI and EEG . However , because of our interest in the startle data , it was important to present white noise probes in at least one of the experiments , and our MRI scanner does not currently have that capability . To overcome that limitation , we collected startle data in the MEG study , which we used to validate the continuous anxiety ratings , which were collected during both studies . Finally , another limitation with the MEG data is that we do not have a measure of within-block movement , making it difficult to determine whether motion differed between the safe and threat blocks; however , it should be noted that any trials contaminated by movement ( i . e . muscle ) artifact were removed . Current translational neuroscientific models of anxiety focus on regions of the canonical fear network as anatomical hubs for anxiety . However , support for these models in humans is often based on studies that focus on these regions a priori ( i . e . seed-based functional connectivity ) , at the expense of rigorous , unbiased , whole-brain approaches . In this work we conduct two separate experiments , using complimentary imaging modalities ( fMRI and MEG ) to assess the effect of elevated state anxiety on functional connectivity and cortical excitability across the entire brain . In these studies , we identify effects in a common region , the IPS , which is a key node in the frontoparietal attention network . These results suggest that threat enhances processing in this region , possibly facilitating attentional processing , leading to increased vigilance . However , these results have broader implications for future research and treatment . First , our results suggest that elevated anxiety in humans is primarily a cognitive state that is not fully captured by the current translational models . Second , because the most prominent region contributing to elevated state anxiety is cortical , it may be possible to target this region with neuromodulatory methods and reduce anxiety . Forty-two ( 28 female; age: M = 28 . 45 , SD = 6 . 23 ) healthy volunteers from the Washington DC metropolitan area were recruited to participate in the current study . Of these , 32 ( 18 female; age: M = 27 . 82 , SD = 5 . 11 ) participated in the magnetoencephalography ( MEG ) study , and 30 ( 19 female; age: M = 28 . 52 , SD = 6 . 75 ) participated in the functional magnetic resonance imaging ( fMRI ) study . Among the subjects included in the final analysis , 18 ( 13 female; age: M = 27 . 67 , SD = 4 . 76 ) participated in both . For the MRI study , two subjects were removed due to technical malfunction with the scanner/data , three subjects were removed because their anxiety ratings were two standard deviations below the mean . For the MEG study , three subjects were removed from the final analysis for sleeping , moving , or not paying attention to the task . One subject withdrew in the middle of the recording . The final sample included 38 total subjects ( MEG N = 28; MRI N = 25 ) : 18 subjects participated in both experiments , 10 participated only in the MEG experiment , and seven participated only in the MRI experiment . Although no explicit power analysis was conducted prior to the experiment , we chose a sample size of approximately 25 participants , to ensure enough power to detect a behavioral threat of shock effect based on previous studies ( Schmitz and Grillon , 2012; Balderston et al . , 2017b , 2016 ) . Following an initial telephone screen , participants visited the National Institutes of Health Clinical Center for a comprehensive screening by a clinician . Inclusion criteria for healthy volunteers were: ( 1 ) no current Axis I psychiatric disorder as assessed by SCID-I/NP ( First et al . , 2012 ) , ( 2 ) no first-degree relative with a known psychotic disorder , ( 3 ) no interfering acute or chronic medical condition , ( 4 ) no brain abnormality on MRI as assessed by a licensed radiologist , ( 5 ) negative urine drug screen , and ( 6 ) right-handedness . All participants gave written informed consent approved by the National Institute of Mental Health ( NIMH ) Combined Neuroscience Institutional Review Board and received compensation for participating . Presentation software package ( version 14 . 6 , Neurobehavioral Systems , Berkeley , CA ) was used to present the stimuli via projection systems in both the MEG ( front ) and MRI ( rear projection ) studies . A 100 ms , 200 Hz train of stimulation was administered via a Digitimer constant current stimulator ( DS7A; Digitimer , Letchworth Garden City , UK ) . The transistor–transistor logic ( TTL ) pulse train used to trigger the train of stimulation was generated via a Grass stimulator ( SD9 , Warwick , RI ) . In the MEG experiment , this shock was delivered to the subjects’ right wrist via two 8 mm Ag/AgCl surface cup electrodes ( EL258-RT , Goleta , CA ) , filled with electrolyte gel ( GEL100 , Goleta , CA ) . In the MRI experiment , the same shock was delivered to the subjects’ right wrist via two 11 mm Ag/AgCl MRI-safe disposable sticker electrodes ( EL508 , Goleta , CA ) . The intensity of the shock could range from 0 mA to 100 mA and was calibrated prior to the experiment to a level that the subject rated as ‘uncomfortable but not painful’ . During the MEG study , subjects were exposed to several presentation of an acoustic white noise , to trigger an acoustic startle reflex used to assess anxiety . In order to avoid artifact due to the magnets and moving metal in traditional headphone drivers , we engineered a custom pneumatic system for generating white noise with air pressure . First we used a 3D printer to create plastic over-the-ear air vortex generators ( vortices ) . Then , we attached these to tubes connected to a solenoid , which was connected to an air tank . When triggered by a TTL pulse , the solenoid allowed air from the tank to pass through tubes to the air vortex generators . In doing so , the air generated a white noise with an intensity proportional to the air pressure released . Therefore , we calibrated the volume of the white noise to 103 dB by adjusting the pressure on the air tank regulator , and testing the intensity with a sound pressure level meter . During the experiment , subjects had continuous access to an online visual analogue scale that they could use to continuously update their anxiety rating during the task . Subjects controlled this scale using the response device provided for each experiment . In the MEG experiment , subjects used a custom-built fiber optic joystick . In the MRI experiment , subjects used a 4-button fiber optic response device ( Current Designs , Philadelphia , PA ) . The acoustic startle reflex was measured during the MEG study via electromyographic ( EMG ) activity of the eyeblink reflex recorded via 2 8 mm Ag/AgCl surface cup electrodes ( EL258-RT , Goleta , CA ) , filled with electrolyte gel ( GEL100 , Goleta , CA ) placed below the right eye over the orbicularis oculi muscle ( Blumenthal et al . , 2005 ) . EMG was recorded at 600 Hz via the MEG system amplifier and analyzed using a custom MATLAB script ( see Source code 1 ) . The EMG signal was extracted from the recordings , bandpass filtered ( 30–500 Hz ) , rectified , and smoothed with a 20 ms time constant . The peak startle/eyeblink magnitude was determined during the 20–100 ms after the onset of the noise presentation . The peaks were then transformed to z-scores and converted to t-scores within-subjects to reduce large inter-individual differences in the overall magnitude of the startle reflex ( Blumenthal et al . , 2005 ) , and to facilitate comparison with the online anxiety ratings . Subjects reported their anxiety level continuously throughout the experiment using the response device . They also received continuous feedback via a colored circle in the center of the screen , updated at 60 Hz ( A colored circle , as opposed to a moving cursor [Schultz et al . , 2012; Balderston and Helmstetter , 2010] , was chosen to minimize eye movements in the MEG study ) . As they updated their rating it updated the color of the circle , which ranged from white to red with 256 possible intermediate hues . Values representing these hues were used to numerically represent the subjects’ current anxiety level . During the MEG study , these ratings were sampled just prior to each startle probe for comparison with the startle magnitudes . Because startle probes were not presented during the MRI study , online ratings were sampled once per repetition ( TR ) . The values were then transformed to z-scores and converted to t-scores within-subjects to facilitate comparison with the startle responses . Prior to each experiment , subjects completed several standard psychological questionnaires , including the Spielberger State-Trait Anxiety Index ( STAI ) ( Spielberger , 1987 ) , the Anxiety Sensitivity Index ( ASI ) ( Peterson and Heilbronner , 1987 ) , the Beck Anxiety Inventory ( BAI ) ( Beck et al . , 1988 ) , and the Beck Depression Inventory ( BDI ) ( Beck and Steer , 1987 ) . At the start of both experiments and after each magnetoencephalography ( MEG ) /functional magnetic resonance imaging ( fMRI ) run , subjects were given a set of affective rating scales: ( 1 ) How anxious are you ( 1 = not anxious , 9 = extremely anxious ) ? ( 2 ) How afraid are you ( 1 = not afraid , 9 = extremely afraid ) ? ( 3 ) How would you rate the intensity of the electrical stimulation ( 1 = not painful at all , 9 = uncomfortable but not painful ) ? We collected four runs , each containing 245 multi-echo EPI images , using a 3T Siemens MAGNETOM Skyra ( Erlangen , Germany ) fMRI system , and a 32-channel head coil . For each image , we collected 32 interleaved 3 mm slices ( matrix = 64 mm×64 mm; FOV = 192 × 192 ) parallel to the AC-PC line ( TR = 2 s; TEs = 12 ms , 24 . 48 ms , 36 . 96 ms; flip angle = 70° ) . These 32 slices covered the entire cerebrum , but did not cover the most posterior parts of the brainstem and cerebellum . Slices were collected with an anterior-to-posterior phase encoding direction . We also collected two , 10 image multi-echo EPI series with the same parameters and the same field of view; however , one of these series was collected with a posterior-to-anterior phase encoding direction , and these series were used to correct for EPI distortion in the phase encoding direction ( Morgan et al . , 2004 ) . We also acquired a multi-echo T1-weighted MPRAGE ( TR = 2530 ms; TEs = 1 . 69 ms , 3 . 55 ms , 5 . 41 ms , 7 . 27 ms; flip angle = 7° ) . We acquired 176 , 1 mm axial slices ( matrix = 256 mm × 256 mm; field of view ( FOV ) = 256 mm × 256 mm ) , which were later co-registered to the EPI images . Functional images were preprocessed and analyzed using the AFNI software package ( see Source code 1 for processing and analysis scripts ) ( Cox , 1996 ) . EPI images for each run and each echo were first reconstructed , despiked ( i . e . single voxel outliers were truncated ) , slice-time corrected , and then deobliqued . Then each volume in the series was registered to the first volume , and skull-stripped . Preprocessed images were then entered into a multi-echo-independent components analysis using the meica . py script distributed with the AFNI software package ( Kundu et al . , 2012 ) . This analysis uses the T2* decay of BOLD signals , measured across the echoes to denoise the timeseries . The analysis first decomposes the timeseries into independent components using FastICA . Then it determines whether signal intensity across echoes decays in a manner consistent with what is expected from BOLD data . Components that fit the model are kept , components that do not ( i . e . components where the signal intensity does not decay across echoes ) are discarded . A new denoised timeseries is then synthesized from the components not discarded . This technique has been previously shown to robustly remove sources of noise corresponding to motion , physiology , and scanner artifact ( Kundu et al . , 2012 ) . The denoised timeseries for all runs are then registered to the first run , scaled , and further denoised using a general linear model with regressors of no interest . Regressors of no interest included the six motion parameters from the volume registration step , up-through third-order polynomials to model baseline drift , and hemodynamic response functions ( HRF ) s corresponding to button presses and shock deliveries . In addition , images where the derivative of the motion regressors from volume registration step had a Euclidean norm above 0 . 5 mm were censored ( ‘scrubbed’ ) from further analyses . All remaining images from the safe and threat blocks were concatonated into separate timeseries; however , as part of the denoising procedure , we removed neural responses related to both button presses and transitions from one block type to another . To correct for geometric distortion of the EPI images , the forward and reverse phase-encoding blips are first averaged across time , skull-stripped , and then rigid-body aligned with the reference image for the other EPI timeseries . Then , the two blips are non-linearly aligned to each other using the ‘plusminus’ flag in the AFNI program 3dQwarp , so that the resulting image is ‘in the middle’ of the two . This reference image can then be registered to the T1 image , and the voxelwise displacement map for the forward blip is then saved , so that it can be applied to the EPI timeseries . Although a standard linear registration approach would have possibly yielded similar results at the group level , the nonlinear approach we used leads to better T1/EPI within-subject registration , and has been shown to perform even better than when using fieldmaps ( Hong et al . , 2015 ) . To align the EPI data to Montreal Neurological Institute ( MNI ) space and to mask out non-grey matter voxels , the T1 data was processed as follows . First , the volumes for the T1 echoes were averaged , and the resulting volume was run through the standard Freesurfer processing pipeline ( Desikan et al . , 2006; Fischl et al . , 2004 ) . Next , the skull-stripped anatomy is non-linearly registered to MNI space using the MNI_avg152T1 template distributed with AFNI . In addition , masks that included all cortical and sub-cortical grey matter atlas regions for each subject . These were warped to MNI space , downsampled to the EPI resolution , and dilated by 1 voxel . A group grey matter mask was created by averaging these binary masks , and thresholding out voxels with less than 2/3 overlap ( Torrisi et al . , 2015 ) . Next the original space , skull-stripped anatomy was aligned to the distortion corrected EPI reference image in two steps . The first was a simple affine transformation . The second was a non-linear transformation using a criterion based on the local Pearson correlation ( Saad et al . , 2009 ) , and the inverse displacement map was saved . Finally , the following warps were applied to the EPI timeseries data: inverse warp from the distortion correction step , affine transformation matrix from the T1 alignment step , the inverse warp from the T1 alignment step , and the non-linear warp from the T1 to MNI transformation . Finally , the EPI data were masked with the group grey matter mask , and blurred within this mask using a 6-mm FWHM Gaussian kernel . To analyze the MRI recordings , we used a global brain correlation ( GBC ) approach ( Cole et al . , 2010 ) . In brief , we correlated each voxel in our grey matter mask with each other voxel in the mask , applied the Fisher’s Z transform , and summed across correlations . The result was a map where the value in each voxel reflected the strength of the correlation between that voxel and the rest of the brain . We conducted this GBC analysis for each subject and each condition , and used these values for further analysis . First , to identify changes in GBC across the entire brain , we averaged across voxels for safe and threat . We then conducted a paired-sample t-test on these averages . Next , to identify changes in GBC at the regional level , we analyzed the voxel-wise GBC for safe and threat . For this , we conducted a voxel-wise paired-sample t-test on the GBC values . Finally , to identify changes in connectivity with highly connected regions , we identified regions of interest ( ROI ) s from the voxel-wise GBC threat > safe t-test , and correlated the timecourse of activity in these ROIs with every voxel in the brain for safe and threat , and applied the Fisher’s Z . We then conducted paired-sample t-tests on the resulting connectivity maps . We used Monte Carlo simulations and a cluster-based method to correct for multiple comparisons across voxels . First , we estimated the smoothness in our residual timeseries using a Gaussian plus mono-exponential shaped function implemented by the ‘-acf’ option in the AFNI program 3dFWHMx , which addresses recent concerns over inflated Type one error in studies using the cluster correction method ( Cox et al . , 2016 ) . We calculated smoothness for each subject , and averaged this across subjects . Next , we simulated 10 , 000 random statistical parametric maps in 3dClustSim with a smoothness matching that of the original timeseries . For each simulation , we thresholded at a voxel-wise alpha of 0 . 005 , and extracted the largest cluster . We then compared our test statistics to the distribution of clusters across all simulations to identify a minimum cluster size threshold of 80 , corresponding to a two-tailed alpha of 0 . 05 . We recorded neuromagnetic activity at 600 Hz from 271 radial first-order gradiometers using a 275 channel CTF-OMEGA whole-head magnetometer ( VSM MedTech , Ltd . , Canada ) . Recording took place in a magnetically shielded room ( Vacuumschmelze , Germany ) , and Synthetic third-order gradient balancing was used for active noise cancellation ( Vrba and Robinson , 2001 ) . MEG recordings were preprocessed and analyzed using the Fieldtrip toolbox in MATLAB ( See Source code 1 for processing and analysis scripts ) ( Oostenveld et al . , 2011 ) . First , movement within runs was checked by comparing the position of the HPI coils from the beginning of the run to the end of the run , and any run with a root mean square movement value above 1 cm was excluded from the analysis . Next , startle probe onsets were identified , and the 2 s window prior to each trial was extracted , demeaned , and detrended . Next , muscle artifacts were identified using the ft_artifact_muscle function in Fieldtrip . Trials where muscle movements were identified were removed . Next the recordings were low-pass filtered with a 90 Hz cutoff , and notch filtered at 60 Hz to remove line noise . The recordings were then downsampled to 300 Hz , and submitted to an independent components analysis . Components were visually inspected , and those with a topography , and timecourse consistent with either blinks or heartbeats were identified for removal ( typically 1–2 per artifact ) . Rejected components were projected out of the dataset , and the singular value decomposition of the covariance matrix was inspected to determine the regularization factor ( lambda ) . A lambda of 5% was found to be sufficient to regularize the covariance matrix for source analysis following the removal of the rejected components . The result was a dataset cleaned for blink and heartbeat artifacts . Because we were specifically interested in alpha oscillations , we identified each subject’s individual alpha frequency ( IAF ) . We began by transforming the timeseries data into the frequency domain using a multi-taper fast Fourier transform ( mtmfft ) based on a set of discrete slepian sequences . In this initial transformation , we used a frequency window of 1 Hz to 20 Hz , and a single taper per frequency . We then averaged these spectrograms across sensors and across trials for each subject , and identified the largest local maxima in the average spectrogram for each subjects . For the majority of subjects ( 24/28 ) , the largest peak occurred in the alpha frequency band ( 8–12 Hz ) . For all other subjects , the average IAF was used for further analyses . Next we conducted a second mtmfft , using a 2 Hz window centered around each subject’s IAF , with two tapers per frequency and averaged the resulting power estimates across frequency for each sensor and each trial . This mean IAF power estimate was used in both the sensor space analyses and the source space analysis . Single subject T1 images were used to generate the forward model for the MEG source analysis . First the T1 images were aligned to a single subject template in MNI space . Next the brain surface was extracted , and a single shell head model was generated from this surface . Then a source model was created using a single subject template with current dipoles placed along a regular 8 mm grid inside the brain surface . The single subject MNI space images were aligned to CTF space ( i . e . coregistered to the sensors ) manually , guided by the vitamin E capsules placed over the fiducial points . The resulting transformation matrix was applied to both the head model and the source model , and alignment between the sensors , head model , and source model was visually inspected . Finally , leadfields were then created using the location of the sensors , head model , and source model dipole locations . Because we were interested in frequency information ( as opposed to time ) , we used the dynamic imaging of coherent sources ( DICS ) ( Gross et al . , 2001 ) technique to localize the sources of our recordings . We began by computing the cross spectral density matrix ( CSD ) from the frequencies of interest from all trials in the analysis . We then estimated the beamformer filter using the CSD , leadfields ( with fixed orientations ) , headmodel , and gradiometer locations . Once estimated , this common filter was applied to the safe and threat conditions independently . We compared IAF power in safe vs . threat . For this , we averaged the IAF power across trials independently for safe and threat for each subject and each sensor , and conducted a paired sample t-test on these averages . For both analyses , we used Monte Carlo simulations and a cluster-based method to correct for multiple comparisons across sensors . We calculated 1000 random permutations , where condition labels were shuffled across subjects . For each permutation , we thresholded the shuffled results at a sensor-level alpha of 0 . 005 , summed the t-value across sensors in each cluster , and extracted the largest summed t-value . We then compared our test-statistic to the distribution of summed t-values and discarded any clusters where the summed t-value was smaller than the summed t-value corresponding to a two-tailed alpha of 0 . 05 . As with the sensor-level analyses , to compare power in the safe and threat conditions , we projected the average IAF power into source space independently for each condition using the common filter calculated from all trials . We then conducted a paired-sample t-test on these power estimates at each voxel within the source model . As with the sensor-space data we used Monte Carlo simulations and a cluster-based method to correct for multiple comparisons across voxels . As before we calculated 1000 random permutations , where condition labels were shuffled across subjects . For each permutation , we thresholded the shuffled results at a source-level alpha of 0 . 005 , summed the t-value across sensors in each cluster , and extracted the largest summed t-value . We then compared our test-statistic to the distribution of summed t-values and discarded any clusters where the summed t-value was smaller than the summed t-value corresponding to a two-tailed alpha of 0 . 05 .	Anxiety disorders affect around one in five Americans , and in many cases people experience anxiety so intensely that they have difficulties performing day-to-day activities . To help these people , it is important to understand how anxiety works . Current research suggests that anxiety disorders are caused when the connections in the brain that control our response to threat are either excessively or inappropriately activated . However , it was not clear what causes the anxiety to last for long periods . To better understand this phenomenon , Balderston et al . studied the brains of over 30 volunteers using two types of measurements called magnetoencephalography and fMRI . In the each experiment , participants experienced periods of threat , where they could receive unpredictable electric shocks . In the first experiment , Balderston et al . measured the brain activity by recording the magnetic fields generated in the brain . In the second experiment , they used fMRI to record changes in the blood flow throughout the brain to measure how the different regions in the brain communicate . The recordings identified a single part of the brain that increased its activity and changed its communication pattern with the other regions in the brain , when people are anxious . This region in a part of the brain called parietal lobe , is also important for processing attention , which suggests that anxiety might make people also more aware of their surroundings . However , this extra awareness might also make it more difficult for people to concentrate . Future studies may be able to stimulate this area of the brain through the scalp to potentially reduce anxiety , as the affected area is close to the skull .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2017	Threat of shock increases excitability and connectivity of the intraparietal sulcus
Climate change is accelerating plant developmental transitions coordinated with the seasons in temperate environments . To understand the importance of these timing advances for a stable life history strategy , we constructed a full life cycle model of Arabidopsis thaliana . Modelling and field data reveal that a cryptic function of flowering time control is to limit seed set of winter annuals to an ambient temperature window which coincides with a temperature-sensitive switch in seed dormancy state . This coincidence is predicted to be conserved independent of climate at the expense of flowering date , suggesting that temperature control of flowering time has evolved to constrain seed set environment and therefore frequency of dormant and non-dormant seed states . We show that late flowering can disrupt this bet-hedging germination strategy . Our analysis shows that life history modelling can reveal hidden fitness constraints and identify non-obvious selection pressures as emergent features . Study of climate effects on phenology have quantified shifts in the timing of developmental transitions such as flowering , bud burst , and migration . These show that warmer temperatures are advancing plant phenology , with bud burst and flowering occurring earlier and the growing season becoming longer ( Menzel and Fabian , 1999; Fitter and Fitter , 2002; Cotton , 2003; Parmesan and Yohe , 2003; Cleland et al . , 2007 ) . This is an adaptive plant response to shifting temperatures , because for many species the flowering date is unaffected by climate change ( Fitter and Fitter , 2002 ) , and plants that couple their phenology to temperature are more likely to have populations resilient to climate change ( Willis et al . , 2008 ) . Why many plants have evolved to behave in this manner is currently unclear . The model plant Arabidopsis thaliana is widely distributed in the Northern hemisphere and can be used to analyze phenology , including control by individual genes ( Wilczek et al . , 2009; Chiang et al . , 2013 ) . In northern Europe , Arabidopsis exhibits a primarily winter annual life history , but in central and southern Europe , summer rapid cycling and summer annuals appear alongside winter annuals ( Thompson , 1994 ) . Importantly , A . thaliana is a ruderal species , primarily colonizing frequently disturbed habitats: the long-term persistence of such populations ( and therefore fitness ) is strongly dependent on seed behaviour and seed bank formation ( Grime et al . , 1981 ) . During the vegetative phase , plants have signalling pathways that couple progression to flowering to temperature and photoperiodic cues , and these pathways have been elucidated genetically under laboratory conditions ( Andrés and Coupland , 2012 ) . These same pathways are subject to positive selection in natural populations ( Toomajian et al . , 2006 ) . However , when late flowering Arabidopsis mutants are analysed under field conditions they delay reproduction by only a few days , leaving the significance or their role unclear ( Wilczek and et al . , 2009; Chiang et al . , 2013 ) . Temperature during seed set also strongly influences life history by modulating seed dormancy after shedding ( Fenner , 1991; Schmuths et al . , 2006; Chiang et al . , 2009; Kendall et al . , 2011 ) , but in contrast to the control of flowering time comparatively little attention has been paid to the role of this process in life history generation and adaptation . Given that the environment during seed set is determined by flowering time , we hypothesized that understanding the role of either flowering time or dormancy control in population life history required a detailed integrated study of both traits . To address this problem , we therefore constructed and parameterized a field-validated model of Arabidopsis life history for the Columbia-0 ( Col-0 ) accession . Previously , photothermal models have been used to predict Arabidopsis flowering time under field conditions ( Wilczek et al . , 2009 ) . In order to predict seed set conditions in the wild , we sought to extend this treatment to include the reproductive phase in order to determine the timing of seed set on plants germinated and grown at different times of the year . To parameterize this model , we first grew plants from bolting to first set seed under a range of temperatures under laboratory conditions ( Figure 1A ) . We found a simple linear relationship between temperature and the number of days from bolting to first seed set , showing that seed set timing depended on ambient temperature . Our data suggested that seed set was much less dependent on photoperiod , with only photoperiods of 8 hr significantly delaying seed set ( Figure 1—figure supplement 1 ) . With these data , we constructed and optimized a thermal time model from first flowering to first seed set , and compared the predicted timing of seed set to that of plants setting seed under field conditions in York , UK in 2012 and 2013 ( Figure 1B ) . The model was a good predictor of seed set timing in the field ( R2 = 0 . 43 ) for seed set in York in 2012 and 2013 , and outperformed models that also included photoperiod components . By combining this with a previously published bolting time model ( Wilczek et al . , 2009 ) , we could predict Col-0 first seed set date for any germination date . 10 . 7554/eLife . 05557 . 003Figure 1 . Flowering time and seed set control constrain mean temperature at seed shedding for winter and spring annuals . ( A ) Laboratory experiments from bolting to first seed set at various constant temperatures used to constrain the seed set thermal time model . Data represent the mean and standard error of a minimum of five independent plants per treatment . ( B ) Field trial testing of the seed set model using nine growings of a minimum of five plants showing mean and standard error of bolting date and time to seed set . Closed circles show data while model prediction is shown by the continuous line . ( C ) Map of Central Europe showing the sites for which temperature data was gathered to simulate Col-0 behaviour . ( D ) Simulation of flowering time ( grey ) using the model from ( Wilczek and et al . , 2009 ) and first seed set model ( black ) using temperature data from Gorsow , Poland . Flowering and first seed set date ( y-axis ) can be determined for each possible germination date ( x-axis ) . Note that seed set dates are similar for germination between September and April . ( E ) Predicted first seed set date ( closed squares ) , predicted mean temperature at first seed set ( open circles ) , and mean annual temperature ( closed circles ) for October/November germination dates at each site . Seed set is shown to be clustered around a mean daily temperature of 14 . 5°C . DOI: http://dx . doi . org/10 . 7554/eLife . 05557 . 00310 . 7554/eLife . 05557 . 004Figure 1—figure supplement 1 . The growth rate of Arabidopsis Col-0 plant between first flowering and first seed set at photoperiods between 8 and 16 hr daylength at 22°C . Data represent mean and standard error of five plants per treatment . DOI: http://dx . doi . org/10 . 7554/eLife . 05557 . 004 The precise location for the collection of Col-0 is uncertain but it is known to originate from central Europe ( Robbelen , 1965; C Dean Personal communication ) . Simulated at the most likely collection site , Gorsow , Poland , a key feature of the model is that germination times from September to April ( winter and spring annuals ) lead to a very similar seed set time in late May ( Figure 1C ) , a behaviour caused in part by growth rate and flowering time control ( Wilczek and et al . , 2009 ) but further enhanced by the seed set responses to temperature . Seed germinating from May to July set seed later in the same summer ( rapid cycle ) , whereas August germinators have a wide range of possible flowering times ( Wilczek and et al . , 2009 ) . To understand how this would vary in response to shifting climate norms , we simulated the flowering and seed set models using climate data across a transect covering 25° of latitude from Arkangel'sk , Russia , to Valencia , Spain ( Figure 1D , E ) . Warmer climates advanced the seed set date , with 1°C rises in annual temperature giving about a 7-day advance in seed set time for winter annual cohorts from late June in Arkhangel'sk to early April in Valencia ( Figure 1E ) . This behaviour has been widely observed in phenological studies , but our analysis also revealed a striking effect on temperature at first seed set: although mean annual temperature at the sites varied between 0 . 5°C and 17 . 5°C , winter annual seed set temperature was maintained at 14–15°C independent of latitude ( Figure 1E ) . Therefore , a key emergent feature of the model is that the acceleration of flowering and seed set timing by warmer temperatures serves to stabilize the temperature range during which seed set begins . Lower temperatures during seed production enhance seed dormancy in most angiosperm species ( Fenner , 1991 ) , but rarely has this response been characterized in detail . Laboratory experiments revealed that Col-0 dormancy appeared related to the mean daily temperature and was not clearly influenced by daily temperature cycles ( Figure 2—figure supplement 1 ) . To understand precisely how Col-0 seed dormancy responds to temperature during seed set , we matured seeds between 10°C and 20°C in the laboratory and germinated them following incubation treatments at temperatures between 4°C and 20°C ( Figure 2 ) . Our results revealed a clear bi-phasic behaviour: At or below 14°C , seeds produced are very dormant and do not germinate at high levels even after prolonged incubations under any temperatures . This is not due to low viability because seeds set at 10°C are viable at harvest ( Kendall et al . , 2011 ) . In contrast , above 14°C , high levels of germination were possible especially where seeds were incubated at similar or lower temperatures than experienced during seed set . At these temperatures , seeds incubated in the dark experienced a transient period of primary dormancy loss during which time germination could occur if the seeds were given light , followed by secondary dormancy induction . This high sensitivity of dormancy to a 1°C increase in seed maturation temperature supports our previous observation of a general increase in the temperature-sensitivity of gene expression in developing seeds relative to vegetative tissues ( Kendall et al . , 2011 ) . 10 . 7554/eLife . 05557 . 005Figure 2 . The germination physiology of Arabidopsis Col-0 in response to temperature during seed set and imbibition . Germination of seed matured between 12°C and 18°C incubated in the dark between 4°C and 20°C ( see legend ) and placed to germinate in the light at 22°C after the indicated time periods . Data points represent mean and standard error of five seed batches . DOI: http://dx . doi . org/10 . 7554/eLife . 05557 . 00510 . 7554/eLife . 05557 . 006Figure 2—figure supplement 1 . Comparison of the relative effects of constant 16°C vs a 24-hr temperature cycle with a mean of 16°C on Col-0 seed dormancy . Dormancy levels of seed lots were revealed by cold stratification in the dark at 4°C ( incubation time ) followed by germination at 22°C in white light . Data points represent the mean and standard error of five biological replicate seed batches . DOI: http://dx . doi . org/10 . 7554/eLife . 05557 . 00610 . 7554/eLife . 05557 . 007Figure 2—figure supplement 2 . Global fit of the seed germination model output ( red ) with the time-series training data ( blue ) for all seed maturation temperatures ( Tm ) and imbibition temperatures ( Ti ) . Predicted decline in primary dormancy ( green ) and increase in secondary dormancy ( black ) are shown alongside germination data and model fit . The key feature of the model is that secondary dormancy varies with TI , but not with TM , and that primary dormancy varies with TM , but not with TI . DOI: http://dx . doi . org/10 . 7554/eLife . 05557 . 00710 . 7554/eLife . 05557 . 008Figure 2—figure supplement 3 . Fit of model to lab-generated data not used to train the model . Data were collected using seed matured at 16°C and incubated at five indicated temperatures . Data represent the mean and standard error of five independent seed batches per test . Model predicted germination is shown in red , predicted primary dormancy in green , and predicted secondary dormancy in black . DOI: http://dx . doi . org/10 . 7554/eLife . 05557 . 00810 . 7554/eLife . 05557 . 009Figure 2—figure supplement 4 . Germination kinetics of field-collected seed batches from 2012 and model prediction of temperature during seed maturation . A ) Seed from five independent growings was sown in the lab and dormancy was assayed as described previously for lab-grown seed at five incubation temperatures . Data represent the mean of at least five independent seed lots per growing . ( B ) The temperature history of five field-set seed lots ( black ) , and the predicted mean seed maturation ( red ) temperature ( red ) by the seed germination model . DOI: http://dx . doi . org/10 . 7554/eLife . 05557 . 009 To allow simulation of the whole life history , we developed a predictive framework for Col-0 germination probability and constructed a germination model based on the following assumptions used previously for similar studies ( Totterdell and Roberts , 1979; Bradford , 2005; Batlla et al . , 2009 ) : ( 1 ) germination is possible in the absence of dormancy; ( 2 ) primary dormancy depth ( dormancy generated during seed maturation ) is fixed at harvest and is lost during imbibition; ( 3 ) the onset of secondary dormancy ( dormancy induced de novo after imbibition ) begins independently during seed imbibition . Primary dormancy loss and secondary dormancy induction were represented as cumulative logistic functions ( see ‘Materials and methods’ ) . We noted that secondary dormancy appeared to be induced faster at warmer imbibition temperatures ( Figure 2 ) , and fixing this relationship in the model revealed that close matches to experimental data could be produced if initial primary dormancy was modelled as a function of the temperature during seed maturation ( see ‘Materials and methods’ and Figure 2—figure supplements 2–4 ) . In fact , a model with four parameters in which primary dormancy loss was dependent linearly on maturation temperature and secondary dormancy induction was dependent exponentially on imbibition temperature could be parameterised to provide a good fit to experimental data generated between 12 and 20°C ( R2 = 0 . 84; Figure 2—figure supplement 2 ) and could match germination data in an experimental series set at 16°C and not used to train the model ( Figure 2—figure supplement 3 ) . In order to determine whether our model could predict the germination of seed set under field conditions , we collected five seed batches during 2012 in a field site in York , UK and germinated them under multiple temperature regimes in the laboratory ( Figure 2—figure supplement 4 ) . Behaviour of these lots was qualitatively similar to that observed in the laboratory . We then fitted the parameterised model to each field germination data set , generating a prediction for the seed maturation temperature for each seed lot . This was then compared to the temperature history of each batch in the field ( Figure 2—figure supplement 4 ) . In each case the model-predicted temperature was close to the daily mean temperature in the last days before harvest for each seed batch , as has been shown previously in barley ( Gualano and Benech-Arnold , 2009 ) . Therefore , our model captures field-relevant relationships between maturation temperature and germination behaviour . To understand the germination potential of seeds set at different times of the year , we ran the model for the locations in our transect ( Figure 3 ) . The germination model predicts behaviour with two principle states obvious from the data: Firstly , mean temperatures below 14°C during seed set strongly inhibit germination . Secondly , above 15°C the interaction between maturation temperature and imbibition temperature generally permits high germination rates . This is because faster gain of secondary dormancy after imbibition at warmer temperatures is offset by the lower level of primary dormancy induced during seed maturation . Warmer climates have a longer summer window of high germination until at very warm locations the model predicts germination inhibition at the height of summer , due to fast secondary dormancy induction . Our models show that the predicted transition from dormancy to germination takes place as the seed set temperature rises above 14°C , coinciding with the time of first seed set for winter annual cohorts , regardless of location ( Figure 3 ) . Therefore , simulated Col-0 phenology generates a coincidence of seed set timing with germination physiology which is independent of climate over 25° of latitude , demonstrating that this aspect of Col-0 phenology is strongly buffered against variation in temperature . Our model predicts that the low dormant progeny will enter a rapid cycle if the climate permits , flowering and setting seed later the same summer ( Figure 3 ) . This switch therefore has the potential to control the proportion of seeds entering the seed bank and those that emerge immediately for a rapid cycle generation for production of further seeds . 10 . 7554/eLife . 05557 . 010Figure 3 . Simulation of the interaction of Col-0 life history stages at sub-arctic to sub-tropical sites . Germination dates are given on the x-axis . On the y-axis flowering time ( red line ) and first seed set ( black line ) are shown alongside the simulated progeny germination frequency ( blue-hued heat map ) based on the mean temperature over 1 week before shedding ( seed maturation temperature ) and temperature after shedding ( imbibition temperature ) . Behaviour of winter annuals can be observed using September–November germination dates , and ability to complete a summer rapid cycle can be ascertained by progression to seed set in the same year for summer germination dates . The model is simulated for European locations on the transect introduced in Figure 1 , and for the more sub-tropical locations of the Canary Islands and Cape Verde ( Sal ) , both locations at which Arabidopsis thaliana populations have been described . DOI: http://dx . doi . org/10 . 7554/eLife . 05557 . 01010 . 7554/eLife . 05557 . 011Figure 3—figure supplement 1 . The coincidence between seed set timing and the dormancy state transition is preserved during artificial warming and cooling simulations , based on 2°C increments from the mean temperature series in Gorsow , Poland . Predicted germination is shown in the greyscale heat-map , bolting and first seed set timing in green and red , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 05557 . 011 Given the high sensitivity of Col-0 seed behaviour to temperature ( Figure 2 ) , we hypothesized that moderately delayed flowering could affect progeny life histories . To test this hypothesis , we ran the model using previously published parameter sets for simulating the gigantea ( gi ) and vernalisation insensitive 3 ( vin3 ) ( Sung and Amasino , 2004; Toomajian et al . , 2006; Wilczek and et al . , 2009 ) for Gorsow , Poland ( Figure 4 ) , although in principle our model behaviour is relevant to large areas of central Europe ( Figure 1; Figure 3 ) . GI confers a weak late flowering in field conditions of 7–10 days during summer while vin3-1 mutants generate a longer delay of 25–50 days ( Sung and Amasino , 2004; Wilczek and et al . , 2009 ) . Thus , these are representative of any natural variants with mild and more severe delays in flowering affecting temperature or photoperiod sensitivity . This is more informative than considering strong FRI alleles because these do not substantially delay flowering of winter annuals ( Wilczek and et al . , 2009; Chiang et al . , 2013 ) . Both gi and vin3 mutants were predicted to set seed later in the year than wild type ( Wilczek and et al . , 2009 ) , producing a greater proportion of seeds above the 14 . 5°C threshold and theoretically giving rise to more seeds capable of immediate germination . Given the time of year , these germinating seeds would be committed to a rapid cycling life history ( Figure 1D ) . Assuming that seed set lasts for 1 month and follows a normal distribution , simulated across our transect the model predicts that gi mutations could double the number of low-dormant seeds in northern and central Europe but would have little effect at lower latitudes ( Figure 4B ) . Mean temperature at shedding was raised from 14°C to 15°C for gi mutants at sensitive sites ( Figure 4A , B ) . For vin3 the model predicts a rise in the mean temperature at maturation of almost 5°C , enough to commit most seeds to rapid cycling and severely curtail entry into the soil seed bank ( Figure 4C ) . Our analysis therefore suggests that small variations in flowering time may have dramatic effects on progeny life history in Arabidopsis by affecting seed dormancy , providing evidence for a previously unrecognized mechanism through which flowering time can affect plant fitness . This conclusion is supported by a recent study that shows that changing flowering time affects seed dormancy under field conditions more than flowering time itself ( Chiang et al . , 2013 ) . 10 . 7554/eLife . 05557 . 012Figure 4 . The predicted effect of genetic variation in flowering time on seed set temperature and progeny behaviour . ( A ) Time to bolting for the gi-201 and vin3-1 mutants lines was calculated using previously published parameter sets ( 7 ) for Gorsow , Poland and overlayed on ten year average mean temperature ( blue ) and model-predicted germination ( black ) , assuming that both GI and VIN3 affect germination only indirectly via flowering time . ( B ) Predicted mean daily temperature at first seed set for winter annual Col-0 ( closed circles ) , gi ( open circles ) , and vin3 ( closed triangles ) across the European climate transect ( Figure 1D ) , using the flowering time and seed set models . Note that sensitivity of life history parameters to GI increases with increasing latitude with wider variation in annual photoperiod . ( C ) Assuming germination is spread evenly over 1 month after first seed set , predicated germination of the whole progeny of Col-0 , gi and vin3-1 winter annual seeds across the transect . Later flowering has the potential to lead to large changes in the number of low-dormant seeds due to the extreme sensitivity of seed set to temperature in this range ( Figure 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05557 . 012 We have revealed that Arabidopsis Col-0 seed dormancy is remarkably sensitive to a 1°C rise in mean temperature during seed set from 14°C to 15°C ( Figure 2 ) . However , until now the significance of the response of primary dormancy to the temperature during seed maturation has not been understood . A central feature of Arabidopsis phenology is the ability to generate genetically identical cohorts of seeds with contrasting life histories , such as winter annuals and rapid cyclers ( Montesinos-Navarro et al . , 2012 ) . Our data and models show that in a striking coincidence winter annual seeds are set when the mean temperature is approximately 14–15°C , the temperature at which a major switch from the production of dormant to non-dormant seeds occurs . This inevitably divides progeny into at least two distinct cohorts: obligate germinators which become summer rapid cyclers , and seeds which enter the soil seed bank even if initially exposed to light ( Figure 5 ) . Notably , a similar transition has been observed in Capsella bursa-pastoris seeds: in this case there is a transition in seed coat colour during seed set that alters dormancy and correlates with flowering time ( Toorop et al . , 2012 ) . In Arabidopsis seed coat , pigmentation is affected by the temperature during seed set ( MacGregor et al . , 2015 ) , again demonstrating control by the mother plant . Similar bet-hedging germination strategies have been described frequently across flowering plants suggesting wide general relevance ( Childs et al . , 2010 ) , but further studies are required to understand whether this is a common strategy for Arabidopsis in locations where temperatures are warm enough to permit life history variation . In addition , our study only considers annual temperature and photoperiod variation and other factors , notably water availability , are also known to strongly affect germination and growth rates . 10 . 7554/eLife . 05557 . 013Figure 5 . Scheme to show the role of winter annual flowering time control in soil seed bank formation . Seed bank persistence requires that seed entry rate ( production of dormant seeds ) exceeds exit rate , the sum of germination , death or deep burial . Width or arrows indicates relative flux variation with flowering time . Early flowering plants produce larger proportions of dormant seeds , increasing seed bank size . Later flowering plants produce fewer dormant seeds but more rapid cyclers , potentially producing more seeds from the next generation . DOI: http://dx . doi . org/10 . 7554/eLife . 05557 . 013 Because seed bank loading is a key component of fitness in ruderal plant populations ( Grime , 1988 ) , control of progeny behaviour likely generates a major selection pressure on flowering time control . Genetic analyses of Arabidopsis fitness have frequently considered yield to be paramount , but seed behaviour and seed bank dynamics are central to long-term population persistence: in Norway the mean lifespan of Arabidopsis in the seed bank was estimated at between 1 and 8 years ( Lundemo et al . , 2009 ) , but work to understand genetic contributions to seed bank dynamics is in its infancy . Col-0 germination in autumn or spring gives a similar flowering date in May . This suggests that Col-0 can operate equally well as a summer or winter annual . This prediction is supported by field emergence data , which shows that seed set in the laboratory at 15°C or seed set in the field conditions germinate in the soil in both autumn and spring emergence windows ( Figure 6 ) . This bet-hedging strategy can therefore be relevant to both summer and winter annual accessions . Previously it has been shown that seeds of the Cape Verde Island accession germinate only in an autumn emergence window in central England ( Footitt et al . , 2011 ) , whereas accession Bur-0 shows only a spring emergence window ( Footitt et al . , 2013 ) . However , our work shows that winter and spring annual behaviour does not necessarily require different genotypes and are not mutually exclusive strategies . Both spring and autumn germination windows have also been described in coastal but not montane Spanish populations ( Montesinos et al . , 2009 ) , suggesting that the strategy employed by Col-0 is of wide relevance in some ecological contexts . 10 . 7554/eLife . 05557 . 014Figure 6 . Field emergence time in 2013/14 of Col-0 seed set in the laboratory at 15°C or set in the field in York in spring 2013 . Data represent the total percentage emergence at 2 weekly intervals of 500 seeds sown for each experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 05557 . 014 In the case of Col-0 , winter annual behaviour appears to require ( 1 ) a strong slowing of growth rates under low temperatures , showing that growth rate can be as important as leaf number in determining timing of the floral transition in natural populations , and ( 2 ) strong tendency towards fast secondary dormancy induction combined with an autumn germination window . In contrast , strong FRI and FLC alleles and other late flowering alleles also confer flowering delays primarily on summer germinants ( Wilczek and et al . , 2009 ) . This achieves winter annual behaviour with a second mechanism that permits early seed germination . The increased prevalence of strong FLC alleles at northern latitudes ( Shindo et al . , 2006; Li et al . , 2014 ) is good evidence that an advantage of this latter strategy is maximisation of energy capture in short growing seasons through early seedling establishment , rather than generation of winter annual behaviour itself . At mid-latutides stronger FRI/FLC alleles have the potential to push back seed set of summer germinants into autumn ( Wilczek and et al . , 2009 ) , and during autumn the temperature again passes back through the critical 14–15°C range . This may then stratify the behaviour of progeny seeds of these cohorts . In addition to flowering time , latitudinal clines in seed dormancy have also been described , whereby accessions form more northerly latitudes exhibit lower primary dormancy levels ( Atwell et al . , 2010 ) . Given that decreasing temperature during seed maturation strongly increases primary dormancy , our analysis suggests that these opposing genetic and environmental trends could offset each other in field conditions allowing maintenance of progeny behavioural strategies . Variation at the DELAY OF GERMINATION1 ( DOG1 ) locus is a major determinant of natural variation in primary seed dormancy ( Bentsink et al . , 2006; Chiang et al . , 2011 ) and temperature-regulation of DOG1 transcript levels during seed maturation ( Chiang et al . , 2011; Kendall et al . , 2011; Chiang et al . , 2013 ) may therefore play an important role in adaptation of seed dormancy to changing temperatures . Recent articles also show that it is possible to model Arabidopsis populations over several generations using thermal time models ( Donohue et al . , 2014; Burghardt et al . , 2015 ) . The process reveals theoretical interactions between primary dormancy and flowering time that can recreate aspects of Arabidopsis phenology observed previously under field conditions at different latitudes , including the ability of Col-0 to adopt winter annual , summer annual , and rapid cycling life histories described here . Our analysis also shows that germination behaviour not included in these models such as temperature regulation of primary dormancy depth and secondary dormancy kinetics are necessary to understand life history control . Combining these two complimentary approaches has the potential to enable a new level of understanding of the diversity of life history strategy in annual plants and their genetic basis . A further feature of Col-0 phenology is that the bet-hedging strategy is stable over latitude throughout Europe , except in the far north where temperature is too cool to generate germinating cohorts . It is also stable in respect to simulated warming ( Figure 3—figure supplement 1 ) , except in respect to time of year of seed set . This surprising lack of local adaptation could facilitate population stability in changing environments or enable rapid colonization of new territories , if timing of seed set was under no further constraints . Understanding the generality of this strategy is therefore of clear importance . Our work suggests that integration of life history modelling with behaviour of genetic variants has the potential to reveal fitness tradeoffs across the whole life history and identify non-obvious selection pressures as emergent features . For seed production , Col-0 plant were grown in sanyo MLR 350 growth cabinets at 80–100 µmol m² s−1 white light at 22°C in 16 hr light 8 hr dark cycles until bolting . At bolting , plants were transferred to the indicated seed maturation environment . For temperature manipulations photoperiod was maintained constant at 16 hr light 8 hr dark , and for photoperiod manipulations the temperature remained constant at 22°C . Seeds were harvested when approximately 50% of siliques had dehisced . Each batch was then sieved through a 250-µm mesh ( Fisher Scientific , Basingstoke , UK ) to exclude poorly filled seeds . Approximately , 25–50 freshly harvested seeds from each plant were sown onto 0 . 9% water agar plates , and at least five plants per treatment were used to generate biological replicate seed batches . Plates were wrapped in foil to prevent light from reaching the seeds . A single un-stratified plate was used as the zero time point and the remaining plates were wrapped in foil to exclude light , and incubated in growth cabinets at 4°C , 8°C , 12°C , 15°C , or 20°C . After 3 , 7 , 14 , 21 , 28 , and 42 days , a single plate from each stratification treatment was removed and placed in 12 hr white light dark cycles at 80 µmol m² s−1 and 22°C for 7 days to permit germination of non-dormant seeds , which was scored as radical protrusion . Occasionally , seeds had germinated during the dark incubation phase and were easily identified by elongated hypocotyls . It was presumed that these seeds were non-dormant at the start of the experiment and were therefore discarded from further analysis . Five independent growings of plants were raised at a field site between October 2011 and July 2013 . Col-0 seeds were germinated at 22°C after stratification at 4°C for 2–3 days , transplanted into John Innes seed compost ( Levington ) in P40 trays , and kept moist for 1 week in Sanyo cabinet operating at 15°C to harden off . Trays were then moved to the field site , a walled garden within the University of York campus grounds , and dates of bolting , first and last mature seed were recorded for each individual . Slug pellets were applied when necessary , and fences around the perimeter excluded vertebrate herbivores . Seeds from five of these batches were harvested and used for germination time series experiments of the same format as described above . To compare weather station readings with soil level air temperature , LogTag TRIX-8 temperature loggers ( LogTag Recorders Ltd , Auckland , New Zealand ) were used . Loggers were placed inside trays containing plants in order to shield the sensor from direct sunlight . Loggers were set to record the temperature every 10 min , and mean hourly temperatures were calculated from these readings . Seedling emergence experiments were performed as described previously ( Footitt et al . , 2011 ) . Briefly , 500 seeds from a mixed batch of lab grown seed set at 15°C or field grown seed in York UK in spring 2013 were sown in June 2013 in pots of John Innes seed compost , and pots were sunk into the ground at our field site at the University of York , UK . Once every two weeks each pot was exhumed , seed and soil spread on a tray and exposed to natural light for 5 min . Germinated seeds were removed and scored before replacing the soil and reburying the pot . Daily average , maximum and minimum temperatures for the period January 2011 to August 2013 were collected from the University of York Department of Electronics weather station archive ( see sources , below ) . The weather station was situated approximately 300 m away from the field site , and approximately 21 m above ground level . This was preferred to soil temperature measurements because it more closely reflects temperature recordings by weather stations used for life history simulations . An additional 14 locations were selected based on a transect across continental Europe spanning the habitat of natural accessions of Arabidopsis thaliana and close to the believed site of collection for Col-0 . Temperature data spanning 10 years were collected ( www . geodata . com ) , and mean values for daily average , daily minimum and daily maximum temperatures were calculated for each day of the year . Times of sunrise and sunset at the nearest available location to each weather station were also collected , however since these do not vary significantly year to year , only 1 year of data was used . Photothermal models require temperature estimates to a resolution of 1 hr . Therefore , estimates of hourly temperature were produced from daily average maximum and minimum temperatures by assuming daily minimum between the hours of 21:00 and 02:00 , daily maximum between 09:00 and 14:00 , and average daily temperature for the remaining hours . These hourly temperature estimates , alongside daily sunrise and sunset times were used as the basis for simulations . A simple model in which the probability that any random seed in a population will germinate P ( G ) is mutually exclusive to the probability that it is dormant P ( D ) was defined as: ( 1 ) P ( G ) =1−P ( D ) . The idea of simultaneous independent loss of primary dormancy and induction of secondary dormancy ( Totterdell and Roberts , 1979; Batlla et al . , 2009 ) was used to explain germination behaviour in Arabidopsis seed populations . The individual probabilities of both primary dormancy P ( Dp ) and secondary dormancy P ( Ds ) were considered to be independent , and thus the total probability of dormancy P ( D ) was derived using the general disjunction rule as follows: ( 2 ) P ( D ) = P ( Dp∪ Ds ) = P ( Dp ) +P ( Ds ) − P ( Dp ) P ( Ds ) . Population-based threshold models ( Lundemo et al . , 2009 ) have shown that variation in germination timing could be due to differences in base water potential ( ψb ) . This parameter is defined as the minimum water potential required for germination and is normally distributed within seed populations . The idea of using ψb as a threshold value means that for any environmental water potential ( ψ ) , only seeds with ψb values exceeding ψ are capable of germination . Dormancy breaking or inducing treatments have also been shown to alter the mean ψb in seed populations . This translates to increasing proportions of seeds either losing or gaining dormancy as the population mean moves in relation to the threshold . If the mean dormancy were to change with a constant rate , a cumulative distribution curve would emerge from plotting these percentages over time . This concept was applied to both primary and secondary dormancy processes . It was assumed that seed populations have initially high mean values of primary dormancy and low mean values of secondary dormancy . Over time the mean primary dormancy would decrease , resulting in a cumulative reduction in the percentage of primary dormant seeds , while the opposite trend was presumed to occur for secondary dormancy . A cumulative distribution function of dormancy in general could be defined as the probability that a random seed has a dormancy value D less than or equal to the threshold d and can be found by integrating the probability density function , or normal distribution f ( x ) between the limits −∞ and d and as follows: ( 3 ) F ( d ) =P ( D≤d ) =∫−∞df ( x ) dx . In this definition , it is the threshold d which varies rather than the population mean; however , the two concepts are mathematically equivalent . Logistic functions were chosen to reproduce the desired S-shape of a cumulative distribution with only a small number of parameters . This approach conveniently negates any need to determine whether seeds are primary or secondary dormant , measure dormancy thresholds , population parameters , or have any knowledge about the agents causing dormancy . The following equations describe logistic functions which were used to model the probabilities of primary and secondary dormancy over time x: ( 4 ) P ( Dp ) = 11+eRp ( x+Ap ) , ( 5 ) P ( Ds ) =1− [11+eRs ( x+As ) ] , where Rp and Rs are the rates of primary dormancy loss and secondary dormancy induction; Ap and As are offset parameters . Because a standard logistic curve passes through x = 0 at y = 0 . 5 , Ap and As were chosen such that the curves were repositioned so that P ( Dp ) crosses x = 0 at 0 . 99 and P ( Ds ) crosses x = 0 at 0 . 01 . The magnitude of the two offset parameters required to position the curves were found by rearranging equations ( Fitter and Fitter , 2002; Cotton , 2003 ) and then substituting values of x = 0 and either P ( Dp ) = 0 . 01 or P ( Ds ) = 0 . 99 . Training data were generated by setting Col-0 seeds at 12°C , 13°C , 14°C , 15°C , 17°C and 18°C and incubating seeds at temperatures between 4°C and 20°C for up to 8 weeks in the dark before transferring to the light at the times indicated for germination at 22°C ( Figure 2 ) . Any seeds that germinated in the dark prior to light exposure were excluded from the analysis . In this data set there appeared a clear trend for an increase in the rate of secondary dormancy induction with higher incubation temperatures . The model was fitted using the fit function of MATLAB and allowed to optimise Rp independently for each germination time-series . This revealed a clear trend such that lowering maturation temperature decreased Rp; we therefore produced nine different models consisting of all possible combinations of linear , logistic , and exponential functions to model rates of primary and secondary dormancy loss or induction with maturation and incubation temperature respectively and fitted to the entire training data set using the fit function in the MATLAB curve fitting toolbox ( Mathworks ) . The best model based on the fit with the training data was a linear model of primary dormancy loss with maturation temperature ( Tm ) and an exponential model of secondary dormancy induction with incubation temperature ( Ti ) shown in Equations 6 , 7 ( Table 1 ) and the fit to data shown in Figure 2—figure supplement 2 . ( 6 ) Rp=aTm+b , ( 7 ) Rs=cedTi . 10 . 7554/eLife . 05557 . 015Table 1 . Comparisons of R2 and parameter number of the nine germination models tested with linear , exponential or logistic relationship between Rp and Rs with temperatureDOI: http://dx . doi . org/10 . 7554/eLife . 05557 . 015Rp modelRs modelNumber of parametersTotal R²ExponentialExponential40 . 71ExponentialLogistic50 . 62ExponentialLinear40 . 64LogisticExponential50 . 83LogisticLogistic60 . 82LogisticLinear50 . 75LinearExponential40 . 83LinearLogistic50 . 81LinearLinear40 . 73 Combining Equations 1 , 2 with Equations 4–7 results in a conceptually simple probabilistic model of germination with only 4 parameters ( a , b , c , d ) which determine the rates of primary dormancy loss and secondary dormancy induction with temperature . Optimised parameter values are shown in Table 2 . 10 . 7554/eLife . 05557 . 016Table 2 . Optimised final parameter values for the germination and seed set modelsDOI: http://dx . doi . org/10 . 7554/eLife . 05557 . 016ParameterValueGermination model A1 . 56 B−21 . 79 C0 . 05 D0 . 18Seed set model Tb ( base temperature ) 5 . 25 Threshold5370 A good fit was achieved by optimising model parameters using the training data ( Figure 2—figure supplement 2 ) ; however , a better test of the model is to assess its ability to predict data not used in its construction . Data obtained from maturing seeds at 16°C were used as a validation test ( Figure 2—figure supplement 3 ) . This produced a good fit to data ( R2 = 0 . 88 ) and therefore the model could predict the germination of seed lots produced and incubated at different temperatures in the lab not used to train the data . To understand whether the model was successful at predicting germination in seed lots from real field situations , the model was fitted against data sets generated from five field-grown populations ( Figure 2—figure supplement 4 ) . The parameters and incubation temperatures were either fixed or known , and the MATLAB fit function was used to derive a predicted value for Tm for each lot which could be compared to temperature data logged from the field sites between flowering and seed set ( Figure 2—figure supplement 4 ) . In each case , the predicted Tm appeared close to that during the final few days of seed set , suggesting that this period is critical in determining dormancy depth in Arabidopsis , as previously determined in sorghum . Photothermal time models integrate temperature and photoperiod information over time using a function such as the one below ( Borthwick et al . , 1943; Thomas and Raper , 1976 ) . ( 8 ) Th=∫abf ( x ) dx . Photothermal model of seed development rate over time x , is integrated from the point of bolting a , up to the point of first seed maturity , b . This permits the calculation of Th , the total thermal time experience required to achieve the production of mature seeds . In variable environments , it is necessary to approximate the above integral with a summation of the function over discrete time intervals . The summation function becomes as follows: ( 9 ) Th=∑t=1nθ ( t ) , where θ ( t ) is the thermal time accumulated at time t , and Th is the cumulative total between bolting at t = 1 , and seed maturity t = n . If Th is known in advance , this concept can be used to determine the value of n , and hence the time required to produce mature seeds in any particular environment . To develop the best model of developmental rates of Col-0 seed set was measured in multiple temperature regimes ( Figure 1 ) and photoperiod regimes ( Figure 1—figure supplement 1 ) . The thermal time model was as follows: ( 10 ) θ ( t ) ={ ( T ( t ) −Tb ) , T ( t ) ≥Tb0 , T ( t ) <Tb , where θ ( t ) is the number of thermal time units accumulating during the hour beginning at time t , T ( t ) is the average temperature during hour t , and Tb is the genotype-specific base temperature . We also compared a photothermal time model: ( 11 ) θ ( t ) ={ ( T ( t ) −Tb ) ×P ( t ) , T ( t ) ≥Tb0 , T ( t ) <Tb , where θ ( t ) is the number of photothermal time units for the hour beginning at time t , T ( t ) is the average temperature ( °C ) during the time interval , and Tb is the base temperature , which is a genotype-specific constant . P ( t ) is a measure of daylight . In order to parameterise the models seed was set under laboratory conditions at a range of temperatures between 8°C and 25°C , and at photoperiods between 8 hr light periods and 16 hr light periods ( Figure 1A ) . Optimal parameters were calculated using the fit function of MATLAB . The performance of each model was evaluated by comparing the predictions of seed development times with nine independent recorded seed development times in the field . These data were collected between October 2011 and July 2013 and closely matched the predictions for each model . The photothermal model ( R2 = 0 . 28 ) did not outperform the simple thermal time model ( R2 = 0 . 42 ) : therefore , we elected to use a simple thermal time model of Arabidopsis seed set ( Figure 1 ) .	Plants adjust when they grow , develop flowers and produce , or ‘set’ , seeds in response to changes in temperature and day length . It is therefore unsurprising that climate change alters the timing of these important events in plants' lives; for example , many plants are adapting to rising temperatures by flowering earlier and growing for longer . The environmental signals that control when a plant flowers , and the genes that underlie this process , have been well studied in the model plant Arabidopsis thaliana . This plant's ability to quickly colonize and thrive in disturbed habitats—including agricultural land , construction sites , and waste ground—is partly because some of its seeds lie dormant in the soil , for up to several years , before they start to grow . Whether or not a seed undergoes a period of dormancy is controlled by the temperature that the seeds experienced when they were developing; this in turn is influenced by earlier events , such as when the flowers first developed , and when the plant first started to grow from its seed ( a process called germination ) . To try to understand these complex interactions , Springthorpe and Penfield developed a computational model of the major events in the life of an Arabidopsis plant . Data collected from Arabidopsis plants that normally germinate in winter and spring were then used to check whether the model could accurately represent what happens in nature . The analysis confirmed that the timing of seed setting depends mostly on the environmental temperature . Springthorpe and Penfield then showed that plants both flowered and set seed earlier in response to increases in temperature , so that the seeds were shed precisely when the temperature was between 14°C and 15°C . Springthorpe and Penfield discovered that rise in the average temperature when a plant set seed from 14°C to 15°C had a dramatic effect on the seeds . Almost all of the seeds that developed below 14°C became dormant , while very few of the seeds that developed above 15°C became dormant . From their findings , Springthorpe and Penfield predict that the temperature control of flowering time has evolved to constrain when seeds are set and ensure that plants produce a mixture of seeds: some that will become dormant , and some that will not . Their findings also show that modelling the whole life history of an organism has the potential to reveal strategies that are not obvious when studying single events in isolation . If the model was extended to include genetic variation across populations of plants , this approach could give new insights into how individual genes help plants adapt to weather and climate .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "plant", "biology" ]	2015	Flowering time and seed dormancy control use external coincidence to generate life history strategy
The mammalian suprachiasmatic nucleus ( SCN ) drives daily rhythmic behavior and physiology , yet a detailed understanding of its coordinated transcriptional programmes is lacking . To reveal the finer details of circadian variation in the mammalian SCN transcriptome we combined laser-capture microdissection ( LCM ) and RNA-seq over a 24 hr light / dark cycle . We show that 7-times more genes exhibited a classic sinusoidal expression signature than previously observed in the SCN . Another group of 766 genes unexpectedly peaked twice , near both the start and end of the dark phase; this twin-peaking group is significantly enriched for synaptic transmission genes that are crucial for light-induced phase shifting of the circadian clock . 341 intergenic non-coding RNAs , together with novel exons of annotated protein-coding genes , including Cry1 , also show specific circadian expression variation . Overall , our data provide an important chronobiological resource ( www . wgpembroke . com/shiny/SCNseq/ ) and allow us to propose that transcriptional timing in the SCN is gating clock resetting mechanisms . The suprachiasmatic nucleus ( SCN ) of the hypothalamus is the seat of the principal circadian clock in mammals . Entrained by photic information and other external stimuli , this group of neurons maintains 24-hr rhythms of physiology and behaviour by synchronising molecular oscillators in the brain and peripheral tissues ( Welsh et al . , 2010; Ko and Takahashi , 2006; Maywood et al . , 2007 ) . The transcriptional output of the SCN is thus critical for this fundamental biological adaptation to night and day ( Panda et al . , 2002; Zhang et al . , 2014b; Li et al . , 2015 ) . An early microarray study of the SCN identified 650 genes displaying circadian oscillations ( Panda et al . , 2002 ) . However , this number will have been limited by the number of genes surveyed , the low signal-to-noise ratio of certain genes ( e . g . Per2 ) , and the use of experimental samples likely to include a large proportion of the surrounding hypothalamic tissue . The study of such heterogeneous samples will result in reduced amplitudes of oscillations for genes that cycle asynchronously in different brain regions ( Zhang et al . , 2014b ) . A recent study estimated that over half of all protein-coding genes show circadian oscillations in the mouse , although only 642 could be identified in hypothalamic samples ( Zhang et al . , 2014b ) . Furthermore , many circadian transcriptomic surveys of mammalian systems have utilised constant lighting conditions to focus on the endogenous , free-running circadian clock and have not assessed more physiologically relevant light/dark ( LD ) cycles ( Zhang et al . , 2014b; Panda et al . , 2002; Hughes et al . , 2009; Hurley et al . , 2014 ) . To identify circadian processes driven by photic cues , we combined RNA-seq with laser-capture microdissection ( LCM ) of the SCN to provide a comprehensive and tissue-specific transcriptomic investigation of the master pacemaker . These data reveal many hundreds of novel cycling transcripts and provide evidence that the temporal variation in a specific group of genes plays a role in modulating light-induced phase resetting of the circadian clock . LCM accurately isolated the entire SCN for transcriptional profiling ( Figure 1—figure supplement 1A ) . Pooled dissected tissue from five adult male mice provided one of three replicates for each of six timepoints over a 12:12 LD cycle ( ZT2 , 6 , 10 , 14 , 18 and 22 ) . Sequencing RNA in each replicate generated up to 133 million paired-end reads of which 45–65% were subsequently mapped to the mouse genome allowing for gene expression level estimation . 10 . 7554/eLife . 10518 . 003Figure 1 . RNA-seq data of laser-capture microdissection ( LCM ) suprachiasmatic nucleus ( SCN ) samples recapitulate the cycling of clock gene expression and identify known and novel SCN enriched genes . ( A ) Heatmap showing mean expression of key clock genes in the SCN across circadian time . The majority ( 8/10 ) show significant cycling ( DESeq2 and JTK Cycle; q < 0 . 05 ) and peak at the expected timepoint denoted by a yellow box ( Ko and Takahashi , 2006 ) . ( B ) Statistically significant Gene Ontology ( GO ) enrichments for SCN enriched genes ( q < 0 . 05 ) . Bars represent the number of genes of a given category expected by chance among SCN enriched genes ( grey ) versus those observed ( black ) . q-values: q < 0 . 05 , q < 0 . 01 , q < 0 . 001 . ( C ) In situ hybridisation ( ISH ) of four SCN enriched genes at ZT6 in the wild-type brain . Scale bar: 0 . 5 mm ( low magnification ) ; 0 . 2 mm ( high magnification ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10518 . 00310 . 7554/eLife . 10518 . 004Figure 1—figure supplement 1 . Generation of the SCN transcriptome over 24 hr . ( A ) Representative Nissl-stained brain sections before and after laser-capture microdissection ( LCM ) of the SCN . Scale bar: 0 . 5 mm . ( B ) Principal component analysis ( PCA ) plot showing separation of RNA-seq data according to timepoint and sequencing batch . Ellipses delineate the 95% confidence intervals for each timepoint . ( C ) Dot plot showing the Pearson’s correlation between RNA-seq and quantitative polymerase chain reaction ( qPCR ) values for all the annotated genes surveyed ( excluding those housekeeping genes used to normalise the data ) ; 43/59 genes show a high concordance between methods ( r >0 . 4 ) . ( D ) Scatter plot showing a trend for increased correlation of cycling gene expression measured in the two different platforms for increased dynamic range . Gene expression values with larger dynamic ranges tend to be less susceptible to extrinsic factors such as random noise and batch effects which therefore contribute less . ( E ) Heatmap showing the mean expression levels for 17 twin-peaking genes across circadian time according to ( i ) RNA-seq or ( ii ) qPCR from LCM SCN samples . DOI: http://dx . doi . org/10 . 7554/eLife . 10518 . 00410 . 7554/eLife . 10518 . 005Figure 1—figure supplement 2 . Use of publically available RNA-seq data to identify SCN enriched genes . ( A ) PCA plot showing successful separation of RNA-seq data according to tissue but , importantly , not according to study . Adjacent numbers indicate the study from which data were collected ( 1 , Azzi et al . ( 2014 ) ; 2 , Barbosa-Morais et al . ( 2012 ) ; 3 , Merkin et al . ( 2012 ) ; 4 , Brawand et al . ( 2011 ) ; 5 , Zhang et al . , ( 2014b ) ; 6 , Menet et al . ( 2012 ) ) . Ellipses delineate the 95% confidence intervals for each tissue . The first principal component ( PC1 ) explains nearly 80% of the expression variance in samples and displays a large separation between liver and brain tissues signifying substantial transcriptomic differences between these disparate tissues . The second principal component ( PC2; approximately 10% ) in part explains variation among different regions of the brain . Samples from the same tissue ( e . g . cerebellum ) cluster together more closely than tissue samples from the same study ( labelled by number ) but different brain region . This indicates that RNA-seq batch effects are small relative to biological variation . Of the non-SCN brain tissues samples , the published ‘hypothalamus’ RNA-seq dataset ( Zhang et al . , 2014b ) shows the greatest similarity to the SCN samples . ( B ) Fold-change ( FC ) enrichment of SCN marker ( purple ) and control gene ( red ) expression in the LCM SCN samples relative to mean brain fragments per kilobase of exon per million ( FPKM ) levels from other transcriptomic studies in ( A ) . These SCN markers display large FC enrichment compared to general markers of neurons ( e . g . Snap25 ) , oligodendrocytes ( e . g . Mbp ) and housekeeping genes ( e . g . ActB , Gapdh ) . ( C ) Heatmap showing mean expression levels of the 146 SCN enriched genes across all other transcriptomic studies in ( A ) . Numbers indicate study from which data were collected as in ( A ) . These genes show the highest expression in the SCN followed by the hypothalamus , as expected , given the location of the SCN within the hypothalamus . DOI: http://dx . doi . org/10 . 7554/eLife . 10518 . 005 Global expression differences were apparent between timepoints , as revealed by principal component analysis ( PCA ) ( Figure 1—figure supplement 1B ) . Of 10 key circadian-clock genes , 8 showed the classic sinusoidal circadian profile ( q < 0 . 05 ) with a zenith and nadir at the expected timepoints ( Figure 1A ) ; this verifies the temporal expression of our LCM SCN transcriptome dataset and indicates that any sequencing batch effects are minimal . Furthermore , quantitative ( q ) PCR of the LCM SCN samples showed a high degree of concordance with the RNA-seq expression data ( Figure 1—figure supplement 1C–D ) . Our SCN transcriptomes are also separable from those in other brain regions and non-CNS tissues ( Zhang et al . , 2014b; Brawand et al . , 2011; Merkin et al . , 2012; Menet et al . , 2012; Barbosa-Morais et al . , 2012; Azzi et al . , 2014 ) ( Figure 1—figure supplement 2A ) . In addition , four key SCN markers ( Avp , Vip , Grp and Six6 ) showed a far greater expression enrichment in the SCN over other brain regions ( Figure 1—figure supplement 2B ) . In summary , these data provide the first identification of SCN enriched genes either on a genome-wide scale or over a 24-hr period and allow an anatomically precise and comprehensive examination of the SCN transcriptional output over the full day . To allow further exploration into the dataset , we provide a user-friendly web application ( www . wgpembroke . com/shiny/SCNseq/ ) . A total of 146 genes were identified whose expression is highly enriched in the SCN ( Materials and methods; analysis of variance ( ANOVA ) , q < 0 . 05 ) of which 4 out of 4 were confirmed using in situ hybridisation ( ISH ) ( Figure 1—figure supplement 2C; Figure 1C ) . These genes included many known SCN markers , as well as others not previously implicated in circadian biology or in SCN function ( Figure 1C ) . Their significant enrichments for the Mouse Phenotype term ‘abnormal circadian rhythm’ ( q = 1 . 2×10-4 ) and for Gene Ontology ( GO ) annotations relating to signalling pathways ( q < 0 . 05; Figure 1B ) are consistent with the SCN’s role as the master pacemaker . Enrichment for the flagellum annotation and G-protein coupled receptor ( GPCR ) signalling pathways ( Figure 1B ) may reflect the roles of primary cilia as GPCR signalling hubs ( Omori et al . , 2015 ) . These results thus suggest that GPCR-mediated neuropeptide signalling via cilia may be important for SCN function . We next identified 4569 genes ( 24% of 18 , 889 Ensembl-annotated coding or non-coding models expressed ( ≥1 read ) in over six samples ) whose expression oscillated in a sinusoidal manner ( Figure 2—figure supplement 1A , B ) according to both JTK Cycle and DESeq2 approaches ( Materials and methods ) . Importantly , this number of cycling genes is sevenfold higher than previously reported in the SCN ( Panda et al . , 2002 ) . This large increase is likely to reflect a combination of factors: first , the significantly improved anatomical enrichment of the SCN tissue we have surveyed; second , the greater sensitivity and specificity of the statistical approaches we have applied ( Hughes et al . , 2010 ) ; and finally , our sampling approach under LD as opposed to the constant dark ( DD ) conditions utilised in earlier studies ( Panda et al . , 2002 ) . For example , we are able to detect cycling genes that may be physiologically relevant to light exposure or light-induced transcriptional cascades that are undetectable as cycling under DD ( Ueda et al . , 2002; Leming et al . , 2014 ) . Our new collection of SCN cycling genes were enriched for GO terms relating to key processes such as translation and mitochondrion-related genes , in line with previous findings ( Figure 2—figure supplement 1C; Panda et al . , 2002 ) . De novo transcription has been shown to drive rhythmicity of approximately 20–30% of all cycling genes ( Koike et al . , 2012; Menet et al . , 2012 ) , which illustrates the importance of post-transcriptional regulation in shaping the circadian transcriptome . Interestingly , we observe novel enrichments among cycling genes in terms relating to RNA splicing and ribonucleoprotein ( RNP ) complexes ( Figure 2—figure supplement 1C ) which suggests that these processes may establish oscillatory expression patterns ( Wang et al . , 2013 ) . Other post-transcriptional mechanisms , such as miRNA level fluctuation , which were not assessed in this experiment , could also contribute to initiating these patterns . 10 . 7554/eLife . 10518 . 006Figure 2 . Gene clustering identifies a novel twin-peaking synaptic gene module with a potential role in phase resetting . ( A ) Relative expression of each gene present in the identified twin-peaking module over 24 hr . The black line represents the module eigengene – the first principal component of the gene expression matrix for this module . ( B ) Top three statistically significant ( q < 0 . 05 ) Gene Ontology ( GO ) ( biological process ( BP ) , cellular component ( CC ) , molecular function ( MF ) ) and Kyoto Encyclopedia of Genes and Genomes ( KEGG ) enriched terms for the 496 significantly fluctuating twin-peaking genes . q-values: q < 0 . 05 , q < 0 . 01 , q < 0 . 001 . ( C ) Illustration of the theoretical phase response curve ( PRC ) in mouse , which demonstrates the phase shifting potential when a light pulse is applied at different times ( black line ) ( Golombek and Rosenstein , 2010 ) . Our module eigengene for the twin-peaking module ( blue dotted line ) indicates that when its expression lies above the threshold level ( black horizontal line ) , the phase of the organism is able to shift . The greater the expression over this threshold , the larger the phase shift . Conversely , when expression is low and below this threshold , the suprachiasmatic nucleus ( SCN ) is resistant to phase shifting induced by light pulses . This suggests the presence of susceptibility windows ( shown by green bars ) when light impulses are able to shift the phase of the SCN . ( D ) Circadian gating of light signalling pathway , adapted from ( Iyer et al . , 2014 ) . Twin-peaking gene-encoded proteins identified by laser-capture microdissection ( LCM ) SCN RNA-seq are indicated in blue . Photic information is transmitted to the SCN via glutamatergic signalling through the retinohypothalamic tract ( RHT ) . ( i ) Brain-derived neurotrophic factor ( BDNF ) gates glutamatergic synaptic transmission by regulating either presynaptic release or membrane channel activity . ( ii ) Voltage-gated potassium channels hyperpolarise the SCN neurons allowing photic input from the RHT to elicit a significant response . ( iii ) Glutamate binding to glutamate receptor , ionotropic , N-methyl D-aspartate ( GRIN ) receptors induces calcium influx inducing CaMKII autophosphorylation . ( iv ) Active pCaMKII phophorylates neuronal nitric oxide synthase ( nNOS ) , which generates nitric oxide ( NO ) , producing differential effects according to time of day . ( v ) In early night ( ZT14 ) , we observe peaking of genes relating to small guanosine triphosphate ( GTPase ) activation and actin-based cytoskeletal remodelling . At this timepoint , F-actin levels are at their highest and an influx of Ca2+ depolymerises the F-actin affecting components of the mitogen-activated protein ( MAP ) kinase pathway . Ca2+ also induces intracellular Ca2+ release though the opening of the endoplasmic reticulum calcium channel RyR2 . ( vi ) In late night ( ZT22 ) , NO activates guanylyl-cyclase ( GC ) , which increases cyclic guanosine monophosphate ( cGMP ) levels activating protein kinase G ( PKG ) . ( vii ) Phosphorylation of cAMP response element-binding protein ( CREB ) leads to the transcription of PER1 , which contributes to the shifting of the circadian phase . DOI: http://dx . doi . org/10 . 7554/eLife . 10518 . 00610 . 7554/eLife . 10518 . 007Figure 2—figure supplement 1 . Identification of 4569 cycling genes in the SCN . ( A ) Heatmap showing the expression levels in each biological replicate for the 4569 significantly cycling genes in the SCN across circadian time ordered by peak phase ( DESeq2 and JTK Cycle , q < 0 . 05 ) . ( B ) Polar plot showing the distribution of peak phases for the identified 4569 cycling genes binned in 2-hr windows . Cycling genes tend to peak during transcriptional ‘rush hours’ near to light transitions ( shown by black lines ) , as previously described ( Zhang et al . , 2014b ) . ( C ) Top three statistically significant GO ( BP , CC , MF ) and KEGG enrichments for the 4569 cycling genes . q-values: q < 0 . 05 , q < 0 . 01 , q < 0 . 001 . ( D ) Number of other tissues in which the 4569 cycling genes also cycle . Colours indicate the distribution of these cycling genes across the different tissues . ( E ) Number of the 4569 cycling genes that are also significantly cycling in other tissues from other studies ( Zhang et al . , 2014b ) ordered by fold-change ( FC ) enrichment ( right ) . Tissues which show significantly greater LCM SCN cycling genes are denoted with an asterisk ( q < 0 . 05 , q < 0 . 01 , q < 0 . 001 ) ; as expected , the greatest enrichment in overlap was with the SCN results from the microarray study by Panda et al . ( 2002 ) . ( F ) Statistically significant GO and KEGG enrichments for the 1034 cycling genes peaking at ZT14 . q-values: q < 0 . 05 , q < 0 . 01 , *q < 0 . 001 . ( G ) Distribution of peak phases for significantly cycling genes with GO annotations ‘actin cytoskeleton organisation’ ( n = 25 ) and ‘small GTPase regulator activity’ ( n = 29 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10518 . 007 Approximately half of the SCN cycling genes ( 2451/4569 ) showed no evidence for sinusoidal expression in 12 other tissues , including the whole hypothalamus , based on published circadian expression data ( JTK Cycle; q < 0 . 05; Figure 2—figure supplement 1D ) ( Zhang et al . , 2014b ) . These genes included those essential for normal SCN function , including Prok2 , Rorb and Nts , which were also validated by quantitative polymerase chain reaction ( qPCR ) ( Figure 1—figure supplement 1C ) . This indicates that many genes cycle exclusively in the SCN , or only cycle under a 12:12 LD lighting schedule rather than under the constant dark conditions used previously ( Zhang et al . , 2014b ) . Comparing our results under LD with published DD datasets may inflate the perceived number of genes cycling solely in the SCN , however , as there are genes likely to cycle under LD exclusively as shown by microarray studies in lower organisms ( Ueda et al . , 2002; Leming et al . , 2014 ) . Six of the nine genes which cycled significantly across all 12 other tissues ( q < 0 . 05 ) also cycled in the LCM SCN samples ( Arntl , Dbp , Nr1d2 , Per2 , Per3 and Tsc22d3 ) suggesting a core set of universally cycling genes ( Figure 2—figure supplement 1D ) . In support of this notion our LCM SCN cycling gene list significantly overlapped with those cycling in 8 of 12 other tissues ( q < 0 . 05 ) , as well as those cycling in an SCN microarray study ( Zhang et al . , 2014b; Panda et al . , 2002 ) ( Figure 2—figure supplement 1E ) . As expected , the SCN , hypothalamus and cerebellum showed the greatest overlap of their circadian transcriptomes ( Figure 2—figure supplement 1E ) . Previous studies only considered circadian gene expression following a sinusoidal pattern . Alternative and potentially functionally important expression profiles may thus have been overlooked . Using Weighted Correlation Network Analysis ( WGCNA ) , which does not assume a single mode of temporal variation , we identified a single module of 766 genes , which instead of adopting a sinusoidal profile , displayed two peaks of expression at ZT14 and ZT22 ( Figure 2A ) . From this ‘twin-peaking’ module 496 genes individually exhibited statistically significant temporal fluctuations ( DESeq2 , q < 0 . 05 ) , of which 17 of 17 were validated by qPCR ( Figure 1—figure supplement 1E ) . This profile is also distinct from previously described 12-hr harmonics ( Cagampang et al . , 1998a; 1998b; Hughes et al . , 2009 ) , and prior to this , only single genes from individual studies were shown to possess similar expression patterns ( Shinohara et al . , 1993; Cagampang et al . , 1998c; 1998d ) . For example , isoforms of protein kinase C ( Prkcb and Prkcg ) exhibit a twin-peaking profile in rat SCN similar to that observed in our RNA-seq data ( Cagampang et al . , 1998c ) . This twin-peaking transcriptional profile , particularly for this large number of genes , was thus unexpected . Functional annotations related to synaptic transmission , calcium signalling and gated channel activity were greatly and significantly enriched among these twin-peaking genes ( q < 0 . 05; Figure 2B ) , suggesting a circadian component to establishing the electrophysiological environment of the SCN . The peaking of expression at ZT14 and ZT22 coincides with phase shifting ‘susceptibility windows’; time periods in which the photic input permits phase shifting ( or ‘resetting’ ) of the clock ( Ding et al . , 1994; Iyer et al . , 2014 ) . This is demonstrated by the phase response curve ( PRC ) : a light impulse at ZT14 will initiate a phase delay of the clock whereas a light impulse at ZT22 will initiate a phase advance ( Figure 2C ) ( Golombek and Rosenstein , 2010; Ding et al . , 1994 ) . At all other timepoints we assessed under an LD cycle the twin-peaking genes exhibited low expression levels exactly when the clock is resistant to phase shifts . This suggests a possible transcriptional gating mechanism in which a threshold level of gene expression needs to be exceeded for light to activate this phase shifting pathway ( Ding et al . , 1994; Gillette and Mitchell , 2002; Iyer et al . , 2014 ) . Transcriptional profiling at a higher temporal resolution would be required to confirm the elevated expression of twin-peaking genes above this critical threshold at timepoints when light-induced phase shifts are possible ( e . g . ZT12; Figure 2C ) . Genes previously known to participate in this light-induced phase shifting pathway display this twin-peaking pattern ( Figure 2D ) , and their pharmacological inhibition ( e . g . for Grin2a/b , Camk2a , Nos1 , Ryr2 ) blocks light-induced phase shifts ( Ebling et al . , 1991; Yokota et al . , 2001; Ding et al . , 1998; 1994; 2007; 2010 ) . Many genes whose expression peaks singly at ZT14 have functions related to actin cytoskeleton reorganisation and/or to small GTPase signalling ( Figure 2—figure supplement 1F–G ) . This supports the hypothesis that actin depolymerisation is important for phase delays but not phase advances ( Gerber et al . , 2013; Cabej , 2013 ) . Taken together , our data indicate that oscillations in the SCN transcriptome play an important role in gating the differential responses to light , a fundamental circadian process . The complex relationship between photic and non-photic cues and their relative roles in phase adjustment must also be considered , however . Light exposure during constant conditions can still modify the processing of non-photic signals by the SCN ( Challet and Pevet , 2003 ) . For example , novelty-induced running in the subjective mid-day induces a phase advance in mice , although a 1-hr light pulse administered after access to running wheels can significantly attenuate this affect ( Biello and Mrosovsky , 1995 ) . More work is required to determine how the timing or relative level of gene expression mediates these light-induced behavioural changes . A total of 3187 multiexonic long intergenic non-coding RNA ( lincRNAs; >200 nucleotides ) loci were identified from the SCN transcriptome data . Considered together , they displayed evidence of significant sequence conservation , had lower expression and fewer exons than protein-coding genes in line with lncRNAs identified in other mammalian tissues ( Zhang et al . , 2014a; Cabili et al . , 2011; Ponjavic et al . , 2007 ) , but did not display correlated expression with that of the nearest protein-coding gene suggesting these lincRNAs are often not functioning as enhancer ( e ) RNAs ( Figure 3—figure supplement 1A–D ) . Of these lincRNAs , 341 ( 11% ) displayed significant temporal fluctuation , including lincRNA NONCO7761 ( 3100003L05Rik ) whose presence in the SCN was confirmed by ISH at ZT6 ( Figure 3D ) as well as by qPCR to peak at both ZT6 and ZT18 ( Figure 3E ) . In addition , 28 of these fluctuating lincRNAs peaked at ZT14 and ZT22 ( Figure 3B–C ) , and thus may modulate or be modulated by the transcriptional or post-transcriptional regulation of protein-coding genes of the twin-peaking synaptic module described earlier . 10 . 7554/eLife . 10518 . 008Figure 3 . Identification of multiexonic lincRNAs whose expression fluctuates in the suprachiasmatic nucleus ( SCN ) . ( A ) Proportion of the 3187 identified lincRNAs which ( i ) have been previously annotated ( according to Ensembl or UCSC ) , ( ii ) are not annotated in these resources but have been identified in other studies ( Belgard et al . , 2011 , Ramos et al . , 2013 ) or ( iii ) are unique to this study . ( B ) Proportion of the 3187 identified lincRNAs which show significant fluctuating expression patterns ( twin-peaking module and DESeq2 q < 0 . 05 , or JTK cycle q < 0 . 05 and DESeq2 q < 0 . 05 ) ; protein-coding genes are shown for comparison . ( C ) Heatmap showing the expression levels in each biological replicate for the 341 fluctuating lincRNAs . ( D ) In situ hybridisation ( ISH ) of NONCO7761 ( 3100003L05Rik ) at ZT6 in the mouse brain showing low yet distinct expression in the SCN . Scale bar: 0 . 5 mm ( low magnification ) , 0 . 2 mm ( high magnification ) . ( E ) RNA-seq and quantitative polymerase chain reaction ( qPCR ) show fluctuating mean expression ± SD of the lincRNA NONCO7761 ( 3100003L05Rik ) over 24 hr . DOI: http://dx . doi . org/10 . 7554/eLife . 10518 . 00810 . 7554/eLife . 10518 . 009Figure 3—figure supplement 1 . Properties of identified SCN lincRNAs . ( A ) Density plot comparing the distribution of expression levels of the 3187 identified lincRNAs ( red ) and protein-coding genes ( blue ) ; lincRNAs tend to be significantly lower expressed than protein-coding genes ( p < 2 . 2×10-16 , Kolmogorov–Smirrnov ( KS ) test ) . ( B ) Cumulative frequency graph showing the number of exons possessed by protein-coding genes and the identified 3187 lincRNA loci . Despite this study only considering multiexonic lincRNAs , protein-coding genes tend to possess significantly more exons ( p < 2 . 2×10-16 , KS test ) . ( C ) Violin plot showing phastCons conservation scores for the exonic regions of the 2796 identified lincRNAs with available conservation scores; these lincRNAs showed significantly less conservation than protein-coding genes but significantly greater conservation than two proxies for neutral evolution: ( i ) transposable elements within a 50-kb window of the lincRNA locus ( p < 2 . 2×10-16 , KS test ) , ( ii ) 500-bp of sequence up- or downstream of the lincRNA locus selected randomly within a 1- to 5-kb window ( p < 2 . 2×10-16 , KS test ) . ( D ) Violin and box plot showing no significant difference ( p = 0 . 45 , KS test ) in the Pearson’s correlation of a lincRNA’s expression with that for its genomically most proximal protein-coding gene ( n = 2261 , expression correlation not possible if closest gene is not expressed ) compared to a randomly selected protein-coding gene . DOI: http://dx . doi . org/10 . 7554/eLife . 10518 . 009 We also identified 1013 novel exons – transcribed regions of the transcriptome which have not been annotated by Ensembl or UCSC , but contribute to alternative transcripts in a known gene . Of these , we focused on the novel exon of the core clock gene Cry1 which appears to represent an alternative transcriptional start site ( TSS ) ( Figure 4A ) . Despite its large size ( 322 bp ) and high expression , the novel exon present in the first intron of Cry1 had remained elusive , yet is confirmed by reverse transcription polymerase chain reaction ( RT-PCR ) from SCN tissue ( Figure 4C ) . The novel and canonical Cry1 isoforms are expressed in antiphase , as confirmed by qPCR from LCM SCN samples ( Figure 4D ) , indicating an unanticipated mode of temporal regulation in the SCN for this core clock gene . 10 . 7554/eLife . 10518 . 010Figure 4 . Identification of a novel antiphase Cry1 isoform . ( A ) Genomic region of mouse chromosome 10 displaying the novel exon ( blue bar ) for the clock gene Cry1 , which appears to represent an alternate transcriptional start sites ( TSS ) . Peaks ( black ) signify total expression of the exon whereas representative RNA-seq spliced reads ( black ) indicate exonic splice junctions . ( B ) Schematic diagram showing the 5’ gene structure of the canonical ( red ) and novel ( blue ) Cry1 isoforms . The novel exon ( exon 1b ) splices into the subsequent second exon using two different exon boundaries . Dotted arrows denote the position of PCR primers used to amplify Cry1 by RT-PCR . ( C ) RT-PCR from SCN RNA at ZT6 confirms the presence of novel Cry1 TSS capable of producing two different isoforms as labelled . The identity of all PCR products was confirmed by sequencing . ( D ) Quantitative polymerase chain reaction ( qPCR ) of the canonical and novel Cry1 TSS show antiphase cycling expression ( mean ± SD ) patterns . DOI: http://dx . doi . org/10 . 7554/eLife . 10518 . 010 In conclusion , this first temporal analysis of the SCN using RNA-sequencing reveals the presence of a twin-peaking transcriptional module with a suspected function in circadian control , and identifies thousands of novel transcripts , including a novel Cry1 isoform , which may play important roles in SCN function . All animal experimentation has been approved by the UK Home Office and the University of Oxford Ethical Review Board . Male C3H/HeH mice at 10–12 weeks of age were group housed in light-tight chambers equipped with light emitting diode ( LED ) lighting at 150 lx at the cage floor . After 7 days of acclimatisation , mice were singly housed for 7 days prior to tissue harvesting . Brains were removed at one of six Zeitgeber times ( ZT ) and immediately frozen on dry ice in optimal cutting temperature ( OCT ) mounting media ( VWR , Lutterworth , United Kingdom ) . Eighteen frozen coronal sections at 15 μM were cut spanning the entire rostral to caudal region of the SCN and mounted onto polyethylene naphthalate ( PEN ) slides ( Carl Zeiss Ltd . , Cambridge , United Kingdom ) . LCM was carried out from dehydrated , Nissl-stained sections using the PALM system ( Carl Zeiss Ltd . , Cambridge , United Kingdom ) as previously described ( Dulneva et al . , 2015 ) . Tissue from five animals was pooled to generate one biological replicate sample ( 15 mice in total were used per ZT ) and total RNA was purified using the RNeasy micro kit ( Qiagen , Manchester , United Kingdom ) . RNA quality was determined using an RNA Picochip ( Agilent , Stockport , United Kingdom ) , with RNA Integrity Number ( RIN ) values over 8 for all samples and average yield of approximately 10 ng per replicate . Approximately 1 ng of RNA was amplified using the SMARTer protocol ( Takata Clontech Bio Europe SAS , France ) , and the resulting libraries were 100-bp paired-end sequenced on the HiSeq ( Illumina , Fulbourn , United Kingdom; Wellcome Trust Centre for Human Genetics Sequencing Core ) . The average read pair count obtained was ∼35 M per technical replicate ( ∼100 M per biological replicate ) . One SCN , six brain and five liver transcriptome datasets from six independent studies ( GSE54124 , GSE30352 , GSE54652 , GSE41637 , GSE41338 and GSE36874 ) were downloaded from the Gene Expression Omnibus ( GEO ) database to allow comparison of expression between tissues . PCA revealed separation of datasets according to tissue rather than experimental factors ( Figure 1—figure supplement 2A ) suggesting negligible batch effects and therefore differences in expression are likely due to biological , rather than technical differences . All sequencing data was processed as follows unless otherwise specified . To retain only high quality reads an in-house script was used to: ( i ) trim ends of reads with a Phred base quality score below 3 , ( ii ) remove specified over-represented adapter or contaminating sequences , ( iii ) remove reads with a length below 50 bp ( or 40 bp if initial sequencing length was shorter ( Merkin et al . , 2012; Barbosa-Morais et al . , 2012 ) and ( iv ) remove any reads with an average Phred quality score below 30 . Reads were multimapped against the Genomic Mapping and Alignment Program for mRNA and EST sequences ( GMAP; version 2012-07-20; Wu and Watanabe , 2005 ) and processed mm10/GRCm38 . 70 reference genome using Genomic Short Nucleotide Alignment Program ( GSNAP; Wu and Nacu , 2010 ) with the option to consider novel splice sites . Data are deposited in GEO accession number GSE72095 . Further details of all genes surveyed are shown in Supplementary file 2 . De novo transcript assembly was conducted using cufflinks v2 . 0 . 2 ( Trapnell et al . , 2010 ) allowing identification of 6610 novel exons . These were required not to overlap but were within a 10-kb window of any feature in the most recent Ensembl or UCSC gene annotation sets . Of these exons , only those ( i ) with a reciprocal overlap of <25% with any retroposed pseudogene ( ucscRetroAli6 ) , ( ii ) with a reciprocal overlap of <50% with any transposable element , ( iii ) with over 20 spliced reads in the novel exon , ( iv ) with over 10 spliced reads per 100 bp of the novel exon , ( v ) with over 20 reads that splice into a known transcript and were ( vi ) shorter than 3000 bp , were retained providing a robust set of 1013 novel exons . Further details of all novel exons surveyed are shown in Supplementary file 4 . Identification of lincRNAs was conducted through the use of the CGAT NGS pipelines rnaseqtranscripts . py and rnaseqlncrna . py ( Sims et al . , 2014 ) . The first pipeline identifies transfrags using cufflinks and retains those present in at least two samples . The second predicts lncRNAs by removing transfrags which overlap protein-coding exons . These lncRNAs are then assessed for coding potential using the coding potential calculator ( CPC; Kong et al . , 2007 ) and removed if annotated as ‘coding’ ( CP score >1 ) . Only the intergenic ( >2 kb from any protein-coding gene ) , multiexonic lncRNAs with expression ( ≥1 read ) in over six biological replicates were chosen for further investigation . Further details of all lincRNAs surveyed are shown in Supplementary files 3 and 5 . To determine the expression level for each gene in each sample , aligned reads were quantified using either HTSeq ( version 0 . 5 . 4p1; Anders et al . , 2015 ) or cuffNorm . Both methods used a GTF file generated from combining both the most recent Ensembl annotation and the identified set of lincRNAs . FPKM values were generated using cuffQuant and normalised with cuffNorm , whereas count tables were generated using HTseq and normalised using DESeq2 ( Love et al . , 2014 ) . To identify SCN enriched genes , genes were required to show significantly greater expression ( q < 0 . 05 , Benjamini–Hochberg ) in the brain than in liver and significantly greater expression ( q < 0 . 05 , Benjamini–Hochberg ) in both this study’s SCN dataset and the published SCN dataset ( Azzi et al . , 2014 ) relative to the brain using ANOVA in R . FPKM values were then averaged for each study and quantile normalised . Fold-change ( FC ) enrichment scores were calculated by averaging the FPKM values for each study for a particular tissue and then comparing this average to the mean FPKM value for another tissue . A gene was defined as being SCN enriched if it showed a threefold enrichment in the SCN over brain samples . The R packages JTK_CYCLE ( v2 . 1; Hughes et al . , 2010 ) and DESeq2 ( v 1 . 6 . 3 ) were used together to identify genes whose expression significantly altered over time with a sinusoidal oscillatory pattern . DESeq2 was used to perform a likelihood ratio test to determine the genes whose expression significantly altered over time ( q < 0 . 05 , Benjamini–Hochberg ) and the JTK_CYCLE was used to identify genes whose expression followed 24-hr periodic waveforms ( q < 0 . 05 , Bonferroni ) . To identify genes with non-sinusoidal oscillatory patterns , signed WGCNA ( Langfelder and Horvath , 2008 ) was used to detect modules based upon gene coexpression . Modules were merged if their similarity was greater than 0 . 3 according to dendrogram height . Of the resultant 23 modules only the ‘lightsteelblue1’ module was investigated further because it was the only non-sinusoidal module whose expression pattern was not influenced by anomalous expression of a single sample at a particular timepoint . Genes of this module were defined as fluctuating if their expression significantly altered over time ( q < 0 . 05 , Benjamini–Hochberg; DESeq2 ) . Enrichment analysis was performed using the Database for Annotation , Visualization and Integrated Discovery ( DAVID ) Functional Annotation Tool v6 . 7 ( Huang da et al . , 2009 ) . DAVID incorporates Fisher’s exact test and was used to determine significant ( q < 0 . 05 , Benjamini ) GO terms and Kyoto Encyclopedia of Genes and Genomes ( KEGG ) pathways . GO terms were passed through REVIGO to prune redundant terms with a semantic similarity of 0 . 5 , 0 . 7 or 0 . 9 , depending on the length of the GO term ( Supek et al . , 2011 ) . Mammalian Phenotype ( MP ) enrichment was determined using the Mouse Genome Informatics ( MGI ) database ( Eppig et al . , 2015 ) . For validation of results from RNA-seq , qRT-PCR was performed using the BioMark system ( Fluidigm ) as according to the two-step single-cell gene expression protocol using EvaGreen as described in the Real-Time PCR Analysis User Guide ( PN 68000088 , Fluidigm ) . For the process , 666 ng of starting RNA was used for each sample , which underwent 18 cycles of specific target amplification ( STA ) followed by a tenfold dilution . SCN CT values were normalised using the housekeeping genes ActB , Gapdh and eight genes which showed the most stable expression in the SCN according to sequencing data ( Ankrd40 , Nkiras1 , Sar1a , Smarce1 , Snrpn , Tpgs2 , Tsn and Ywhaz ) . Primer sequences are listed in Supplementary file 1 . For RT-PCR of novel exons of Cry1 , total RNA was purified from dissected SCN tissue and used to prepare cDNA ( Revertaid , Fermentas ) . Primers representing previously annotated ( 5’ GTGAGGAGGTTTTCTTGGAAG 3’ ) and novel ( 5’ CTTCTAGGGAATTGCGACTG 3’ ) exons were used with a common reverse primer ( 5’ CTGGGAAATACATCAGCTGG 3’ ) to amplify products by RT-PCR followed by visualisation on an agarose gel and sequencing . The genomic coordinates of the novel exon are: chr10: 85 , 183 , 342 to 85 , 183 , 832 ( mm10 ) with splicing into the following exon at 85 , 183 , 342 or 85 , 183 , 510 . Brains from male mice representing ZT6 were isolated as above and frozen sections at 15 μM were cut and mounted onto Superfrost slides ( VWR ) . Regions of the selected transcripts were cloned into pCR4-TOPO ( Life Technologies , Paisley , United Kingdom ) for DIG-labelled riboprobe synthesis: Avp ( 1–490 bp of accession number NM_009732 . 1 ) , Vip ( 191–715 bp of NM_011702 . 1 ) , Scg2 ( 844–1287 bp of NM_009129 . 1 ) , Grpr ( 1287–1741 bp of NM_008177 . 1 ) and NONCO7761 ( 3100003L05Rik; 32–583 bp of NR_045907 ) . Slide hybridisation and riboprobe detection was carried out as described previously ( Chodroff et al . , 2010 ) .	The daily cycles of life in mammals are driven by a small region of the brain called the suprachiasmatic nucleus ( or SCN ) . The SCN receives signals from sunlight and other environmental factors to help coordinate most aspects of daily biological activity and behaviour . To work correctly , it is essential that the SCN switches certain genes on and off at exactly the right time . However , many questions remain over the identity of these genes and how their levels of activity change during a 24-hour period . When a gene is active ( or “being expressed” ) , it is used as a template to build the molecules of RNA that are needed to make proteins and to help to control how cells work . Pembroke et al . have now sequenced the RNA molecules made in the SCN of mice ( which plays the same role as the equivalent human brain region ) over a 24-hour period . The mice spent half of each day in the light , and half in the dark . This revealed that the expression levels of over a quarter of all the genes that are found in the SCN fluctuate over a 24-hour period . One particular group of genes peak in activity twice a day; Pembroke et al . suggest that these genes are important for controlling how an animal can adjust its body clock to light . Further research is now needed to find out which of the newly discovered fluctuating genes play the most important roles in daily activity rhythms , and which might play a part in disease .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "neuroscience", "tools", "and", "resources", "genetics", "and", "genomics" ]	2015	Temporal transcriptomics suggest that twin-peaking genes reset the clock
Voltage-sensing domains ( VSDs ) underlie the movement of voltage-gated ion channels , as well as the voltage-sensitive fluorescent responses observed from a common class of genetically encoded voltage indicators ( GEVIs ) . Despite the widespread use and potential utility of these GEVIs , the biophysical underpinnings of the relationship between VSD movement and fluorophore response remain unclear . We investigated the recently developed GEVI ArcLight , and its close variant Arclight' , at both the single-molecule and macroscopic levels to better understand their characteristics and mechanisms of activity . These studies revealed a number of previously unobserved features of ArcLight's behavior , including millisecond-scale fluorescence fluctuations in single molecules as well as a previously unreported delay prior to macroscopic fluorescence onset . Finally , these mechanistic insights allowed us to improve the optical response of ArcLight to fast or repetitive pulses with the development of ArcLightning , a novel GEVI with improved kinetics . Accurately measuring membrane potential is crucial for understanding the activity of excitable cells . Traditionally , these measurements have been performed using various configurations of electrodes , but these can present numerous challenges such as mechanical disruption of cells and tissues and an inability to measure from many sites simultaneously with high spatial resolution ( Salzberg , 1989 ) . It has been found that the fluorescence of many membrane-bound organic dyes changes as a function of membrane potential ( Tasaki et al . , 1968; Cohen et al . , 1974; Loew et al . , 1979; Davila et al . , 1973 ) . Accordingly , this fluorescence can serve as an optical reporter of transmembrane voltage and render electrodes unnecessary . While this solves many problems associated with the use of electrodes , these potentiometric dyes stain cell membranes indiscriminately , leading to the labeling of glia and other tissue as well as neurons ( Jin et al . , 2010 ) . Fortunately , recent advances in optogenetics have provided an alternative to these techniques through the use of genetically encoded fluorescent voltage indicators ( GEVIs ) ( Baker et al . , 2008; Jin et al . , 2010 ) . These fluorescent proteins are expressed under the control of a desired genetic promoter and thus can be targeted to a chosen cell type . Once translated , the GEVIs are trafficked to the cell membrane where they sense changes in membrane potential and transduce this into a change in their fluorescence , thereby providing an optical readout of cellular electrical excitability . Many different designs have been employed to engineer GEVIs ( Siegel and Isacoff , 1997; Sakai et al . , 2001; Dimitrov et al . , 2007; Kralj et al . , 2011; Akemann et al . , 2012 ) . To date , most of these consist of a protein voltage-sensing domain ( VSD ) from a voltage-sensitive ion channel or phosphatase fused to one or more green fluorescent protein ( GFP ) derivatives . However , it remains largely unclear how movement of a VSD couples into changing fluorescence in an attached GFP , limiting the extent to which rational design can aid in the development and optimization of these GEVIs . To help address this knowledge gap , we have conducted an extensive biophysical investigation of a voltage-sensitive phosphatase-based GEVI , ArcLight ( Jin et al . , 2012 ) . We chose this molecule because it is one of the most promising GEVIs developed thus far with demonstrated utility in intact brain tissue ( Cao et al . , 2013 ) . ArcLight is comprised of the VSD from the Ciona intestinalis voltage-sensor containing phosphatase ( Ci-VSP ) ( Murata et al . , 2005 ) fused to a variant of GFP . This voltage-sensing domain is highly similar to the VSD of voltage-gated ion channels; all contain a bundle of four transmembrane helices , with the fourth segment ( S4 ) containing positively charged amino acids that behave as the primary sensors of voltage changes . The fluorescence of Arclight changes by more than 30% in response to a 100 mV change in membrane potential ( Jin et al . , 2012 ) , an extremely large voltage-sensitive change compared to most other fluorescent proteins with S4-voltage sensors . In this work , we first demonstrate that ArcLight fluorescence responds to motion of the attached voltage sensor and then investigate the characteristics of the fluorescence of ArcLight' , a close homolog of ArcLight , at the single-molecule level . Although single GEVI molecules appeared to function normally , these traces unexpectedly displayed a significant degree of noise at millisecond timescales . Further investigations into this noise revealed that ArcLight' fluorophores possess intrinsic noise due to internal dynamics of the GFP moiety even in the absence of the voltage sensor , and that this noise appears to be a general feature of many GFP constructs including eGFP . Reducing or eliminating this fluctuation noise of the fluorophore could be a novel path towards improving this class of GEVIs . Finally , we investigated the behavior of ArcLight fluorescence at the macroscopic level and used this investigation to develop a novel voltage indicator with improved characteristics . ArcLight demonstrates distinct advantages over many other genetically-encoded voltage indicators , including large signal size relative to background , relatively low spectral bandwidth requirements compared to many FRET-based sensors ( Mishina et al . , 2014; Akemann et al . , 2012; 2013; Tsutsui et al . , 2013 ) , high quantum yield compared to archaerhodopsin based sensors ( Flytzanis et al . , 2014; Gong et al . , 2013; Hochbaum et al . , 2014; Kralj et al . , 2012 ) , and a demonstrated success in multiple biological preparations ( Jin et al . , 2010; 2012; Cao et al . , 2013; Leyton-Mange et al . , 2014 ) . However , ArcLight’s ability to respond to rapid changes in membrane potential is limited by its slow kinetics ( Jin et al . , 2012 ) . To date , improvements in the kinetics of ArcLight have relied on switching residues from the Ciona intestinalis voltage-sensing domain ( Ci-VSD ) to residues from analogous voltage-sensitive phosphatases found in other species , especially Gallus gallus and Danio rerio ( Piao et al . , 2015; Han et al . , 2013 ) . Our mechanistic investigations into the link between ArcLight voltage sensor movement and fluorescence revealed an avenue of improvement for ArcLight as a voltage indicator . Using the cut-open oocyte voltage-clamp technique , we found that ArcLight fluorescence likely monitors a protein transition subsequent to gating charge movement , but that accelerating gating kinetics using a previously-reported mutation ( Lacroix and Bezanilla , 2012 ) can nonetheless significantly speed up the fluorescence response . We termed this fast ArcLight derivative 'ArcLightning' . As expected from the accelerated kinetics of its Ci-VSD , ArcLightning expression in mammalian cells resulted in rapid and large voltage-sensitive fluorescence changes . This work shows that mutations discovered during biophysical studies of voltage sensor behavior can be used to rationally design improved GEVIs . Our first goal was to validate that ArcLight fluorescence changes are driven by voltage sensor conformational changes , rather than by a different mechanism . While a direct response to membrane potential is unlikely as the fluorophore does not reside in the membrane electric field , it is conceivable that , for instance , the fluorophore is responding to localized pH changes near the membrane due to voltage-driven proton fluxes across the membrane . Using the cut-open oocyte voltage-clamp technique ( Stefani and Bezanilla , 1998 ) we were able to simultaneously measure the gating current of ArcLight ( Figure 1A ) and its fluorescence ( Figure 1B ) from the same oocyte ( Cha and Bezanilla , 1998 ) . We then utilized the well-described mutations of the R217 residue of the Ci-VSP voltage sensor ( Dimitrov et al . , 2007; Villalba-Galea et al . , 2013 ) to shift the voltage-dependence of the voltage sensor and observe the effects on the fluorescence . In ArcLight , this residue is mutated to a glutamine ( R217Q ) , which places the midpoint of voltage sensor response at approximately -20 mV . Returning this residue to arginine ( R217R ) shifts the midpoint of gating charge movement to positive potentials while replacing the glutamine with glutamate ( R217E ) shifts the response to even more negative potentials ( Figure 1C ) . Crucially , the fluorescence response mirrored these changes , indicating that ArcLight fluorescence signals are caused by voltage sensor movement rather than by direct influence of membrane potential or by voltage-driven ion concentration changes . 10 . 7554/eLife . 10482 . 003Figure 1 . ArcLight fluorescence responds to voltage sensor movements . ( A ) Family of gating currents from an oocyte injected with ArcLight R217Q . The holding potential is -80 mV , and the pulses range from -160 mV ( darkest red ) to +140 mV ( darkest blue ) in 20 mV increments . ( B ) Fluorescence traces simultaneously acquired with gating currents in A . Colors indicate identical membrane potentials as in A . ( C ) Normalized gating-charge-versus-voltage ( Q-V ) curves ( triangles; cyan – R217R , grey – R217Q , purple – R217E ) and fluorescence-versus-voltage ( F-V ) curves ( circles; blue – R217R , black – R217Q , red – R217E ) for ArcLight . For clarity of comparison , the fluorescence is plotted as the change in fluorescence over the background fluorescence , multiplied by -1 ( i . e . , -ΔF/F0 ) . Each Q-V curve was normalized by fitting a Boltzmann function to the data and scaling the limiting values of the function to 0 and 1 . These scaling factors were then used to normalize each respective F-V curve . Error bars represent 95% confidence intervals of the mean . N = 5 for R217R , 7 for R217Q , and 6 for R217E . DOI: http://dx . doi . org/10 . 7554/eLife . 10482 . 003 Our next goal was to determine whether single GEVI molecules showed any unexpected behaviors that were being masked by ensemble averages . To do this , we visualized voltage-dependent fluorescence from single GEVI molecules using total internal reflection fluorescence microscopy ( TIRFM ) . This technique presented an electrophysiological challenge , as TIRFM can only be performed on an oocyte once the vitelline membrane has been removed . The bare plasma membrane of the oocyte seals to the glass coverslip producing high access resistance to the bath electrodes and , consequently , a poor voltage clamp . Using an electrochromic voltage-sensitive small molecule dye ( di-8-ANEPPS ) , we measured the speed of the voltage clamp on the underside of a peeled oocyte placed directly on a glass coverslip ( Figure 2—figure supplement 1 ) . As expected , the clamp was very slow due to the high access resistance . To remedy this situation , we coated glass coverslips with a polymer cushion . This cushion created a larger , conductive aqueous space between the oocyte and the glass that was thin enough to allow successful TIRFM . This was found to dramatically improve voltage clamp speed ( Figure 2—figure supplement 1 ) . This advance allowed us to successfully voltage clamp the molecules observable in TIRFM . Having overcome this challenge , we turned our attention to observing single molecules of ArcLight' ( see Methods section for details ) . Briefly , ArcLight' is a previously published variant of ArcLight ( Jin et al . , 2012 ) with two mutations in the GFP domain that seemed to improve its performance as a single-molecule fluorophore . Oocytes expressing low concentrations of ArcLight' had their vitelline membranes removed mechanically and were placed on polymer-coated coverslips and imaged with TIRFM . At the site of cRNA injection ArcLight' density appeared high and single molecules could typically not be resolved . However , imaging from fields further from the site of injection showed distinct , punctate sources of light consistent with small clusters and single molecules of ArcLight’ ( Figure 2A ) . As expected , many of these points showed clear changes in fluorescence in response to applied voltage changes , suggesting that they came from functional GEVI molecules ( Figure 2B ) . To confirm that these single-molecules were behaving normally , traces from many different fluorescent spots were summed together ( Figure 2C ) ; these summed traces recapitulated the macroscopic response of ArcLight' to changes in membrane potential indicating that their function was uncompromised . Although the single-molecules appeared to be operating normally , they consistently displayed an interesting phenomenon: a large degree of noise that did not appear to be shot noise . Whereas shot noise has no correlation from data point to data point , the single-molecule traces fluctuated between relatively distinct fluorescence levels , sometimes remaining at a single level for many milliseconds ( Figure 2B , C ) . To our knowledge , there have been no prior reports of GEVIs displaying fluctuations on these millisecond timescales . Accordingly , we initially hypothesized that these fluctuations were reflective of voltage sensor conformational state changes . To test this , we bacterially expressed isolated ArcLight' GFP domains without attached voltage sensors and observed them as single molecules . To our surprise , these isolated GFP domains displayed roughly similar fluctuations to those observed in the full ArcLight molecules ( Figure 3A ) , indicating that these fluctuations originate in the fluorophore itself . To confirm that this fluctuation noise was distinct from shot noise , we computed the autocorrelation functions for fluorescence traces from both an isolated ArcLight' GFP domain and from a region of background fluorescence from the same image ( Figure 3B , C ) . As expected , the background fluorescence displayed no significant autocorrelation , suggesting that the noise in this trace is primarily due to shot noise or other white noise processes . By contrast , the ArcLight' fluorescence showed significant correlation over tens of milliseconds . Finally , we performed a similar experiment on wild-type eGFP and also observed significant fluctuation noise ( Figure 3D ) . This result , consistent with the GFP photophysics literature , strongly suggests that this is not a pathology unique to the ArcLight' fluorophore , but rather is one that likely affects many current GEVIs ( see Discussion ) . This observation may lead to a novel avenue for improvement of GEVI signals based on the attenuation or elimination of this noise . 10 . 7554/eLife . 10482 . 004Figure 2 . Voltage-sensitive fluorescence can be recorded from single ArcLight' molecules . ( A ) An image of the surface of an oocyte expressing ArcLight' in single-molecule concentrations . Square image field is 50 μm × 50 μm . Height in the z-axis corresponds to fluorescence intensity . This image is an average of 800 frames . Background correction has been performed using morphological filtering . ( B ) Three example traces are shown of single-molecule ArcLight' fluorescence in response to a three second depolarizing pulse from -90 mV to +60 mV . Traces were filtered at 40 Hz and had a linear baseline subtraction applied . ( C ) The upper of the two traces shows an example trace of single-molecule ArcLight' fluorescence from -120 mV to +120 mV for 600 ms ( red dashed lines mark the beginning and end of the depolarizing pulse ) . The lower of the two traces is the average of 67 traces . This summation demonstrates that the single-molecule data here recapitulates the macroscopic voltage-sensitive fluorescence response of Arclight . Both traces were filtered at 100 Hz . A linear baseline subtraction was applied to the summed trace . ADU is analog-to-digital units , the output unit of the EMCCD camera . DOI: http://dx . doi . org/10 . 7554/eLife . 10482 . 00410 . 7554/eLife . 10482 . 005Figure 2—figure supplement 1 . A polymer cushion improves clamp speed under a peeled oocyte . The speed of the voltage clamp on the bottom side of an oocyte is observed by monitoring the fluorescence of di-8-ANEPPS in the oocyte membrane . When a peeled oocyte is placed directly on a clean glass coverslip , it seals against the glass and creates a slow voltage clamp ( red trace ) . However , coating the coverslip with a water-soluble polymer ( polyethylenimine ) decreases access resistance under the oocyte and thus improves clamp speed in this region ( blue trace ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10482 . 00510 . 7554/eLife . 10482 . 006Figure 3 . Fluorescence from an isolated ArcLight' GFP domain and eGFP show significant fluctuation noise . ( A ) Fluorescence from a single isolated GFP domain from ArcLight' with no attached voltage sensor . The fluorescence displays fluctuations between different levels and does not appear to be Poisson-distributed . The trace was filtered at 100 Hz . ( B ) Background fluorescence from a region of an image with no GFP molecule displays no significant autocorrelation at any non-zero lag . This suggests that most noise in this regime is due to white processes such as shot noise . ( C ) The fluorescence from a single isolated GFP domain of ArcLight' displays significant autocorrelation . This suggests that the fluorescence moves between distinct levels corresponding to different conformational , chemical , or electronic states of the GFP domain . Blue bars in B and C represent approximate 95% confidence intervals for a Gaussian-distributed white noise process . No filtering was applied to the data used for the autocorrelations . ( D ) Single molecules of wild-type eGFP also display fluctuation noise that appears qualitatively similar to that of ArcLight' GFP domains , suggesting that this noise may be present in many GFP derivatives . The trace was filtered at 100 Hz . ADU is analog-to-digital units , the output unit of the EMCCD camera . DOI: http://dx . doi . org/10 . 7554/eLife . 10482 . 006 Having investigated the characteristics of single ArcLight fluorophores , we turned our attention towards understanding the macroscopic response of this molecule . As discussed previously , altering the set point of the voltage-dependence of the Ci-VSP voltage-sensing domain by mutating the R217 residue to either a Q or an E correspondingly alters the set point of the ArcLight fluorescence change ( Figure 1C ) . Indeed , the change in ArcLight fluorescence at steady state was not statistically significantly different from the degree of change in Ci-VSD conformation as measured by gating charge movement . We also tested whether ArcLight fluorescence kinetics are the same as the kinetics of the gating charge movement ( see Methods for details ) . In both 'on' responses where the membrane was held at a negative potential and then pulsed to depolarized potentials ( Figure 4A ) and 'off' responses where the membrane was held at a positive potential and then pulsed to hyperpolarized potential ( Figure 4B ) , we observed that ArcLight fluorescence kinetics are considerably slower than those of voltage sensor movement ( Figure 4C , D ) . The fastest component of fluorescence during the 'on' response was approximately twice as slow as the gating current kinetics , and the 'off' fluorescence kinetics are more than ten times slower than voltage sensor gating . Indeed , neither component of fluorescence response appears to correlate well with any component of gating charge movement ( Figure 4—figure supplement 1 ) . Overlaying simultaneously acquired gating charge and fluorescence traces , along with an estimated measure of membrane potential , makes the slowed response of ArcLight fluorescence especially apparent ( Figure 4—figure supplement 1 ) . These observations raised the question of how gating charge movement is kinetically linked to ArcLight fluorescence . Interestingly , there is a distinct delay between the induction of a voltage change and the initiation of the resulting fluorescence change from ArcLight ( Figure 4—figure supplement 2 , Figure 4E , F ) . This delay shows a marked voltage-dependence and is present in both on and off fluorescence transitions . Both the delays and fluorescence kinetics show pronounced voltage dependencies , with τ-V curves which are slowest at intermediate potentials and which speed up towards limiting values at extreme potentials . This pattern is reminiscent of gating current τ-V curves , and suggests that voltage sensor movement is partially responsible for determining ArcLight fluorescence kinetics . In fact , the delays of the fluorescence response align quite well with the gating current kinetics ( Figure 4E ) . In addition to wild-type ArcLight , we also investigated the kinetics of fluorescence response in relation to gating current kinetics for the R217R and R217E constructs ( Figure 4—figure supplement 3 ) . These two mutants generally behaved similarly to the R217Q construct with fluorescence lags of a comparable speed to VSD movement and a considerably slower fluorescence response . 10 . 7554/eLife . 10482 . 007Figure 4 . ArcLight fluorescence partially follows the kinetics of voltage sensor movement . ( A ) ArcLight fluorescence response to an 'on' pulse protocol with a holding potential of -120 mV with 200 ms pulses ranging from +120 mV to -140 mV by 20 mV intervals . ( B ) As in A , but in response to an 'off' pulse protocol pulsing from a holding potential of +40 mV to 200 ms pulses ranging from -160 mV to +40 mV by 20 mV intervals . ( C ) ArcLight on gating kinetics ( blue ) are faster than and do not correlate strongly with ArcLight fluorescence kinetics ( green ) . Kinetics were obtained from gating currents and fluorescence changes recorded simultaneously from the same oocyte . ( D ) ArcLight off gating kinetics ( blue ) are much faster than and do not correlate with ArcLight fluorescence kinetics ( green ) . ( E ) Upon depolarization , ArcLight fluorescence change shows a distinct lag ( see Figure 4—figure supplement 1 ) . This lag is voltage-dependent , becoming shorter with more extreme changes in membrane potential . ArcLight gating current kinetics from the on pulse ( blue ) correlate quite well with the ArcLight fluorescence lag ( red ) . ( F ) ArcLight gating current kinetics from the off pulse ( blue ) correlate with ArcLight fluorescence lag ( red ) better than they do with ArcLight fluorescence kinetics ( shown in D , green ) . N = 4 for on pulse protocol data , 5 for off pulse protocol data . DOI: http://dx . doi . org/10 . 7554/eLife . 10482 . 00710 . 7554/eLife . 10482 . 008Figure 4—figure supplement 1 . ArcLight fluorescence changes are slower than integrated gating charge kinetics . ( A ) Both gating charge movement kinetics ( blue , obtained by integrating gating currents ) and fluorescence changes ( green ) of ArcLight are well-fit by a double-exponential function ( circles – faster time constant; squares – slower time constant ) . All fluorescence kinetics are slower than all gating charge kinetics . ( B ) The slow kinetics of the ArcLight fluorescence response are emphasized by overlaying traces for estimated membrane potential ( black ) , integrated gating charge ( blue ) , and fluorescence ( green ) . The upper group of three traces is for a pulse to +40 mV , while the lower group of three traces is for a pulse to -40 mV . The holding potential is -120 mV for both sets of traces . Estimated membrane potential was calculated by integrating an uncompensated capacitive transient acquired during pulse protocols identical to those used to record the respective gating charge and fluorescence traces . Gating charges were normalized by the maximum charge reached during the +40 mV acquisition . Estimated membrane potential and fluorescence traces were each normalized to their respective integrated gating charge traces . DOI: http://dx . doi . org/10 . 7554/eLife . 10482 . 00810 . 7554/eLife . 10482 . 009Figure 4—figure supplement 2 . The onset of ArcLight fluorescence change lags behind voltage change onset . ( A ) Following the onset of the voltage change ( vertical red line ) , a measurable lag occurs prior to the onset of ArcLight fluorescence change . Traces are shown from a holding potential of -80 mV and range from +140 mV ( darkest blue ) to -160 mV ( darkest red ) . ( B ) Exponential fits to the fluorescence response provide a quantitative measure of this delay , taken at the point where the fit crosses the abscissa . For example , at +120 mV ( fluorescence trace in blue ) the exponential fit ( solid pink line ) crosses the abscissa shortly after 5 ms ( the intersection is marked by a dashed vertical pink line ) . This is about 2 ms after voltage onset , marked by a vertical red line as in A . Similarly , at 0 mV ( fluorescence trace in green , darkened from the trace in A for improved visual clarity ) the exponential fit ( solid dark green line ) crosses the abscissa ( intersection marked by a dashed vertical dark green line ) at about 9 ms , or 6 ms after voltage onset . DOI: http://dx . doi . org/10 . 7554/eLife . 10482 . 00910 . 7554/eLife . 10482 . 010Figure 4—figure supplement 3 . The R217R and R217E ArcLight mutants show similar behaviors to R217Q , but with shifted voltage dependence . ( A ) Gating current weighted time constants ( blue ) , fluorescence lags ( red ) , and the fast time constants of the fluorescence response ( green ) were calculated for ArcLight R217R as described above for wild-type ArcLight . The holding potential was -80 mV . This construct follows the same general pattern as the R217Q ArcLight: the fluorescence lags are of similar speed to voltage sensor movement but the fluorescence response itself is considerably slower . ( B ) As in A , but for ArcLight R217E . The holding potential was -80 mV . This construct also displays a pattern similar to wild-type ArcLight . DOI: http://dx . doi . org/10 . 7554/eLife . 10482 . 010 To investigate these behaviors further , we tested whether altering the kinetics of Ci-VSD movement altered the kinetics of either the slow ArcLight fluorescence change or the lag before fluorescence change . By mutating the residue I126 in the S1 segment of the voltage-sensing domain of Ci-VSP to a phenylalanine , gating current activation kinetics were previously shown to be accelerated roughly forty-fold while deactivation currents were accelerated roughly eighty-fold ( Lacroix and Bezanilla , 2012 ) . We hypothesized that making the same mutation in ArcLight might speed up fluorescence responses in addition to gating currents . As the I126F mutation also shifted the Ci-VSP Q-V to more hyperpolarized potentials , the Q at position 217 in ArcLight was mutated back to R to keep the voltage-dependence at roughly physiological levels ( Figure 5A ) . Since Ci-VSP R217R gating currents are about four-fold slower than R217Q gating currents ( Villalba-Galea et al . , 2013 ) , we would expect that ArcLight I126F Q217R would display gating current kinetics about 10 times faster in response to depolarizing pulses and 20 times faster in response to hyperpolarizing pulses than wild-type ArcLight . In practice , we observe a roughly six to twelve-fold increase in speed ( Figure 5B , C Figure 5—figure supplement 1 ) . Accordingly , if ArcLight fluorescence kinetics track directly with the kinetics of its voltage sensing domain , the fluorescence changes of ArcLight I126F should be about 8 times faster than wild-type ArcLight during both depolarizations and hyperpolarizations . However , while I126F did accelerate the kinetics of ArcLight fluorescence , it did so in an interesting manner ( Figure 5D , E ) . Specifically , while wild-type ArcLight fluorescence kinetics show a pronounced dependence on voltage , I126F kinetics are quite flat over most of the physiological voltage range . Furthermore , as the wild-type kinetics speed up towards extreme membrane potentials , they typically seem to approach the near-constant value of I126F . Interestingly , the lags before fluorescence movement seem to show a very similar pattern ( Figure 5F , G ) . An additional feature is the remarkable acceleration of I126F fluorescence kinetics upon hyperpolarization compared to those from ArcLight ( Figure 5E ) . These observations provide some insight into the biophysical basis of ArcLight fluorescence sensing ( see Discussion ) . Due to its improved kinetic response versus wild-type ArcLight , we have termed the I126F Q217R mutant 'ArcLightning' . 10 . 7554/eLife . 10482 . 011Figure 5 . Accelerated gating kinetics speed up fluorescence , up to a point . ( A ) The voltage-dependence of fluorescence change of ArcLight I126F Q217R occurs over a physiological voltage range , as taken from an on protocol identical to that in Figure 4 . ( B ) The gating current kinetics of ArcLight I126F Q217R ( squares ) obtained from the on pulse protocol are faster than those measured from ArcLight ( circles ) . ( C ) The gating current kinetics of ArcLight I126F Q217R ( squares ) obtained from an off pulse protocol identical to that in Figure 4 are faster than those measured from ArcLight ( circles ) . ( D ) The fluorescence kinetics of ArcLight I126F Q217R ( squares ) in response to a depolarizing pulse are generally faster than those measured from ArcLight ( circles ) . A notable exception is at more positive potentials , e . g . +80 mV , where the kinetics of the two constructs are essentially equal . ( E ) The fluorescence kinetics of ArcLight I126F Q217R ( squares ) obtained in response to a hyperpolarizing pulse are much faster than those measured from ArcLight ( circles ) . ( F ) The fluorescence lag of ArcLight I126F Q217R ( squares ) during on pulses is briefer than the lag seen in ArcLight’s fluorescence ( circles ) . ( G ) As in F , but the fluorescence lag upon an off pulse . On and off protocols were as in Figure 4 , and n = 5 for all ArcLight I216F Q217R data . DOI: http://dx . doi . org/10 . 7554/eLife . 10482 . 01110 . 7554/eLife . 10482 . 012Figure 5—figure supplement 1 . Substitution of I126F and Q217R into ArcLight induces a roughly constant acceleration of gating current . ( A ) Ratio of ArcLight gating current kinetics to I126F Q217R gating current kinetics across all voltages in response to an 'on' protocol . ( B ) As in A , but in response to an 'off' protocol . The on and off protocols were identical to those described in Figure 4 . Briefly , depolarizations used a holding potential of -120 mV , with 200 ms pulses from -140 mV to +120 mV; hyperpolarizations used a holding potential of +40 mV , with 200 ms pulses from +40 mV to -140 mV . N = 4 for ArcLight activation kinetics and 5 for all other kinetics . DOI: http://dx . doi . org/10 . 7554/eLife . 10482 . 012 Given its faster response relative to ArcLight , we suspected that ArcLightning might be useful as an improved GEVI . When expressed in HEK cells and measured at 19°C , ArcLightning showed signals which were substantially faster than those of ArcLight , particularly when returning to hyperpolarized potentials ( Figure 6A ) . The acceleration of kinetics observed from ArcLight to ArcLightning in oocytes is recapitulated in mammalian cells ( Figure 6—figure supplement 1 ) . The fast component of ArcLightning fluorescence is faster than the fast component of ArcLight fluorescence at physiological membrane potentials , and the kinetics of the two constructs converge to similar values at extreme potentials where the ArcLight gating is faster . Furthermore , over much of the physiological voltage range , the fractional amplitude of the fast component is larger in ArcLightning than in ArcLight . The increased speed of return from a depolarizing pulse is particularly noticeable in the case of repetitive stimuli ( Figure 6B ) . In this case , ArcLightning is able to return to baseline fluorescence in between consecutive voltage steps , providing an accurate readout of the size of each step . ArcLightning also retains its improved kinetics at 35°C . In response to trains of short pulses with durations emulating action potentials at 67 Hz , ArcLightning is still able to recover to baseline in between pulses ( Figure 6C ) . ArcLight , by contrast , does not recover in time and thus shows an apparent increase in magnitude of each successive voltage pulse , despite them all being of equivalent magnitude . Although the larger net signal size of ArcLight may remain advantageous in some situations , the faster kinetics of ArcLightning should help in the resolution and analysis of many fast phenomena . 10 . 7554/eLife . 10482 . 013Figure 6 . ArcLightning displays improved fluorescence response to repetitive pulses in mammalian cells . ( A ) ArcLightning ( pink ) expressed in HEK cells displays moderately faster kinetics in response to depolarization and much faster kinetics in response to hyperpolarization than ArcLight ( blue ) at 19°C . Notice that ArcLightning develops a slower component only at potentials exceeding +50 mV . ( B ) In response to a train of 100 ms pulses , separated by 100 ms of holding at -80 mV , ArcLightning ( pink ) displays an improved ability to measure discrete events compared to ArcLight ( blue ) . All traces in A and B had no baseline subtraction or filtering . ( C ) At 35°C , both ArcLight ( blue ) and ArcLightning ( pink ) follow 5 ms , 100 mV pulses separated by 10 ms at the resting potential to mimic action potentials at approximately 67 Hz . ArcLightning provides the additional advantage of abolishing the sloping baseline of fluorescence that ArcLight produces in response to repetitive pulses as a result of its slow off kinetics . Both traces had a single-exponential baseline subtracted and were low-pass filtered at 300 Hz . DOI: http://dx . doi . org/10 . 7554/eLife . 10482 . 01310 . 7554/eLife . 10482 . 014Figure 6—figure supplement 1 . ArcLightning is faster than ArcLight in mammalian cells . ( A ) The on kinetics of ArcLightning ( pink squares ) are faster than those of ArcLight ( blue circles ) at physiological voltages . As seen in oocytes , kinetics became similar at highly depolarized potentials , suggesting an underlying rate-limiting step of fluorescence change that is not highly voltage-dependent . ( B ) The fractional amplitude of the fast time constant in response to depolarization is higher in ArcLightning ( pink squares ) than in ArcLight ( blue circles ) . Thus , in addition to being faster , the first component of ArcLightning comprised a greater percentage of its total fluorescent response . ( C ) As in A , but for kinetics in response to hyperpolarizations to baseline following the on pulse . ( D ) As in B , but for fractional amplitude of the fast time constant of responses to hyperpolarizing pulses . Kinetics of ArcLightning responding to hyperpolarization were similar to or faster than those of ArcLight , and the fractional amplitude of the fast time constant was larger for ArcLightning than for ArcLight . To obtain these kinetics , an 100 ms on pulse was made from a holding potential of -80 mV , ranging from +100 mV to -120 mV in increments of 20 mV . Activation kinetics were measured from the on pulse; deactivation kinetics were measured from the off as the potential returned to -80 mV . N = 5 for all data . DOI: http://dx . doi . org/10 . 7554/eLife . 10482 . 014 Here we report three advances in our understanding of GEVIs in general and ArcLight in particular . First , we show that the ArcLight' GFP domain possesses considerable intrinsic fluctuation noise on the millisecond timescale , and that this noise is also present in wild-type eGFP . Second , we demonstrate that for ArcLight , the transitions of the voltage-sensing domain align with the transitions of the fluorescence signal , although the fluorescence changes are significantly slower than the voltage sensor changes . Finally , we used a point mutation discovered during biophysical investigations of Ci-VSP to design ArcLightning , a novel version of ArcLight with accelerated movements of the voltage-sensing domain and improved fluorescence response times . Our single-molecule analysis of ArcLight' revealed a previously unrecognized behavior of the GEVI: an inherent noise , or fluctuation in the fluorescence signal . This noise appears to not be Poisson-distributed , and thus is not likely simply attributable to shot noise from a single fluorophore . Furthermore , we observed similar fluctuations in single molecules of eGFP . There have been prior reports of fluorescence fluctuations in various GFP derivatives which generally fall into two categories: millisecond and microsecond-scale fluctuations detected with fluorescence correlation spectroscopy ( Hess et al . , 2004; Haupts et al . , 1998; Bosisio et al . , 2008 ) , and much slower blinking detected in imaging studies of gel-immobilized fluorophores that displayed on- and off- dwell times that are typically hundreds of milliseconds and seconds , respectively ( Dickson et al . , 1997; Peterman et al . , 1999; Garcia-Parajo et al . , 2000 ) ; the latter of these modes of fluctuation appears to be quite different from what we observe . We have also , however , found one prior report of fluctuation in an immobilized GFP mutant on comparable timescales to what we observe ( Moerner et al . , 1999 ) , further emphasizing that this noise may be a general feature of GFP derivatives . Crucially , however , this noise has to our knowledge never been verified or acknowledged in a GEVI . This is important as this noise adversely impacts the use of GEVIs in two ways . First , while the fluctuations themselves are not visible in macroscopic recordings due to the large number of fluorophore signals being averaged together , they very likely sum together in some manner to raise the noise level of macroscopic recordings . Second , molecules in the dark state do not contribute useful 'signal' photons to help overcome 'noise' photons delivered by autofluorescence and other sources . Both of these effects likely serve to decrease the signal-to-noise ratio observed for a given set of excitation parameters as compared to a non-fluctuating GEVI , thus limiting the size of phenomena that can be resolved . Accordingly , the reduction or elimination of this noise would likely be a significant benefit to GEVIs , thus providing a new direction for future GEVI optimization . In addition , prior work has shown that pulsed excitation light can significantly increase the amount of signal derived from fluorophores that visit dark states ( Mejía-Alvarez et al . , 2003; Donnert et al . , 2007 ) . By showing that the ArcLight' fluorophore makes frequent fluctuations , this work motivates investigations into whether an appropriately-modulated excitation can improve the signal from this and other GEVIs . Despite its popularity as a GEVI , the mechanism by which ArcLight transduces the motion of the VSD into a change in the fluorescence of the attached eGFP remains largely unknown ( Han et al . , 2013; Han et al . , 2014 ) . However , the data presented here provides some insight into this process . The gating current kinetics of ArcLight look qualitatively similar to those from Ci-VSP R217Q ( Figure 1A ) with roughly equal kinetics and V1/2 ( 11 ms versus 8 ms , -7 mV versus -16 mV ) ( Figure 4C , Figure 1C ) ( Villalba-Galea et al . , 2013 ) . This suggests that the attachment of the fluorescent protein is producing at most a small change in load on the voltage sensor movement , as compared to the natural load of the linker and the phosphatase in Ci-VSP R217Q . However , in ArcLight the fluorescence changes are substantially slower than the simultaneously measured gating currents ( Figure 4C ) , and contain a measurable lag prior to onset ( Figure 4E , Figure 4—figure supplement 1 ) . This initially suggested a three-state minimal model of fluorescence with two state transitions: the first carrying gating charge movement and the second corresponding to fluorescence change ( Figure 7—figure supplement 1 ) . The lag would be accounted for by the time required for a significant population to reach the second state , while a smaller rate constant on the second transition than the first transition would cause the fluorescence response to be slower than gating charge movement . To test this three-state model , we accelerated the gating charge movement of ArcLight with the I126F mutation . If the three-state model ( Figure 7—figure supplement 1 ) was accurate , we would expect the lag preceding fluorescence change to decrease roughly in proportion to the quickening in gating current kinetics while maintaining a similar τ-V dependence . Instead , we observed that both the lag and the fluorescence kinetics become only weakly dependent on voltage . Furthermore , the value they take is approximately the same value that wild-type ArcLight reaches at extreme potentials when its voltage sensor is moving maximally fast . In particular , the lag preceding fluorescence does not speed up by nearly the same factor that the gating current speeds up . These observations are more consistent with a four-state minimal model ( Figure 7 ) . Here , as in the previous model , the first transition carries the gating charge and possesses voltage-dependent kinetics . The second transition accounts for the lag and possesses kinetics which do not strongly depend on voltage . Finally , the third transition is responsible for the change in fluorescence and also has nearly voltage-independent kinetics . Thus , in wild-type ArcLight , the first transition contributes significantly to the overall observed time constants , causing the observed lags and fluorescence kinetics to behave as if they were voltage-dependent . In ArcLightning , however , the first transition is so fast that it does not contribute significantly and the next two transitions essentially become rate-limiting . This results in the observed values , which do not depend strongly on voltage . 10 . 7554/eLife . 10482 . 015Figure 7 . A minimal four-state model of ArcLight kinetics . The first transition carries the gating charge movement while the third transition carries the fluorescence change and the second transition carries no observable changes . The kinetics of the first transition depend strongly on voltage , while the kinetics of the second two are only weak functions of voltage . Thus , if k1≫k2 , k3 , the second two transitions will become rate-limiting , and both the lag and fluorescence kinetics will appear nearly voltage-independent , as observed when gating is accelerated with the I126F mutation or in wild type ArcLight when gating becomes fast due to pulsing to extreme potentials . This model explains why wild-type ArcLight fluorescence kinetics and lags approach the same values as the I126F mutant at extreme potentials . DOI: http://dx . doi . org/10 . 7554/eLife . 10482 . 01510 . 7554/eLife . 10482 . 016Figure 7—figure supplement 1 . A minimal three-state model of ArcLight kinetics . The first transition carries the gating charge movement while the second transition carries the fluorescence change . If k1>k2 , the fluorescence would be slower than the gating currents , as observed . The lag would be caused by the delay in having a sufficient population reach state S1 such that they can begin making the second transition and undergoing fluorescence change . This model was eventually unable to account for all observed phenomena , prompting the creation of a four-state model ( see Figure 7 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10482 . 01610 . 7554/eLife . 10482 . 017Figure 7—figure supplement 2 . ArcLight fluorescence may be influenced by the relaxed state . ( A ) The Q-V curve obtained from a protocol of hyperpolarizing pulses from a +40 mV holding potential ( red squares ) is shifted to more negative potentials compared to a Q-V curve obtained from depolarizing pulses from a -120 mV holding potential ( blue squares ) , as predicted if relaxation is occurring . The fluorescence response of ArcLight follows Q-V curves from holding potentials of both -120 mV ( cyan circles ) and +40 mV ( pink circles ) . Data was normalized as described in the Methods . ( B ) The fluorescence response to variable depolarizing pulses following holding at -120 mV ( cyan ) is faster than the response to variable hyperpolarizing pulses following a 200 ms pulse to +40 mV ( black ) . Applying a holding potential of +40 mV slows the subsequent fluorescence response upon hyperpolarization still further ( pink ) , consistent with fluorescence responses slowing from an entrance to the relaxed state . On and off protocols are identical to those described for Figure 4 , except for the +40 mV pulse protocol , which used a -80 mV holding potential for the remainder of the five second cycle period . N = 4 for holding at -120 mV , 5 for holding at +40 mV , and 6 for pulsing to +40 mV . DOI: http://dx . doi . org/10 . 7554/eLife . 10482 . 017 Taken together , these data suggest that ArcLight fluorescence changes are slower than gating currents primarily due to an intrinsic delay in the fluorophore or linker , rather than the fluorescence accurately reporting on late VSD movements such as relaxation ( Villalba-Galea et al . , 2008 ) . There are two pieces of evidence in favor of this intrinsic delay . First , ArcLight still shows fluorescence responses at extreme hyperpolarized potentials where the VSD is unlikely to enter the relaxed state . Second , speeding up VSD movement does not make the fluorescence response arbitrarily fast . Rather , there seem to be voltage-independent intrinsic delays in fluorescence response which become rate-limiting once VSD movement is fast enough ( Figure 5 ) . Thus , at least some component of the ArcLight response is likely due to VSD activation coupled with intrinsic fluorophore delay . While ArcLight fluorescence does not seem to be a direct readout of VSD relaxation , the relaxed state may nonetheless play a role in the very slow response of fluorescence to hyperpolarizing pulses following a prolonged depolarization ( Figure 4D ) . This is suggested by two observations: first , ArcLight’s fluorescence follows the leftward Q-V shift in gating current that is a hallmark of the relaxed state in Ci-VSP ( Villalba-Galea et al . , 2008 ) ( Figure 7—figure supplement 2 ) . Second , ArcLight’s deceleration of off kinetics increases as the duration of the depolarizing pulse increases ( Figure 7—figure supplement 2 ) , as occurs in the relaxation process of voltage-gated potassium channels ( Lacroix et al . , 2011 ) . Thus , exit of the VSD from the relaxed state may be linked to the extremely slow fluorescence kinetics . Finally , VSD relaxation explains the comparatively flat kinetics observed in ArcLight gating currents during positive holding potentials . Since these holding potentials dramatically left-shift the V1/2 of the ArcLight Q-V to values well outside experimentally-accessible values ( Figure 7—figure supplement 2 ) , the corresponding observed τ-V curve ( Figure 4F ) represents only a single tail of the true τ-V , and thus appears to be relatively constant . ArcLightning is unique when compared to many previously developed fluorescent probes . Its speed is not as rapid as ElectricPk ( Barnett et al . , 2012 ) , but it has a much larger voltage-sensitive fluorescence response . As opposed to ArcLightning , both ArcLight and the more recent voltage indicator ASAP1 have significant slow components in their off response ( Jin et al . , 2012; St-Pierre et al . , 2014 ) . Accordingly , further study of the interactions between VSD movement and fluorescence response , including the contribution of the relaxed state , of voltage-sensitive phosphatase derived GEVIs should prove useful ( Villalba-Galea et al . , 2009 ) . Improved understanding of this mechanism may lead to additional probes that accomplish what ArcLightning accomplishes: rapid kinetic responses in response to hyperpolarization that prevent the drifting baselines observed in response to repetitive pulses in similar probes ( Piao et al . , 2015 ) and a roughly linear voltage-sensitivity across the physiological voltage range that could allow for the visualization of both subthreshold and action potential activity . FRET-based voltage indicators , including VSFP , Mermaid , and Butterfly indicators have recently achieved many or all of these characteristics ( Knöpfel et al . , 2015 ) . Many of these probes have slightly larger fluorescence changes than ArcLightning and comparable kinetics , and the recently developed chimeric VSFP Butterflies appear particularly promising ( Mishina et al . , 2014 ) . However , ArcLightning and other monochromatic GEVIs have the distinct advantage of freeing other spectral channels for use in measuring other parameters , such as cell morphology or changes in calcium concentration . Additionally , as ArcLighting requires the recording of only one fluorophore’s emission , it is more compatible with simpler optical setups . In addition to the ability demonstrated here to accelerate the voltage-sensitive response of ArcLight , the previously reported I126F mutation ( Lacroix and Bezanilla 2012 ) may also be applicable to other GEVIs such as ASAP1 ( St-Pierre et al . , 2014 ) and the Butterfly indicators ( Akemann et al . , 2015 ) . Finally , the single-molecule and macroscopic fluorimetric techniques presented here will improve our future capabilities to mechanistically understand VSP-derived GEVIs and to engineer new and improved generations of these probes . ArcLight for Xenopus laevis expression was created in the lab by fusing Venus to Ci-VSP in the SP64T plasmid vector at the appropriate location and mutating to the super ecliptic pHluorin A227D to generate ArcLight or ArcLight' ( Jin et al . , 2012 ) . ArcLight' was identical to ArcLight but with two point mutations ( L64F and T65S ) which return the fluorophore to the wild-type GFP scaffold rather than the eGFP scaffold; thus , ArcLight' contains the ecliptic pHluroin A227D fluorophore rather than the super ecliptic pHluorin A227D fluorophore . The two constructs were shown to have nearly identical voltage-dependence ( Jin et al . , 2012 ) , and in our hands ArcLight' appeared to perform better in single-molecule experiments . ArcLight' GFP domains and eGFP were expressed in E . Coli in the pQE-32 plasmid vector using previously-described methods ( Negro et al . , 1997 ) . ArcLight for mammalian expression was a gift from Vincent Pieribone ( Addgene plasmid #36856 ) ( Jin et al . , 2012 ) . Mutations I126F and Q217R were generated in these constructs using site-directed mutagenesis by polymerase chain reaction and subsequently verified by sequencing . For oocyte expression , DNA was prepared using the NucleoSpin Plasmid kit ( Macherey-Nagel , Bethlehem , PA ) and linearized with NotI ( New England Biolabs , Ipswich , MA ) . Linearized cDNA was transcribed to RNA with the mMESSAGE mMACHINE Sp6 kit ( Life Technologies , Carlsbad , CA ) . Oocytes were injected with either 0 . 25 to 1 ng of RNA ( single-molecule recordings ) or 50 ng of RNA ( macroscopic recordings ) and incubated at 16°C in solution containing ( in mM ) 96 NaCl , 2 KCl , 1 . 8 CaCl2 , 1 MgCl2 , 10 HEPES , at pH 7 . 4 with 10 mg/L of gentamicin . Recordings were made 1-–4 days following injection . For mammalian cell expression , DNA was prepared using the NucleoBond Xtra Midi Plus kit ( Macherey-Nagel ) . DNA was then transfected into HEK293 cells that had been previously plated on glass coverslips at low density , using Lipofectamine LTX reagent ( Thermo Fisher Scientific , Waltham , MA ) . HEK cells were incubated at 37°C with 5% CO2 in DMEM medium with HEPES and no phenol red ( Thermo Fisher Scientific ) for three to four days prior to recording . Glass slides were prepared ahead of time by incubation in piranha solution ( 70% H2SO4 at 18M / 30% H2O2 at 30% w/w , both from Sigma-Aldrich , St . Louis , MO ) followed by copious rinsing with water and storage under water in a 50 mL tube until use . When ready to use , a slide was removed and dried under a stream of nitrogen , and a chamber cut from cured Sylgard 184 ( Dow Corning , Midland , MI ) was placed on the dry slide . A solution of 200 mM polyethylenimine ( Mw ≈ 750 , 000 , Sigma-Aldrich ) in SOS ( 96 mM NaCl , 2 mM KCl , 1 mM MgCl2 , 1 . 8 mM CaCl2 , 20 mM HEPES ) was placed into the chamber and left for 30 minutes , followed by extensive rinsing with fresh SOS . This recording chamber was filled with SOS for recording . To allow TIRF microscopy , oocytes were placed in a high osmolality ‘shrinking’ solution prior to recording and the vitelline membranes were mechanically removed ( Sonnleitner et al . , 2002 ) . Oocytes were then placed into the recording chamber and mounted on the microscope . ArcLight fluorescence from single molecules was recorded by an Evolve 128 EMCCD camera ( Photometrics , Tucson , Arizona ) attached to a home-built TIRF setup based on an Olympus IX71 inverted microscope ( Center Valley , Pennsylvania ) with a 60X/1 . 45 NA microscope objective ( Olympus ) . Excitation was provided by a 473 nm DPSS laser ( Shanghai Dream Lasers Technology Co . , Ltd . , Shanghai , China ) . Typical excitation intensities were around 150 W/cm2 . Fluorescence was observed through a T495lpxt dichroic and ET500lp emission filter ( Chroma Technology Corp . , Bellows Falls , VT ) . Voltage clamp was performed in a two-electrode configuration with a Warner Instruments OC-725A amplifier ( Hamden , Connecticut ) . Both the electrophysiological and optical equipment were controlled using in-house software . For single-molecule recordings , the recording chambers were cooled to approximately 14°C by cooling the microscope objective lens assembly with recirculating chilled water . The bacterially-expressed ArcLight' GFP domain and eGFP were incubated on PEGylated glass slides possessing a low density of Cu-NTA chelates ( Cu_01 , MicroSurfaces , Inc . , Englewood , NJ ) as described previously ( Holtz et al . , 2007 ) . Since the proteins contained poly-histidine tags , they efficiently bound the copper chelate while a subsequent rinse with fresh buffer removed excess proteins . As this point , the slides were imaged using TIRFM with the same optical configuration as described above . Simultaneous recordings of ArcLight gating currents and fluorescence responses were performed using the cut-open oocyte voltage-clamp technique ( Stefani and Bezanilla , 1998 ) in combination with a photodiode to measure temporal changes in fluorescence emission ( Cha and Bezanilla , 1998 ) . Gpatch , an in-house program , controlled an SB6711 digital signal processor-based board ( Innovative Integration , Simi Valley , CA ) with an A4D4 conversion board ( Innovative Integration , Simi Valley , CA ) . Oocytes were held under voltage-clamp with a Dagan CA-–1B amplifier ( Minneapolis , MN ) and filtered at 2–5 kHz depending on the sampling rate . ArcLight emission fluorescence was collected through an Olympus LUMPlan FL N 40X/0 . 8 NA water-immersion objective by a PIN-020A photodiode ( UDT Technologies , Torrance , CA ) , amplified by a patch clamp amplifier L/M-EPC-7 by LIST Medical Electronic ( Darmstadt , West Germany ) with a filter of 10 kHz , and then integrated over each sampling period using a home-built integrator circuit . ArcLight was excited via a ThorLabs LED controller triggering a 470 nm LED ( ThorLabs , Newton , New Jersey ) that was passed through a filter cube housing a 480/40 excitation filter , a 505 long-pass dichroic , and a 535/50 emission filter ( Chroma , Bellows Falls , VT ) . All recordings were performed at around 19°C , with an external solution containing ( in mM ) 120 N-methyl-D-glucamine/methanesulfonic acid ( NMG/MES ) , 10 HEPES , and 2 Ca ( OH ) 2 and an internal solution containing ( in mM ) 120 NMG-MES , 10 HEPES , and 2 EGTA . Both solutions were set to pH 7 . 5 . Current microelectrodes pulled on a Flaming/Brown micropipette puller ( Sutter Instruments , Novato , CA , model P-87 ) were filled with 3 M KCl and had a resistance of ~0 . 2–0 . 9 M . HEK cells were patch-clamped using an Axon Instruments Axopatch 200A amplifier ( Molecular Devices , Sunnyvale , CA ) . Signals were filtered ( Frequency Devices , Ottawa , Illinois , model 950L8L ) and digitized with an SBC-6711-A4D4 data acquisition board ( Innovative Integration ) . Patch pipettes with resistances of roughly 5 MΩ were pulled on a CO2 laser micropipette puller ( Sutter Instruments , model P-2000 ) . The bath temperature was raised and maintained using a home-built system with an Omega CNPt series temperature controller ( Stamford , CT ) . Data analysis was performed in MATLAB ( The MathWorks , Inc . , Natick , MA ) , as well as in-house software . Weighted time constants ( τw ) of gating current were taken using double exponential fits to the decay phase of the gating current . Kinetics of fluorescence traces were first taken from double exponential fits to the trace; when fit with two exponentials , fluorescence traces obtained from off pulses in oocytes only had one meaningful time constant , and thus were refit with one exponential . For clarity of comparison , all kinetics were reported as τ1 , which denotes the faster and larger amplitude component of the fluorescence traces . Q-V curves were calculated by normalizing the integral of the gating current following a sloped baseline subtraction to remove leak current . When holding at +40 mV the Q-V and F-V curves did not approach saturation in the hyperpolarized direction . Therefore , we first normalized the Q-V by the total gating charge moved in the same oocyte during a depolarizing pulse from a holding potential of -120 mV as these traces saturate on both ends and total gating charge is conserved ( Villalba-Galea et al . , 2008 ) . The normalization factor found in the Q-V was used to normalize the F-V when holding at +40 mV .	Nerve cells , or neurons , transmit information using changes in the voltage across their cell membranes . In the brain , these neurons work together in complex networks , and so understanding how the brain processes information will require neuroscientists to analyze voltage changes in many neurons at the same time . To achieve this , scientists have developed genetically-encoded voltage indicators ( or GEVIs ) . These commonly feature a fluorescent protein attached to a voltage-sensitive protein; when the voltage-sensitive protein moves in response to changes in electrical activity , the amount of light emitted by the fluorescent protein also changes . Treger , Priest and Bezanilla have now studied the characteristics of a popular GEVI called ArcLight by recording how fluorescence and voltage are related , both in single molecules and in groups of millions of molecules . This revealed that the fluorescence response of ArcLight does not occur instantly when a voltage change occurs . Instead the indicator fluoresces after a short delay . This delay corresponds with how quickly the voltage-sensitive protein responds . The fluorescence of a close relative of ArcLight also rapidly flickers , which deteriorates the signal quality . Using this knowledge Treger , Priest and Bezanilla engineered the voltage-sensitive protein of ArcLight to develop a new variant of the indicator , named ArcLightning . Tests revealed that ArcLightning responds much faster than ArcLight to voltage changes in neurons , although the flicker of the fluorescent protein likely remains . ArcLightning should prove to be a valuable tool for analyzing how neurons work together in living animals , but the flicker of the fluorescent protein suggests that there is further room for improvement . The rational design method used to develop ArcLightning could also be applied to improve other recently developed voltage indicators .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics", "neuroscience" ]	2015	Single-molecule fluorimetry and gating currents inspire an improved optical voltage indicator
Cells must appropriately sense and integrate multiple metabolic resources to commit to proliferation . Here , we report that S . cerevisiae cells regulate carbon and nitrogen metabolic homeostasis through tRNA U34-thiolation . Despite amino acid sufficiency , tRNA-thiolation deficient cells appear amino acid starved . In these cells , carbon flux towards nucleotide synthesis decreases , and trehalose synthesis increases , resulting in a starvation-like metabolic signature . Thiolation mutants have only minor translation defects . However , in these cells phosphate homeostasis genes are strongly down-regulated , resulting in an effectively phosphate-limited state . Reduced phosphate enforces a metabolic switch , where glucose-6-phosphate is routed towards storage carbohydrates . Notably , trehalose synthesis , which releases phosphate and thereby restores phosphate availability , is central to this metabolic rewiring . Thus , cells use thiolated tRNAs to perceive amino acid sufficiency , balance carbon and amino acid metabolic flux and grow optimally , by controlling phosphate availability . These results further biochemically explain how phosphate availability determines a switch to a ‘starvation-state’ . Cells utilize multiple mechanisms to sense available nutrients , and appropriately alter their internal metabolic state . Such nutrient-sensing systems assess internal resources , relay this information to interconnected biochemical networks , and control global responses that collectively reset the metabolic state of the cell , thereby determining eventual cell fate outcomes ( Jeong et al . , 2000; Förster et al . , 2003; Zaman et al . , 2008; Broach , 2012; Cai and Tu , 2012; Ljungdahl and Daignan-Fornier , 2012 ) . However , much remains unknown about how cells sense and integrate information from multiple nutrient inputs , to coordinately regulate the metabolic state of the cell and commit to different fates . In this context , the metabolic state of the cell is also closely coupled with mRNA translation . Protein synthesis is enormously energy consuming , and therefore must be carefully regulated in tune with nutrient availability ( Warner , 2001 ) . Generally , overall translational capacity and output increases during growth and proliferation ( Jorgensen et al . , 2004 ) , and decreases during nutrient limitation ( Wullschleger et al . , 2006 ) . Signalling processes that regulate translational outputs ( such as the TORC1 and PKA pathways ) are well studied ( Wullschleger et al . , 2006; Zaman et al . , 2008; Broach , 2012; González and Hall , 2017 ) . Notwithstanding this , little is known about how core components of the translation machinery might directly control metabolic outputs , and thus couple metabolic states with physiological cellular outcomes . tRNAs are core components of the translation machinery , and are extensively modified post-transcriptionally ( Björk et al . , 1987; Phizicky and Hopper , 2010 ) . Some tRNA modifications are required for tRNA folding , stability , or the accuracy and efficiency of translation ( Phizicky and Hopper , 2010 ) . However , the roles of many of these highly conserved modifications remain unclear . One such modification is a thiolation of uridine residue present at the wobble-anticodon ( U34 ) position of specifically glu- , gln- and lys- tRNAs ( s2U34 ) ( Gustilo et al . , 2008; Phizicky and Hopper , 2010 ) . In yeast , this is mediated by a group of six enzymes- Nfs1 , Tum1 , Uba4 , Urm1 , Ncs2 and Ncs6 , which are evolutionarily conserved ( Nakai et al . , 2008; Leidel et al . , 2009; Noma et al . , 2009 ) . These enzymes incorporate a thiol group derived directly from an amino acid ( cysteine ) , and replace the oxygen present at the 2-position of U34 with sulfur ( Schmitz et al . , 2008; Leidel et al . , 2009; Noma et al . , 2009 ) . Surprisingly , these thiolated tRNAs appear to have a relatively minor role in general translation , as seen in multiple studies ( Rezgui et al . , 2013; Zinshteyn and Gilbert , 2013; Klassen et al . , 2016; Chou et al . , 2017 ) with modest roles in enhancing the efficiency of wobble base codon-anticodon pairing ( Yarian et al . , 2002; Rezgui et al . , 2013 ) . Contrastingly , tRNA thiolation appears to directly alter cellular metabolism , but this connection has remained largely unexplored . The suggestive connections to metabolism come from disparate studies . The loss of tRNA thiolation results in hypersensitivity to oxidative agents , and the TORC1 inhibitor rapamycin ( Fichtner et al . , 2003; Goehring et al . , 2003a; Goehring et al . , 2003b; Laxman and Tu , 2011; Scheidt et al . , 2014 ) , suggesting a role for thiolated tRNAs in maintaining metabolic homeostasis . More pertinently , several studies have observed that a loss of this modification alters amino acid homeostasis , and the amino-acid starvation regulator Gcn4 is induced ( Laxman et al . , 2013; Zinshteyn and Gilbert , 2013; Nedialkova and Leidel , 2015 ) . Further , thiolated tRNAs are required to maintain metabolic cycles in yeast ( Laxman et al . , 2013 ) . Finally , the amounts of thiolated tRNAs reflect the intracellular availability of sulfur-containing amino acids ( cysteine and methionine ) , and couple the sensing of amino acid sufficiency with growth ( Laxman et al . , 2013 ) . These studies all hint that a core function of this tRNA modification may be to integrate the sensing of amino acid availability ( primarily methionine/cysteine ) , with the coordinate regulation of overall metabolic state , in order for the cell to appropriately commit to growth . Yet , how thiolated tRNAs regulate metabolism , and the extent to which this may control cellular outcomes remains entirely unaddressed . In this study , by directly analyzing different metabolic outputs , we identify the metabolic nodes that are altered in tRNA thiolation deficient cells . We find that tRNA thiolation regulates central carbon and nitrogen ( amino acid ) metabolic outputs , by controlling flux towards storage carbohydrates . In tRNA thiolation deficient cells , overall metabolic homeostasis is altered , with carbon flux diverted away from the pentose-phosphate pathway and nucleotide synthesis axis , and towards storage carbohydrates trehalose and glycogen . This thereby alters cellular commitments towards growth and cell cycle progression . Counter-intuitively , we discover that this metabolic-state switch in cells lacking tRNA thiolation is achieved by down-regulating a distant metabolic arm of phosphate homeostasis . We biochemically elucidate how regulating phosphate balance can couple amino acid and carbon utilization towards or away from nucleotide synthesis , and identify trehalose synthesis as a pivotal control point for this metabolic switch . Through these findings we show how tRNA thiol-modifications regulate overall metabolic homeostasis , integrating nutrient inputs to enable optimal growth . We further present a general biochemical explanation for how inorganic phosphate homeostasis regulates commitments to different arms of carbon and nitrogen metabolism , thereby determining how cells commit to a ‘growth’ or ‘starvation’ state . Earlier studies had observed an increased expression of amino acid biosynthetic genes , and an activation of the amino acid starvation responsive transcription factor Gcn4 , in cells lacking tRNA thiolation ( Laxman et al . , 2013; Zinshteyn and Gilbert , 2013; Nedialkova and Leidel , 2015 ) . These studies therefore suggested that tRNA thiolation-deficient cells were amino-acid starved . We investigated this surmise , by directly measuring free intracellular amino acids in wild-type ( WT ) and tRNA thiolation mutant cells , using two distinct thiolation pathway mutants ( uba4Δ and ncs2Δ ) . In our studies , we used a prototrophic yeast strain grown in synthetic minimal medium without supplemented amino acids , in order to minimize any confounding interpretations coming from supplemented amino acids/nucleobases in the medium . Using quantitative , targeted LC-MS/MS approaches , we compared relative amounts of amino acids in WT and thiolation mutants . Unexpectedly , we observed a substantial increase in the intracellular pools of free amino acids in thiolation mutants ( Figure 1A ) . This shows that the thiolation deficient cells are not amino acid starved , but instead accumulate amino acids . We next correlated these actual amino acid amounts with the abundance of Gcn4 . Gcn4 is the major amino acid starvation responsive transcription factor , and is induced upon amino acid starvation ( Hinnebusch , 1984; Hinnebusch , 2005 ) ( Figure 1B ) . We measured Gcn4 protein in WT and thiolation deficient cells , and contrarily observed increased Gcn4 protein in thiolation mutants ( Figure 1C ) . Further , GCN4 translation was correspondingly higher in thiolation mutants ( Figure 1—figure supplement 1A ) ( as also seen earlier in Zinshteyn and Gilbert , 2013; Nedialkova and Leidel , 2015 ) . This increased GCN4 translation in the thiolation mutants was also Gcn2- and eIF2α phosphorylation-dependent ( Figure 1—figure supplement 1B and C ) . These observations comparing actual amino acid amounts in cells with the activity of Gcn4 therefore present a striking paradox . As canonically understood , Gcn4 is induced upon amino acid starvation , while Gcn4 translation and protein decrease when intracellular amino acid amounts are restored ( Hinnebusch , 1984; Hinnebusch , 2005 ) . Contrastingly , in the results observed here , despite the high amino acid amounts present in the tRNA thiolation mutants , the Gcn2-Gcn4 pathway remains induced . We therefore concluded that the metabolic node regulated by tRNA thiolation , resulting in an apparent amino acid starvation signature , cannot be at the level of amino acid biosynthesis and availability . We therefore considered the possible metabolic outcomes of amino acid utilization , and hypothesized that an alteration in amino acid utilization could be a source of this metabolic rewiring . In particular , amino acids are the sole nitrogen donors for de novo nucleotide synthesis ( Figure 1D ) . Since amino acids accumulated in thiolation mutants , we explored the possibility that this was due to reduced de novo nucleotide synthesis . We first measured steady-state nucleotides in WT and thiolation mutants , and observed decreased steady-state levels of nucleotides ( NMPs , as well as ATP ) in thiolation mutants ( Figure 1—figure supplement 2A and B ) . To unambiguously determine if these decreased nucleotide amounts in the thiolation mutants were due to reduced nucleotide synthesis , we adopted a metabolic flux-based approach we had developed earlier ( Walvekar et al . , 2018b ) . In such an approach , 15N-labelled ammonium sulfate can be provided as a pulse to cells growing with ammonium sulfate as a sole nitrogen source , and label incorporation via glutamine and aspartate into newly formed nucleotides can be measured . Notably , the incorporation of 15N-label into nucleotides ( GMP , AMP and CMP ) decreased in thiolation mutants relative to WT cells ( Figure 1E ) , indicating reduced flux towards nucleotide synthesis . As an internal control , the 15N-label incorporation into amino acids ( aspartate and glutamine ) themselves were not affected in thiolation mutants ( Figure 1—figure supplement 2C ) , ruling out amino acid synthesis defects . These data therefore show that nitrogen incorporation from amino acids to nucleotides decrease in the thiolation mutants , resulting in decreased nucleotides . We further addressed this pharmacologically , using a purine-analog , 8-azaadenine , which acts as a pseudo-feedback inhibitor of nucleotide biosynthesis . Consistent with the decreased nucleotide levels observed , thiolation mutants exhibited increased sensitivity to 8-azaadenine ( Figure 1—figure supplement 2D ) . Collectively , these data show that the loss of tRNA thiolation decreases nucleotide biosynthesis , with a corresponding accumulation of amino acids . Notably , early studies have shown that nucleotide limitation can itself directly induce Gcn4 ( Rolfes and Hinnebusch , 1993 ) , suggesting that the increased Gcn4 amounts could be due to this . We also asked if the decreased nucleotide synthesis was due to reduced expression of nucleotide biosynthetic genes . We observed that the expression of candidate genes in this pathway were increased in thiolation mutants ( Figure 1—figure supplement 2E ) , diminishing the possibility of reduced nucleotide biosynthetic capacity as a reason for decreased nucleotides . Indeed , increased mRNA levels of nucleotide biosynthetic genes observed in thiolation mutants may be due to feedback upregulation in response to reduced nucleotides , which is also a well-established phenomenon ( Davis and Ares , 2006; Kwapisz et al . , 2008 ) . Collectively , despite increased intracellular pools of amino acids , tRNA thiolation deficient cells exhibit signatures of amino acid starvation , including decreased nucleotide biosynthesis . These data therefore suggest that the tRNA thiolation pathway is important for cells to appropriately balance amino acid utilization for nucleotide synthesis . Despite amino acids being non-limiting in thiolation deficient cells , flux towards nucleotide synthesis was decreased . This observation was in itself puzzling , and the reason was not obvious . We therefore asked if carbon metabolism was instead rewired in the thiolation mutants . Our reasoning was as follows: while amino acids are the sole nitrogen donors for nucleotide synthesis , the carbon backbone for nucleotides is derived from central carbon metabolism ( Figure 2A ) . We reasoned that since the decreased nucleotide synthesis was not due to amino acid limitation , this could instead be due to a metabolic shift where carbon flux is routed away from nucleotide synthesis . Carbon derived from glucose is converted to glucose-6-phosphate , and then is typically directed towards glycolysis and the pentose phosphate pathway ( PPP ) . The PPP , along with one-carbon folate metabolism provides the necessary carbon precursors ( including ribose sugars ) for nucleotide synthesis ( Figure 2A ) ( Boyle , 2005; Hosios and Vander Heiden , 2018 ) . However , if glucose-6-phosphate is instead diverted towards storage carbohydrates trehalose and glycogen ( Figure 2A ) , this can result in reduced flux into the arm of glucose metabolism leading towards nucleotide synthesis . To assess if such a metabolic rewiring might happen in tRNA thiolation mutants , we pulsed [U-13C6]-labelled glucose to growing WT or thiolation deficient cells , and measured label incorporation into nucleotides as the end-point readout . Here , labelled carbons will only be present in newly synthesized nucleotides , and the label can only come from the pulsed labelled glucose , through the PPP and one-carbon metabolic pathways ( Figure 2A ) . We observed significantly decreased carbon label incorporation towards new nucleotide synthesis , as shown for GMP and AMP , as well as label incorporation into ADP and ATP in the thiolation mutants ( Figure 2B , Figure 2C , and Figure 2—figure supplement 1A ) . This result is also consistent with the decreased nucleotide synthesis based on amino acid derived nitrogen assimilation , observed earlier ( Figure 1E ) . Experimental note: Given how rapidly the pulsed carbon label saturates in glucose medium for early glycolytic and PPP intermediates ( Heerden et al . , 2014 ) , using nucleotide synthesis as a read-out of this arm of carbon metabolism is a more reliable , quantitative indicator of carbon flux . Here , we reliably obtained nucleotide information ( ensuring that label incorporation was not saturated ) by pulsing cells with 13C-glucose and quenching/processing metabolites within 5 min . By this time , the 13C-glucose label incorporation into glycolytic and pentose phosphate pathway intermediates was already near-saturation , and hence differences in these metabolites could not be seen ( Figure 2—figure supplement 1B ) . We therefore carried out experiments where cells were quenched and metabolites extracted within 2 min of 13C-glucose addition . Here , a significant decrease in 13C-label incorporation into newly synthesized ribose-5-phosphate ( a late PPP metabolite ) was observed ( Figure 2D ) . This is entirely consistent with results obtained with nucleotides in Figure 2B and C . Summarizing , these data show that carbon ( glucose ) flux towards nucleotide synthesis was reduced in the thiolation mutants . The end-point readouts of the alternative metabolic arm where glucose-6-phosphate is diverted away from the PPP are trehalose and glycogen ( as shown in Figure 2A ) . To examine if this arm of metabolism is altered in the thiolation mutants , thereby resulting in the decreased carbon flux towards nucleotides ( shown earlier ) , we first estimated the steady-state amounts of trehalose and glycogen with a biochemical assay . Here , we observed a marked increase in these metabolites in the thiolation mutants ( Figure 2E and Figure 2—figure supplement 1C ) . Subsequently , we directly estimated flux towards trehalose synthesis . For this , we used a similar experiment as described earlier , where [U-13C6]-labelled glucose was pulsed and newly formed labelled trehalose was measured in a metabolic flux experiment , to test if this arm of the pathway is altered . Here , we observed a strong increase in the synthesis of trehalose ( as measured by label incorporation into M + 6 and M + 12 mass isotopomers of trehalose ) in the tRNA thiolation mutants ( Figure 2F ) . Collectively , these results show that cells lacking tRNA thiolation rewire metabolic outputs towards the synthesis of storage carbohydrates , and away from nucleotide synthesis , suggesting a ‘starvation-like’ metabolic state . Notably , this occurs despite the absence of glucose ( carbon ) or amino acid ( nitrogen ) limitation in the thiolation mutants . Earlier studies have noted a coupling of tRNA thiolation with methionine/sulfur amino acid availability ( Laxman et al . , 2013 ) . Here , the amount of thiolated tRNAs increase with increasing sulfur amino acids and vice versa . Further , in thiolation mutants , proteins involved in methionine salvage and biosynthesis increase even when methionine was in excess , suggesting a mis-sensing of this amino acid in these mutants . Finally , Uba4 is itself regulated by methionine amounts , decreasing sharply with slight methionine-limitation ( Laxman et al . , 2013 ) . Separately , studies ( Tu et al . , 2005; Laxman et al . , 2013 ) show that the amounts of thiolated tRNAs are highest in cells entering a ‘growth state’ ( with high ribosomal biosynthesis ) . In this context , we also recently defined a methionine induced anabolic program , where abundant methionine triggers increased carbon flux ( particularly PPP flux ) towards nucleotide synthesis ( Walvekar et al . , 2018b ) . Given our observations here in standard medium , where thiolation mutants showed decreased carbon flux from glucose towards nucleotides , we further investigated this coupling of tRNA thiolation with methionine availability . The prediction is that if tRNA thiolation enables cells to fully respond to abundant methionine , then thiolation mutants will exhibit a reduced methionine response . As a result , when methionine is supplemented , thiolation mutants will show reduced carbon incorporation into nucleotides ( i . e . nucleotide synthesis ) , compared to WT cells . To test this , WT or thiolation mutant ( uba4Δ ) cells grown in standard glucose minimal medium were supplemented with 2 mM methionine , and then pulsed with 13C-glucose ( as described earlier ) for 15 min . Subsequently , we compared carbon-label incorporation into newly synthesized nucleotides . Here , we expectedly observed a sharp increase in carbon-label incorporation into nucleotides in WT cells supplemented with methionine ( Figure 2G , and Figure 2—figure supplement 1D ) . Notably , while the extent of this methionine-dependent induction of carbon flux into nucleotides was significantly reduced , it was not completely abolished in the thiolation mutant ( uba4Δ ) , indicating the involvement of additional as yet unknown pathways ( Figure 2G , and Figure 2—figure supplement 1D ) . Together , these data reiterate that tRNA thiolation allows cells to fully integrate amino acid ( methionine ) sensing , with a routing of carbon ( glucose ) flux towards new nucleotide synthesis , collectively indicative of a growth state . Furthermore , in a converse experiment , we subjected WT cells to brief inorganic sulfur/sulfur amino acid limitation , to determine the metabolic signature of cells in this condition ( experimental design shown in Figure 2—figure supplement 2A ) . Cells subjected to a brief shift to sulfur starved medium showed a sharp decrease in sulfur amino acid-related metabolites ( Figure 2—figure supplement 2B ) . We measured the steady-state amounts of other amino acids , nucleotides ( AMP ) , trehalose and the level of Gcn4 translation . We observed increased steady-state amino acid amounts ( Figure 2—figure supplement 2C ) , reduced nucleotides ( Figure 2—figure supplement 2D ) , increased trehalose levels ( Figure 2—figure supplement 2E ) , and strong Gcn4 induction ( Figure 2—figure supplement 2F ) upon sulfur limitation . These data strikingly resembled the metabolic state of the thiolation mutants . Thus , WT cells subject to sulfur amino acid limitation phenocopied the metabolic signature of tRNA thiolation mutants . Dissecting physiological roles of such a fine-tuning of metabolic outputs can be challenging , and this has been the case for tRNA thiolation mutants . However , a simple yeast system , termed ‘yeast metabolic cycles’ or metabolic oscillations , has been effective in identifying regulators that couple metabolism with cell growth/cell-division ( Tu et al . , 2005; Slavov and Botstein , 2011 ) . In continuous , glucose-limited cultures , yeast cells exhibit robust metabolic oscillations , which are tightly coupled to the cell division cycle , and where DNA replication and cell division are restricted to a single temporal phase ( Tu et al . , 2005; Chen et al . , 2007 ) . In an earlier study we had observed that tRNA thiolation mutants exhibit abnormal metabolic cycles ( Laxman et al . , 2013 ) . This was reminiscent of phenotypes exhibited by mutants of cell division cycle regulators ( Chen et al . , 2007 ) . We therefore asked if tRNA thiolation coupled metabolic and cell division cycles . To test this , we sampled the cells at regular intervals of time during the metabolic cycles , and determined their budding index . While WT cells showed synchronized cell cycle progression , tRNA thiolation mutants showed asynchronous cell division ( Figure 3A ) , suggesting a de-coupling of metabolic and cell division cycles . Given our earlier data showing a metabolic rewiring away from nucleotide synthesis in thiolation mutants , we hypothesized that tRNA thiolation controlled normal cell cycle progression by regulating the balance between nucleotide synthesis , and storage carbohydrate synthesis . To test this directly , we arrested cells in G1-phase using alpha factor , synchronously released them into the cell cycle by washing away the alpha factor , and monitored cell cycle progression by flow cytometry ( Figure 3B ) . 30 min post-release from G1-arrest , we observed delayed cell cycle progression and accumulation of cells in the S-phase in tRNA thiolation deficient cells . Further , using time lapse live-cell microscopy we found that the duration of the S-G2/M phase was longer in thiolation mutants ( Figure 3C and Figure 3D ) . Since nucleotides are required for DNA replication during the S-phase of the cell cycle , we reasoned that that this S-phase delay is due to decreased flux towards nucleotide synthesis in thiolation mutants ( shown earlier ) . To investigate this , we examined the sensitivity of WT and thiolation deficient cells to hydroxyurea ( HU ) , which inhibits the ribonucleotide reductase ( RNR ) enzyme and arrests cells in the S-phase . Thiolation mutants exhibited increased sensitivity to HU ( Figure 3E and Figure 3—figure supplement 1A ) . As a control , to rule out any abnormal activation of DNA damage readouts in the thiolation mutants as a cause for this phenotype , we also examined activity of the Rad53 checkpoint pathway . Indeed , the observed HU sensitivity was not due to any defect in the activation of the Rad53 checkpoint pathway in thiolation mutants ( Figure 3—figure supplement 1B ) . Summarizing , these results show that tRNA thiolation-mediated regulation of metabolic homeostasis , leading towards regulated nucleotide synthesis , is required for appropriately coupling metabolic state with normal cell cycle progression . Thus far , it remains unclear why the loss of tRNA thiolation results in this distinct metabolic switch , where carbon and amino acid flux is diverted away from nucleotide synthesis and into storage carbohydrates . This therefore suggests a deeper , non-intuitive regulatory check-point underpinning the overall metabolic rewiring towards a ‘starvation-like’ state in tRNA thiolation mutants . In order to identify what this controlling bottleneck might be , we identified transcriptional and translational changes in thiolation mutants relative to WT by performing RNA-seq and Ribo-seq ( Ingolia et al . , 2009 ) based on methods described earlier ( Weinberg et al . , 2016; McGlincy and Ingolia , 2017 ) . Notably , as detailed in the Materials and methods section , while collecting cells for RNA and Ribo-seq , we avoided the use of cycloheximide , as described earlier ( Weinberg et al . , 2016 ) , to minimize biases in our interpretations of ribosome profiling datasets due to the use of this translational inhibitor as described elsewhere ( Hussmann et al . , 2015 ) . Instead , cells were rapidly harvested by filtration and lysed as described . Ribosome profiling datasets were generated for WT cells as well as two distinct thiolation pathway mutants ( uba4Δ and ncs2Δ ) , in biological triplicates . The correlations among the biological replicates of ribosome footprint reads , and also among RNA-seq reads were all excellent ( R > 0 . 97 ) , as shown in Figure 4—figure supplement 1A ) . Further , figures ( Figure 4—figure supplement 1B; Figure 4—figure supplement 1C ) show transcript and ribosome footprint read correlations ( R2 in all comparisons > 0 . 8 ) , as well as read-length distributions . Using these datasets , we compared global gene expression , as well as ribosome footprints of WT cells with the uba4Δ and ncs2Δ thiolation mutants ( Figure 4A ) . Notably , comparing WT cells with the thiolation mutants ( uba4Δ and ncs2Δ ) , we surprisingly found exceptional correlation for transcripts , as well as ribosome footprints ( R2 >0 . 97 , and p<=2 . 2×10−16 for all datasets ) ( Figure 4A and B ) . These data surprisingly revealed that there are very little gene expression or translation changes observed in the thiolation mutants ( complete data in Supplementary file 4 ) . Fewer than ~30 genes were up or downregulated at a two-fold change cutoff ( arbitrarily used to illustrate the point ) , compared to WT cells ( Figure 4A and B ) , with a false discovery rate ( FDR ) of 0 . 05 . Furthermore , we observe only modest increases in ribosome-densities at codons recognized by thiolated tRNAs – AAA , CAA and GAA in the uba4Δ and ncs2Δ cells ( Figure 4—figure supplement 2A ) . Collectively , these extensive analysis show that the loss of tRNA thiolation has minimal effects on translational outputs in vivo , and any of the changes observed in the translation rates largely corresponded to changes at the transcriptional level . Given the lack of large-scale changes at the transcriptional and translational levels , but robust changes in cellular metabolism in the thiolation mutants , we focused our analysis on changes in expression levels of genes involved in metabolic pathways . We first examined several general amino acid control ( GAAC ) response genes , including the Gcn4 targets in amino acid biosynthetic pathways . Expectedly , we found these to be transcriptionally upregulated in the thiolation deficient cells ( Figure 4C , and Figure 4—figure supplement 3C ) . Additionally , as expected , we observed increase in GCN4 translation itself in the thiolation mutants , with no change in GCN4 transcripts ( Figure 4—figure supplement 2B and C ) . These data collectively corroborate our earlier data from Figure 1 , and agrees with previous reports ( Zinshteyn and Gilbert , 2013; Nedialkova and Leidel , 2015 ) . Also consistent with our earlier data , most nucleotide biosynthesis genes showed an increase in mRNA and ribosome-footprint abundances in the thiolation deficient cells ( Figure 4—figure supplement 3A; Figure 4—figure supplement 3C ) . In these datasets , there were also no obvious changes in central carbon metabolism genes in thiolation mutants . Notably , genes related to either trehalose/glycogen biosynthesis , or the PPP , showed negligible changes in either mRNA or ribosome-footprint abundances in the thiolation mutants ( Figure 4D , and Figure 4—figure supplement 3C ) . This further reiterates that the rewiring observed in thiolation deficient cells was largely driven by metabolic flux . Finally , we also found small decreases in the transcription and translation rates of all large and small subunit genes of ribosomes in uba4Δ and ncs2Δ cells at the translational level ( Figure 4—figure supplement 3B and C ) , consistent with earlier observations ( Laxman et al . , 2013; Nedialkova and Leidel , 2015 ) . However , in the course of this extensive functional analysis , we observed that an unusual group of ~20 genes were strongly downregulated in the thiolation mutants , both at the transcript and ribosome-footprint levels ( Figure 4E ) . Although these do not obviously group into a single category based on gene-ontology ( GO ) , we noted that these were functionally related to an important metabolic node . These genes are all part of the PHO regulon , which regulates phosphate homeostasis in cells ( Ljungdahl and Daignan-Fornier , 2012; Secco et al . , 2012 ) . This downregulation of these PHO-related genes was exceptionally significant ( p<10−7 ) , compared to other genes across the genome , both at the level of transcript abundances and ribosome-footprints ( Figure 4E , Figure 4F , and Figure 4—figure supplement 3C ) . Also notably , the only unaltered gene transcripts/ribosome footprints in the PHO regulon in the thiolation mutants , viz . PHO2 , PHO4 , PHO80 , PHO85 , PHO87 , PHO90 and PHO91 , are transcription factors , cyclins/cyclin-dependent kinases or low affinity phosphate transporters that are not transcriptionally/translationally regulated , but are regulated at the level of their activity ( Lemire et al . , 1985; Toh-e and Shimauchi , 1986; Madden et al . , 1988; Yoshida et al . , 1989; Madden et al . , 1990; Schneider et al . , 1994; Ogawa et al . , 1995; Lenburg and O'Shea , 1996; Auesukaree et al . , 2003 ) . Finally , we also observed that some genes related to phospholipid metabolism were downregulated in the thiolation mutants ( Figure 4—figure supplement 3C ) . Collectively , these data unexpectedly revealed a strong downregulation of genes related to phosphate homeostasis in the tRNA thiolation mutants . Inorganic phosphate ( Pi ) homeostasis is complex , but critical for overall nutrient homeostasis ( Ljungdahl and Daignan-Fornier , 2012; Secco et al . , 2012 ) . The PHO regulon comprises of several genes that respond to phosphate starvation , and maintains internal phosphate levels by balancing transport of Pi from the external environment , from within vacuolar stores , and the nucleus ( Figure 5A ) ( Ljungdahl and Daignan-Fornier , 2012; Secco et al . , 2012 ) . Extensive studies have defined global cellular responses to phosphate limitation ( Ogawa et al . , 2000; Wykoff and O'Shea , 2001; Boer et al . , 2003; Boer et al . , 2010; Saldanha et al . , 2004; Gresham et al . , 2011; Levy et al . , 2011; Choi et al . , 2017; Gurvich et al . , 2017 ) . In general , the PHO response is very sensitive to phosphate limitation , and is induced rapidly to restore internal phosphate levels . Extended phosphate starvation switches cells to an overall metabolically starved state . Our observed reduction in PHO-related transcripts and ribosome-footprints in the tRNA thiolation mutants was striking . We therefore first biochemically validated our results from the ribosome-profiling data , to more systematically investigate the extent of PHO downregulation . For this , we first measured protein amounts of Pho12 and Pho84 ( two arbitrarily selected PHO genes which are downregulated in the thiolation mutants , as shown in Figure 4E ) , in WT cells and thiolation mutants grown in the same conditions used for ribosome profiling analysis . Amounts of these proteins were substantially reduced in uba4Δ and ncs2Δ cells ( Figure 5B ) . We further compared Pho12 and Pho84 protein levels in WT cells and thiolation mutants under conditions of phosphate-limitation . Interestingly , we observed that even under these PHO inducible conditions , thiolation mutants exhibited substantially reduced levels of these PHO proteins compared to WT cells ( Figure 5C and Figure 5—figure supplement 1 ) , although the PHO response was induced . This suggested a constitutive dampening ( but not shut-down and absence ) of the PHO response in the thiolation mutants . In order to more quantitatively estimate this dampening of the PHO response in the thiolation mutants , we utilized a robust assay to measure acid phosphatase activity . This represents the enzymatic activity of the PHO acid phosphatases , including Pho5 , Pho11 , Pho12 and the more constitutively expressed Pho3 . Using this assay , we observed significantly reduced PHO activity in thiolation mutants ( Figure 5D ) . Notably , this also quantitatively revealed that the thiolation mutants are unlike cells lacking Pho4 ( PHO induction absent ) , or lacking Pho80 ( constitutively extremely high PHO induction ) . These data reveal that the thiolation mutants are effectively in a constitutively phosphate-limited state , due to a dampened PHO response . These cells will therefore have a constitutively altered phosphate homeostasis , with reduced phosphate availability . We therefore asked if phosphate starvation in WT cells itself resembles the metabolic hallmarks of the thiolation mutants . We first biochemically estimated amounts of trehalose in WT cells with or without phosphate starvation , and found a robust increase in trehalose upon phosphate starvation ( Figure 6A ) , much like the thiolation mutants . We next measured Gcn4 ( protein ) in WT cells , with or without phosphate starvation . Here , we observed a strong induction in Gcn4 protein upon phosphate starvation ( Figure 6B ) . Further , like the Gcn4 response in the thiolation mutants , the Gcn4 induction in phosphate-starved WT cells was Gcn2-dependent ( Figure 6—figure supplement 1 ) . In addition to these data shown here in WT cells with phosphate starvation , earlier studies in response to phosphate limitation , have also observed high amino acid and low nucleotide levels under these conditions ( Boer et al . , 2010; Klosinska et al . , 2011 ) . These data are strikingly consistent with the metabolic profile observed in thiolation mutants , and suggests that the induction of Gcn4 observed in response to phosphate limitation might be due to reduced nucleotide levels . Notably , a metabolic signature of phosphate starvation is a depletion in ATP levels ( Boer et al . , 2010 ) . We have already demonstrated not just reduced ATP levels in the thiolation mutants , but carbon flux through nucleotide synthesis , including ATP synthesis was also reduced in thiolation mutants ( shown earlier in Figure 2 ) . Finally , prior studies with phosphate limitation have shown that ribosomal genes are slightly repressed with phosphate limitation ( Saldanha et al . , 2004 ) . In the thiolation mutants , where phosphate homeostasis is affected due to the dampened PHO response , we also observed a small decrease in ribosomal genes ( as shown earlier in Figure 4—figure supplement 3B; Figure 4—figure supplement 3C ) . Thus , these data from WT cells starved of phosphate strikingly phenocopied the tRNA thiolation mutants . Finally , we used the acid phosphatase activity assay ( described earlier ) as a robust read-out for the extent of the PHO response , to compare activities in WT cells , cells lacking tRNA thiolation , and cells lacking Gcn4 ( individually or in combination ) . This was done to test if the Gcn4 induction was upstream or downstream of the PHO response . Notably , the loss of GCN4 in cells lacking thiolation ( uba4Δ gcn4Δ ) also resulted in significantly decreased acid phosphatase activity ( Figure 6C ) , similar to the thiolation mutants . However , the loss of GCN4 alone had no effect on acid phosphatase activity . This suggests that the Gcn4 induction is downstream of the dampened PHO response in the thiolation mutants . Collectively , these results strongly suggest that effective phosphate limitation is responsible for the metabolic state switch exhibited by the thiolation deficient cells . This observed downregulation of phosphate metabolism in the thiolation deficient cells is striking . Nonetheless , it is not immediately obvious biochemically how this relates to re-routing carbon towards storage carbohydrates , and decoupling amino acid metabolism from nucleotide synthesis . Perplexingly , in our transcript and translation analysis , no other metabolic arms were similarly decreased in the thiolation deficient cells , and only the amino acid biosynthesis arm ( dependent on Gcn4 ) increases , which we have addressed earlier . Notably , while earlier studies have hinted that phosphate limitation results in a shift towards storage carbohydrates ( Lillie and Pringle , 1980; Boer et al . , 2003; Boer et al . , 2010 ) , this more extensive metabolic rewiring has not been carefully analyzed , and a biochemical explanation for this is missing . We wondered if some overlooked aspect within this biochemical process could explain why a perturbation in phosphate homeostasis connects to the synthesis of storage carbohydrates trehalose and glycogen , as is also seen in tRNA thiolation mutants . To address this , we carefully examined all the metabolic nodes altered in the tRNA thiolation mutants , evaluating necessary co-factors and products of each pathway , and looking for possible connections to phosphate . Here , we noted an apparently minor , largely ignored output in the arm of carbon metabolism , where glucose-6-phosphate is routed towards trehalose synthesis . The first step of trehalose synthesis is the formation trehalose-6-phosphate ( T-6-P ) , carried out by trehalose-6-phosphate synthase ( Tps1 ) . This is followed by the dephosphorylation of T-6-P by Tps2 ( De Virgilio et al . , 1993 ) , forming trehalose ( Figure 7A ) . We noted that this Tps2-dependent second step is accompanied by the release of free , inorganic phosphate ( Pi ) ( Figure 7A ) . Canonically , these two steps are viewed as an apparently futile trehalose cycle during glycolysis , regenerating glucose , in order to maintain balanced glycolytic flux ( Heerden et al . , 2014; van Heerden et al . , 2014 ) . However , we reasoned that if the availability of inorganic phosphate is limiting , a shift to trehalose synthesis can be a way by which cells can liberate Pi , and restore phosphate levels . For this to be generally true , the prediction is that during phosphate starvation , WT cells must accumulate trehalose in order to recover phosphate . As shown earlier , this is exactly what is observed in WT cells limited for phosphate ( Figure 6A ) , and in the tRNA thiolation mutants ( Figure 2E; Figure 2F ) which are effectively phosphate limited due to a reduction in the PHO genes . Given the central role of phosphate , cells utilize all means possible to restore internal phosphate ( Ljungdahl and Daignan-Fornier , 2012 ) . Therefore it is experimentally challenging to study changes in phosphate homeostasis in cells . However , we directly tested the hypothesis that trehalose synthesis is a direct way for cells to restore internal phosphate in tRNA thiolation mutants , by utilizing cells lacking TPS2 . These cells cannot complete trehalose synthesis , and importantly cannot release phosphate ( Figure 7A ) . We first measured the intracellular Pi levels in WT cells , thiolation mutants ( uba4Δ ) , cells lacking Tps2p ( tps2Δ ) , and cells lacking tRNA thiolation as well as Tps2p ( uba4Δ tps2Δ ) . pho85Δ cells were used as a control , since they exhibit intrinsically higher intracellular Pi levels ( Liu et al . , 2017 ) . We first observed that in cells lacking TPS2 ( tps2Δ ) intracellular Pi levels were lower ( ~75–80% ) relative to WT cells ( Figure 7B and Figure 7—figure supplement 1A ) . This suggests that while other pathways ( phosphate uptake , glycerol production and vacuolar phosphate export ) remain relevant , Tps2p-mediated Pi release by dephosphorylation of trehalose-6-P is itself important for maintaining internal phosphate levels . Importantly , uba4Δ cells had only slightly reduced intracellular Pi levels ( ~90% ) relative to WT cells ( Figure 7B and Figure 7—figure supplement 1A ) . This is consistent with the prediction that due to reduced PHO expression in these cells , phosphate homeostasis is altered , but they can compensate for phosphate availability through increased trehalose synthesis . Contrastingly , the cells lacking both thiolation and Tps2 ( uba4Δ tps2Δ ) showed a dramatic reduction in Pi levels ( ~65% ) , compared to either of their single mutants . This striking reduction in Pi levels in these cells is consistent with the predicted outcome , where an inability to release phosphate from trehalose ( tps2Δ ) is also coupled with reduced expression of phosphate assimilation genes ( uba4Δ ) . Next , we tested possible genetic interactions between tps2Δ and thiolation mutants ( uba4Δ and ncs2Δ ) by assessing relative growth . In our genetic background , tps2Δ cells exhibit slightly slower growth at 37°C . Notably , in cells lacking both TPS2 and tRNA thiolation ( tps2Δ uba4Δ or tps2Δ ncs2Δ ) , we observed a strong synthetic growth defect , in conditions of low phosphate as well as normal phosphate ( Figure 7C and Figure 7—figure supplement 1B ) . This is entirely consistent with the proposed role of Tps2p in maintaining phosphate balance in thiolation mutants . Finally , if we completely imbalance phosphate homeostasis in cells , using cells lacking PHO80 , individual mutants of either pho80Δ or thiolation deficient cells show minimal growth defects , but double mutants ( pho80Δ uba4Δ or pho80Δ ncs2Δ ) show a severe synthetic growth defect ( Figure 7D ) . Collectively , our results suggest that altering phosphate homeostasis by decreasing PHO activity regulates overall carbon and nitrogen flow . Cells can therefore deal with decreased phosphate availability by diverting carbon flux away from nucleotide biosynthesis , and towards Tps2-dependent trehalose synthesis and Pi release . This restores phosphate , while concurrently resulting in an accumulation of amino acids , and a reduction in nucleotide synthesis . In this study , we highlight two related findings- a direct role for a component of translational machinery , U34 thiolated tRNAs , in regulating cellular metabolism by controlling phosphate homeostasis; and a biochemical rationale for how phosphate availability regulates flux through carbon and nitrogen metabolism . An integrative model emerges from our studies , explaining how high amounts of thiolated tRNAs reflect a ‘growth state’ , while reduced tRNA thiolation reflect a ‘starvation state’ ( Figure 8 ) . Cells can use tRNA thiolation to sense overall nutrient sufficiency and appropriately modulate metabolic outputs , using phosphate homeostasis as the metabolic control point ( Figure 8 ) . In this model , tRNAs are thiolated in tune with methionine and cysteine availability , as has been demonstrated earlier ( Laxman et al . , 2013 ) . Separately , the presence of these sulfur amino acids themselves reflect an overall amino acid sufficiency state , leading to an anabolic program capable of sustaining cell growth and proliferation ( Walvekar et al . , 2018b ) . Methionine up-regulates both amino acid synthesis and carbon flux leading towards nucleotide synthesis , and therefore growth ( Walvekar et al . , 2018b ) . Collectively therefore , in conditions of methionine sufficiency ( and therefore amino acid sufficiency ) , where tRNAs are maximally thiolated , cells direct carbon flux towards nucleotide biosynthesis , coupled with amino acid utilization ( as shown in Figures 1 and 2 ) . Accordingly , at this level of metabolic coupling , thiolated tRNAs sense amino acids , and ensure appropriate nucleotide levels for growth and cell cycle progression ( as shown in Figures 2 and 3 ) . On the other hand , the loss of thiolated tRNAs results in an inability of cells to fully sense and integrate these nutrient cues , rewiring carbon and nitrogen flux away from nucleotide synthesis and instead towards storage carbohydrates . This switches cells to a ‘starvation-like state’ . This metabolic rewiring mediated by tRNA thiolation is achieved not by directly regulating enzymes in these arms of carbon metabolism , but instead by down-regulating the PHO regulon . This constricts intracellular phosphate availability ( as shown in Figures 4 and 5 ) . A result of this dampened PHO response , and constriction in available free phosphate , is that in order to restore phosphate , cells divert glucose flux towards Tps2-mediated trehalose synthesis , which concurrently releases Pi ( as shown in Figures 6 and 7 ) . Thus , while the trehalose shunt and phosphate recycling restores phosphate levels , this is at the cost of decreased nucleotide biosynthesis , and delayed cell cycle progression . Effectively , the loss of tRNA thiolation rewires cells to a starved metabolic state . Collectively , tRNA thiolation appropriately regulates metabolic outputs by controlling phosphate homeostasis , thereby enabling cells to commit to growth ( Figure 8 ) . Intriguingly , this correlation of tRNA thiolation with growth and rewired metabolism is emerging in cancer development ( McMahon and Ruggero , 2018; Rapino et al . , 2018 ) , suggesting possibly conserved metabolic roles for these modified tRNAs . Insight into a deeper coupling of discrete metabolic arms emerges from this study , suggesting how cells can fully integrate carbon , nitrogen , sulfur and phosphate inputs for optimal growth . Earlier studies have noted a strong correlation of phosphate starvation with decreased nucleotide synthesis , and ATP availability ( Boer et al . , 2010; Klosinska et al . , 2011 ) . Reduced ATP synthesis is a hallmark of phosphate starvation ( Boer et al . , 2010 ) , along with increased trehalose synthesis ( discussed in a subsequent paragraph ) . Through this metabolic rewiring , reduced flux through the PPP and one-carbon/folate cycle can be inferred . A reduction in these metabolic pathways will reduce not just nucleotide synthesis , but also the production of NADPH , and cellular reductive biosynthetic capacity ( Fan et al . , 2014; Hosios and Vander Heiden , 2018 ) . Notably , the reductive costs in terms of NADPH utilized to assimilate sulfates into sulfur amino acids ( methionine and cysteine ) , and their subsequent metabolites , are themselves extremely high ( Thomas and Surdin-Kerjan , 1997; Kaleta et al . , 2013; Walvekar et al . , 2018b ) . Indeed , this coupling of NADPH production and methionine availability is also observed in the converse direction ( as noted earlier in the text ) , where the presence of methionine increases PPP flux ( Walvekar et al . , 2018b ) , and mutants in the PPP pathway are methionine auxotrophs ( Thomas et al . , 1991 ) . Therefore , an amplified regulatory response in relation to tRNA thiolation can be imagined . In such a response , in the presence of methionine , there is an increase in PPP and one-carbon flux and nucleotide synthesis , which is amplified by the maximally thiolated tRNAs by ensuring sufficient phosphate availability and an intact PHO response . Conversely , reduced sulfate assimilation in turn could limit thiolation , and thereby amplify the regulatory response , again using phosphate homeostasis as a control point . Indeed , the metabolic state of the tRNA thiolation mutants ( and the phenotypes associated ) are consistent with what is expected in a cell with reduced NADPH availability and reduced reductive biosynthetic capacity . Our data suggest a subtler , more integrative role for tRNA thiolation in optimally sensing overall nutrient sufficiency , and modulating overall metabolic responses leading to a growth-sustaining state . More generally , our findings identify a biochemical reaction , the trehalose shunt , that explains how phosphate homeostasis determines the extent of carbon and nitrogen flux towards nucleotide synthesis . While it is textbook knowledge that inorganic phosphate is important for glucose homeostasis ( Mason et al . , 1981; Boyle , 2005; Heerden et al . , 2014; van Heerden et al . , 2014 ) , the biochemical connection of phosphate balance to carbon and nitrogen flux remains poorly explained . Studies have observed that trehalose increases upon phosphate starvation ( Lillie and Pringle , 1980; Klosinska et al . , 2011 ) , and TPS2 is upregulated ( Ogawa et al . , 2000 ) . In these conditions , central carbon metabolism is down , and phosphate limitation is a ‘general starvation’ cue ( Brauer et al . , 2008; Boer et al . , 2010; Gurvich et al . , 2017 ) . Why this occurs has not been immediately apparent . Here , identifying the trehalose shunt as a way to restore phosphate balance , explains these observations . These data also explain earlier observations from pathogenic fungi , showing that that trehalose synthesis determines flux through the pentose phosphate pathway , and nitrogen metabolism ( Wilson et al . , 2007 ) . Additionally , phosphate starvation results in better cell survival in limited nutrient conditions , and also promotes efficient recovery when nutrients become available ( Gurvich et al . , 2017 ) . Since trehalose accumulation and utilization are respectively tightly coupled with exit from and re-entry into the cell division cycle ( Shi et al . , 2010; Shi and Tu , 2013 ) , we propose a dual role for trehalose synthesis during phosphate-limitation . During phosphate limitation , trehalose synthesis concurrently releases inorganic phosphate , which restores phosphate balance in the cell and diverts flux away from nucleotide synthesis and growth . When phosphate is no longer limiting , cells can liquidate trehalose to re-enter the cell division cycle , enabling rapid recovery . Thus , this study adds an important biochemical function to the many roles played by this versatile metabolite . These include cell survival during desiccation and freezing ( Calahan et al . , 2011; Erkut et al . , 2011; Erkut et al . , 2016; Tapia and Koshland , 2014; Tapia et al . , 2015 ) , the ability to act as a protein chaperone ( Tapia and Koshland , 2014 ) , and as a membrane protectant ( Abusharkh et al . , 2014 ) . A modified tRNA is an unusual but effective mechanism to coordinately integrate sensing of overall nutrient sufficiency , and regulate metabolic homeostasis . While previous studies have observed decreased phosphate-related transcripts in tRNA thiolation deficient cells ( Leidel et al . , 2009; Nedialkova and Leidel , 2015; Chou et al . , 2017 ) , possible roles for phosphate in tRNA thiolation mediated function have been ignored . Furthermore , tRNAs undergo other conserved modifications in the U34 position: 5-methoxycarbonylmethyluridine ( mcm5 U34 ) , 2-thiouridine ( s2 U34 ) , 5-methoxycarbonylmethyl-2-thiouridine ( mcm5s2 U34 ) , 5-methylaminomethyluridine ( mnm5U34 ) ( Phizicky and Hopper , 2010 ) . In this study we focus only on how the s2 U34 modification ( which is derived from sulfur amino acids ) regulates cellular metabolic state . Interestingly , the other U34 modifications all require s-adenosyl methionine ( SAM ) , and SAM is itself directly derived from sulfur amino acid metabolism ( Thomas and Surdin-Kerjan , 1997 . Mutants of all these related U34 tRNA modifications show similar metabolic phenotypes as the thiolation mutants ( Zinshteyn and Gilbert , 2013; Nedialkova and Leidel , 2015; Chou et al . , 2017; Han et al . , 2018 ) , and have a down-regulated PHO response ( Chou et al . , 2017 ) . This raises the possibility that these U34-tRNA and other tRNA modifications derived from amino acid metabolism use similar mechanisms , of controlling phosphate availability in order to regulate metabolic homeostasis . A primordial role of such tRNA modifications might therefore be to appropriately sense overall amino acid sufficiency ( with methionine as a sentinel growth signal; Walvekar et al . , 2018b ) , and modulate metabolic states towards growth , regulating phosphate availability as a means to achieve this . Co-opting tRNAs ( which are the translation components most closely linked to amino acids ) to control metabolic states can therefore be an efficient means to ensure appropriate commitments to growth and proliferation , and maximize cellular fitness . Concluding , here we discover that a sulfur amino acid-dependent tRNA modification ( thiolated U34 ) enables cells to appropriately balance amino acid and nucleotide levels and regulate metabolic state , by controlling phosphate homeostasis . More generally , we suggest how phosphate homeostasis can impact flux through different arms of carbon and nitrogen metabolism . The prototrophic CEN . PK strain of Saccharomyces cerevisiae was used in all the experiments ( van Dijken JP et al . , 2000 ) . All the strains used in this study are listed in Supplementary file 1 . For all experiments , cells were grown overnight at 30°C in rich media ( 1% yeast extract , 2% peptone , 2% dextrose ) , washed once and subsequently sub-cultured in minimal media ( 0 . 67% yeast nitrogen base with ammonium sulfate , without amino acids , 0 . 1% glucose ) unless specified . Phosphate-limited medium was prepared as described previously ( Klosinska et al . , 2011 ) except that 0 . 1% . glucose was used unless specified . The only source of phosphorus in phosphate-limited media ( low Pi ) was KH2PO4 , which was present at a concentration of 0 . 15 mM . In high phosphate media ( high Pi ) , KH2PO4 was present at a concentration of 7 . 5 mM with 0 . 1% . glucose unless specified . In no phosphate media ( no Pi ) , KH2PO4 was completely absent and 0 . 1% . glucose was used . Complete medium was high Pi media supplemented with amino acids and 0 . 1% glucose . Amino acid concentrations were used as described previously ( Sherman , 2002 ) . Sulfur-rich medium was minimal media ( 0 . 67% yeast nitrogen base with ammonium sulfate , without amino acids ) with 2% glucose . Sulfur-starved medium was prepared as described previously with 2% glucose ( Kankipati et al . , 2015 ) Sulfur amino acid limited medium was minimal media ( 0 . 67% yeast nitrogen base with ammonium sulfate , without amino acids ) supplemented with all amino acids at a final concentration of 2 mM with methionine and cysteine being completely absent and 2% glucose . For Gcn4 , total eIF2α , P- eIF2α , Pho12 , Pho84 and Rad53 protein levels , cells were grown overnight in rich media , washed once and subsequently sub-cultured in minimal media at an initial OD600 of 0 . 1 and grown till the OD600 reached 0 . 8–1 . 0 . For Pho84 and Pho12 protein levels in high and low Pi media , cells were grown overnight in rich media , washed once and subsequently sub-cultured in high and low Pi media at an initial OD600 of 0 . 1 and incubated for 5 hr at 30°C . For Gcn4 protein levels in high and no Pi media , cells were grown overnight in rich media , washed once and subsequently sub-cultured in high and no Pi media at an initial OD600 of 0 . 1 and incubated for 8 hr at 30°C . Cells were harvested by centrifugation and protein was isolated by trichloroacetic acid ( TCA ) precipitation method . Briefly , cells were resuspended in 400 μl of 10% trichloroacetic acid and lysed by bead-beating three times . The precipitates were collected by centrifugation , resuspended in 400 μl of SDS-glycerol buffer ( 7 . 3% SDS , 29 . 1% glycerol and 83 . 3 mM Tris base ) and heated at 100°C for 10 min . The lysate was cleared by centrifugation and protein concentration was determined by using a bicinchoninic acid assay ( 23225 , Thermo Fisher ) . Equal amounts of samples were electrophoretically resolved on 4–12% pre-cast Bis-tris polyacrylamide gels ( NP0322BOX , Invitrogen ) . Anti-HA ( 12CA5 , Roche ) was used to detect Gcn4-HA , anti-phospho eIF2α ( Ser51 ) ( 9721S , Cell Signalling Technology ) was used to detect phospho-eIF2α , anti-FLAG ( F1804-5MG , Sigma-Aldrich ) was used to detect total eIF2α ( eIF2α-FLAG ) , Pho12-FLAG and Pho84-FLAG , anti-Rad53 yC-19 ( sc-6749 , Santa Cruz Biotechnology ) was used to detect phosphorylated Rad53 protein . Horseradish peroxidase-conjugated secondary antibodies ( mouse and rabbit ) were obtained from Sigma-Aldrich . For Western blotting , standard enhanced chemiluminescence reagent ( GE Healthcare ) was used . Coomassie brilliant blue R-250 was used to stain gels for loading control . For steady state amino acids and nucleotides levels , cells were grown overnight in rich media , washed once and subsequently sub-cultured in minimal media at an initial OD600 of 0 . 1 and grown till the OD600 reached ~0 . 8 . ~10 OD600 cells were quenched with 60% methanol at −40°C , and metabolites were extracted , as explained in detail elsewhere ( Walvekar et al . , 2018a ) . For steady state amino acids , nucleotides and sulfur-containing metabolite levels in sulfur-rich and sulfurstarved conditions , wild-type cells were grown overnight in rich media , washed once and subsequently sub-cultured in sulfur-rich and sulfur-starved conditions at an initial OD600 of 0 . 25 and incubated for 3 hr at 30°C . ~5 OD600 cells were quenched with 60% methanol at −40°C , and metabolites were extracted . For 15N-label incorporation in amino acids and nucleotides , cells were grown overnight in rich media , washed once and subsequently sub-cultured in minimal media ( 0 . 67% yeast nitrogen base without amino acids and ammonium sulfate , 0 . 1% glucose , 20 mM ammonium sulfate ) at an initial OD600 of 0 . 1 and grown till the OD600 reached 0 . 5 . At this point , 15N2-ammonium sulfate ( 299286 , Sigma-Aldrich ) was added to reach a ratio of 50% unlabeled to 50% fully labelled ammonium sulfate . Metabolites were extracted from ~6 OD600 cells . For 13C- label incorporation in nucleotides and other central carbon metabolites , cells were grown overnight in rich media , washed once and subsequently sub-cultured in minimal media at an initial OD600 of 0 . 1 and grown till the OD600 reached 0 . 5 . For experiments where 13C- label incorporation into nucleotides was measured , in medium supplemented with or without additional methionine , cells were grown overnight in rich media , and subsequently sub-cultured in fresh rich media at an initial OD600 of 0 . 2 and grown till the OD600 reached 1 , washed once and shifted to minimal media , 2% glucose with or without 2 mM methionine for 1 hr . After this time , [U-13C6] glucose ( CLM-1396-PK , Cambridge Isotope Laboratories ) was added to reach a ratio of 50% unlabeled to 50% fully labelled glucose . Metabolites were extracted from ~6 OD600 cells . Extensive metabolite extraction protocols are described ( Walvekar et al . , 2018a ) . Metabolites were analyzed using LC-MS/MS method as described in Walvekar et al . ( 2018a ) . Standards were used for developing multiple reaction monitoring ( MRM ) methods on Thermo Scientific TSQ Vantage Triple Stage Quadrupole Mass Spectrometer or Sciex QTRAP 6500 . All the parent/product masses relevant to this study are listed in Supplementary file 3 . Amino acids were detected in the positive polarity mode . For nucleotide measurements , nitrogen base release was monitored in the positive polarity mode . Trehalose was detected in the negative polarity mode . For PPP metabolites and other triose phosphates , phosphate release was monitored in the negative polarity mode . Metabolites were separated using a Synergi 4µ Fusion-RP 80A column ( 100 × 4 . 6 mm , Phenomenex ) on Agilent’s 1290 infinity series UHPLC system coupled to the mass spectrometer . For positive polarity mode , buffers used for separation were- buffer A: 99 . 9% H2O/0 . 1% formic acid and buffer B: 99 . 9% methanol/0 . 1% formic acid ( Column temperature , 40°C; Flow rate , 0 . 4 ml/min; T = 0 min , 0% B; T = 3 min , 5% B; T = 10 min , 60% B; T = 11 min , 95% B; T = 14 min , 95% B; T = 15 min , 5% B; T = 16 min , 0% B; T = 21 min , stop ) . For ADP and ATP separation and detection , buffers used for separation were- buffer A: 5 mM ammonium acetate in H2O and buffer B: 5 mM ammonium acetate in 100% methanol , and metabolites were measured in positive polarity mode . Alternately , buffers used for separation were- buffer A: 5 mM ammonium acetate in H2O and buffer B: 100% acetonitrile ( Column temperature , 25°C; Flow rate: 0 . 4 ml/min; T = 0 min , 0% B; T = 3 min , 5% B; T = 10 min , 60% B; T = 11 min , 95% B; T = 14 min , 95% B; T = 15 min , 5% B; T = 16 min , 0% B; T = 21 min , stop ) , and negative polarity mode was used . The area under each peak was calculated using Thermo Xcalibur software ( Qual and Quan browsers ) and AB SCIEX MultiQuant software 3 . 0 . 1 . For all spotting assays , cells were grown overnight in rich media , washed once and subsequently sub-cultured in minimal media at an initial OD600 of 0 . 2–0 . 25 and grown till the OD600 reached 0 . 8–1 . 0 . Cells were harvested by centrifugation , washed once with water and 10 μl sample for each suspension was spotted in serial 10-fold dilutions . For 8-aza adenine and hydroxyurea sensitivity assays , cells were spotted onto minimal media plates containing 250 and 300 μg/ml 8-aza adenine ( A0552 , TCI chemicals ) or 150 mM hydroxyurea ( H8627 , Sigma-Aldrich ) and incubated at 30°C . For control plates without drug , images were taken after 1–2 days and for drug containing plates after 4–5 days . For genetic interaction analysis with Tps2 , cells were spotted onto high and low Pi media plates with 2% glucose . Plates were incubated at 30°C and 37°C . For genetic interaction analysis with Pho80 , cells were spotted onto minimal media plates with 0 . 1% and 2% glucose . Plates were incubated at 30°C . For trehalose and glycogen measurements in wild-type and thiolation mutants , cells were grown overnight in rich media , washed once and subsequently sub-cultured in minimal media at an initial OD600 of 0 . 1 and grown till the OD600 reached 0 . 8–1 . 0 . For trehalose measurement in high and no Pi media , cells were grown overnight in rich media , washed once and subsequently sub-cultured in either high or no Pi media at an initial OD600 of 0 . 1 and incubated for 8 hr at 30°C . For trehalose measurement in sulfur-rich and sulfur-starved conditions , wild-type cells were grown overnight in rich media , washed once and subsequently sub-cultured in sulfur-rich and sulfur-starved conditions at an initial OD600 of 0 . 25 and incubated for 5 hr at 30°C . Cells were harvested by centrifugation and washed with ice-cold water . Cells were lysed in 0 . 25 M sodium carbonate by incubating at 95–98°C for 4 hr . Subsequently , added 0 . 15 ml 1M acetic acid and 0 . 6 ml of 0 . 2 M sodium acetate to bring the solution to pH 5 . 2 . Trehalose and glycogen were digested overnight using trehalase ( T8778 , Sigma-Aldrich ) and amyloglucosidase ( 10115 , Sigma-Aldrich ) respectively . Glucose released from these digestions was measured using a Glucose ( GO ) Assay Kit ( GAGO20 , Sigma-Aldrich ) . The concentration of released glucose ( μg/ml ) was determined from the standard curve and plotted . Statistical significance was determined using Student T-test ( GraphPad Prism 7 ) . Continuous chemostat cultures to establish the YMC were performed as described previously ( Tu et al . , 2005 ) . An overnight batch culture of prototrophic CEN . PK strain ( van Dijken JP et al . , 2000 ) grown in rich medium was used to inoculate working volume of 1L in the chemostat . At 20 min time-intervals , cells were fixed with 2% paraformaldehyde , and imaged under a bright-field microscope . ~200 cells from each time point were sampled , and budding cells were counted manually . Cell cycle synchronization and flow cytometry analysis bar1Δ::Hyg , uba4Δ::NAT bar1Δ::Hyg and ncs2Δ::NAT bar1Δ::Hyg cells were grown overnight in minimal media and subsequently sub-cultured in minimal media at an initial OD600 of 0 . 05 and grown till the OD600 reached 0 . 2 . Cells were harvested by centrifugation , washed with water and resuspended in the same medium containing 10 μg/ml of α-factor ( GenScript ) . Cells were kept at 30°C for 3 hr till complete G1 arrest was observed by light microscopy . Subsequently , 5 ml culture was harvested by centrifugation , washed with water and fixed with 70% ethanol for G1-arrested population . Remaining culture was synchronously released into the cell cycle by washing away the α-factor . Cells were collected at different intervals of time post G1 release , fixed with ethanol , treated with RNaseA ( R4875 , Sigma-Aldrich ) and a protease solution ( P6887 , Sigma-Aldrich ) as described ( Haase and Reed , Cell cycle , 2002 ) . Cells were stained with SYTOX green ( S7020 , Invitrogen ) and analyzed on BD FACS Verse flow cytometer . Cells were grown overnight in rich media , washed once and subsequently sub-cultured in minimal media at an initial OD600 of 0 . 1 and grown till the OD600 reached 0 . 4–0 . 5 . 1 . 5% agar pads ( 50081 , Lonza ) were prepared containing minimal media . The pad was cut into small pieces after it solidified . 2 μl of the cell suspension was placed on the agar pad , which was inverted and placed in a glass bottom confocal dish ( 101350 , SPL Life Sciences ) for imaging . Phase-contrast images were captured after every 3 min’ interval for total of 360 mins on ECLIPSE Ti2 inverted microscope ( NIKON ) and 60X oil-immersion objective . Images were stacked and analyzed using ImageJ software . Statistical significance was determined using a Student T-test ( GraphPad Prism 7 ) . Cells were grown overnight in rich media and subsequently shifted to minimal media till the OD600 reached 0 . 5–0 . 8 . Cells were rapidly harvested by filtration and lysed , as described in detail ( Weinberg et al . , 2016; McGlincy and Ingolia , 2017 ) . For both transcriptome and ribosome profiling analyses , three biological replicates each for WT and tRNA thiolation mutant cells ( uba4Δ and ncs2Δ ) were included . Total RNA and ribosome-protected fragments were isolated from the cell lysates and RNA-seq and ribosome profiling were performed , as described ( Weinberg et al . , 2016; McGlincy and Ingolia , 2017 ) , with minor modifications . Separate 5’ and 3’ linkers were ligated to the RNA- fragment instead of 3’ linker followed by circularization ( Subtelny et al . , 2014 ) . 5’ linkers contained four random nt unique molecular identifier ( UMI ) similar to a five nt UMI in 3’ linkers . During size-selection , we restricted the footprint lengths to 18–34 nts . Matched RNA-seq libraries were prepared using RNA that was randomly fragmentation by incubating for 14 min at 950C with in 1 mM EDTA , 6 mM Na2CO3 , 44 mM NaHCO3 , pH 9 . 3 . RNA-seq fragments were restricted to 18–50 nts . Ribosomal rRNA were removed from pooled RNA-seq and footprinting samples using RiboZero ( Epicentre MRZH116 ) . cDNA for the pooled libraries were PCR amplified for 16 cycles . RNA-seq and footprinting reads were mapped to the yeast transcriptome using the riboviz pipeline ( Carja et al . , 2017 ) . Sequencing adapters were trimmed from reads using Cutadapt 1 . 14 ( Martin , 2011 ) ( --trim-n -e 0 . 2 –minimum-length 24 ) . The reads from different samples were separated based on the barcodes in their 3’ linkers using fastx_barcode_splitter ( FASTX toolkit , Hannon lab ) with utmost one mismatch allowed . UMI and barcodes were removed from reads in each sample using Cutadapt ( --trim-n -m 10 u 4 u −10 ) . Trimmed reads that aligned to yeast rRNAs and tRNAs were removed using HISAT2 v2 . 1 . 0 ( Kim et al . , 2015 ) . Remaining reads were mapped to a set of 5812 genes in the yeast genome ( SGD version R64-2-1_20150113 ) using HISAT2 . Only reads that mapped uniquely were used for all downstream analyses . Codes for generating processed fastq and gff files were obtained from riboviz package ( https://github . com/shahpr/RiboViz; Carja et al . , 2017 ) . Gene-specific fold-changes in RNA and footprint abundances were estimated using DESeq2 packages in R ( Love et al . , 2014 ) using default log-fold-change shrinkage options . Changes in ribosome-densities ( translation efficiencies ) were estimated using the Riborex package in R ( Li et al . , 2017 ) . The complete transcript/ribosome footprint datasets are available at GEO ( number GSE124428; link: https://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE124428 ) . Cells were grown overnight in rich media , washed once and subsequently sub-cultured in high and low Pi media ( for experiment related to wild-type and uba4Δ ) or complete media ( for experiment related to wild-type , uba4Δ , gcn4Δ and uba4Δ gcn4Δ ) at an initial OD600 of 0 . 1 and grown till the OD600 reached 0 . 5–0 . 6 . 8 OD600 cells were collected , washed once and resuspended in sterile water to final OD600 of 16 . Acid phosphatase activity was assayed , as described previously with some modifications ( Huang and O'Shea , 2005 ) . Briefly , 450 μl of each cell suspension was added to 200 μl of 20 mM p-nitrophenyl phosphate ( PNPP , N4645 , Sigma‐Aldrich ) in 0 . 1M sodium acetate , pH-4 . 2 , mixed and incubated at room temperature for 30 min . To stop the reaction , 200 μl of the reaction mixture was withdrawn and added to 200 μl of ice cold 10% trichloroacetic acid . To this reaction , 400 μl of saturated sodium carbonate solution ( 2M , pH-11 . 5 ) was added , mixed and centrifuged at 3000 rpm for 10 min . 80 μl of the supernatant was transferred in technical triplicates to a 96-well plate and liberated p-nitrophenol was determined by measuring OD420 on a plate reader . Phosphatase activity was measured in units expressed as OD420/OD600 × 1000 . Statistical significance was determined using a Student T‐test ( GraphPad Prism 7 ) . Cells were grown overnight in rich media , washed once and subsequently sub-cultured in minimal media at an initial OD600 of 0 . 1 and grown till the OD600 reached 0 . 8–1 . 0 . Free intracellular phosphate levels were determined , as described previously ( McNaughton et al . , 2010 ) . Briefly , cells were harvested by centrifugation and washed twice with ice-cold water . Cells were lysed by resuspending in 200 μl 0 . 1% triton X-100 and vortexed for 5 min with glass beads . The lysate was cleared by centrifugation and protein concentration was determined by using bicinchoninic acid assay ( 23225 , Thermo Fisher ) . 30 μg of whole cell lysate was used for measurement of free intracellular phosphate levels using ammonium molybdate and ascorbic acid colorimetric assay as described ( Ames , 1966 ) . Potassium dihydrogen phosphate solution was used for standard curve ( 0 to 500 µM KH2PO4 ) . The amount of phosphate was expressed as µM Pi . Statistical significance was determined using a Student T-test ( GraphPad Prism 7 ) . WT , uORF1* and uORF4* Gcn4-luciferase reporter plasmids ( SL148 , SL149 and SL150 ) were generated by PCR amplification of a 777 bp fragment of WT , uORF1and uORF4 Gcn4 constructs ( GeneArt , Thermo Scientific ) and subsequent cloning in SL147 plasmid ( Supplementary file 2 ) . SL147 was generated by cloning luciferase cDNA from pGL3-Basic ( Addgene ) in pSL80 plasmid ( Wu and Tu , 2011 ) . For luciferase assay in wild-type , thiolation mutants , gcn2Δ and double mutants , cells transformed with Gcn4-luciferase reporter plasmids were grown overnight in rich media in presence of G418 . Cells were subsequently sub-cultured in the same media without selection till logarithmic phase and then shifted to minimal media . After 4–5 hr of growth ( OD600 of 0 . 8–1 . 0 ) , cells were collected , washed twice with ice-cold lysis buffer ( 1X PBS , 1 mM PMSF ) . For luciferase assay in wild-type and gcn2Δ , in high and no Pi media , cells transformed with Gcn4-luciferase reporter plasmid were grown overnight in rich media in presence of G418 . Cells were subsequently sub-cultured in high and no Pi media supplemented with amino acids , 2% glucose and incubated for 8 hr at 30°C , cells were collected , washed twice with ice-cold lysis buffer ( 1X PBS , 1 mM PMSF ) . For luciferase assay in sulfur amino acid limited condition , wild-type cells transformed with Gcn4-luciferase reporter plasmid were grown overnight in rich media , and subsequently sub-cultured in rich media at an initial OD600 of 0 . 2 and grown till the OD600 reached 0 . 4–0 . 5 and half of the cells were collected for luciferase assay . Remaining cells were washed once and subsequently shifted to sulfur amino acid limited condition and incubated for 1 hr at 30°C . Lysis was done by vortexing cells for 5 min in presence of glass beads . Lysates were cleared by centrifugation and protein concentration was determined by using a bicinchoninic acid assay ( 23225 , Thermo Fisher ) . Luciferase assay was performed using luciferase assay system kit ( E1500 , Promega ) and activity was measured using a Sirius luminometer ( Tiertek Berthold ) . Data output provided as Relative Light Units per sec ( RLU/s ) was used to determine relative luciferase activity . Statistical significance was determined using a Student T-test ( GraphPad Prism 7 ) . Cells were grown overnight in rich media , washed once and subsequently sub-cultured in minimal media at an initial OD600 of 0 . 1 and grown till the OD600 reached 0 . 8–1 . 0 . Cells were harvested by centrifugation , and RNA was isolated by hot phenol beating method ( Collart and Oliviero , 2001 . 6 μg of total RNA was used for DNase I ( AM2238 , Thermo Fisher ) treatment . cDNA was synthesized with random primers ( 48190011 , Thermo Fisher ) and SuperScript II reverse transcriptase ( 18064014 , Thermo Fisher Scientific ) according to the manufacturer’s protocol . cDNA quantification was done by real-time PCR on an ViiA 7 Real-Time PCR System ( Thermo Fisher ) using Maxima SYBR Green/ROX qPCR Master Mix ( K0222 , Thermo Fisher ) . ACT1 was used as an internal normalization control . All qRT-PCRs were performed in triplicates using at least two independent biological RNA samples . Statistical significance was determined using a Student T-test ( GraphPad Prism 7 ) . Total cellular ATP was measured using the ATP determination kit ( A22066 , Thermo Fisher ) according to the manufacturer’s protocol . Briefly , 5 µl sample or 5 µl of different concentrations of ATP standard solution ( 0 to 5 µM ATP ) was added to 50 µl of assay solution . Luminescence was measured directly after addition of the sample to assay solution using a Sirius luminometer ( Tiertek Berthold ) . ATP concentrations in samples were calculated from the ATP standard curve and relative levels were plotted . Statistical significance was determined using a Student T-test ( GraphPad Prism 7 ) .	The building blocks of all cells are made from a handful of chemical elements , including carbon , nitrogen , sulfur and phosphorus . To grow optimally , cells need to regulate their metabolism – in other words , the biochemical reactions that keep them alive – based on the availability of these elements . As a result , cells have evolved various mechanisms to sense when usable forms of these elements are present . Proteins are chains of building blocks known as amino acids , which are assembled with the help of molecules called transfer ribonucleic acids , or tRNAs for short . Some of these molecules can be modified by attaching sulfur-containing chemical tags known as thiol groups to make “thiolated tRNAs” . Research has shown that , when there was more of an amino acid known as methionine around , the cells made more thiolated tRNA . These previous studies also suggested that mutant cells lacking thiolated tRNAs might have altered carbon and nitrogen metabolism . Yet , it remained unclear what exactly was leading to this metabolic rewiring . Now , Gupta et al . have combined several biochemical and genetics approaches to study the role of thiolated tRNAs in yeast . The experiments revealed that mutant cells lacking thiolated tRNAs were unable to properly sense the levels of methionine and other amino acids , which are the cell’s major source of nitrogen . These mutant cells were also found to have a reduced level of phosphorous-containing compounds known as phosphates , which are involved in numerous biological processes . Gupta et al . showed that reducing the level of phosphates caused carbon that is normally used to make chemicals required for growth to be re-routed towards making carbohydrates to store energy instead . This is similar to what happens when the cells are starving , showing that a ‘squeeze’ on internal phosphates metabolically rewires cells into a state that is like starvation . These findings show how modified tRNAs can use the availability of amino acids to alter the cell’s metabolism by altering how much phosphate is present . In doing so , the thiolated tRNAs essentially allow the cell to decide whether it has enough of the right nutrients to grow . These findings may also have implications for human health , since errors in coordinating metabolism are responsible for certain medical conditions including several cancers . Finally , technical challenges mean many questions remain unanswered about how phosphate levels are regulated within cells . These new findings point to a pressing need to understand phosphate metabolism as a prerequisite to better understand how cells regulate their overall metabolism .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "genetics", "and", "genomics" ]	2019	A tRNA modification balances carbon and nitrogen metabolism by regulating phosphate homeostasis
Eukaryotes and prokaryotes last shared a common ancestor ~2 billion years ago , and while many present-day genes in these lineages predate this divergence , the extent to which these genes still perform their ancestral functions is largely unknown . To test principles governing retention of ancient function , we asked if prokaryotic genes could replace their essential eukaryotic orthologs . We systematically replaced essential genes in yeast by their 1:1 orthologs from Escherichia coli . After accounting for mitochondrial localization and alternative start codons , 31 out of 51 bacterial genes tested ( 61% ) could complement a lethal growth defect and replace their yeast orthologs with minimal effects on growth rate . Replaceability was determined on a pathway-by-pathway basis; codon usage , abundance , and sequence similarity contributed predictive power . The heme biosynthesis pathway was particularly amenable to inter-kingdom exchange , with each yeast enzyme replaceable by its bacterial , human , or plant ortholog , suggesting it as a near-universally swappable pathway . Despite over 2 billion years of divergence , eukaryotes and prokaryotes still share hundreds of genes ( Theobald , 2010; O'Brien et al . , 2005; Brown and Doolittle , 1997; Martin and Müller , 1998 ) . Though these ancient genes are identifiable as orthologs at the sequence level , the preservation of original protein function across such deep timescales has not been systematically explored . The function of certain genes could potentially become frozen in place in the course of evolution , sheltered from lineage-specific functional alterations introduced by mutations , gene fusions , and non-orthologous gene displacements . Such functionally frozen genes would in principle be able to substitute for their least-diverged ortholog in any other species . Searching for such gene replaceability between species thus serves to test a core assumption of the ortholog-function conjecture: that orthologs retain ancestral function ( Gabaldón and Koonin , 2013 ) . This conjecture forms the basis of most modern biomedical research and is widely used to predict new gene function across organisms ( Lee et al . , 2007 ) . There are many individual examples of genes from one species functioning for their orthologous counterparts in a different species ( Cherry et al . , 2012; Heinicke et al . , 2007 ) , but this trend has only recently begun to be explored systematically , with several large-scale studies substituting human genes for yeast genes and confirming that many human orthologs can successfully replace their yeast counterparts ( Kachroo et al . , 2015; Sun et al . , 2016; Hamza et al . , 2015 ) . At the level of evolutionary divergence of yeast and humans , such data demonstrate widespread functional conservation , even after 1 billion years of divergence . The ability of human genes to functionally replace their yeast orthologs is not strongly predicted by the similarity of sequences , but rather at the level of specific pathways or processes , wherein all genes in a process or pathway tend to be similarly replaceable , or not ( Kachroo et al . , 2015 ) . However , in the timescale of evolution , yeast and humans are relatively similar – both eukaryotes that share thousands of genes and the majority of their core biological processes . Data on eukaryote – prokaryote functional gene replacement are sparse ( Heinicke et al . , 2007 ) . These cross-domain replacements represent a maximum test of the ability of genes to retain their ancestral function across time . Eukaryotic and bacterial genes have been , for the most part , evolving independently since at least the archaeal ancestor of eukaryotes endosymbiotically acquired its bacterial mitochondrion . In eukaryotes , the function of these genes would have had to survive the development of vastly different genome structures , cell division modalities , cell wall compositions , and subcellular compartmentalizations which occurred during eukaryogenesis . Prokaryotic and eukaryotic orthologs also diverged significantly at the amino acid sequence level ( O'Brien et al . , 2005 ) and evolved distinct expression patterns and codon usages ( Sharp et al . , 1993; Bulmer , 1991 ) . Nonetheless , eukaryotes and bacteria are known to use many of the same orthologs to perform the same metabolic enzymatic reactions ( Jardine et al . , 2002; Peregrin-Alvarez et al . , 2003 ) . Thus , in order to more systematically determine the replaceability of orthologs across such deep timescales , we asked in this study how many conserved E . coli genes can successfully substitute for their yeast orthologs . We focused on those genes that are essential for viability in yeast , allowing us to assay for the complementation of otherwise lethal growth defects . We analysed many features of the proteins and ortholog pairs to identify which properties best explained replaceability , finding that replaceability was often determined at the level of specific pathways and processes , with all genes in a pathway or process similarly replaceable . Start codon choice and eukaryote-specific subcellular localization were also critical determinants of replaceability . We discovered that certain core biological processes have remained largely unchanged since the last common ancestor of bacteria , yeast , and humans . In particular , heme biosynthesis pathway enzymes appear to be generally exchangeable between prokaryotic and eukaryotic organisms , broadly retaining ancestral functions across the tree of life over 2 billion years of independent evolution , even when accompanied by evolved changes in enzyme subcellular localization . We focused our efforts on the set of genes with 1:1 orthology between E . coli and yeast and that are known to be essential for yeast growth in standard laboratory conditions ( Figure 1A ) . Each E . coli open reading frame ( ORF ) was cloned into a single-copy yeast centromeric ( CEN ) plasmid under the transcriptional control of a constitutive GPD promoter . Complementation assays were carried out using two types of conditionally essential yeast alleles , consisting of temperature-sensitive ( TS ) haploid and heterozygous diploid deletion strains . In the case of the heterozygous diploid deletion strains , the respective yeast gene null allele could be genetically segregated via sporulation , allowing selection for haploid yeast with the null allele ( selected for in the presence of the antibiotic G418 ) or the wild-type yeast gene ( in the absence of G418 ) ( Figure 1B , Top panel ) . In the case of TS haploid yeast strains , the temperature sensitive yeast proteins functioned normally at the permissive temperature ( 25°C ) but could be conditionally inactivated at the non-permissive temperature ( 36°C ) in order to test for gene replaceability ( Figure 1B , Bottom panel ) . Overall , we could perform informative complementation assays for 51 of the 58 orthologs , as shown for the examples in Figure 1B . 10 . 7554/eLife . 25093 . 003Figure 1 . Many E . coli genes efficiently complement lethal growth defects in their yeast counterparts . ( A ) Yeast and E . coli share hundreds of genes , 58 of which are essential in yeast and have clear 1:1 orthologs in either species . E . coli genes were cloned into a yeast expression vector under the control of a GPD promoter . 51 of these 58 E . coli genes provided informative assays for replaceability in yeast . Initial results from these complementation assays revealed that 25 of 51 ( ~49% ) E . coli genes could functionally replace their orthologous yeast counterparts . ( B ) Complementation assays were performed in two different yeast strain backgrounds , as shown for representative assays . In the case of a yeast strain with a temperature-sensitive allele of the yeast gene Sc-cdc8 , cells carrying the empty vector control grow at the permissive-temperature ( 25°C , yeast protein active ) but not the restrictive-temperature ( 36°C , yeast protein inactive ) , unlike cells expressing the E . coli ortholog ( Ec-tmK ) , indicating that the E . coli gene can functionally replace the yeast gene . In the case of yeast heterozygous diploid ( Sc-ths1Δ/Sc-THS1 ) deletion strain , cells are sporulated and haploid progeny grown on selective medium ( -Ura -Arg -His -Leu + Can ) in the absence ( yeast gene present ) or presence of G418 ( 200 μg/ml ) ( yeast gene absent ) . Cells expressing the E . coli ortholog ( Ec-thrS ) grow on G418-containing medium , unlike cells carrying the empty vector control , indicating successful complementation . ( C ) Haploid yeast gene deletion strains carrying plasmids expressing functionally replacing E . coli genes ( red solid-lines ) generally exhibit comparable growth rates to the wild type parental yeast strain BY4741 ( black dotted-lines ) . The empty vector control ( grey solid-line ) showed no such growth rescue in the presence of G418 . Mean and standard deviation plotted with N = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 25093 . 00310 . 7554/eLife . 25093 . 004Figure 1—figure supplement 1 . Complementation assays performed in a 96-well format in two different yeast strain backgrounds ( Supplementary file 1 ) . ( A and B ) Magic marker heterozygous diploid deletion yeast strains expressing E . coli genes were sporulated and the sporulation mix was spotted on magic marker agar medium ( -Ura -Arg -His -Leu + Can ) with ( yeast gene absent ) or without ( yeast gene present ) G418 ( 200 μg/ml ) . ( C ) Temperature-sensitive haploid yeast strains expressing E . coli genes grown at permissive temperature ( 25°C ) ( yeast protein active ) and at restrictive temperature ( 36°C ) ( yeast protein inactive ) on -Ura agar medium with G418 ( 200 μg/ml ) . Empty vector containing yeast cells were used as negative control for the experiment . ( D ) Haploid yeast gene deletion strains carrying plasmids expressing functionally replacing E . coli genes ( red solid-lines ) generally exhibit comparable growth rates to the wild type parental yeast strain BY4741 ( black dotted-lines ) as grown in YPD liquid medium in the presence of G418 ( 300 μg/ml ) . Mean and standard deviation plotted with N = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 25093 . 00410 . 7554/eLife . 25093 . 005Figure 1—figure supplement 2 . Constitutive plasmid expression of yeast genes efficiently replaced the corresponding genomic copies for 6 non-replaceable alleles . Bacterial orthologs of the yeast genes , Sc-RRP3 , Sc-PGS1 , Sc-SRP54 , Sc-PCM1 and Sc-HSP60 did not show functional replacement when expressed from a constitutive GPD promoter . We expressed the corresponding yeast genes in a similar fashion under the control of the constitutive GPD promoter . All the tested yeast genes functionally replaced the corresponding yeast gene deletions . Empty vector containing yeast cells were used as negative control for the experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 25093 . 005 Of the 51 E . coli genes tested , 25 successfully complemented lethal growth defects in the corresponding yeast strains ( Figure 1—figure supplement 1A , B and C; Supplementary file 1 ) . In nearly all cases , despite plasmid-based expression of the complementing genes , the bacterialized strains grew comparably to the parental , wild type yeast strain , in both synthetic defined medium ( SD -Ura + G418 ) ( Figure 1C ) and rich medium ( YPD + G418 ) ( Figure 1—figure supplement 1D ) . We further verified complementation specificity by testing for plasmid loss ( see Materials and methods and ( Supplementary file 1 ) and sequence verifying all clones . We have previously demonstrated that plasmid-borne copies of yeast genes complemented their corresponding heterozygous diploid deletion alleles at a high rate ( 100% for 29 strains tested in Kachroo et al . , 2015 ) , but as an additional control , we repeated this test for six yeast strains where the E . coli gene failed to rescue , confirming that the corresponding yeast genes were able to complement the growth defect when expressed on a CEN plasmid under the control of the constitutive GPD promoter ( Figure 1—figure supplement 2 and Sc-HEM1 as reported in Figure 4—figure supplement 1 ) . Many eukaryotic orthologs of prokaryotic genes function in specific subcellular compartments absent from prokaryotes , and consistent with this trend , 15 of the 51 tested E . coli genes have mitochondrially-localized yeast orthologs ( Cherry et al . , 2012 ) . Because all but one of these 15 genes were unable to replace their yeast ortholog , we reasoned that lack of mitochondria targeting might account for their failed complementation . We added the mitochondrial localization signal ( MLS ) from the yeast MIP1 gene to each of the 14 non-replaceable E . coli genes and repeated the complementation assays . Four genes could now functionally replace their yeast equivalents ( Figure 2A , Figure 2—figure supplement 1 ) , restoring growth rates to be nearly or fully comparable with the parental strain ( Figure 2B ) . We verified mitochondrial localization by fusing the E . coli Ec-MLS-HscB and Ec-MLS-IlvD proteins with enhanced green fluorescent protein ( EGFP ) and confirming correct trafficking of the EGFP-tagged proteins to yeast mitochondria ( Figure 2C ) . 10 . 7554/eLife . 25093 . 006Figure 2 . The addition of a mitochondrial localization signal ( MLS ) and mutation of start codons from GTG to ATG allows some E . coli genes to swap for their respective yeast orthologs . ( A ) 14 of the 25 non-replaceable E . coli genes were predicted to function in mitochondria in yeast . 4 of 14 were replaceable after adding the MLS at the N-termini of the E . coli genes . Site-specific mutagenesis of E . coli gene start codon from GTG to ATG allowed two to functionally complement the corresponding yeast genes bringing the total number E . coli genes that functionally replace yeast genes to 31 of 51 ( ~61% ) . ( B ) Haploid yeast gene deletion strains carrying mitochondrially localized E . coli genes rescued the growth defect of the yeast gene ( red solid-line ) comparable to the wild type yeast ( black dashed-line ) . The empty vector control ( grey solid-line ) and the yeast cells expressing of E . coli gene without MLS ( blue-solid line ) showed no such growth rescue in the presence of G418 . Mean and standard deviation plotted with N = 3 . ( C ) EGFP-tagged E . coli genes that functionally replaced the yeast gene function were imaged after MitoTracker red staining . EGFP-tagged Ec-MLS-HscB and Ec-MLS-IlvD ( green ) show colocalization with MitoTracker red stained mitochondria ( red ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25093 . 00610 . 7554/eLife . 25093 . 007Figure 2—figure supplement 1 . Some E . coli genes require a yeast mitochondrial localization signal to efficiently replace . The magic marker heterozygous diploid deletion yeast strains carrying empty vector or E . coli gene with or without MLS were sporulated and the sporulation mix was plated on magic marker agar medium ( -Ura -Arg -His -Leu + Can ) with or without G418 ( 200 μg/ml ) . E . coli genes Ec-rpiL , Ec-ilvC , Ec-ilvD and Ec-hscB without an appropriate mitochondrial localization signal cannot complement the corresponding yeast gene deletions Sc-mnp1 , Sc-ilv5 , Sc-ilv3 and Sc-jac1 . However , expression of E . coli genes with yeast MLS efficiently rescued the growth defect of the corresponding yeast gene deletions . DOI: http://dx . doi . org/10 . 7554/eLife . 25093 . 007 Bacterial genes also occasionally lack a standard ATG start codon , with ~14% of all E . coli ORFs employing an alternative start codon ( Blattner et al . , 1997 ) . Three of the tested non-replaceable E . coli genes used a GTG start codon while one used ATT . We therefore used site-directed mutagenesis to introduce canonical ATG start codons , then re-assayed for complementation . After changing their start codons to ATG , two of these four E . coli genes , Ec-rcsC and Ec-tadA , could now replace their yeast orthologs ( Figure 2B ) . Overall , after accounting for mitochondrial localization and alternative start codons and combining results from all assays , a total of 31 out of 51 tested E . coli genes could successfully replace their essential yeast orthologs ( Figure 2 ) . Thus , in a majority ( 61% ) of our tests , both the current day prokaryotic and eukaryotic proteins must have retained their critical ancestral functions such that the prokaryotic proteins could carry out the essential roles of their eukaryotic orthologs well enough to support yeast cell growth . In one-fifth of the cases , replaceability depended on proper subcellular localization or start codon choice to express the prokaryotic gene in the proper eukaryotic context . Given that we observed both replaceable and non-replaceable genes , we sought to determine properties of the tested genes that best explained successful replacements . We considered 22 features of the tested genes , including protein lengths , interactions , sequence similarities , codon usages , and expression levels . We calculated the predictive utility of each feature as the area under a Receiver Operating Characteristic curve ( AUC ) ( Figure 3A; Supplementary file 2 ) . Notably , the extent of protein sequence similarity between orthologs was not a highly predictive feature . A large portion of the tested E . coli and yeast orthologs showed only 20–30% identical amino acid sequences and roughly half of these genes were replaceable; in contrast , the three most divergent orthologs replaced , each showing less than 20% identity ( Figure 3B ) . As we observed a non-monotonic relationship between sequence identity and replaceability , potentially explained by replaceability differences among different functional categories of genes , we tested for the enrichment of particular GO Biological Process ( defined by Gene Ontology Slim annotations ( Ashburner et al . , 2000 ) or KEGG categories ( Kanehisa and Goto , 2000 ) within the individual bins of sequence identity in Figure 3B . Aside from an enrichment in glucose metabolism genes ( 3 of the 7 ) in the 40–50% identity range , we did not find evidence for strong pathway-specific biases that would explain the observed relationship between sequence identity and replaceability . We did observe moderate predictive power for some measures of codon bias , especially those related to codon optimality within E . coli , and less so for codon optimality within a yeast context; more highly optimized E . coli codon usage correlated with a lower replaceability rate . 10 . 7554/eLife . 25093 . 008Figure 3 . Replaceability of E . coli genes is a modular phenomenon . ( A ) Several quantitative properties of the tested genes were assessed for their ability to predict replaceability , measured as the area under a receiver operating characteristic curve ( AUC ) . Having a high fraction of interaction partners that replace was the most predictive property tested , suggesting that the ability to replace is a modular phenomenon whereby genes functioning together are similarly able to replace . A Random Forest classifier constructed with all attributes boosted the maximum AUC to 0 . 79 . ( B ) As shown in ( A ) , sequence similarity was not the most predictive feature . The fraction of replaceable genes in given ranges of similarity was variable , with the vast majority of orthologs being 20–30% identical , a range in which roughly half of proteins replaced . ( C ) Mapping of replaceability status onto yeast GO slim annotations revealed that GO categories have varying rates of replaceability , with core metabolic processes ( e . g . energy metabolism , nucleobase metabolism ) being largely replaceable while more specialized processes ( e . g . protein assembly , membrane transport ) were less so . DOI: http://dx . doi . org/10 . 7554/eLife . 25093 . 008 Instead , the strongest predictive features related to specific pathways and processes , much as we and others have observed for successful humanization of yeast ( Kachroo et al . , 2015; Sun et al . , 2016; Hamza et al . , 2015 ) . This trend was most evident in the observation that a gene was more likely to replace ( or not ) if it had a higher fraction of interaction partners that also replaced ( or not ) . Consequently , different biological processes ( as defined by GO ) displayed varied replaceability , with metabolic processes being largely replaceable , while processes known to be divergent , including ribosomal processing , were much less replaceable ( Figure 3C ) . This trend suggests an explanation for why optimized E . coli codons predicted worse replaceability , as E . coli genes with optimized codons predominantly tend to be highly expressed ribosomal and translational proteins ( Saikia et al . , 2016 ) . This is thus consistent with the notion that replaceability is determined at the level of the pathway or process , with codon choice and gene expression levels reflecting functional constraints of that process . Combining all of these features into a single predictor ( after accounting for mitochondrial localization and alternative start codons ) , using a random forest classifier , improved our predictive power to a 0 . 79 AUC ( Figure 3A ) , demonstrating that the features we investigated provide moderately orthogonal predictive information . Nearly all the genes that we tested from the heme biosynthesis pathway were replaceable by their E . coli orthologs , which in combination with the evidence that replaceability was determined at the level of processes , led us to investigate the heme pathway in more depth . Most of the enzymatic reactions in the heme biosynthesis pathway are identical between E . coli and yeast , but there are clear differences in the way this pathway functions between the species ( Heinemann et al . , 2008 ) . First , heme biosynthesis pathway precursors differ: Yeast condense succinyl-CoA and glycine to produce delta-aminolevulinate in a single enzymatic step catalyzed by Sc-Hem1 , while E . coli produces delta-aminolevulinate in two steps using glutamyl-tRNA as a precursor ( Yin and Bauer , 2013 ) . Second , the bacterial heme pathway is largely cytosolic but in yeast it is partitioned between the mitochondria and cytosol ( Figure 4A ) . We thus next considered these two key pathway differences in more detail . As a control , we first expressed the corresponding yeast genes on plasmids either under the control of constitutive GPD or the native yeast promoter ( Ho et al , 2009 ) to test the effect of constitutive expression on functional replaceability . Except for Sc-HEM4 , which showed toxicity when expressed constitutively , all the other yeast genes showed functional replaceability irrespective of the mode of expression ( Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 25093 . 009Figure 4 . Bacterialization of yeast heme biosynthesis pathway genes at their native loci . ( A ) A schematic of the yeast heme pathway shows the beginning of the pathway in mitochondria using succinyl-CoA and glycine as precursors . The subsequent enzymatic reactions are cytosolic up until the penultimate and ultimate reactions which are mitochondrial . ( B ) Growth kinetics of CRISPR-Cas9 engineered yeast heme pathway genes replaced with the corresponding bacterial genes at their native yeast loci show efficient replaceability in both BY4741 ( red solid-line ) and BY4742 ( blue solid-line ) yeast strains . The wild type BY4741 growth curve is shown as a comparison ( black dotted-line ) . Mean and standard deviation plotted with N = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 25093 . 00910 . 7554/eLife . 25093 . 010Figure 4—figure supplement 1 . Constitutive or native plasmid-based expression of the yeast heme biosynthesis genes generally efficiently complemented growth defects in the corresponding yeast gene deletion strains . Heterologous expression of yeast genes Sc-HEM1 , Sc-HEM2 , Sc-HEM3 , Sc-HEM4 , Sc-HEM12 , Sc-HEM13 , Sc-HEM14 and Sc-HEM15 under the control of constitutive GPD promoter or native promoter efficiently rescued the growth defect of the corresponding yeast gene deletions respectively except in the case of Sc-HEM4 . Sc-HEM4 , when expressed constitutively , resulted in toxicity in the presence of the yeast gene at the native locus and did not complement the function in the absence of the yeast gene . This toxicity was relieved when the yeast gene was expressed under the control of the native yeast promoter . DOI: http://dx . doi . org/10 . 7554/eLife . 25093 . 01010 . 7554/eLife . 25093 . 011Figure 4—figure supplement 2 . Ec-hemA and Ec-hemL carry out the initial reaction in E . coli heme biosynthesis and are both required to complement Sc-HEM1 deletion in yeast , and non-orthologous yeast genes are replaced by E . coli genes that carry out the identical reaction . ( A ) Expression of heme pathway genes of E . coli , Ec-hemA or Ec-hemL , individually cannot complement the lethal growth defect of the deletion of Sc-HEM1 gene in yeast . Co-expression of Ec-HemA and Ec-HemL efficiently rescued the growth defect of Sc-hem1 gene deletion in yeast . ( B ) Growth curves of yeast strains with deletions of Sc-hem4 and Sc-hem14 genes ( grey solid-line ) show functional replaceability ( red solid-line ) by the non-orthologous E . coli genes Ec-hemD and Ec-hemG that carry out identical enzymatic reactions to the corresponding yeast genes . The wild type BY4741 growth curve is shown as a comparison ( black dotted-line ) . The empty vector control ( grey solid-line ) showed no such growth rescue in the presence of G418 . ( C ) Growth curve of engineered yeast strain Sc-hem14Δ::Ec-hemG; Sc-hem15Δ::Ec-hemH in YPD medium harboring E . coli genes at the native yeast loci . The strain displayed a growth defect ( red solid-line ) compared to the wild type BY4741 strain ( black dotted-line ) . Mean and standard deviation plotted with N = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 25093 . 01110 . 7554/eLife . 25093 . 012Figure 4—figure supplement 3 . The penultimate and ultimate heme pathway enzymes in yeast are replaceable by their bacterial orthologs , in spite of mis-localizing to the plasma membrane . EGFP-tagged Ec-HemG and Ec-HemH localize to the plasma membrane in yeast . The EGFP-tagged proteins do not localize to the mitochondria since no clear co-localization is observed with the Mitotracker red stain . EGFP-tagged Ec-HemG and Ec-HemH expression ( red solid-line ) efficiently rescue the growth defects of the respective yeast gene deletions ( Sc-hem14 and Sc-hem15 ) ( pink dotted-line ) comparable to the wild type yeast ( black dotted-line ) . Empty vector control is incapable of rescuing the growth defect of the deletion strains ( grey dotted-line ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25093 . 01210 . 7554/eLife . 25093 . 013Figure 4—figure supplement 4 . Confirmation of CRISPR-Cas9 mediated bacterialized yeast strains . ( A ) Schematics of the yeast heme pathway gene loci carrying functionally replaceable E . coli genes while retaining their native promoters and terminators . The arrows indicate the primers used to confirm the replacement ( refer to Supplementary file 3 ) . ( B ) PCR amplification of expected size was obtained for each individual bacterialized yeast strains . DOI: http://dx . doi . org/10 . 7554/eLife . 25093 . 013 In our initial screen , the E . coli ortholog of Sc-HEM1 , Ec-kbL , failed to replace the yeast gene , an observation consistent with prior data showing that Ec-kbL does not take part in E . coli heme biosynthesis , but rather carries out an unrelated but mechanistically-similar oxido-reductase reaction involved in L-threonine degradation ( UniProt Consortium , 2015; Mukherjee and Dekker , 1990 ) . Instead , a two-step enzymatic reaction by E . coli proteins Ec-HemA and Ec-HemL produces the heme precursor , delta-aminolevulinate ( Schauer et al . , 2002; Ilag and Jahn , 1992 ) . Since the initial steps of the pathway are localized to the mitochondria , we added the Sc-MIP1 MLS to the 5’ ends of these genes and expressed them simultaneously in the Sc-HEM1 heterozygous diploid deletion strain . Co-expression of the two E . coli genes successfully replaced yeast gene function ( Figure 4—figure supplement 2A ) . Additionally , two enzymes , Ec-HemD and Ec-HemG , were not identified as orthologs between E . coli and yeast , despite carrying out identical reactions to Sc-Hem4 and Sc-Hem14 , respectively . Expression of these non-orthologous but functionally analogous E . coli genes in the respective yeast deletion strains showed that they were indeed able to successfully replace the yeast genes ( Figure 4—figure supplement 2B ) . For these enzymes , the key determinants for successful replacement are thus their enzymatic reactions , rather than any other aspects of the genes . Sc-Hem14 and Sc-Hem15 carry out the final two steps in yeast heme biosynthesis and are localized to the mitochondria ( Cherry et al . , 2012; Koh et al , 2015 ) ( Figure 4A ) . Both genes were replaceable by the E . coli genes carrying out the analogous reactions , Ec-HemG ( Figure 4—figure supplement 2B ) and Ec-HemH ( Figure 1C ) , despite the lack of targeting sequences for mitochondrial localization . As E . coli lack mitochondria , and Ec-HemG and Ec-HemH are both predicted to localize to the plasma membrane in E . coli ( Papanastasiou et al . , 2013 ) , we thus assayed their localization in yeast when expressed as EGFP-fusion proteins . Strikingly , both localized to the yeast plasma membrane ( Figure 4—figure supplement 3 ) . In spite of failing to localize to the yeast mitochondria , the bacterialized strains grew well compared to wild type yeast ( Figure 4—figure supplement 3 ) , suggesting that mitochondrial localization is not an absolute requirement for their functions , as many heme pathway intermediates are cytosolic . However , concurrent bacterialization of both yeast genes resulted in a viable but defective yeast strain ( Figure 4—figure supplement 2C ) , suggesting that the fitness cost of mis-localizing both proteins is not tolerated well , potentially due to cumulative effects of reduced efficiency of the bacterial proteins , altered allosteric regulation in yeast , or the accumulation of heme precursors in the wrong compartment ( cytosol ) ( Yin and Bauer , 2013 ) . Because heterologous expression using a constitutive promoter could be compensating for more subtle functional differences , we also wished to measure complementation after placing the bacterial orthologs under control of the native yeast gene regulation . We thus used CRISPR/Cas9-based precision genome engineering to genomically replace each of the heme biosynthesis pathway genes in turn in yeast ( except Sc-HEM12 ) with its respective E . coli counterpart , from start to stop codon , while retaining the native promoters , terminators , and chromosomal context of the yeast genes ( Figure 4B , Figure 4—figure supplement 4 ) . All strains but two grew comparably to the wild-type; the Sc-hem14∆::Ec-hemG and Sc-hem15∆::Ec-hemH strains showed modest growth defects ( Figure 4B ) . Because these two yeast proteins are known to be mitochondrially localized ( Cherry et al . , 2012 ) , we re-engineered each of the Ec-hemG and Ec-hemH ORFs into the yeast chromosome such that each gene’s native yeast MLS was retained ( Sc-hem14∆::Ec-MLS-hemG and Sc-hem15∆::Ec-MLS-hemH ) . The addition of the yeast MLS to each E . coli ORF completely ameliorated growth defects from the ORFs alone ( Figure 4B ) . Thus , the yeast heme biosynthesis pathway appears entirely replaceable , one gene at a time , by their corresponding bacterial genes , whether expressed constitutively from plasmids or directly integrated into chromosomes under native yeast transcriptional regulation . The extent of replaceability strongly suggests that ancestral functions in these genes ( with the obvious exception of the non-orthologous steps ) have remained intact and unaltered , at least in terms of critical , enzymatic functionality . Mitochondrial localization of several of the enzymes , while needed to fully recover growth rates , is not essential for viability . Ec-hemH and Sc-HEM15 encode ferrochelatase , the enzyme responsible for adding iron to the porphyrin ring of protoporphyrin IX to produce protoheme ( Figure 4A ) . In the course of constructing the CRISPR-edited yeast strains , we noticed that the Sc-hem15∆::Ec-hemH yeast strain turned pink on a standard YPD agar medium upon prolonged incubation of 3–4 days ( Figure 5A ) . This phenotype was consistent across all independently obtained , sequence verified yeast clones . The pink phenotype decreased dramatically in the Sc-hem15∆::Ec-MLS-hemH strains in which Ec-HemH was correctly localized to the mitochondria by addition of an MLS . 10 . 7554/eLife . 25093 . 014Figure 5 . Mislocalization of the bacterialized ferrochelatase enzyme identifies a porphyria-like phenotype in yeast . ( A ) Bacterialization of the ultimate yeast gene in the heme biosynthesis pathway results in a distinct pink colony phenotype on YPD agar medium . In contrast , wild type BY4741 strain colonies appear as creamy-white . ( B ) Acetate-extracted secreted products from the pink Sc-hem15Δ::Ec-hemH strains show strongly enhanced fluorescence at 635 nm ( excitation 399 nm ) , comparable to a protoporphyrin IX standard and unlike a heme standard or extracts from the parental BY4741 strain . The introduction of an MLS to the bacterialized yeast strain ( Sc-hem15Δ::Ec-MLS-hemH ) significantly reduced protoporphyrin IX secretion , while deletion of the MLS from the native yeast locus in strain Sc-ΔMLS-HEM15 caused several strains to increase protoporphyrin IX secretion . DOI: http://dx . doi . org/10 . 7554/eLife . 25093 . 01410 . 7554/eLife . 25093 . 015Figure 5—figure supplement 1 . Absorbance ( top ) and emission ( bottom ) spectra of extracts obtained from acetate ( left ) and pyridine ( right ) extraction of the wild type or bacterialized yeast colonies grown on YPD medium . Purified protoporphyrin IX ( red solid-line ) or heme ( yellow solid-line ) were used as standards . Extract from the bacterialized Sc-hem15Δ::Ec-hemH yeast strain ( dark blue-line ) matched with that of the protoporphyrin IX standard . Bacterialized ScHEM15Δ::Ec-MLS-hemH yeast strain ( orange solid-line ) showed significantly reduced peak for protoporphyrin IX . Extracts from wild type BY4741 ( black-line ) and BY4742 ( light blue solid-line ) were used as controls . DOI: http://dx . doi . org/10 . 7554/eLife . 25093 . 01510 . 7554/eLife . 25093 . 016Figure 5—figure supplement 2 . Deletion of protoporphyrinogen oxidase , Sc-HEM14 , in the Sc-hem15Δ::Ec-hemH strain suppressed the porphyria-like pink phenotype . Top row from left show growth spots of the BY4741 wild type , Sc-hem15Δ::Ec-hemH and Sc-hem15Δ::Ec-MLS-hemH yeast strains . Bottom row from left show corresponding strains harboring Sc-hem14 deletion . DOI: http://dx . doi . org/10 . 7554/eLife . 25093 . 016 We speculated that the pink phenotype was likely due to aberrant accumulation of porphyrin intermediates , presumably leading to their secretion , as we observed that the pigment could be washed off the cells . Therefore , we chemically extracted the pink pigment from Sc-hem15∆::Ec-hemH , Sc-hem15∆::Ec-MLS-hemH and wild type yeast cells ( Materials and methods ) and performed fluorescence spectroscopy to determine that the pigment likely corresponds to protoporphyrin IX ( Figure 5B , Figure 5—figure supplement 1 ) . In order to determine whether protein mis-localization contributed to the phenotype , we removed the MLS from the native yeast gene . Several clones of the Sc-ΔMLS-HEM15 yeast strain displayed similar extracellular pigment ( Figure 5B , Figure 5—figure supplement 1 ) . These results suggest that mislocalized plasma membrane-bound Ec-HemH in yeast does not convert protoporphyrin IX to protoheme efficiently , resulting in the accumulation and secretion of protoporphyrin IX . We further tested this line of reasoning by deleting the gene for the preceding step in the pathway , Sc-HEM14 , which encodes the enzyme protoporphyrinogen oxidase and is responsible for making protoporphyrin IX . Using CRISPR , we deleted the Sc-HEM14 ORF in wild type BY4741 , Sc-hem15Δ::Ec-HemH , and Sc-hem15Δ::Ec-MLS-HemH strains . Consistent with protoporphyrin IX being the pink pigment in the Sc-hem15Δ::Ec-HemH strain , the Sc-hem15Δ::Ec-HemH hem14Δ strain lost the pink phenotype , even after growing for 6 days . Moreover , we observed that all strains carrying the hem14Δ allele were in fact significantly paler than even wild type BY4741 cells , presumably reflecting extensive protoporphyrin IX depletion in these cells ( Figure 5—figure supplement 2 ) . In humans , disrupting heme biosynthesis leads to the disease porphyria , and the secretion of porphyrin intermediates is specifically observed in a subtype known as protoporphyria ( Bloomer et al . , 1998 ) , wherein reduced activity of the human heme pathway protein Hs-FECH leads to accumulation and subsequent secretion of protoporphyrin IX into surrounding tissues . Our data suggest that yeast protein localization and protoporphyrin secretion phenotypes might in the future be exploited to investigate disease-causing mutations in human Hs-FECH , even in cases where disease variants do not show any discernible growth defect in yeast . The data above show that genes in the yeast heme biosynthesis pathway can be replaced by their bacterial counterparts , extending earlier studies demonstrating that some heme biosynthesis genes can also be humanized ( Kachroo et al . , 2015; Sun et al . , 2016; Schauer and Mattoon , 1990 ) . Given the ancient conservation of this pathway , we sought to further expand our investigation of its replaceability by swapping the corresponding genes from the plant Arabidopsis thaliana into yeast . In plants , heme biosynthesis enzymes form precursors for chlorophyll , and the pathway is largely chloroplast-localized , in contrast to compartmentalization of the heme biosynthetic pathway between the mitochondria and cytosol in many other eukaryotes ( UniProt Consortium , 2015; Ashburner et al . , 2000; Mochizuki et al . , 2010 ) . Nonetheless , the fact that Arabidopsis ferrochelatase was cloned by complementing a mutant yeast phenotype suggests that other heme pathway genes might also successfully replace the yeast genes ( Smith et al . , 1994 ) . The first enzymatic step in the plant heme biosynthetic pathway is similar to bacteria , a two-step reaction using glutamyl-tRNA as a substrate ( Figure 6A and B ) ( Ilag et al . , 1994 ) . We expressed both plant genes , At-HEMA1 and At-GSA2 , simultaneously and were able to functionally replace the corresponding yeast gene function . Neither protein , when individually expressed , could functionally replace the yeast gene ( Figure 6—figure supplement 1A ) . 10 . 7554/eLife . 25093 . 017Figure 6 . Yeast heme biosynthesis pathway enzymes can be successfully replaced by orthologs or analogs from bacteria , plants , and humans , in spite of alterations to subcellular localization . Enzymatic steps of extant bacterial and eukaryotic heme biosynthesis pathways are identical save for the starting metabolites and conversion to delta-aminolevulinate; bacteria also exhibit non-orthologous gene displacement of several enzymes . Heme biosynthesis occurs in the bacterial cytoplasm and inner membrane , the human and yeast in mitochondria and cytoplasm , and the plant in chloroplast and cytoplasm . In spite of these localization changes over evolution , most of the defects in growth rate and viability conferred by heme pathway mutations in yeast can be complemented by introduction of the corresponding ( A ) bacterial genes , ( B ) plant genes ( except for At-HemE ) , and ( C ) human genes . Yellow indicates a replaceable gene , blue non-replaceable . DOI: http://dx . doi . org/10 . 7554/eLife . 25093 . 01710 . 7554/eLife . 25093 . 018Figure 6—figure supplement 1 . Heme biosynthesis genes from Arabidopsis thaliana and Glycine max generally efficiently replace their counterparts in yeast , except in the case of ΔSc-Hem12 . ( A ) Expression of heme pathway genes from Arabidopsis thaliana , At-HEMA1 or At-GSA2 , individually cannot complement the lethal growth defect of the deletion of Sc-hem1 gene in yeast . Co-expression of At-HEMA1 and At-GSA2 rescued the growth defect of Sc-hem1 gene deletion in yeast . ( B ) Haploid yeast gene deletion strains carrying plasmids expressing functionally replacing Arabidopsis ( red or blue solid-lines ) and ( B’ ) Glycine max ( Gm-HEMG ) heme pathway genes ( red solid-line ) generally exhibit comparable growth rates to the wild type parental yeast strain BY4741 ( black dotted-line ) as grown in magic marker liquid medium in the presence of G418 ( 200 μg/ml ) . ( B’’ ) Native At-HEMC with chloroplast localization signal ( CLS ) showed poor replaceability in yeast ( red solid-line ) . Removal of the CLS from At-HEMC allowed efficient rescue of the corresponding yeast gene deletion , ΔSc-Hem3 ( blue solid-line ) . ( B’’’ ) However , neither the expression of Arabidopsis proteins At-HEME1 or At-HEME2 ( with or without CLS ) alone nor their co-expression could functionally rescue the corresponding yeast gene deletion , ΔSc-Hem12 . Wild type BY4741 haploid strain is plotted for comparison ( black dotted-line ) . Strains carrying empty vector were used as controls ( grey solid-line ) . Mean and standard deviation plotted with N = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 25093 . 01810 . 7554/eLife . 25093 . 019Figure 6—figure supplement 2 . Heme biosynthesis enzymes normally localized to plant chloroplasts or human mitochondria localize to the mitochondria when expressed in yeast . ( A ) EGFP-tagged penultimate At-PPOX1-EGFP and ultimate At-FC1-EGFP proteins localize to mitochondria in yeast . Green fluorescence proteins co-localized with Mitotracker red-stained mitochondria . In certain cases , At-FC1-EGFP formed aggregates . Expression of EGFP-tagged plant genes , At-PPOX1-EGFP and At-FC1-EGFP ( red solid-line ) , efficiently rescue the growth defect of the corresponding yeast gene deletions ( pink dotted-line ) . The over-expression of the tagged proteins is not toxic to the wild type yeast strain ( grey dotted-line ) . The growth rescue by plant genes is as efficient as the wild type BY4741 yeast strain ( black dotted-line ) . Mean and standard deviation plotted with N = 3 . ( B ) The EGFP-tagged last three heme pathway genes from humans localize to mitochondria in yeast . The green fluorescence co-localized with the Mitotracker red-stained mitochondria in yeast . Expression of EGFP-tagged human genes , Hs-PPOX-EGFP , Hs-FECH-EGFP and Hs-CPOX-EGFP ( red solid-line ) , efficiently rescue the growth defect of the corresponding yeast gene deletions ( pink dotted-line ) . The over-expression of the tagged proteins is not toxic to the wild type yeast strain ( grey dotted-line ) . The growth rescue by the human genes is as efficient as the wild type BY4741 yeast strain ( black dotted-line ) . Mean and standard deviation plotted with N = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 25093 . 01910 . 7554/eLife . 25093 . 020Figure 6—figure supplement 3 . Human heme biosynthesis genes efficiently replace their yeast counterparts . Functional replacement of human genes in yeast . ( A ) Expression of Hs-UROS in Sc-hem4 heterozygous diploid deletion yeast strain resulted in toxicity post-sporulation as seen by the lack of growth on either magic marker agar medium with ( yeast gene present ) or without G418 ( yeast gene absent ) . ( B ) This toxicity was relieved by replacing the human Hs-UROS at the native yeast locus . Growth curve of the humanized yeast Sc-hem4Δ::Hs-UROS strain ( red-solid line ) showed comparable growth to the wild type yeast BY4741 ( black dotted-line ) . ( C ) Expression of human Hs-UROD ( a human orfeome clone with G303V mutation ) in Sc-hem12 heterozygous diploid deletion yeast strain did not complement the growth defect of the yeast gene as shown by plating the post sporulation mix on magic marker medium with or without G418 . Reverting the sequence to the wild type Hs-UROD gene resulted in efficient rescue of the growth defect of the corresponding yeast gene . ( D ) Expression of human genes , Hs-PPOX , Hs-UROD , Hs-ALAS1 ( red solid-line ) and Hs-ALAS2 ( blue solid-line ) , efficiently rescue the growth defect of the corresponding yeast gene deletions ( grey solid-line ) , Sc-hem14 and Sc-hem1 , respectively . The rescue was largely comparable to the wild type BY4741 yeast strain ( black dotted-line ) . Strains carrying empty vector were used as controls ( grey solid-line ) . Mean and standard deviation plotted with N = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 25093 . 020 In Arabidopsis , unlike for the case of E . coli , a majority of genes in the heme biosynthesis pathway have acquired lineage-specific amplifications , resulting in two co-orthologs for each single yeast gene ( Figure 6B ) . In these cases , we tested both co-orthologs individually for replaceability; all replaced successfully , with the exception of one case where only one replaced ( At-CPX1 replaced while At-CPX2 did not ) , and one case where neither replaced ( At-HEME1 and At-HEME2 ) ( Figure 6B , Figure 6—figure supplement 1B , B’’’ ) . Because the plant heme biosynthesis pathway builds precursors for chlorophyll synthesis ( Tanaka et al . , 2011; Papenbrock et al . , 1999 ) , this pathway , especially the penultimate step producing protoporphyrin IX , is the target of many commercial herbicides . Both Arabidopsis paralogs that we tested , At-PPOX1 and At-PPOX2 , could efficiently complement the yeast gene responsible for this critical step , Sc-HEM14 ( Figure 6—figure supplement 1B ) . To confirm the generality of these results , we further tested the soybean ( Glycine max ) ortholog Gm-HEMG in yeast . As for each of the Arabidopsis paralogs , the single soybean ortholog also successfully complemented the Sc-hem14 deletion growth defect ( Figure 6—figure supplement 1B’ ) . It is noteworthy that plant heme biosynthesis genes harbor chloroplast localization sequences ( UniProt Consortium , 2015 ) , and we did not remove these for our complementation experiments . We speculated that the chloroplast leader peptides might be recognized and localized by the yeast mitochondrial localization machinery , so we constructed EGFP-fusions of the plant enzymes and assayed their localization by fluorescence microscopy . EGFP fusions of At-PPOX1 and At-FC1 showed clear mitochondrial localization in yeast ( Figure 6—figure supplement 2A ) . At-FC1 additionally showed amorphous aggregates in some yeast cells , suggesting localization might occasionally be imperfect . Nonetheless , both EGFP-tagged genes were able to efficiently rescue the growth defect of the corresponding yeast gene deletion ( Figure 6—figure supplement 2A ) . Thus , these plant chloroplast localization signals appear to be recognized and processed as mitochondrial localization signals in yeast . These findings suggested that plant versions of cytosolic yeast heme pathway proteins could potentially be mis-localizing to the mitochondria in yeast ( Figure 4A ) . Indeed , At-HEMC only weakly replaced the yeast gene , Sc-HEM3 . We found that removing the chloroplast localization signal ( CLS ) from At-HEMC markedly enhanced its ability to functionally replace its yeast ortholog ( Figure 6—figure supplement 1B’’ ) . In contrast , neither of two Arabidopsis paralogs , At-HEME1 and At-HEME2 , could functionally replace their yeast ortholog , Sc-HEM12 , even after removing their CLS sequences , or even when co-expressed in the yeast strain ( Figure 6—figure supplement 1B’’’ ) . We speculate that there could be several other reasons why complementation failed , including unknown intermediate reactions , required localization in a special compartment ( e . g . chloroplast ) or different transcriptional/translational regulation in plants that might contribute to the lack of functional replaceability . Earlier studies have shown successful replacement of the yeast heme biosynthesis genes by their human orthologs Hs-ALAD ( Schauer and Mattoon , 1990 ) , Hs-HMBS , Hs-CPOX and Hs-FECH ( Kachroo et al . , 2015 ) , while Hs-UROS expression resulted in toxicity and Hs-UROD failed to replace its yeast ortholog ( Kachroo et al . , 2015; Sun et al . , 2016 ) . We , therefore , sought to complete tests of the remaining human genes in the pathway . In the case of Hs-UROS , we reasoned that toxicity was due to expression from the heterologous constitutive promoter ( Figure 6—figure supplement 3A ) . Indeed , similar to the results obtained with the yeast version of this gene ( Figure 4—figure supplement 1 , Sc-HEM4 ) , we found that toxicity could be abrogated by inserting the human gene at the native yeast chromosomal locus , thus providing native yeast gene expression and regulation for the human ORF ( Figure 6—figure supplement 3B ) . This suggests that , at least in yeast , this step is regulated transcriptionally for optimal function . We also found that the human ORFeome clone of Hs-UROD contained a mutation ( G303V ) that when reverted to wild-type sequence allowed it to replace the yeast gene ( Figure 6—figure supplement 3C and D ) , and we additionally confirmed that human Hs-PPOX could complement the severe growth defect of the yeast Sc-hem14 deletion strain ( Figure 6C , Figure 6—figure supplement 3D ) . Finally , in humans , the initial step of heme biosynthesis is identical to that of yeast ( Sc-HEM1 ) but is encoded by two co-orthologs , Hs-ALAS1 and Hs-ALAS2 . We found that both of these human genes could individually replace the yeast gene function ( Figure 6C , Figure 6—figure supplement 3D ) . The subcellular localization of heme biosynthesis differs slightly between humans and yeast , such that the last three proteins in the human heme biosynthesis pathway are mitochondrially localized , as opposed to only the last two in yeast ( Grandchamp et al . , 1978; Ferreira et al . , 1988 ) . As all three of these genes replaced , we tested if the human genes were localized to the mitochondria in yeast . Indeed , EGFP-tagged Hs-FECH , Hs-PPOX , and Hs-CPOX all localized to mitochondria in yeast ( Figure 6—figure supplement 2B ) and efficiently rescued the growth defect of the corresponding yeast gene deletion ( Figure 6—figure supplement 2B ) , confirming that the human mitochondrial localization signal is recognizable by the yeast localization machinery . Thus , across our attempts to humanize , plantize , and bacterialize this pathway , the presence of mitochondrial leader peptides from the human genes and the chloroplast leader peptides from the plant genes , as well as the absence of bacterial leaders , all overrode the native yeast localization of the heme biosynthesis pathway . However , the pathway function was largely resilient to these effects , with the exception of protoporphyrin IX accumulation in the mislocalized bacterialized strains ( Figure 5 ) . As illustrated in Figure 7 , the heme pathway has had a complicated evolutionary trajectory in eukaryotes due to endosymbiotic events , which has served to increase its similarity between bacteria and eukaryotes ( Kořený and Oborník , 2011 ) . During eukaryogenesis , early eukaryotes adopted a large portion of the bacteria-like heme biosynthesis pathway of their endosymbiont mitochondria . The subsequent endosymbiotic acquisition of chloroplasts along the plant lineage ( Oborník and Green , 2005 ) resulted in redundancy between mitochondrial-origin and chloroplast-origin portions of their heme biosynthesis pathways , a state that can be observed today in Euglena , a non-plant , photosynthetic eukaryote with more recently acquired chloroplasts ( Kořený and Oborník , 2011 ) . Over time , plants kept the chloroplastic system and lost most of the mitochondrial system . These evolutionary transfers may have been possible due the apparent modularity of the heme pathway , which we observe in its high tolerance for substituting genes or enzymatic functions across species . 10 . 7554/eLife . 25093 . 021Figure 7 . The complex evolutionary history of the heme biosynthesis pathway is reflected in high replaceability across species . In eukaryotes , heme biosynthesis enzymes have been replaced historically by endosymbiosis events from bacteria , leading to higher similarity across these lineages , while the archaeal pathway appears to be more divergent ( Storbeck et al . , 2010 ) . Following the endosymbiosis of the cyanobacterial chloroplast , plants adopted most of the chloroplast-derived heme biosynthesis genes , losing many ancestral eukaryotic heme pathway genes ( Oborník and Green , 2005 ) . Yeast and humans both retain the predicted ancestral eukaryotic heme biosynthesis pathway . While enzymatic steps are mostly shared between yeast , plants , bacteria , and humans , localization of individual proteins differs substantially between species . Asterisks indicate results curated from literature . DOI: http://dx . doi . org/10 . 7554/eLife . 25093 . 021 Our data demonstrate that despite 2 billion years of divergence from their last common ancestor , heme biosynthesis genes are still carrying out a conserved and necessary function that can be swapped into yeast with minimal effect on growth and irrespective of orthology and subcellular localization . Taking these data together with literature studies showing successful replacement of the E . coli Ec-hemG gene by the plant or human Hs-PPOX gene ( Lermontova et al . , 1997; Dailey and Dailey , 1996; Narita et al . , 1996 ) , and that introducing the protoporphyrinogen oxidase from Bacillus subtilis into plants improves yields ( Lee et al . , 2000 ) , heme biosynthesis thus appears to be a pathway whose genes are freely exchangeable between bacteria , plants ( with the exception of At-HEME ) , humans , and yeast ( Figure 7 ) . In conclusion , in order to discern whether orthology strictly confers function across deep evolutionary distances , we systematically tested E . coli genes with 1:1 orthology to essential yeast genes for their ability to functionally replace their yeast counterparts . We discovered that ~61% ( 31/51 ) of the tested E . coli and yeast genes still retain ancestral function to a sufficient extent that the bacterial genes efficiently replace their yeast equivalents . Eukaryote-specific features such as subcellular localization ( 4 of 14 ) and proper start codon usage ( 2 of 4 ) were critical for swappability for some of the E . coli orthologs . Our analysis of replaceable/non-replaceable orthologous pairs revealed that amino acid sequence similarity was not the most important property , consistent with a general trend for sequence conservation to often more strongly reflect other attributes of protein function ( e . g . , abundance and protein-specific functional constraints ) ( Jordan et al . , 2002; Wang and Zhang , 2009 ) . Rather , the top predictors of replaceability were features attributed to specific gene modules . These results largely agree with previously published work on humanization of yeast genes ( Kachroo et al . , 2015; Hamza et al . , 2015; Sun et al . , 2016 ) , suggesting that functional replaceability is predominantly determined at the level of pathways and processes , even across very large evolutionary distances . As our assays can be considered a form of forced horizontal gene transfer , our results provide support for the ‘complexity hypothesis’ ( Jain et al . , 1999 ) , which posits that informational ( transcription , translation , etc . ) genes are less likely to be horizontally transferred than those genes that are operational ( metabolism , housekeeping , etc . ) . Consistent with this expectation , we see metabolism-associated genes replacing more often than those involved in ‘informational’ processes like transcription or translation . In the course of these studies , we found that heme biosynthetic reactions were entirely replaceable across the prokaryote-eukaryote divide , despite non-orthologous functional displacement and lack of eukaryotic subcellular localization by native E . coli genes ( Figure 7 ) . Although the archaeal pathway is considerably diverged , our studies across bacteria and eukaryotes showed a high degree of replaceability: Plant heme biosynthesis enzymes functionally replaced yeast enzymes in all but one reaction . Swaps of the corresponding human enzymes into yeast in this and prior studies all suggest that heme biosynthesis is a near universally replaceable pathway . Our results thus demonstrate that orthologous genes carry out similar functions that allow for their ability to functionally replace each other across even the 2 billion year evolutionary rift separating prokaryotes and eukaryotes from their last common ancestor . These swaps allow engineering of orthologous pathways in model organisms highly amenable to genetic perturbations , like yeast and bacteria , for further characterization . Refer to Supplementary file 3 for all the primers used in this study . Gene replaceability was tested using available yeast strains from two yeast strain collections , the temperature-sensitive ( TS ) collection ( Li et al . , 2011 ) and the heterozygous diploid deletion magic marker collection ( Pan et al . , 2004 ) , as follows: Genes with 1:1 orthology between yeast and E . coli were obtained from the Inparanoid 8 webserver ( Sonnhammer and Östlund , 2015 ) and filtered to an only yeast-essential set . Orthologs to these selected yeast genes in human and Arabidopsis were downloaded from Inparanoid 8 and further refined by comparison to orthology calculations by eggNOG4 . 5 ( Huerta-Cepas et al . , 2016 ) , OMA ( Altenhoff et al . , 2015 ) , and reference to the evolutionary history of the heme pathway in photosynthetic organisms ( Oborník and Green , 2005 ) . The predictive power of each feature was calculated as the area under the receiver-operator characteristic curve ( AUC ) while treating each feature as an individual classifier . Each feature was sorted in both ascending and descending directions , retaining the direction providing an AUC > 0 . 5 . To assess significance , a shuffling procedure was performed as follows: For each feature , the replaceable/non-replaceable status of each ortholog pair was shuffled ( retaining the original ratio of replaceable to non-replaceable assignments ) , and the AUC was calculated . The shuffling procedure was carried out 1000 times for each feature , and the mean AUC values and their standard deviations are reported . A Random Forest classifier was constructed using all features and evaluated using 10-fold cross-validation . The random forest was constructed to have no maximum tree depth , and ties between similarly good attributes were broken randomly . The combined classifier was implemented using the Weka data-mining software ( Frank et al . , 2004 ) . Yeast cultures expressing GFP-tagged bacterial , plant , or human genes were grown to an optical density ( OD ) of ~1 , then 500 μl of the culture washed with 1X PBS , and mitochondria fluorescently labeled by adding 100 nM MitoTracker Red CMXRos ( Invitrogen ) . The cells were incubated in the dark on a mildly shaking platform for 20 min at room temperature , then washed twice with 1X PBS and resuspended in 15 μl of 1X PBS for imaging by confocal microscopy , using a Zeiss LSM 710 confocal microscope with a Plan-Apochromat 63×/1 . 4 oil-immersion objective . Yeast strains were either pre-cultured in liquid YPD or -Ura Dextrose selective medium for 2 hr or overnight respectively . The culture was diluted in YPD or -Ura Dextrose medium to an OD of ~0 . 1 in 100 or 150 μl total volume in a 96-well plate . Plates were incubated in a Synergy H1 shaking incubating spectrophotometer ( BioTek ) , measuring the optical density every 15 min over 48 hr . Growth curves were performed in triplicate for each strain by splitting the pre-culture into three independent cultures for each 48–60 hr time course . Bacterialized Ec-hemH yeast strains were grown on YPD as lawns or large patches for 5 days ( the phenotype manifests after several days of growth ) . Clumps of cells about 5–7 mm in diameter were collected with a toothpick and first suspended in water , then pelleted at 15 , 000 g for 30 s . This created a distinctive pale yellow yeast pellet , with the red pigment appearing in a small clump on top . The water was removed while carefully avoiding disruption of the red pigment pellet , after which we performed extractions with two different methods . The first method , based on Bassel et al . ( Bassel et al . , 1975 ) , was to add 1 ml pyridine to each pellet , spinning down at 15 , 000 g for 30 s and recovering only the liquid fraction ( cell debris would pellet down while the red pigment migrated into the liquid pyridine phase ) . The second referred to as ‘acetate extraction’ in this text , was to extract with a 3:1 ethyl acetate:glacial acetic acid solution as described in Pretlow and Sherman ( 1967 ) . We then measured the absorbance of the extractions in a transparent plastic 96-well plate on the ( Synergy H1 from BioTek ) on wavelengths from 223 nm to 998 nm , with 1 nm steps . We measured fluorescence on the same instrument by exciting at 399 nm and measuring emission at 450 nm to 699 nm with 1 nm step . The spectra were compared with those shown in Bark et al . ( 2010 ) . We also obtained protoporphyrin IX ( Sigma-Aldrich , P8293-1G ) and hemin B ( Sigma-Aldrich , 51280–1G ) and suspended these in acetate and pyridine to closely resemble the chemistry of our extractions . These solutions were measured alongside the extractions themselves as standards , in order to further confirm the identity of the molecules we detected . Genomic editing and replacement of yeast ORFs is described in greater detail at Bio-protocol ( Akhmetov et al . , 2018 ) . Using CRISPR , we deleted the Sc-HEM14 ORF in wild type BY4741 , Sc-hem15Δ::Ec-HemH , and Sc-hem15Δ::Ec-MLS-HemH strains . Specifically , we co-transformed the plasmid expressing Cas9 and gRNA targeting the yeast Sc-HEM14 gene with a 200 bp oligonucleotide repair template comprising 100 bp each of sequence matching the 5' and 3' UTRs of the Sc-HEM14 gene and selected for growth on SD-Ura medium . The resulting hem14Δ strains were confirmed by PCR using primers outside the region of homology . Supplementary file 3 provides relevant primers and oligos .	All life on Earth – from bacteria to human beings – can be traced back to a common ancestor that lived over three billion years ago . As a result , modern-day organisms share many essential parts of life’s molecular machinery , such as certain genes and proteins . Yet there are also vital differences that allow scientists to divide almost all living things into one of two groups , known as prokaryotes and eukaryotes . Prokaryotes are all simple , single-celled organisms , such as bacteria; while eukaryotes include more complex organisms , such as plants , animals and fungi . Scientists have previously found that eukaryotes and prokaryotes have hundreds of genes in common , even though they have evolved separately for over two billion years . As different species evolve , however , their genes mutate and change , potentially affecting the way they work . So , although scientists can recognize equivalent genes between species , they are not sure if they work the same way as they did in the species’ ancient ancestors . To investigate this , one-by-one Kachroo , Laurent et al . replaced over 50 genes in baker’s yeast ( a eukaryote ) with their equivalent gene from E . coli bacteria ( a prokaryote ) . If the yeast cells grow healthily after the gene is replaced , it means that that gene works in a similar way in both bacteria and yeast . That , in turn , suggests it is likely that the genes work as they did in the last common ancestor of bacteria and yeast . The experiments found that most of the tested E . coli genes ( 61% to be precise ) could successfully replace equivalent genes in yeast cells . Moreover , genes often work together in groups , and Kachroo , Laurent et al . found that genes in some groups were more successfully replaced than others . For example , nearly every gene that is important for producing a molecule called heme could be freely swapped from bacteria , plants and humans into yeast . This group of genes has probably worked the same way in different species for billions of years . Understanding why genes sometimes change how they work is an important question for scientists studying evolution , but this knowledge has other uses . For example , people need heme to , amongst other things , carry oxygen in their blood , and a mutation in a gene in the heme production pathway causes a disease called porphyria . Scientists could replace genes in yeast cells to better model the disease in humans , leading to a better understanding of its causes and more efficient development of new drugs .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology", "computational", "and", "systems", "biology" ]	2017	Systematic bacterialization of yeast genes identifies a near-universally swappable pathway
Phase transitions of linear multivalent proteins control the reversible formation of many intracellular membraneless bodies . Specific non-covalent crosslinks involving domains/motifs lead to system-spanning networks referred to as gels . Gelation transitions can occur with or without phase separation . In gelation driven by phase separation multivalent proteins and their ligands condense into dense droplets , and gels form within droplets . System spanning networks can also form without a condensation or demixing of proteins into droplets . Gelation driven by phase separation requires lower protein concentrations , and seems to be the biologically preferred mechanism for forming membraneless bodies . Here , we use coarse-grained computer simulations and the theory of associative polymers to uncover the physical properties of intrinsically disordered linkers that determine the extent to which gelation of linear multivalent proteins is driven by phase separation . Our findings are relevant for understanding how sequence-encoded information in disordered linkers influences phase transitions of multivalent proteins . There is growing interest in intracellular phase transitions that lead to the formation of membraneless bodies that are collectively known as biomolecular condensates ( Banani et al . , 2017; Shin and Brangwynne , 2017 ) . These are two- or three-dimensional assemblies that comprise of multiple proteins and RNA molecules and lack a surrounding membrane . Biomolecular condensates are associated with a range of cellular functions including cell signaling ( Su et al . , 2016 ) , ribosomal biogenesis ( Feric et al . , 2016; Zhu and Brangwynne , 2015; Mitrea et al . , 2016 ) , cytoskeletal regulation ( Li et al . , 2012; Banjade and Rosen , 2014 ) , stress response ( Parry et al . , 2014; Munder et al . , 2016; Ramaswami et al . , 2013; Riback et al . , 2017 ) , cell polarization ( Saha et al . , 2016; Nott et al . , 2015 ) , and cytoplasmic branching ( Lee et al . , 2015 ) . It has been proposed that the protein components of biomolecular condensates can be classified as scaffolds versus clients ( Banani et al . , 2017; Banani et al . , 2016 ) . Scaffolds are thought to drive phase transitions , whereas client molecules preferentially partition from the cytoplasm or nucleoplasm into condensates ( Banani et al . , 2016; Wheeler et al . , 2016 ) . Scaffold proteins that drive phase transitions have distinct features , the most prominent being multivalency of folded domains or Short Linear amino acid Motifs ( SLiMs ) ( Banani et al . , 2017; Li et al . , 2012; Brangwynne et al . , 2015; Csizmok et al . , 2016; Kato et al . , 2012 ) . Valency quantifies the number of interaction domains or SLiMs . Ligands of multivalent proteins can be other multivalent proteins or polynucleotides . The simplest multivalent proteins are linear polymers that consist of multiple protein-protein/protein nucleic acid interaction domains or SLiMs connected by intrinsically disordered linkers that lack specific interaction motifs ( Figure 1a ) . Linear multivalent proteins may be classified as associative polymers ( Tanaka , 2011; Semenov and Rubinstein , 1998 ) , with specific intra- and intermolecular associations being mediated by non-covalent interactions amongst domains or motifs . Unlike generic homopolymers where the interactions are isotropic , uniform , and typically short-range ( Flory , 1942a; Flory , 1974 ) , the interactions involving associative polymers span a range of length scales and can be directional in nature ( Brangwynne et al . , 2015; Tanaka , 2011 ) . This includes a hierarchy of so-called weakly polar interactions involving charges , dipoles , and quadrupoles ( Nott et al . , 2015; Brangwynne et al . , 2015; Burley and Petsko , 1988; Brady et al . , 2017; Lin et al . , 2016 ) , hydrogen bonds , screened charge-charge interactions ( Pak et al . , 2016 ) , and hydration-mediated interactions ( Boeynaems et al . , 2017; Schneider et al . , 2002; Pochan et al . , 2003 ) . This hierarchy of interactions will enable non-covalent interactions known as physical crosslinks that involve associative domains/motifs that enable the formation of system-spanning networks known as gels ( Semenov and Rubinstein , 1998; Schneider et al . , 2002; Pochan et al . , 2003; Rubinstein and Colby , 2003 ) . Associative polymers can undergo two types of reversible gelation transitions . These are gelation without phase separation or gelation driven by phase separation ( Tanaka , 2011; Semenov and Rubinstein , 1998; Rubinstein and Semenov , 1998 ) . Our work focuses on the differences between the distinct gelation transitions and the molecular determinants of these differences in linear multivalent proteins . Gelation without phase separation refers to a switch from a solution of dispersed monomers and oligomers – a sol – to a system-spanning network – a gel ( Figure 1b ) . This networking transition is characterized by a concentration threshold , known as the percolation threshold ( Broadbent and Hammersley , 1957 ) that defines the gel point ( Flory , 1941 , 1942b; Stockmayer , 1943 ) . If the bulk concentration of associative domains/motifs is below the gel point , then the multivalent proteins form a sol . For concentrations above the gel point , the multivalent proteins are incorporated into a system-spanning network known as a physical gel . Physical gels ( referred to hereafter as gels ) are defined by specific , reversible non-covalent interactions , that represent physical crosslinks between protein modules/SLiMs and their ligands ( Su et al . , 2016; Tanaka , 2011; Falkenberg et al . , 2013 ) . Therefore , a gel is a percolated network characterized by system-spanning reversible physical crosslinks . Accordingly , the average extent of crosslinking will determine the network structure including the free volume or porosity , and average stiffness of the gel . Conversely , the timescales for making and breaking crosslinks will determine the rheological properties of gels ( Tanaka , 2011 ) . This definition of a gel , which is based on Flory’s work ( Flory , 1974 ) , is also consistent with criteria outlined by Almdal et al . ( Almdal et al . , 1993 ) . It is important to clarify that our definition of a gel does not conflate gels with solids nor does it suggest that gels have to be pathological states of matter . Polymer solutions can also undergo phase separation ( Brangwynne et al . , 2015; Flory , 1942a; Pak et al . , 2016; Huggins , 1942 ) . Above a saturation concentration , the polymer solution will undergo phase separation by separating into a dense polymer-rich phase that coexists with a dilute liquid that is deficient in polymers ( Flory , 1942a; Huggins , 1942 ) . The formation of two coexisting phases characterized by phase separation represents a condensation or density transition , with the dense phases forming spherical droplets ( Figure 1c ) . Given the three-way interplay among polymer-solvent , solvent-solvent , and polymer-polymer interactions , a necessary condition for phase separation is that inter-polymer attractions are more favorable , on average , when compared all other interactions ( Brangwynne et al . , 2015; Tanaka , 2011; Rubinstein and Colby , 2003 ) . Interestingly , phase separation of associative polymers such as linear multivalent proteins will promote gelation if the concentration of interaction domains within the dense phase is above the gel point ( Figure 1c ) . To understand this conceptually , we shall denote cg as the gel point whilst csl and csh will respectively denote the saturation concentrations of the coexisting dilute and dense phases that result from phase separation . For gelation without phase separation the gel point lies below the saturation concentration for phase separation ( cg >csl ) . In contrast , if csl <cg < csh the gel point lies below the saturation concentration for phase separation and the concentration within the coexisting dense droplet is above the gel point . In this scenario , the system will undergo gelation driven by phase separation thus resulting in droplet-spanning networks . What are the molecular determinants of gelation with and without phase separation ? We answer this question by focusing on linear multivalent proteins with folded domains interspersed by disordered linkers . Specifically , using computer simulations and theoretical analysis we show that for linear multivalent proteins of fixed binding affinity between modules and valence , the disordered linkers determine the preference for gelation driven by phase separation as opposed to gelation without phase separation . This behavior is determined by the sequence-specific properties of linkers , which can be quantified in terms of a single parameter known as the effective solvation volume ( ves ) . The effective solvation volumes reflect the average volumes occupied by linkers , referenced to the volume occupied if that linker lacked a bias to be well solvated or poorly solvated ( Rubinstein and Colby , 2003 ) . When residues prefer to interact with solvent several additional layers of solvent effectively bloat them , and so the linker becomes expanded . When residues prefer to interact with other residues they have less volume for the solvent , and so the linker becomes compact . The effective solvation volume ( ves ) of a linker can be pictured in terms of the impact that a linker has on bringing together interaction modules that are connected to either end ( see Figure 2 ) . Qualitatively , we can think about this in terms of a hypothetical outwards force that acts on the two interaction modules at either end of the linker . When ves is positive , the linker is highly expanded and this outwards force repels the two interaction modules , driving them apart . A positive ves is realized because the linker is self-repelling , carving for itself a large volume in space for favorable interactions with the solvent . When ves is negative , the linker is compact , and the hypothetical force pulls the two interaction modules inward , driving them close together . A negative ves is realized because the solvent is squeezed out , the linker is self-attractive , and this causes the interaction domains to be pulled towards one-another . When ves is close to zero , the linker does not have strong interaction preferences and mimics a passive tether . Accordingly , both expanded and compact linker conformations are equally likely . The hypothetical outwards/inward force is negligible – the preferences for compact versus expanded conformations cancel one another – and the interaction modules meander around in three-dimensional space with respect to one another , restrained only the connectivity of the linker . A value of ves ≈ 0 is realized due to a counterbalancing of attractive and repulsive interactions in the linker . Formally , the effective solvation volume of a linker is quantified in terms of the solvent-mediated pairwise interactions between pairs of linker residues and the details are discussed in Appendix 1 . If the linker sequence is such that there are net attractions between all pairs of residues , then ves will be negative and this will be true for linkers that form compact globules . Conversely , if there are net repulsions between all pairs of residues , then the residues prefer to be solvated and ves will be positive . This is the case for so-called self-avoiding random coil ( SARC ) linkers . Finally , if the effects of inter-residue attractions offset the effects of inter-residue repulsions , then ves ≈ 0 and this is the scenario for so-called Flory random coil ( FRC ) linkers . The effective solvation volume is directly proportional to the second virial coefficient denoted as B2 ( 32 , 41 ) . Negative , zero , or positive values of ves correspondingly imply negative ( attractive interactions ) , zero ( non-interacting ) , or positive ( repulsive interactions ) values of B2 . Therefore , ves can be inferred using either atomistic simulations ( as shown in this work ) or via measurements of B2 as shown by Wei et al . ( Wei et al . , 2017 ) . For generic homopolymers , the sign and magnitude of ves are determined by the effective chain-solvent interactions , which in turn depend on the chemical makeup of the chain . For proteins , the interplay between chain-chain and chain-solvent interactions is specified by the amino acid sequence , whereby the composition and patterning of a disordered linker will determine the balance of chain-chain and chain-solvent interactions ( Das et al . , 2015; Holehouse et al . , 2017; Martin et al . , 2016 ) . Therefore , the effective solvation volume of a disordered linker is determined directly by its primary sequence . To set the stage for our investigations , we first performed proteome-wide bioinformatics analysis combined with all-atom simulations to quantify conformational consequences of sequence-specific effective solvation volumes of disordered linkers in naturally occurring multi-domain human proteins . This analysis shows that the sub-proteome of linear multivalent proteins comprises of linkers of varying lengths that span a range of effective solvation volumes , from significantly negative to significantly positive values . Using coarse-grained numerical simulations and analytical theories we then show that the type of gelation transitions that linear multivalent proteins undergo is directly determined by the physical properties of linkers , which include the lengths of linkers and their sequence-specific effective solvation volumes . We first sought to obtain accurate and efficient estimates of the effective solvation volume ( ves ) for a large set of disordered segments . For this we used all-atom simulations , which have a proven track record of describing sequence-specific conformational properties of intrinsically disordered proteins ( Das et al . , 2015; Martin et al . , 2016; Vitalis and Pappu , 2009a2009; Das et al . , 2016 ) . Although a formal and rigorous calculation of ves is technically possible using these simulations , this approach is computationally expensive and non-trivial for large numbers of sequences . Recognizing that the effective solvation volume directly determines the global dimensions of a linker , we used the ensemble-averaged conformational properties to calculate a proxy for ves ( Mao et al . , 2013 ) . Specifically , we leverage the profile of inter-residue distances to determine how a given linker sequence deviates from a sequence-specific theoretical reference that recapitulates ves = 0 , which is the Flory Random Coil ( FRC ) ( Holehouse et al . , 2015 ) . These profiles ( Figure 3a ) describe the average spatial separation between all pairs of residues as a function of their separation along the polypeptide sequence . We obtained sequence-specific inter-residue distance profiles by performing all-atom Metropolis Monte Carlo simulations using the ABSINTH implicit solvent model and forcefield paradigm ( Vitalis and Pappu , 2009b ) as described in the methods section . Figure 3a shows the calculated inter-residue distance profiles for fourteen distinct sequences , each of length 40 residues . Details of the sequences are shown in ( Table 1 ) . Figure 3a illustrates changes to the inter-residue distance profiles as a function of changes to the fraction of charged residues . Figure 3a also shows the inter-residue distance profile for a reference FRC linker . Sequences with positive ves will have inter-residue distance profiles that lie above the FRC reference . Conversely , sequences with negative ves will have profiles with uniformly smaller inter-residue spatial separations for given sequence separations when compared to the FRC reference . Accordingly , Figure 3a shows that sequences deficient in charged residues are expected to have negative ves values , whereas sequences enriched in charges are expected to have positive ves values . Since inter-residue distance profiles are direct manifestations of sequence-specific effective solvation volumes ( Mao et al . , 2013 ) , we use these profiles to calculate a parameter ∆ that serves as a proxy for sequence-specific ves values . This parameter is defined as the mean signed difference between the sequence-specific inter-residue distance profile and the corresponding profile for a FRC reference . In Figure 3b we plot the calculated ∆ values against the fraction of charged residues for the fourteen disordered sequences from Figure 3a . The value of ∆ can be negative , equal to zero , or positive and this depends on whether the value of ves is negative , zero , or positive , respectively . Sequences that form compact globules have negative values of ves and negative values of ∆ . For the sequences examined , this is true when the fractions of charged residues is below 0 . 3 . Within an interval between 0 . 3 and 0 . 5 for the fraction of charged residues , sequences mimic the FRC limit , where ves ≈ 0 . This is manifest as –0 . 1 ≤ ∆≤0 . 1 . Sequences that prefer chain-solvent interactions to intra-chain interactions will be expanded relative to the FRC limit . This leads to positive values of ves and corresponds to values of ∆ that are greater than 0 . 1 . We extended our analysis of sequence-specific effective solvation volumes to naturally occurring disordered linkers in multi-domain proteins within the non-redundant human proteome . Using a stringent set of criteria ( see Materials and methods section ) we identified approximately 100 linear multivalent proteins from the non-redundant human proteome ( 20 , 162 sequences ) and extracted 226 unique linker regions ( see Materials and methods for details ) . For each of the 226 linkers we performed all-atom simulations to quantify the sequence-specific values of ∆ . The 226 sequences span a range of lengths ( Figure 3c ) . We calculated the distribution of ∆ values for all linkers using results from all-atom simulations ( Figure 3d ) . This distribution shows that sequences of naturally occurring disordered linkers span the entire range of ∆ values . Of the 226 unique linker sequences , approximately 30% have negative effective solvation volumes ( ∆ < –0 . 1 ) whereas 38% have sequences defined by ∆ values in the range –0 . 1 ≤ ∆≤0 . 1 , implying that they will have near zero effective solvation volumes and are mimics of FRC linkers . Finally , 30% of linkers are characterized by ∆ values greater than 0 . 1 , which means that their effective solvation volumes are positive . The limiting form of a positive effective solvation volume linker is the self-avoiding random coil or SARC for which ∆ ≈ 0 . 5 . The key finding is that disordered linkers come in a range of sequence flavors , and 68% have a positive or near positive effective solvation volume . Supplementary file 1 summarizes key details regarding the naturally occurring linkers , including the protein name , UniProt identifier ( Finn et al . , 2014 ) , the value of ∆ , and Gene Ontology ( GO ) annotations . The linkers are derived from multivalent proteins associated with a range of different functions . The proteins we identified were significantly enriched for RNA/DNA binding and RNA localization , as assessed by PANTHER-GO enrichment analysis ( Mi et al . , 2017 ) ( p<0 . 005 ) . This is of particular relevance , given that many micron-sized biomolecular condensates contain protein and RNA molecules ( Banani et al . , 2017 ) . With this analysis in hand , our next goal was to understand how different types of linkers might modulate the gelation transitions and overall phase behavior of linear multivalent proteins . For linkers with negative effective solvation volumes the linkers serve as additional drivers of phase separation ( Crick et al . , 2013 ) . These attractive linkers should be thought of as separate interaction domains and are hence distinct from regions that modulate the phase behavior of interaction domains . Therefore , we focused our studies on disordered linkers with near zero or positive effective solvation volumes ( ves ≥ 0 ) . Numerical simulations of phase transitions require the inclusion of hundreds to thousands of distinct multivalent proteins and a titration of a spectrum of protein concentrations . Furthermore , phase transitions are characterized by sharp changes to a small number of parameters , and the observation of these sharp transitions is computationally intractable with all-atom simulations . Therefore , we developed and deployed coarse-grained lattice models to study the impact of linkers on phase transitions . Parameters of the lattice models are summarized in Table 1 of the Materials and methods section . Lattice models afford the advantage of a discretized conformational search space ( Feric et al . , 2016 ) . This enables significant enhancements in computational efficiency . Key features of lattice models are the mapping of real protein architectures onto lattices and the design of an interaction model ( Feric et al . , 2016 ) . The design of our simulation setup was inspired by the synthetic poly-SH3 and poly-PRM system studied by Li et al ( Li et al . , 2012 ) . The general framework of our lattice model has been extended to other systems including branched multivalent proteins ( Feric et al . , 2016 ) , and is transferable through phenomenological or machine learning approaches ( Ruff et al . , 2015 ) to any system of multivalent proteins and polynucleotides We modeled each multivalent poly-SH3 and poly-PRM protein using a coarse-grained bead-tether model ( Figure 4 ) . A single lattice site was assigned to each SH3 domain . This sets the fundamental length scale in our simulations . Each PRM comprises of approximately 10-residues , thus giving it the approximate dimensions of a single SH3 domain . Therefore , each PRM was also assigned to a single lattice site . Previous all-atom simulations showed that the spatial dimensions of a single SH3 domain corresponds to ~7 linker residues , if ves ≥ 0 ( 54 ) . Therefore , the linker length can be written as N ≈ 7 n where n is the number of lattice sites that span the linker and N is the number of linker residues . All simulations were performed on 3-dimensional cubic lattices with periodic boundary conditions . Individual SH3 domains and PRMs can bind to one another and form a 1:1 complex with an intrinsic binding energy of –2kBT . Here , kB is Boltzmann’s constant and T is temperature . This intrinsic affinity reproduces measured dissociation constants for SH3 domains and PRMs ( Li et al . , 2012 ) . We start with two stylized linkers namely , Flory random coil ( FRC ) linkers and the self-avoiding random coil ( SARC ) linkers . FRC linkers correspond to chains with ves = 0 . We model FRC linkers as implicit linkers ( Figure 3a ) – the linkers have a fixed length and tether the domains together , but do not occupy any volume on the lattice . Practically this is realized by imposing a cubic infinite square well potential to ensure that the lattice spacing between tethered interaction domains does not exceed n , which is the linker length in terms of the number of lattice sites . For the SARC linkers with positive ves , we use explicit linkers as shown in Figure 3b . A SARC linker of length n has n beads , where each bead is constrained to occupy vertices adjacent to its nearest neighbor beads on the lattice . Each explicitly modeled linker bead occupies a finite volume corresponding to one lattice site . Phase separation results from a change in density . We quantify a parameter ρ , which we define as the ratio of Rlattice to Rgproteins . Here , Rlattice is the radius that we would obtain if all proteins were uniformly dispersed across the lattice ( Figure 5 ) . Conversely , Rgproteinsis the actual ensemble-averaged radius of gyration over the spatial dimensions of the SH3 , PRM , and linker beads ( Figure 5 ) . For a system that has undergone phase separation , the parameter ρ will be >1 . ρ is directly related to the relative density of the proteins and measures the extent of spatial clustering of domains and linker residues . If ρ is equal to one , then the proteins are uniformly dispersed through the lattice . We quantify gelation in terms of the fraction of molecules in the system that are part the single largest cluster . This is denoted as ϕc ( Figure 5 ) . We analyze each configuration of multivalent proteins to detect the formation of connected clusters . Within each configuration , each molecule is a node . An edge is drawn between two nodes if an SH3 domain from one molecule interacts with a PRM from another molecule . The connected cluster with the largest number of nodes is designated as the largest cluster and the number of molecules corresponding to this cluster is recorded . This quantity is calculated across the entire ensemble of configurations in order to generate an ensemble averaged value of ϕc for the system of interest . As a result of the finite surface tension associated with droplet formation and the precautions taken to reach convergence ( see Materials and methods ) , we find that the single largest cluster absorbs all other clusters , thus giving rise to a true two-phase system as pictured in Figure 5a . We performed a series of Monte Carlo simulations using a coarse-grained lattice model for poly-SH3 and poly-PRM systems of valence 3 , 5 , and 7 and all combinations of these valencies . Unless otherwise specified , in all of our simulations , the linker length n was set to five lattice sites , approximately 35 residues . This linker length corresponds to the main mode in the distribution of linker lengths shown in Figure 3c . The first row of plots in Figure 6 shows how ϕc changes for different simulated systems and provides a quantification of gelation . Each sub-plot in Figure 6a shows the value of ϕc as a function of the concentrations of SH3 domains and PRMs for a particular combination of PRM and SH3 domain valence . Figure 6a establishes two distinctive features of multivalent systems: For a given combination of SH3 and PRM valencies , we observe a sharp increase in the values of ϕc as the concentrations of SH3 domains and PRMs increase . This behavior is consistent with the expected features of a sol-gel transition . Second , as valence increases , there is a lowering of the module concentrations at which ϕc increases sharply . Figure 6b shows results for ϕc obtained for poly-SH3 and poly-PRM systems with SARC linkers . Here , five beads were modeled explicitly for each of the linkers between SH3 domains and PRMs . Although most systems show a sharp increase in ϕc past a threshold SH3/PRM concentration , the concentrations at which the transitions are realized are at least an order of magnitude higher than those observed for the systems with FRC linkers . The differences between FRC and SARC linkers are summarized in Figure 6c , which shows how ϕc changes with module concentrations for the symmetric 3:3 , 5:5 , and 7:7 systems along the diagonals for equal ratios of SH3 domains and PRMs . The x: y designation refers to the valence of SH3 domains: the valence of PRMs . The value of ϕc changes sharply with concentration and this change becomes sharper as the valence increases . For a given valence , ϕc increases more sharply and this sharp change happens at lower module concentrations for proteins with FRC as opposed to SARC linkers . This analysis shows that the effective solvation volumes of linkers can have a profound impact on sol-gel transitions . The bottom row in Figure 6 shows how ρ changes for each of the multivalent systems and provides a quantification of phase separation . Figure 6d , which summarizes the results for FRC linkers , shows sharp changes to ρ as valence increases . This recapitulates the observations in Figure 6a for ϕc indicating that changes to connectivity are coupled to changes in density . This is illustrated in plots for the 7:7 , 7:5 , 5:7 , and 5:5 systems . In contrast , the 5:3 , 3:5 , and 3:3 systems show gelation transitions with negligible changes to ρ . In the highly asymmetric 7:3 and 3:7 systems , the changes in ρ are considerably less pronounced when compared to changes in ϕc . In each simulation , the initial conditions correspond to the multivalent proteins being randomly dispersed across the cubic lattice ( see Video 1 ) . The movie and comparative analysis of results in Figure 6a and d provide visual support for the suggestion that systems with FRC linkers undergo phase separation plus gelation . Figure 6e shows the results obtained for poly-SH3 and poly-PRM systems with SARC linkers . The results provide a striking contrast to the results obtained for proteins with FRC linkers ( see Video 2 ) . None of the systems show discernible changes to ρ . This implies that gelation occurs only when the concentrations are large enough to enable networking through random encounters . The positive effective solvation volumes of SARC linkers suppress phase separation and these systems undergo gelation without phase separation . Figure 6f summarizes the distinctions between FRC and SARC linkers by plotting ρ versus the concentration of modules for the symmetric cases with equal ratios of SH3 domains and PRMs . For SARC linkers , ρ ≈ 1 across the entire concentration range for ( solid curves ) . This emphasizes the suppression of phase separation for systems with SARC linkers . For proteins with FRC linkers , the values of ρ increase sharply above unity beyond system-specific critical concentrations . Representative post-equilibration configurations for 7:7 systems with FRC and SARC linkers of length five are shown in Figure 7 . Both snapshots correspond to values of ϕc being above the gel point . The bounding box corresponds to the volume of the simulation cell and provides perspective regarding the change in density and networking within the system . In Figure 7a , a dense ( high ρ ) spherical droplet , which is a gel ( ϕc is above the percolation threshold ) , coexists with a dilute sol of well-dispersed proteins . In contrast , Figure 7b shows how a system spanning network , that is , gelation occurs in the absence of phase separation . If the linkers are short , then irrespective of the effective solvation volume , the formation of a physical crosslink between a pair of multivalent proteins will increase the probability that a second crosslink can form between the same pair of proteins . In this scenario , there is positive local cooperativity , in that the apparent affinities will increase ( Jencks , 1981 ) but the network cannot grow because the apparent valence is lower than the actual valence . In the limit of positive local cooperativity , phase separation and gelation are suppressed because collective interactions amongst the molecules are weakened in favor of forming network terminating dimers and oligomers . This scenario corresponds to infinite negative global cooperativity . In this scenario , there will neither be gelation nor phase separation . It the linkers go beyond a system-specific length , the domains will become independent of one another . Here , the extent of crosslinking and the gel point are determined entirely by the valence of domains and the intrinsic affinities between domains . This is the limit of classical Flory-Stockmayer theories ( Flory , 1941; Flory , 1942b; Stockmayer , 1943 ) with zero local cooperativity . The linkers are passive tethers that generate multivalency , but they do not make any other contributions to the transitions of multivalent systems . In the limit of zero local cooperativity , gelation occurs without phase separation and the apparent valence equals the actual number of domains , implying zero global cooperativity . For intermediate linker lengths , the signs and magnitudes of the effective solvation volumes of linkers will determine the overall phase behavior . Disordered linkers with negative or near zero ves values can enable phase transitions characterized by positive global cooperativity because they can drive density transitions of multivalent proteins . These linkers can be confined to small volumes , when compared to the volume of the entire system . This derives from the preference for chain-chain interactions ( ves < 0 ) or indifference for chain-chain versus chain-solvent interactions ( ves ≈ 0 ) . Increased concentrations of domains within confined volumes realized by density transitions will enable networking transitions because the gel point is now lower than the concentration of domains within the dense phase . If a multivalent protein contributes to growth of a network by forming a crosslink with a free domain on a protein that has already formed a crosslink with another protein , then the increased crosslinking enables gelation . These collective effects can also increase the apparent affinities between domains ( as in the first scenario ) thereby increasing the concentration of interaction domains . Increased crosslinking enables a networking transition whereas increased concentration of domains enables a density transition . The regime of positive global cooperativity corresponds to the regime where gelation is driven by phase separation . Linear multivalent proteins with large positive effective solvation volume linkers ( ves >> 0 ) will engender negative global cooperativity because the linkers prefer to be solvated and will resist confinement within droplets . In this sense , linkers with large positive effective solvation volumes are analogous to solubilizing tags . Additionally , due to their large positive effective solvation volumes , the linkers act as obstacles that inhibit interactions between domains . These linkers decrease the apparent affinity between interaction domains and reduce the degree of crosslinking . Accordingly , the ability to concentrate multivalent proteins is weakened , and so is the ability to grow a system-spanning network via a connectivity transition . In the scenario of negative global cooperativity , phase separation is suppressed and gelation is realized at bulk concentrations that are considerably higher than the Flory-Stockmayer limit . As a reminder , linkers do not make any contribution to determining the gel point in the Flory-Stockmayer limit ( Flory , 1974 , 1941 , 1942b; Stockmayer , 1943 ) , only the valence and intrinsic affinities matter . To summarize , gelation driven by phase separation will lead to positive global cooperativity , and enable the formation of a percolated network at bulk concentrations that are considerably smaller than the Flory-Stockmayer limit . Systems with zero or negative global cooperativity undergo gelation without phase separation and sol-gel transitions occur at or above the Flory-Stockmayer limit . To put the ideas described above on a quantitative footing and enable comparisons across different systems we calculated the percolation threshold in terms of ϕc , and we designate this as ϕcc . We then use the value of ϕcc to quantify the gel point cg . The gel point is the concentration threshold beyond which the system crosses the percolation threshold . The methods for computing ϕcc for a system with prescribed values for the valence and the binding energy between interaction domains , as well as the calculation of the gel point from ϕcc , are described in the methods section . We introduced a dimensionless parameter c* to quantify the magnitude and type of cooperativity that characterizes phase transitions of linear multivalent proteins . The parameter c* is defined as the ratio of cg , sim to cg , FS , that is , c* = ( cg , sim/cg , FS ) . Here , cg , sim is the gel point quantified in simulations with linkers of specified length and effective solvation volume . It is defined as the lowest concentration of modules at which ϕc>0 . 17 . This is the percolation threshold for our system of finite-sized linear multivalent proteins ( see Materials and methods section ) . In contrast , cg , FS is the gel point obtained from Flory-Stockmayer theories ( Flory , 1974 , 1941 , 1942b; Stockmayer , 1943 ) . Therefore , the value of cg , FS provides an important touchstone for quantifying the influence of linkers on phase transitions , and provides a measure of the deviation from the mean-field behavior expected of long inert linkers . For a system with positive global cooperativity , c<1; for a system with zero global cooperativity , c=1; and for a system with negative global cooperativity , c>1 . The value of c quantifies the joint effects on changes to the apparent affinities of interaction modules and the extent of crosslinking . We quantified the impact of linker lengths on the degree and magnitude of cooperativity for FRC linkers . Figure 8a shows a plot of c* as a function of linker lengths for 3:3 , 5:5 , and 7:7 systems with FRC linkers . The profile of c* is non-monotonic . In the short linker limit ( n ≤ 2 ) the value of c* is greater than one . These linkers are too short and therefore complexes terminate in dimers of poly-SH3 and poly-PRM proteins . This is the regime of positive local and negative global cooperativity where phase transitions do not occur . For multivalent proteins with a valance of 5 or 7 and linker lengths in the range 3 ≤ n < 12 ( or 21 ≤ N ≤ 84 , where N is the number of linker residues ) , the value of c* is less than one , and the lowest values of c* are realized for linkers of length 3 < n < 6 . FRC linkers within a defined length range engender positive global cooperativity and for linker lengths in this optimal range , positive global cooperativity increases with increasing valence . This is the regime where phase separation promotes gelation and c* is less than 1 . Positive global cooperativity weakens with increasing linker lengths . Accordingly , for long linker lengths , c* converges to one implying that the domains interact independently when the FRC linkers are sufficiently long . This is the regime of zero global cooperativity where gelation occurs without phase separation in accord with the predictions of Flory-Stockmayer theory ( Stockmayer , 1943 ) . Figure 8b shows a plot of c* as a function of linker lengths for 3:3 , 5:5 , and 7:7 systems with SARC linkers . Here , c* is greater than one for all the linker lengths . This is a signature of negative global cooperativity . Linkers with positive effective solvation volumes suppress phase separation and shift the gel point to higher concentrations when compared to the threshold predicted by Flory-Stockmayer theories . Explicit linkers also lower the apparent affinity through negative global cooperativity because their positive effective solvation volumes promote solvation thus diminishing productive associations among domains . This becomes less of an issue as the linkers become longer . If one corrects the intrinsic affinity to account for the weakened apparent affinity , then the convergence of the systems with long linkers to the Flory-Stockmayer limit is recovered ( not shown ) . However , the profiles do not change qualitatively and this points to fundamental differences between systems with FRC versus SARC linkers . The analysis in Figure 8 has ramifications for drawing inferences from the proteome-wide analysis summarized in Figure 3 . We find that the values of ∆ and linker length are essentially uncorrelated . This is not surprising because the main determinant of the effective solvation volumes is the sequence/amino acid composition and not the length of the linker . This point is underscored in the analysis summarized in Figure 3 . Our analysis of linker sequences in linear multivalent systems shows that approximately 30% of all linkers in the inventory will have 50 or fewer residues and ∆ values less than 0 . 1 ( Supplementary file 1 ) . These linkers are the most likely candidates for enabling gelation driven by phase separation in linear multivalent proteins . Approximately , 18% of all linkers have fewer than 50 residues and ∆ values greater than 0 . 1 . These are the most likely candidates for weakening phase separation and sequences with large positive values of ∆ will drive gelation without phase separation . The remainder of the linkers , ~50% in all , are longer than 50 residues and these are unlikely to be major modulators of gelation transitions since the analysis in Figure 8 suggests that these linkers cross into the Flory-Stockmayer limit , where the interaction modules become independent of one another . Figure 9 shows the phase diagram that we computed from concentration dependent simulations for a 5:5 system and a hybrid five-site linker . This phase diagram is shown in the two-parameter space of the concentration of domains along the abscissa and increasing intrinsic affinities along the ordinate . For affinities below 3kBT , the system undergoes a continuous transition from a sol to a gel and the green dashed line demarcates the sol-gel line . The gels correspond to system-spanning networks that percolate through the entire simulation volume . The critical point for this system , shown as a red asterisk , is defined jointly by a critical interaction affinity ( 3kBT ) and a critical module concentration ( ~10–3polymers/voxel ) . Above the critical point , the system undergoes gelation driven by phase separation . As the interaction affinity increases above 3kBT , the system separates into two coexisting phases namely , a dilute phase , which is a sol , and a dense phase , which is a gel . As an illustration , for an interaction affinity of 4 . 5kBT , the coexisting concentrations that define the two phases are designated as csl and csh , which are respectively the concentrations of dilute and dense phases . Notice that the gel point , cg , defined as the concentration beyond the percolation threshold , ϕc > 0 . 17 , lies within the two-phase regime such that csl < cg < csh . Here , cg is the apparent gel point that is extrapolated by extending the green dashed line in Figure 9 . Accordingly , the density transition , which we quantify as the concentration range above which ρ becomes greater than 1 . 08 , enables gelation because the concentration within the dense phase ( csh ) is higher than the apparent gel point ( cg ) . The result is a droplet-spanning network as pictured in the Figure 7a . The width of the two-phase regime increases with interaction affinity . This implies that phase separation is realized at lower concentrations of the interacting domains and is depicted by a leftward shift of the arm shown in light blue in Figure 9 . Concomitantly the droplet becomes more concentrated and this is depicted by a rightward shift of the arm shown in purple in Figure 9 . Therefore , if the linker sequence is fixed , mutations to interaction domains or SLiMs that increase affinity will enhance phase separation , giving rise to concentrated droplets encompassing gels that coexist with dilute sols . The effective solvation volumes of linkers were titrated by fixing the linker length and changing the number of linker beads that were modeled implicitly versus explicitly . The magnitude of the effective solvation volume is quantified in terms of the number of explicitly modeled beads within each linker . For example , if two out of five linker beads are modeled explicitly , then ves is proportional to the volume of two lattice units as is the case for linkers that yield phase diagrams shown in Figures 9 and 10c . Each of the panels in Figure 10 corresponds to a distinct type of linker , defined by the effective solvation volume , that is , the number of explicitly modeled linker beads for a linker of length five . The results are shown for interaction affinities of modules that range from 2kBT to 5kBT . Progressing from the top left corner to the bottom right corner , we find that the critical point shifts to higher interaction affinities as the effective solvation volumes of linkers increase . If the linkers have more of an FRC-like character , then the phase transitions are likely to fit the description of being gelation driven by phase separation . For a given value of the affinity , the width of the two-phase regime increases as the magnitude of the effective solvation volume decreases . In contrast , the two-phase regime becomes negligibly small as the magnitude of the linker effective solvation volume increases . In fact , for high effective solvation volumes of linkers , the presence of a two-phase regime is discernible only for very high affinities and phase transitions occur mainly via gelation without phase separation . Using numerical simulations , we show that that linear multivalent proteins can undergo two distinct types of transitions namely , gelation without phase separation and gelation driven by phase separation . We also showed that linkers between domains/motifs in linear multivalent proteins are not just passive tethers . In addition to serving as scaffolds for motifs ( Das et al . , 2016; Banjade et al . , 2015 ) , the physical properties of linkers such as their lengths and effective solvation volumes will directly influence the extent to which phase separation promotes gelation ( Semenov and Rubinstein , 1998 ) . The distinction between gelation without phase separation and gelation driven by phase separation was formalized in the theoretical work of Semenov and Rubinstein ( Semenov and Rubinstein , 1998; Rubinstein and Semenov , 1998 ) . In their mean-field model for infinitely long associative polymers , phase separation facilitates gelation for chains with negative , near zero , or mildly positive effective solvation volumes . Phase separation is suppressed as ves becomes positive and extent to which phase separation promotes gelation is modulated by the affinities between associative domains/motifs . Our numerical results summarized in Figures 6–10 are consistent with the theoretical predictions of Semenov and Rubinstein . This is gratifying given that we focus on finite-sized polymers where the simplifications of mean field theories are not necessarily transferrable . We have also shown that the effective solvation volumes of linkers are directly determined by their primary sequences ( Figure 3 ) . Finally , there appears to be an optimal range of linker lengths that supports gelation driven by phase separation for a given interaction affinity between domains . We focused our simulations of phase transitions on linkers with zero or positive ves values . However , as shown in Figure 3d , approximately 30% of linkers in the sub-proteome of linear multivalent proteins have negative ves values . These linkers will be self-attractive . They can also engage in non-specific attractive interactions with interaction domains as well as other linkers of different sequence composition that have negative ves values . Linkers with negative ves values are best thought of as additional interaction sites . Therefore , linkers with negative ves values have two distinct effects: firstly , they lead to be an effective shortening of the linker length due to linker compaction , and secondly they can engage in additional in trans interactions causing an increase in the effective valence . These effects were illustrated in a previous study that was designed to study coexisting dense phases formed by the intrinsically disordered RGG domain of the protein Fibrillarin-1 ( FIB1 ) . There , the RGG domain of FIB1 was modeled using five explicit sticky beads thus conferring an effectively negative ves value on this domain ( Feric et al . , 2016 ) . Linkers with negative ves values are likely to yield significantly more dense droplets when compared to linkers with near zero or positive ves values . This is underscored in recent measurements of intra-droplet concentrations for disordered proteins with positive ( Wei et al . , 2017 ) versus negative ves values ( Simon et al . , 2017 ) . The intra-droplet concentration for the RGG domain of LAF-1 ( 41 ) , which has a positive ves value , is two orders of magnitude smaller than the intra-droplet concentration measured for elastin-like polypeptides ( Simon et al . , 2017 ) , which have negative ves values . Interestingly , the sequences of many low complexity domains that tether RNA recognition modules in proteins such as hnRNP-A1 and FUS have negative ves values . The high density within these droplets might explain why disease-associated mutations within these sequences engender apparently pathological gelation transitions that appear to be aided by conformational changes into beta-sheet-rich fibrils ( Lee et al . , 2016; Molliex et al . , 2015; Patel et al . , 2015; Burke et al . , 2015; Conicella et al . , 2016; Weber and Brangwynne , 2012 ) . In contrast , linkers characterized by mildly negative , zero , or mildly positive ves values might form reasonably dilute droplets and functional gels that suppress pathological transitions ( Li et al . , 2012; Banjade and Rosen , 2014; Riback et al . , 2017; Banani et al . , 2016; Banjade et al . , 2015 ) . Linear multivalent proteins are associative polymers , will undergo gelation with or without phase separation . We speculate that the regulation of cell signaling by phase transitions might predominantly involve gelation driven by phase separation . This is evidenced by the formation of spherical droplets that is driven by specific multivalent proteins comprising of multiple interaction domains or linear motifs ( Su et al . , 2016; Li et al . , 2012; Banjade and Rosen , 2014; Banani et al . , 2016; Jiang et al . , 2015; Bergeron-Sandoval et al . , 2016 ) . The role of phase separation in cell signaling likely reflects the fact that the formation of dense droplets will increase local protein concentrations , facilitating a cooperative amplification in signal transduction as distinct signaling components undergo efficient intermolecular phosphorylation due to the high local concentration ( Su et al . , 2016; Hernández-Vega et al . , 2017; Woodruff et al . , 2017 ) . Sequestration of key proteins into compartments also seems to be an important biological function that is achievable via phase separation ( Shin and Brangwynne , 2017; Riback et al . , 2017 ) . Gelation within a droplet will contribute directly to the droplet sub-structure and to the spatial organization of components within droplets ( Li et al . , 2012 ) . The extent and dynamics of crosslinking within a droplet-spanning gel will directly influence the material properties of droplets ( Tanaka , 2011 ) . These properties include the void volumes , average mesh sizes , local stiffness , dimensionality of the confined space , and rheological properties such as the viscoelastic profiles of membraneless bodies ( Tanaka , 2011; Semenov and Rubinstein , 1998; Rubinstein and Colby , 2003; Rubinstein and Semenov , 1998 ) . A striking example of the functional relevance of gelation was recently reported for S-crystallin proteins that make up the refractive material of the squid lens ( Cai et al . , 2017 ) . The physics of phase separation is insufficient to explain the formation of a gradient of protein volume fractions across the lens . However , the measured features of the squid lens are readily explained using the framework of patchy colloid theory ( Bianchi et al . , 2006; Bianchi et al . , 2008 ) , whereby the polydispersity of disordered loops in S-crystallin determine the extent of physical crosslinking giving rise to gels of different densities across the lens ( Cai et al . , 2017 ) . Gelation without phase separation may also be useful in biology . Halfmann has reviewed functional scenarios where low complexity domains might undergo dynamical glass transitions that can resemble gelation without phase separation ( Halfmann , 2016 ) . The glass transitions of the inactive bacterial cytosol and the transition to ‘solid-like’ materials in fungi as a response to pH induced stresses are examples of sol-gel transitions on the whole cell level that do not have the characteristic hallmarks of accompanying phase separation of specific components ( Parry et al . , 2014; Munder et al . , 2016 ) . Phase separation without gelation requires that the concentration within the dense phase be lower than the gel point ( csl < csh < cg ) . For associative polymers , given the hierarchy of specific interactions that are encoded by the domains/motifs , it is difficult to envisage a scenario where the interactions would be strong enough to drive phase separation without the formation of physical crosslinks . While our work does not explore the dynamics associated with gelation , there are various lines of evidence that under certain scenarios the liquid-to-solid transition observed within droplet may by refectory for biological function ( Patel et al . , 2015; Mateju et al . , 2017 ) . If the formation of gels with solid-like properties is deleterious , then it is likely that active processes within the cell inhibit this transition within dense droplets , such that physical crosslinks are actively sheared ( Mateju et al . , 2017 ) . Such a scenario would be an example of a so-called active liquid ( Protter and Parker , 2016; Brangwynne et al . , 2011 ) or more precisely a non-equilibrium liquid where energy is expended to suppress or limit gelation that would accompany phase separation of multivalent proteins ( Brangwynne et al . , 2015 ) . Competitor molecules such as specific RNA sequences might also enable a shearing of percolated networks ( Lee et al . , 2015 ) , although this has not been formally proven . We further propose that effective scaffolding proteins for gelation driven by phase separation are likely to be linear multivalent proteins with linkers that have low effective solvation volumes ( ves ≈ 0 ) . Proteins with linkers that have large positive ves values are likely to be clients that partition into the droplets formed by the scaffolds ( Banani et al . , 2017 ) . The precise nature of phase transitions might be biologically tunable . For example , the effective solvation volumes of linkers in linear multivalent protein can be tuned through synergistic actions of kinases and phosphatases ( Bergeron-Sandoval et al . , 2016; Kwon et al . , 2013 ) . This will alter the fraction of charged residues along linkers thus enabling an alteration of the phase behavior by altering the effective solvation volumes of linkers . Support for this proposal comes from the observation that the substrates for multisite phosphorylation tend to be enriched in disordered regions with positive effective solvation volumes ( Holehouse et al . , 2017; Martin et al . , 2016 ) . Additionally , posttranscriptional processing of mRNA transcripts via alternative splicing can also be a route for making tissue-specific alterations to linker sequences . Interestingly , transcripts coding for disordered regions are preferentially targeted by tissue-specific splice factors when compared to transcripts for folded domains ( Buljan et al . , 2013; Buljan et al . , 2012 ) . The inventory of linker sequences , shown in Supplementary file 1 , combined with the analysis presented in our numerical simulations , provides a ready-made route to search for candidate linear multivalent proteins that drive gelation driven by phase separation plus gelation versus gelation without phase separation . Clearly , we need detailed experimental and theoretical characterization of phase diagrams of multivalent proteins , with special attention to the intersection of sol-gel lines and the two-phase regime ( Figures 9 and 10 ) . Our work opens the door to designing systems with bespoke sequence-encoded phase diagrams . The interaction matrix includes the following terms: Each interaction domain ( SH3 domain or PRM ) or explicitly modeled linker bead has a finite ves such that each lattice site may have only one domain or linker bead . All other interactions are nearest neighbor interactions such that adjacent sites x and y on the lattice are assigned an interaction energy εxy in units of kBT , where kB is Boltzmann’s constant and T is the simulation temperature . We designate lattice sites occupied by SH3 domains using the letter S; sites occupied by PRMs by the letter P; and sites occupying linker beads by the letter L . In the default model , the interaction energies have the form: uSS = uPP = uLL=uSL = uPL=0 and uSP = –2kBT . Five types of moves were deployed to evolve the system . ( i ) In addition to occupying adjacent lattice sites , two interacting domains are in a bound state if and only if this is specified by the interaction state of the domains . Accordingly , one of the moves randomly changes the interaction state of a domain without changing lattice positions . ( ii ) The torsional state of an end module that is tethered on one side is altered and a new interaction state is chosen at random . This attempts to move the module to a new location that is within tethering range of the linker , which is the maximum allowable length for the linker . If the module is an interaction domain , then this move also changes the interaction state of the domain similar to move 1 . ( iii ) Crankshaft motions are applied to modules tethered on both sides . The module is moved to a new location that is within tethering range of all linkers that connect to the module in question . This is followed by randomly choosing a new interaction state if the module is an interaction domain . ( iv ) This move involves the collective translation of all modules that are part of a connected network . The latter is calculated by analyzing the list of all proteins that are connected through interacting domains . An arbitrary translation in any direction is then attempted . ( v ) Finally , individual chains are allowed to undergo reptation via a slithering motion of a protein by removing an end domain and its linker and appending it to the other end . The domain and linker are placed in a random position that maintains the tether ranges . After the new position has been assigned , the interaction state of the domain is randomly assigned . If a move results in placement of a domain or module on a site that is already occupied , then the move is rejected . For rotational , torsional , crankshaft , and reptation moves , the moves that do not lead to steric overlap with occupied sites are accepted according to a modified Metropolis criterion viz . , min{1 , wexp ( −ΔE ) } . Here , ∆E is the change in the energy of the system that results from the proposed move . The energy is normalized with respect to kBT . The parameter w is set based on the proposed type of move . For rotational moves , w = 1; for torsional and crankshaft moves , w= ( NpNc ) , where Np and Nc are the number of possible interacting states in the proposed and current states , respectively; finally , for reptation moves , w= ( NpVpNcVc ) , where Np and Nc are the number of possible interacting states in the proposed and current states , respectively whereas Vp and Vc are the total number of conformations the domain and linker could be placed in the proposed state and current state respectively . These modifications to the standard Metropolis Monte Carlo acceptance criterion ensure the preservation of microscopic reversibility . The translation of a connected network does not create or destroy interactions , nor does it move the relevant linkers . Therefore , the proposed translational moves are always accepted if the move does not lead to steric overlaps . For a majority of the simulations , except those where finite size artifacts were queried or the binding affinities were titrated , the interaction energy between adjacent sites with SH3 domains and PRMs was set to –2kBT . In every system , there were 2 . 4 × 103 interaction domains . Concentrations of domains were titrated by changing the number of lattice sites . Each simulation was run for 5 × 109 steps and the average over the last half was used to calculate the size of the largest connected network . In order to query the onset of a gelation transition , we quantified the fraction of molecules that make up the largest connected cluster within the system . We designate this as ϕc . The value of ϕc that is associated with crossing the critical concentration for percolation , defined as the gel point , is determined by comparing the largest connected network from a randomly generated network to the critical concentration predicted by Flory-Stockmayer theory . Here , the number of nodes in the random network is set to the number of interaction domains used in the lattice simulations . The random network was generated for stoichiometric concentrations of complementary domains . For each domain of type A , a random number was compared to the gross probability p that an individual domain would be interacting with a domain of type B . If the random number was less than p , a partner was chosen randomly among the domains of type B that do not already have a binding partner . In order to determine how many Monte Carlo steps the simulations should be run for , we tracked the changes in the largest cluster size for simulations near the critical concentration , where convergence is expected to be the slowest . We then ran our simulations for at least an order of magnitude longer than the equilibration time and analyzed the last half of each simulation to obtain the reported values . For select simulation conditions , we ran independent replicas and reproducibly obtained the same cluster sizes ( ± < 1% ) . In order to locate the concentration where ϕc exceeds the gel point , we ran simulations using a variety of different sized lattices , ranging from 50 to 340 lattice units . The range of box lengths was incrementally refined until the threshold at which the gel point was crossed could be distinguished at the resolution of a single lattice unit . Under the rare case of statistical ambiguity with respect to this threshold , we ran multiple independent simulations at each box length at the approximate gel point , and then averaged the results over all simulations at each box length to obtain a statistically accurate expected value . The gel point or more precisely , the percolation threshold for multivalent polymers can be estimated by analytical methods , one of which is based on Flory-Stockmayer theories . Here , the important parameters are the number of interacting modules within the polymers , V , and the fraction of bound modules , x . For a specific multivalent protein that is incorporated into a pre-formed network , the average number of additional proteins recruited into the network is denoted as ε and is expressed as: ε = ( V – 1 ) x . In a system with two types of multivalent proteins a and b , such as the poly-SH3 and poly-PRM system , the average number of proteins that are recruited into a pre-formed network of multivalent proteins and their ligands can be expressed as: ε = εaεb = ( Va – 1 ) xa ( Vb – 1 ) xb . If ε is greater than 1 , then on average , each protein that is incorporated into the network will bring more than one additional protein with it thus expanding the network . This cascades into an infinitely large cluster of proteins . However , if ε is less than 1 then the proteins that are added are more likely to terminate the network rather than propagate it . For our synthetic poly-SH3 and poly-PRM system , we can calculate the fraction of interactions through knowledge of the dissociation constant , Kd . We designate the SH3 domains as a and the PRMs as b . It follows that: ( 1 ) Kd= ( [ a ]−[ ab ] ) ( [ b ]−[ ab ] ) [ ab ]; Here , [a] , [b] , and [ab] are the concentrations of SH3 domains , PRMs , and bound complexes , respectively . The concentration [ab] can be calculated by a simple rearrangement of Equation ( 1 ) , such that: ( 2 ) [ab]= ( [a]+[b]+Kd− ( [a]+[b]+d ) 2−4[a][b] ) 2; Accordingly , ( 3 ) xa=[ ab ][ a ]= ( [ a ]+[ b ]+Kd− ( [ a ]+[ b ]+d ) 2−4[ a ][ b ] ) 2[ a ] , xb=[ ab ][ b ]= ( [ a ]+[ b ]+Kd− ( [ a ]+[ b ]+d ) 2−4[ a ][ b ] ) 2[ b ] , and ε= ( [ a ]+[ b ]+Kd− ( [ a ]+[ b ]+d ) 2−4[ a ][ b ] ) 4[ a ][ b ] ( Va−1 ) ( Vb−1 ) ; We can solve for the percolation threshold or the concentration at the gel point of module a as a function of the concentration of module b by setting ε = 1 . This yields: ( 4 ) [ a ]c=[ b ]+λ2[ b ]−2λKd± ( λ+1 ) [ b ]2 ( λ−1 ) 2−4λKd2λ; Here , λ = ( Va – 1 ) ( Vb –1 ) . The percolation threshold can also be calculated for the situation where [a] = [b] . In this scenario , ( 5 ) [ a ]c=Kdλ ( 1−λ ) 2; We performed simulations of random percolation models that do not account for linkers or the structure of the lattice models . Each simulation takes the valence , the number of multivalent proteins , and the fraction of bound modules as inputs . The value of ϕc is calculated for prescribed values of the fraction of bound modules and these are shown as solid sigmoidal curves in Figure 11 . The theories of Flory ( Flory , 1941 , 1942b ) and Stockmayer ( Stockmayer , 1943 ) can be used to calculate ϕcc analytically for given values of V and the binding energies , as detailed in the Materials and methods section – see Equations ( 1 ) – ( Zhu and Brangwynne , 2015 ) . These are shown as vertical dashed lines in Figure 11 . For a given valence V , the horizontal intercept that passes through intersection of the vertical dashed lines and the solid curve defines the value of ϕcc . We find this value to be ≈ 0 . 17 , irrespective of the valence . The concentration of modules at which ϕc becomes greater than 0 . 17 is taken to be the value of the gel point cg for the system of interest . We can calculate the value of cg directly from our simulations for the multivalent proteins and compare this to the value of cg that is estimated from Flory-Stockmayer theories . We utilized ρ as the order parameter for differentiating between the sol-gel transitions and phase separation . The coexisting concentrations corresponding to the polymer-rich and polymer-poor phases that delineate the two-phase boundary for a given intrinsic affinity between interaction domains were calculated by assuming that the polymer-rich phase is a uniform density sphere and the polymer-poor phase has a uniform density across the remainder of the lattice . The radius of the polymer-rich phase is the radius of the sphere that is the physically relevant root of the equation: ( 6 ) 1225πNTrN5−43NTRg2rN3−925NNL3rN2+ ( NN−NT ) L54+NTL3Rg3=0; Here , NT is the total number of proteins in the simulation , NN is the number of proteins within the largest network , L is the lattice length on a side , Rg is the radius of gyration over all the proteins in the simulation , and rN is the desired radius of the polymer-rich phase . This equation typically admits only one real root that fits within the lattice and this is true for all of our simulations . The phase boundaries were calculated using: ( 7 ) csl= ( NT−NN ) ( L3−4πrN33 ) Nandcsh=3NN4πrN3 . In addition to starting simulations in the random coil state , we also calculated phase diagrams using simulations that were initialized from a dense phase separated state . For each simulation we equilibrated the proteins in the gel state in a box size of 34 lattice units for 5 × 109 steps . The resulting conformation was then used to initialize simulations in a larger box by expanding the lattice boundary to achieve the desired concentration . For proteins that span the periodic boundary , the first domain was used as the reference for picking which protein image to keep . These initial conditions reproduced the critical concentrations as a function of valence and length . We identified 226 disordered linkers in the human proteome associated with multi-domain proteins . Specifically , we defined disordered linkers in multi-domain proteins as regions predicted to be disordered ( Dosztányi et al . , 2005 ) that connected two Pfam domains ( Finn et al . , 2014 ) that were predicted or known to be folded . We then filtered for linkers that were between 15 and 200 residues in length , and sub-selected for individual proteins where two or more linkers were found . For each of these sequences all-atom simulations were run to provide a general picture of the global conformational behavior associated with disordered linkers in the human proteome . In addition to the set of disordered linkers , we also examined fourteen specifically selected sequences , each consisting of 40 residues . These sequences were chosen to enable a titration of conformational properties as a function of the sequence-encoded fraction of charged residues . Sequences of varying charge were extracted randomly from disordered regions in the human proteome . Disordered regions were identified by extracting sequences from the human proteome that were predicted to be disordered by at least five different disorder predictors in the D2P2 database . We required that each stretch have at least 40 consecutive residues that are disordered . We calculated the fraction of residues by tallying the number of ARG , LYS , ASP , and GLU residues in each fragment . For all sequences described we performed atomistic Monte Carlo simulations using the ABSINTH implicit solvation models and forcefield paradigm ( Vitalis and Pappu , 2009b ) . In this approach , polypeptide chains and solution ions are modeled in atomic detail and the surrounding solvent is modeled using an implicit solvation model that accounts for dielectric inhomogeneities and conformation-specific changes to the free energies of solvation . The simulations were performed and analyzed using tools in the CAMPARI modeling suite ( http://campari . sourceforge . net ) . Forcefield parameters were taken from the abs_opls_3 . 2 . prm parameter set . For each of the fourteen sequences , we performed ten independent simulations , each initialized from a distinct self-avoiding conformation . The methods used to evolve the systems and analyze the simulation results are identical to protocols used in previous studies ( Pak et al . , 2016 Martin et al . , 2016; Das et al . , 2016; Das and Pappu , 2013 ) . For simulations of the 226 disordered linkers , five independent simulations per sequence were performed . Each simulation started from a distinct , randomly selected non-overlapping conformation and comprising 5 × 106 equilibration steps and 5 × 106 production steps in 5 mM NaCl . Simulations of the fourteen specifically selected sequences were run for longer to obtain higher resolution statistics .	Our cells contain a variety of structures called organelles that perform specific roles within a cell . Some organelles are surrounded by a membrane , while others float inside the cell as spherical droplets made of proteins . These proteins contain several sticky regions , which are connected by flexible linker proteins . It is thought that the level of stickiness and the number of sticky regions , or domains , determine whether a protein will form a membraneless organelle . Often , proteins with similar sticky domains have different linkers , and until now , it was assumed that the linkers do not have any other purpose than stringing the domains together . To test this further , Harmon et al . used a combination of computer simulations and physics-based theory . In these simulations , the domains were kept the same , but the properties of linkers were changed to see if this would influence how the membraneless organelles are formed . The results showed that depending on the physical properties of the linkers , the proteins could huddle together and form dense spherical gel-like droplets similar to the membraneless organelles , or form open non-spherical gels . When the linkers were short , the proteins do not easily form droplets . Linkers that were sufficiently long but too bulky , lead to non-spherical gels . Compact linkers , however , enabled proteins to huddle and form spherical gels . The spherical droplet-spanning gels required much less protein compared to the open non-spherical gels . This suggests that proteins important for forming membraneless organelles can be distinguished from those that are not based on the properties of their linkers – even when their domains are similar . These findings further scientists’ knowledge of how specific types of proteins form membraneless organelles and will help to understand how membraneless organelles control many key aspects of how a cell works .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "computational", "and", "systems", "biology" ]	2017	Intrinsically disordered linkers determine the interplay between phase separation and gelation in multivalent proteins
Upon fertilization , the genome of animal embryos remains transcriptionally inactive until the maternal-to-zygotic transition . At this time , the embryo takes control of its development and transcription begins . How the onset of zygotic transcription is regulated remains unclear . Here , we show that a dynamic competition for DNA binding between nucleosome-forming histones and transcription factors regulates zebrafish genome activation . Taking a quantitative approach , we found that the concentration of non-DNA-bound core histones sets the time for the onset of transcription . The reduction in nuclear histone concentration that coincides with genome activation does not affect nucleosome density on DNA , but allows transcription factors to compete successfully for DNA binding . In agreement with this , transcription factor binding is sensitive to histone levels and the concentration of transcription factors also affects the time of transcription . Our results demonstrate that the relative levels of histones and transcription factors regulate the onset of transcription in the embryo . In many organisms , early embryonic development is directed exclusively by maternal products that are deposited into the female gamete during oogenesis . Following the clearance of a subset of these products ( Yartseva and Giraldez , 2015 ) , transcription is initiated and the zygotic genome acquires developmental control ( Blythe and Wieschaus , 2015a; Harrison and Eisen , 2015; Lee et al . , 2014; Tadros and Lipshitz , 2009 ) . This handover is referred to as the maternal-to-zygotic transition and the onset of transcription is called zygotic genome activation ( ZGA ) . The absolute time and number of cell cycles required before the first transcripts can be detected is species specific ( Tadros and Lipshitz , 2009 ) . Additionally , from one gene to another the timing of transcriptional activation varies ( Aanes et al . , 2011; Collart et al . , 2014; Harvey et al . , 2013; Heyn et al . , 2014; Lott et al . , 2011; Owens et al . , 2016; Pauli et al . , 2012; Sandler and Stathopoulos , 2016; Tan et al . , 2013 ) . In fact , for some genes the first zygotic transcripts can be detected several cell cycles before the stage that is traditionally defined as the time point of ZGA ( De Renzis et al . , 2007; Heyn et al . , 2014; Skirkanich et al . , 2011; Yang et al . , 2002 ) . In spite of the progress made , it remains unclear how the onset of transcription in embryos is temporally regulated . Several lines of evidence suggest that the absence of transcription during early embryonic development could be due to limited levels of transcription factors ( Almouzni and Wolffe , 1995; Veenstra et al . , 1999 ) . In this scenario , transcriptional activation would occur once a threshold level of these factors is reached . For example , experiments that used the transcriptional activity of injected plasmids as a read-out revealed that an increase in the amount of the potent , heterologous , transcriptional activator GAL4-VP16 can overcome transcriptional repression of its target gene in the early embryo ( Almouzni and Wolffe , 1995 ) . However , it remained unclear whether limited levels of transcription factors contribute to the absence of endogenous transcription in early embryos . Additional support for the limited machinery model came from work showing that an increase in the concentration of the general transcription factor TBP can cause premature transcription from an injected – and incompletely chromatinized – DNA template in Xenopus embryos . This effect was maintained only when non-specific DNA was added to titrate chromatin assembly ( Almouzni and Wolffe , 1995; Veenstra et al . , 1999 ) . These results suggested that low TBP levels may play a role in the absence of transcription during the early stages of Xenopus development , but that increasing TBP alone is not sufficient to cause sustained premature transcription . During the cleavage stages of Xenopus development , TBP levels increase due to translation , which suggests that TBP levels might contribute to the timely activation of transcription during ZGA ( Veenstra et al . , 1999 ) . Transcription factors have recently been identified that are required for the activation of the first zygotically expressed genes in Drosophila ( Zelda ) and zebrafish ( Pou5f3 , Sox19b , Nanog ) ( Harrison et al . , 2011; Lee et al . , 2013; Leichsenring et al . , 2013; Liang et al . , 2008; Nien et al . , 2011 ) . RNA for these factors is maternally provided and their levels increase due to translation during the early cell cycles . This suggests the possibility that an increase in the concentration of these transcription factors might contribute to the shift from transcriptional repression to transcriptional activity . Although transcription factors levels clearly influence transcriptional activity during early embryogenesis , there is evidence to show that the transcriptional machinery is operational prior to ZGA ( Dekens et al . , 2003; Lu et al . , 2009; Newport and Kirschner , 1982a , 1982b; Prioleau et al . , 1994 ) ( see below ) . Thus , the timing of ZGA cannot be solely explained by a requirement to reach a threshold level of transcriptional activators . The finding that a premature increase in the number of nuclei or the amount of DNA resulted in premature transcription of injected plasmids in Xenopus embryos suggested that the transcriptional machinery is fully functional prior to genome activation and led to the excess repressor model ( Newport and Kirschner , 1982a ) . This model postulates that a transcriptional repressor is titrated by binding to the exponentially increasing amount of genomic DNA , until it is depleted first from the soluble fraction , and then from DNA , to allow for the onset of transcription . Related studies in zebrafish and Drosophila have provided further evidence for this model . Endogenous transcription is initiated earlier in zebrafish embryos that accumulate DNA due to a defect in chromosome segregation ( Dekens et al . , 2003 ) , and transcription is delayed in haploid Drosophila embryos compared to diploid embryos , albeit not for all genes ( Lu et al . , 2009 ) . The excess repressor model predicts that the repressor is present in large excess , at relatively stable levels while the genome is inactive , and can bind DNA with high affinity . Core histones fulfill these criteria ( Adamson and Woodland , 1974; Woodland and Adamson , 1977 ) . Moreover , when bound to DNA in the form of nucleosomes , histones can affect DNA accessibility for DNA-binding proteins . To date , two key studies have investigated the role of core histones in the temporal regulation of zygotic transcription in Xenopus embryos ( Almouzni and Wolffe , 1995; Amodeo et al . , 2015 ) . Experiments that used the transcriptional activity of injected plasmids as a read-out revealed that premature transcription caused by an excess of non-specific DNA can be negated by the addition of histones ( Almouzni and Wolffe , 1995 ) . More recently , the level of histones H3/H4 was shown to regulate the level of transcription in Xenopus egg extract and H3 was suggested to play a similar role in the embryo ( Amodeo et al . , 2015 ) . Taken together , these results support the idea that histones play a role in regulating the timing of zygotic transcription . If histones function as repressors according to the original excess repressor model , it would be predicted that a substantial reduction of the histone-density on DNA would cause the onset of transcription ( Amodeo et al . , 2015; Newport and Kirschner , 1982a ) . However , while such a scenario might be possible for typical sequence-specific repressors of transcription , it is unlikely for histones . Histones assemble into histone octamers on DNA to form nucleosomes , the basic building blocks of chromatin . Thus , random depletion of nucleosomes from DNA would severely compromise the integrity of chromatin structure . Taken together , there is support for the idea that histone levels play a role in regulating the timing of zygotic transcription , but it remains unclear how this would mechanistically work . Furthermore , the observation that both activator and histone levels play a role in shifting the balance between repression and activation at genome activation remains to be clarified . Here , we analyze the onset of zygotic transcription in zebrafish embryos . With a quantitative approach , we show that the concentration of non-DNA-bound histones determines the timing of zygotic transcription and that all four core histones are required for this effect . The reduction in nuclear histone concentration that coincides with genome activation does not result in a significant change in nucleosome density , but rather allows transcription factors to successfully compete for DNA binding . In agreement with this , the association of transcription factors with the genome is sensitive to histone levels , and changing the concentration of transcription factors also affects the time of transcription . Our results show that transcription is regulated by a dynamic competition for DNA binding between histones and transcription factors . Transcription begins when the concentration of non-DNA-bound histones in the nucleus has sufficiently dropped so that the transcriptional machinery can outcompete histones for binding to DNA . Experiments in Xenopus embryos led to the hypothesis that histone levels regulate the onset of zygotic transcription ( Almouzni and Wolffe , 1995; Amodeo et al . , 2015 ) . To analyze whether in zebrafish , histones are potential candidates to be excess repressors , we analyzed the relative levels of the core histones—H3 , H4 , H2A and H2B—by Western blot . We found that they are present at relatively stable amounts from 8-cell to 1K stage ( Figure 1E ) . Assuming that at 1K stage there are sufficient histones to wrap all genomes into chromatin , this suggests that histones are in excess relative to the amount of DNA during the earlier stages . Thus , histones could function as excess repressors of the zygotic genome in zebrafish . If histones function as excess repressors in zebrafish embryos , it would be predicted that their level would affect the onset of transcription . To test this , we analyzed the effect of increasing the amount of histones in the embryo on the timing of transcriptional activation . We injected a stoichiometric mixture of the four core histones ( from here on referred to as histone cocktail , HC; see Materials and methods for more details ) into embryos at the 1-cell stage and then analyzed the onset of transcription for the previously characterized set of genes ( Figures 2A and 1B ) . An increase in the amount of histones delayed the onset of transcriptional activation: transcripts were detected at high stage in uninjected embryos , whereas in embryos injected with histone cocktail , transcription was only induced at oblong stage , a complete developmental stage later ( Figure 2B and Figure 2—figure supplement 1A ) . Comparison of gene expression levels in uninjected and injected embryos at high stage ( when transcripts can consistently be detected in uninjected embryos ) revealed that the level of induction is reduced significantly upon injection of the histone cocktail but not upon injection of BSA as a control ( Figure 2B and Figure 2—figure supplement 1A , bar graphs ) . Extending the analysis further , a large set of genes in Nanostring analysis confirmed that the effect we observed is general , and not limited to six genes ( Figure 2C , Figure 2—figure supplement 2 , Figure 2—source data 1 ) . Staging by morphology was corroborated by cell counting , with absolute time between the analyzed stages being constant , confirming that changes in the timing of transcription were not due to effects on cell cycle length or developmental progression ( Figure 2—figure supplement 1B ) . Moreover , the injected histones can be incorporated into chromatin , as indicated by labeling one of them with Cy5 and detecting this label in chromatin when imaging embryos after injection ( Figure 2—figure supplement 1C ) , confirming that they are functional . Together , these data show that an increase in the excess amount of histones in the embryo delays the onset of transcription . 10 . 7554/eLife . 23326 . 006Figure 2 . Increasing the levels of all core histones delays onset of transcription and gastrulation . ( A ) Schematic representation of experimental procedure . Histone cocktail ( HC ) containing ~5800 genomes worth of histones , or BSA was injected into the yolk of 1-cell embryos and qPCR and NanoString analysis was carried out at stages around genome activation . Orange crosses represent the timing of stages used for the analysis . ( B ) Expression of mxtx2 and fam212aa was analyzed by qPCR at early 1K , high , and oblong stage in uninjected , BSA-injected and HC-injected embryos . Bar graphs show the same data , focusing on high stage . Error bars represent SEM ( n ≥ 13 ) . *p<0 . 001 ( two-tailed Student’s t test , compared to BSA control ) . ( C ) Expression of 53 zygotically expressed genes was analyzed by NanoString analysis at high stage in uninjected , BSA-injected and HC-injected embryos . Mean counts of three independent biological replicates are shown . Location of mxtx2 and fam212aa counts is indicated ( See Figure 2—figure supplement 2 for more details ) . ( D ) Relative expression level of mxtx2 and fam212aa at high stage , for embryos injected with BSA , HC , and HC minus H3 , H4 , H2A , or H2B . Error bars represent SEM ( n = 7 ) . p<0 . 001 ( ordinary one-way ANOVA ) . ( E ) Brightfield images of embryos that were not injected , injected with BSA , or injected with HC . Boxed images represent the onset of gastrulation . Scale bar shown for the uninjected 2-cell embryo applies to all treatments except for 24 hpf embryos which have a different scale bar . All scale bars represent 250 μm . hpf , hours post-fertilization . ( F ) Bar graph shows the quantification of the extra time it takes embryos to start gastrulation upon injecting BSA , HC , or HC minus one histone , compared to uninjected embryos . Error bars represent SEM ( n = 27 for BSA , n = 25 for HC , n = 7 for HC minus one histone experiments ) . p<0 . 001 ( ordinary one-way ANOVA with Tukey’s multiple comparison test ) . In B and D , mRNA levels are normalized to the expression of eif4g2α . DOI: http://dx . doi . org/10 . 7554/eLife . 23326 . 00610 . 7554/eLife . 23326 . 007Figure 2—source data 1 . NanoString probe set . DOI: http://dx . doi . org/10 . 7554/eLife . 23326 . 00710 . 7554/eLife . 23326 . 008Figure 2—figure supplement 1 . Increasing the levels of all core histones delays onset of transcription . ( A ) Expression of nnr , vox , sox19a , and grhl3 was analyzed by qPCR at early 1K , high , and oblong stage in uninjected , BSA-injected and HC-injected embryos . Bar graphs show the same data , focusing on high stage . Error bars represent SEM ( n ≥ 13 ) . p<0 . 001 ( two-tailed Student’s t-test , compared to BSA control ) . ( B ) Staging by morphology was verified by cell counting . Each data point represents a single embryo . Error bars represent SEM . ( C ) Confocal microscope images of transgenic fish line Tg ( h2afz:h2afz-GFP ) injected with Cy5 conjugated to H4 . Arrow points at chromatin in dividing cell . Scale bar , 20 μm . ( D ) Relative expression level of nnr , vox , sox19a , and grhl3 at high stage , for embryos injected with BSA , HC , and HC minus H3 , H4 , H2A , or H2B . Error bars represent SEM ( n = 7 ) . p<0 . 05; p<0 . 01; p<0 . 001 ( ordinary one-way ANOVA ) . In A and D , expression is normalized to the expression of eif4g2α . DOI: http://dx . doi . org/10 . 7554/eLife . 23326 . 00810 . 7554/eLife . 23326 . 009Figure 2—figure supplement 2 . Increasing the levels of all core histones delays onset of transcription for a large number of genes . NanoString’s nCounter technology was used to analyze changes in gene expression upon HC injection for a large number of genes ( see Materials and methods for more details ) . From a custom probe set with 84 zygotically expressed genes and 12 controls genes ( see Figure 2—source data 1 ) , 53 genes that are induced at high stage were used for analysis . mRNA was collected from high stage embryos that were uninjected , BSA-injected and HC-injected ( n = 3 ) . ( A ) Fold difference in expression of twelve control genes , for embryos injected with BSA or HC compared to uninjected . Control genes are maternally provided and were previously shown to be stable from 512 cell to dome stage in NanoString analysis ( data not shown ) . Error bars represent SEM ( n = 3 ) . No significant difference was detected between BSA or HC-injected in control genes ( Ordinary one-way ANOVA with Tukey’s multiple comparison test ) , showing that there are no differences in total RNA amount between samples . ( B ) Proportion of genes affected by BSA or HC injection at high stage in NanoString analysis . Error bars represent SEM ( n = 3 ) . A large proportion of genes are down-regulated in HC-injected embryos ( 86% ) compared to BSA-injected embryos ( 4% ) . ( C ) Mean counts of three independent biological replicates in NanoString analysis for uninjected , BSA-injected and HC-injected embryos compared to a high ( left ) and 1K ( right ) standard from uninjected embryos . While uninjected and BSA-injected are statistically similar to the high standard , HC-injected is statistically similar to the 1K standard . Thus , HC-injected embryos that are developmentally at high stage , are transcriptionally delayed by one developmental stage . Error bars represent SEM ( n = 3 ) . *p<0 . 001 ( Ordinary one-way ANOVA with Tukey’s multiple comparison test ) . ( D ) Fold difference in mRNA counts for HC-injected and BSA-injected compared to uninjected embryos at high stage for all genes that were analyzed by qPCR in this study . Error bars represent SEM ( n = 3 ) . p<0 . 05; p<0 . 01; p<0 . 001 ( two-tailed Student’s t-test ratio paired , compared to BSA-injected ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23326 . 009 To test whether the effect we observe upon injecting the histone cocktail required an increase in the level of all four core histones , we next removed one histone at a time from the cocktail . The total protein content was kept constant by raising the level of the other three histones . Removing any histone from the histone cocktail impaired the ability of the histone cocktail to delay the onset of transcription ( Figure 2D and Figure 2—figure supplement 1D ) . These results show that the injection of basic proteins into the embryo per se does not affect the onset of transcription . Moreover , these results argue that the effect of the histone cocktail relies on increasing the amount of all four histones and suggest that histones exert their repressive effect together . Since the onset of zygotic transcription is known to be required for gastrulation ( Kane et al . , 1996; Lee et al . , 2013; Zamir et al . , 1997 ) , we analyzed the effect of injecting the histone cocktail on the onset of gastrulation . Embryos injected with the histone cocktail initiated gastrulation later than uninjected embryos ( Figure 2E ) . Although there was a delay following injection of BSA , it was significantly shorter than that observed with the histone cocktail and appeared to be a non-specific effect of injection ( Figure 2F ) . Following the onset of gastrulation , embryos appeared to develop normally ( Figure 2E , 24 hpf ) . Removing any histone from the histone cocktail reduced the developmental delay we observed upon injecting the histone cocktail ( Figure 2F ) . We note that the developmental delay in the minus-one histone experiments was not reduced to the level observed for BSA injections . We therefore expect that injecting histones has an additional effect on developmental progression that is independent of the delay in transcription . We conclude that the delay in transcription as a consequence of increased histone levels causes a delay in the onset of gastrulation . If the level of histones regulates zygotic genome activation , it can also be predicted that a reduction in histone levels would result in the premature induction of transcription . A large fraction of the histones that is present at the onset of transcription is loaded in the egg already as protein ( Figure 1E ) . Pentraxin three is a soluble pattern recognition molecule that has been shown to rapidly and irreversibly bind to the core histones H3 and H4 ( Bottazzi et al . , 2010; Daigo et al . , 2014 ) . We injected mRNA encoding PTX3 fused to RFP , to reduce the pool of available histones H3 and H4 in the zebrafish embryo ( Figure 3A ) . As expected , total levels of H3 and H4 were not affected upon injection of this fusion construct ( Figure 3B ) . Next , we examined if H4 co-precipitated with RFP-tagged PTX3 ( Figure 3C ) . Indeed , this histone associates with PTX3 in vivo , suggesting that the injection of PTX3 results in a reduction of the soluble amount of histones H3 and H4 in zebrafish cells . A decrease in the soluble amount of histones caused premature transcription activation: transcripts were detected at early 1K stage , while in the uninjected embryos , transcripts were only detected at mid 1K ( Figure 3D and Figure 3—figure supplement 1A ) . We included embryos at mid 1K in this experiment , in order to increase our resolution for detecting changes in transcription . Comparison of gene expression levels in uninjected and ptx3-injected embryos at early 1K stage ( one time-point prior to when genes are first induced in uninjected embryos ) revealed that the level of expression is increased upon injection of ptx3 mRNA ( Figure 3D and Figure 3—figure supplement 1A , bar graphs ) . A control injection with rfp mRNA did not result in co-precipitation with H4 ( Figure 3C ) , nor did it affect the onset of transcription ( Figure 3D and Figure 3—figure supplement 1A ) . Staging by morphology was corroborated by cell counting ( Figure 3—figure supplement 1B ) , with absolute time between the analyzed stages being constant . Taken together , our results provide evidence that the level of core histones in the embryo dictates the timing of transcriptional activation . 10 . 7554/eLife . 23326 . 010Figure 3 . Decreasing the level of histones causes premature transcription . ( A ) Schematic representation of experimental procedure . ptx3-rfp or rfp ( control ) mRNA was injected into the cell of 1-cell embryos and qPCR analysis was carried out at stages around genome activation . Orange crosses represent the timing of stages used for the analysis . ( B ) Western blot analysis of PTX3 , histone H3 and H4 levels at 512-cell , 1K and high stage in uninjected embryos , rfp and ptx3-rfp mRNA-injected embryos . Tubulin was used to control for equal loading . Blots shown are representative examples ( n = 3 ) . ( C ) Western blot analysis for histone H4 after a pull-down using an RFP antibody at 1K stage . Uncoupled beads were used as a negative control . Blot shown is a representative example ( n = 3 ) . ( D ) Expression of mxtx2 and fam212aa was analyzed by qPCR at 512-cell , early 1K , and mid 1K stage in uninjected , rfp mRNA-injected and ptx3-rfp mRNA-injected embryos . Bar graphs focus on early 1K stage . Error bars represent SEM ( n ≥ 4 ) . p<0 . 05; *p<0 . 01 ( two-tailed Student’s t-test , compared to rfp mRNA control ) . mRNA levels are normalized to the expression of eif4g2α . DOI: http://dx . doi . org/10 . 7554/eLife . 23326 . 01010 . 7554/eLife . 23326 . 011Figure 3—figure supplement 1 . Decreasing the level of histones causes premature transcription . ( A ) Expression of nnr , vox , sox19a , and grhl3 was analyzed by qPCR at 512 cell , early 1K and mid 1K in uninjected embryos , embryos injected with rfp ( control ) mRNA and embryos injected with ptx3-rfp mRNA . Bar graphs show the same data , focusing on early 1K stage . Error bars represent SEM ( n ≥ 4 ) . n . s . p>0 . 05; p<0 . 05; *p<0 . 01 ( two-tailed Student’s t-test , compared to rfp mRNA control ) . Expression is normalized to the expression of eif4g2α . ( B ) Staging by morphology was verified by cell counting at the stages used for the analysis . Each data point represents a single embryo . Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 23326 . 011 If histones were to function as excess repressors according to the original excess repressor model , the concentration of non-DNA-bound histones would be predicted to decrease during the cleavage stages of development . To test this prediction , we determined the absolute ( molar ) content and , correspondingly , the number of molecules of core histones in embryos using a quantitative mass spectrometry approach we recently developed ( Kumar et al . , unpublished ) ( Figure 4A and Materials and methods for more details ) . We analyzed embryos ranging from 1-cell to shield stage , when gastrulation is well underway ( Figure 4B and Figure 4—source data 1 ) . We observed an increase in the levels of histone protein until 1K stage , after which levels remained reasonably stable until sphere stage . Then , a rapid increase was observed , which is most likely the result of translation of zygotically produced histone mRNAs . Knowing the absolute numbers of histones per embryo as well as the calculated number of histones required to wrap a genome ( Figure 4B , Figre 4—source data 1 ) , allowed us to derive the number of genomes worth of histones per embryo . From that , we derived the number of excess ( non-DNA bound ) histones per cell in genomes worth of histones , for embryos ranging from 1-cell to dome stage ( Figure 4C and Materials and methods for more details ) . For example , at the 1-cell stage , there are 3098 times more histones per cell than are required to wrap the genome into chromatin . Due to the exponential increase in cell number during cleavage divisions , this number has dropped dramatically at 1K stage ( Figure 4C ) . However , due to the large number of histones that is loaded in the oocyte , as well as the increase in histone level due to translation ( Figure 4B ) , there are still nine genomes worth of non-DNA-bound histones per cell . Moreover , because what matters for protein-binding kinetics is the concentration , we next calculated the concentration of non-DNA-bound histones . Because the cleavage divisions are not accompanied by significant growth ( the total animal cap volume increases by 29% from 128-cell to 1K , Figure 4—figure supplement 1A ) , the decreasing number of histones per cell is accompanied by a decreasing cellular volume , and the concentration of non-DNA bound histones in the cell does not change substantially ( Figure 4D ) . Taken together , this shows that during transcription activation there is still a significant amount of non-DNA-bound histone and that the overall concentration of non-DNA-bound histones in the cell has not decreased by much . 10 . 7554/eLife . 23326 . 012Figure 4 . Onset of transcription coincides with a reduction in nuclear histone concentration . ( A ) Our quantitative mass spectrometry approach . Zebrafish histones were quantified by comparing the abundances of native histone peptides with corresponding isotopically labeled peptides from the chimeric protein; chimeric protein was quantified by comparing the abundance of labeled ( from chimera ) and native ( from standard ) BSA peptides ( see Materials and methods for more details ) . ( B ) Quantification of the number of histone H3 , H4 , H2A , and H2B per embryo at indicated stages by quantitative mass spectrometry . Error bars represent SEM ( n = 3 ) . ( C ) The excess number of histones per cell ( in genomes worth ) was calculated using H2B levels ( Figure 4—source data 1 ) and cell numbers ( Figure 1—source data 1 ) , and by assuming an average of 1 . 5 genomes per cell ( see Materials and methods for more details ) . For better visualization of the data at later developmental stages the values for 1-cell and 8-cell are not shown in the graph but are 3098 and 518 , respectively . Error bars represent SEM ( n = 3 ) . GW , genomes worth of histones . ( D ) The total concentration of non-DNA-bound histones was calculated by dividing the excess genomes worth of histone H2B per embryo by the volume of the animal cap at the respective stages ( Figure 4—figure supplement 1A ) . Error bars represent SEM of animal cap volumes ( n = 3 ) . GW , genomes worth of histones . ( E ) The nuclear concentration of non-DNA-bound histones was calculated from immunofluorescence ( from left to right n = 12 , 12 , 14 , 15 ) combined with live imaging and mass spectrometry data ( see Materials and methods for more details ) . Error bars represent SEM of animal cap volumes ( n = 3 ) . ( F ) Relative differences in H2B intensity between chromatin fractions of 256-cell , 512-cell , 1K , and high stage embryos . Sphere stage embryos were used to determine the linear range of H2B detection ( see also Figure 4—figure supplement 1C ) . Blots shown are representative examples ( n ≥ 3 ) . Plots show observed fold differences in H2B intensity in chromatin fractions comparing indicated stages compared to the differences that would be expected if the intensity were to scale with the amount of DNA ( E , embryo ) . ( G ) Competition model . See text for more details ( TFBS , transcription-factor-binding site ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23326 . 01210 . 7554/eLife . 23326 . 013Figure 4—source data 1 . Quantification of histone number by mass spectrometry . See Materials and methods for details on calculations . DOI: http://dx . doi . org/10 . 7554/eLife . 23326 . 01310 . 7554/eLife . 23326 . 014Figure 4—source data 2 . Two channel recording of H4-sfGFP and PCNA-RFP distributions from 8-cell to oblong stage . H4-sfGFP ( left , green channel ) intensities are transformed to logarithmic scale to compensate for intensity increase due to ongoing translation of mRNA into fluorescent fusion protein . PCNA-RFP ( right , magenta channel ) intensities are linear . Both channels are maximum z-projections , with a view upon the animal cap , time stamps are given in hour:minute format , starting with the first acquired frame . DOI: http://dx . doi . org/10 . 7554/eLife . 23326 . 01410 . 7554/eLife . 23326 . 015Figure 4—figure supplement 1 . Onset of transcription coincides with a reduction in nuclear histone concentration . ( A ) Changes in total animal cap volume , the fraction of the animal cap volume occupied by nuclei , and the size of individual nuclei for indicated stages . Volumes were measured by lightsheet microscopy of embryos injected with mRNA encoding H4-sfGFP and subsequent automated image analysis ( error bars represent SEM , n = 3 embryos; offspring of transgenic PCNA-RFP was used to monitor integrity of imaged nuclei via a second color channel ) . ( B ) Image sequences showing nuclear import of H4-sfGFP fusion protein at 32- to 64- and 128-cell stages . Color scaling was kept constant and linear within each stage . Images show a representative maximum z-projection of a subset of a 3D microscopy stack of one of the embryos used in A . ( C ) The linear range of Western blots was determined using chromatin of sphere-stage embryos . Plotted are observed versus expected fold differences in H2B intensity using different numbers of embryos ( n ≥ 3 ) . Band intensities of test stages ( Figure 4F ) were only used for the analysis when they fell within the linear range . DOI: http://dx . doi . org/10 . 7554/eLife . 23326 . 015 Because transcription takes place in the nucleus , we next wanted to investigate the concentration of non-DNA-bound histones in this compartment of the cell . First , we analyzed the dynamics of histone localization by lightsheet microscopy of living embryos ( Figure 4—figure supplement 1B and Figure 4—source data 2 ) . As expected , we found a close coordination between the formation of nuclei after cell division and the import of histones from the cytoplasm into the nucleus: during each cell cycle , non-DNA-bound histones are concentrated in the nucleus . A direct quantification of non-DNA-bound , endogenous histones in the nucleus is difficult , but by combining lightsheet microscopy measurements of both the nuclear volume fraction and the relative fluorescence intensity of histone H4 in cytoplasm and nucleus with the absolute amount of histone H4 as quantified by mass spectrometry , we were able to calculate the nuclear concentration of non-DNA-bound histones from 256-cell to oblong stage ( Figure 4E , Figure 4—figure supplement 1A and see Materials and methods for more details ) . Importantly , our calculations indicate a decrease in the nuclear concentration of non-DNA-bound histones at the onset of transcription . In combination with our finding that histone levels determine the timing of transcription , this suggests that a decrease in the concentration of non-DNA-bound histones in the nucleus causes the onset of transcription during embryogenesis . We next analyzed whether the decreased concentration of non-DNA bound histones in the nucleus is accompanied by a reduced density of nucleosomes on chromatin . We quantified the amount of histone H2B in the chromatin fraction of embryos ranging from 256-cell to high stage . Comparing the amount of histone H2B between stages revealed that the level of H2B scales with the amount of DNA ( Figure 4F and Figure 4—figure supplement 1C ) . This is in agreement with a previous study in which we found that the density of nucleosomes does not significantly change during genome activation ( Zhang et al . , 2014 ) . Our results reveal that global nucleosome density on DNA does not change during genome activation . Taken together , this suggests that the concentration of non-DNA-bound histones in the nucleus determines the timing of transcription without the need for a significant change in global nucleosome density . Our finding that the concentration of histones in the nucleus determines the onset of transcription without a significant change in global nucleosome density on DNA suggests that a simple depletion model cannot explain a role for histone levels in the timing of zygotic transcription . To explain the effect of histone levels on zygotic genome activation , we hypothesized that the transcriptional machinery ( for simplicity referred to as transcription factors ) competes with nucleosome-forming histones for binding to only a minimal fraction of the total DNA , corresponding to transcription-factor-binding sites ( Figure 4G ) . In such a model , local substitution of nucleosomes by transcription factors allows for transcription to be activated , but will cause only localized changes in nucleosome positioning , and will barely affect the average nucleosome density . Transcription factors would lose the competition for DNA binding in the presence of an excess of histones ( pre-ZGA ) , whereas a reduction of the concentration of non-DNA-bound histones in the nucleus would allow transcription factors to gain access to DNA ( approaching ZGA ) and initiate transcription ( ZGA ) . If competition between nucleosome-forming histones and the transcriptional machinery determines the onset of transcription , it would be predicted that the levels at which transcription factors are present could also affect the timing of zygotic transcription . To test this , we changed the level of Pou5f3 , a transcription factor that has been identified as being required for the activation of a large set of genes during genome activation ( Lee et al . , 2013; Leichsenring et al . , 2013; Onichtchouk et al . , 2010 ) . To analyze the effect on the onset of transcription ( Figure 5A ) , we selected five genes that are activated at the onset of genome activation and that have been identified as Pou5f3 targets ( Figure 1—figure supplement 1A ) ( Onichtchouk et al . , 2010 ) . 10 . 7554/eLife . 23326 . 016Figure 5 . Direct experimental evidence for the competition model using endogenous genes and transcription factors . ( A ) Schematic representation of experimental procedure . Pou5f3 levels were decreased by injecting a morpholino , or increased by injecting pou5f3 mRNA ( in combination with sox19b mRNA ) into the cell of 1-cell embryos . Controls used were a dead-end morpholino and gfp mRNA , respectively . qPCR and ChIP-qPCR analysis was carried out at stages around genome activation . Orange crosses represent the timing of stages used for the analysis . ( B ) Expression of apoeb and dusp6 was analyzed by qPCR at 512-cell , early 1K , mid 1K and high stage in control and Pou5f3 morpholino-injected embryos . The data in the bar graphs focus on the mid 1K stage . Error bars represent SEM ( n ≥ 4 ) . p<0 . 05 ( two-tailed Student’s t-test , compared to control MO ) . ( C ) Expression of apoeb and dusp6 was analyzed by qPCR at 512-cell , early 1K and high stage in uninjected embryos , embryos injected with control mRNA and embryos injected with pou5f3 and sox19b mRNA . Bar graphs focus on the early 1K stage . Error bars represent SEM ( n ≥ 4 ) . p<0 . 01 ( two-tailed Student’s t-test , compared to control mRNA ) . ( D ) Binding of Pou5f3 to its respective binding sites for apoeb and dusp6 ( Leichsenring et al . , 2013 ) and control region was analyzed by ChIP-qPCR at the early 1K stage in embryos injected with pou5f3 + sox19b mRNA or pou5f3 + sox19b mRNA plus histone cocktail . Enrichment of pulled-down fragments was normalized to input . Location of primer sets in respect to the transcription start-site used for ChIP-qPCR analysis are indicated by arrows . A genome control region on chromosome 23 was also used . Error bars represent SEM ( n = 5 ) . p<0 . 01 ( two-tailed Student’s t-test ratio paired , compared to pou5f3 + sox19b mRNA-injected embryos ) . In B and C , mRNA levels are normalized to the expression of eif4g2α . DOI: http://dx . doi . org/10 . 7554/eLife . 23326 . 01610 . 7554/eLife . 23326 . 017Figure 5—figure supplement 1 . Reducing transcription factor levels delays the onset of transcription . ( A ) Pou5f3 morpholino validation . Brightfield images of embryos that were injected with control or Pou5f3 morpholino . Embryos injected with Pou5f3 morpholino arrested at sphere or dome stage , as reported previously ( Burgess et al . , 2002 ) . Scale bar , 250 μm . Western blot of embryos injected with 50 ng pou5f3-2xHA mRNA alone or in combination with 6 ng Pou5f3 moprholino . As expected , the morpholino reduces the expression of Pou5f3 . ( B ) We injected embryos with Pou5f3 morpholino and analyzed the effect on transcription . Shown are the fold changes in relative expression levels of Pou5f3-morpholino-injected embryos compared to control morpholino-injected embryos at high stage for Pou5f3 targets apoeb , dusp6 , klf17 , irx7 , and klf2b , and non-Pou5f3 targets sox19a , grhl3 and gadd45bb . Reduction of Pou5f3 results in decreased transcription for all Pou5f3 target genes analyzed at high stage , confirming that they are regulated by Pou5f3 . Non-targets were not affected . Error bars represent SEM ( n = 4 ) . n . s . p>0 . 05; p<0 . 05; p<0 . 01; p<0 . 001 ( two-tailed Student’s t-test , compared to control morpholino ) . ( C ) Expression of klf17 , irx7 and klf2b was analyzed by qPCR at 512-cell , early 1K , mid 1K and high stage in control and Pou5f3 morpholino-injected embryos . Bar graphs focus on mid 1K stage . Error bars represent SEM ( n ≥ 4 ) . p<0 . 01; *p<0 . 001 ( two-tailed Student’s t-test , compared to control morpholino ) . ( D ) Western blot analysis of embryos injected with sox19b-2xHA mRNA alone and in combination with 2 ng of Sox19b morpholino validate the effect of the Sox19b morpholino . ( E ) Expression of dusp6 was analyzed by qPCR at 512-cell , early 1K , mid 1K and high stage in control and Sox19b morpholino-injected embryos . dusp6 was selected for analysis because it is the only gene from our selected Pou5f3-target genes that is also regulated by SoxB1 ( Lee et al . , 2013 ) . Error bars represent SEM ( n ≥ 4 ) . ( F ) Validation of FoxH1 target genes . To verify that the genes we selected require FoxH1 for their expression , we injected embryos with FoxH1 morpholino and analyzed the effect on transcription . Shown are the fold changes in relative expression levels of FoxH1 morpholino-injected embryos compared to control morpholino-injected embryos at high stage for FoxH1 targets dusp6 , flh , wnt11 and gadd45bb . Reduction of FoxH1 results in decreased transcription for all FoxH1 target genes . ( G ) Expression of dusp6 , flh , wnt11 and gadd45bb was analyzed by qPCR at 512-cell , early 1K , mid 1K and high stage in control and FoxH1 morpholino-injected embryos . All four target genes show a delay in the time when they are first transcribed in the FoxH1 morphants compared to control morpholino-injected embryos . Error bars represent SEM ( n ≥ 4 ) . ( H ) Staging by morphology was verified by cell counting at the stages used for the Pou5f3 morpholino analysis . Each data point represents a single embryo . Error bars represent SEM . Gene expression is normalized to the expression of eif4g2α . DOI: http://dx . doi . org/10 . 7554/eLife . 23326 . 01710 . 7554/eLife . 23326 . 018Figure 5—figure supplement 2 . Increasing transcription factor levels causes premature transcription . ( A ) Pou5f3 overexpression validation . Brightfield images of embryos that were injected with control or pou5f3 mRNA . Developmental defects at 24 hpf resembled the ventralized phenotypes described upon pou5f3-VP16 overexpression ( Belting et al . , 2011 ) . Scale bar , 250 μm . Western blot using an HA antibody shows the protein level of Pou5f3-2xHA and Sox19b-2xHA in embryos at 1K stage after injection at the 1 cell stage . Blot shown is a representative example ( n = 2 ) . ( B ) To test whether Pou5f3 and Sox19b are sufficient to drive expression of the genes we selected , we increased the level of both transcription factors and analyzed the effect on transcription . Shown are the expression levels of apoeb , dusp6 , klf17 , irx7 , and klf2b at high stage for embryos injected with pou5f3 + sox19b mRNA relative to control mRNA-injected . Error bars represent SEM ( n ≥ 4 ) . Increasing the levels of Pou5f3 and Sox19b only increased the relative expression level of apoeb and dusp6 , suggesting that Pou5f3 and Sox19b are not sufficient to drive expression of the other genes . Error bars represent SEM ( n ≥ 4 ) . n . s p>0 . 05; p<0 . 05 ( two-tailed Student’s t-test , compared to control mRNA ) . ( C ) Expression of klf17 , irx7 and klf2b and was analyzed by qPCR at 512-cell , early 1K , and high stage in uninjected , control and pou5f3 + sox19b mRNA-injected embryos . Bar graphs focus on the early 1K stage . Error bars represent SEM ( n ≥ 4 ) . n . s p>0 . 05 ( two-tailed Student’s t-test , compared to control mRNA ) . ( D ) Staging by morphology was verified by cell counting at the stages used for the analysis . Each data point represents a single embryo . Error bars represent SEM . In B and C , expression is normalized to the expression of eif4g2α . DOI: http://dx . doi . org/10 . 7554/eLife . 23326 . 018 To reduce the level of Pou5f3 , we used a previously characterized morpholino ( Burgess et al . , 2002 ) and confirmed its effect by analyzing the morphology of injected embryos and the effect of the morpholino on the translation of injected RNA encoding Pou5f3 ( Figure 5—figure supplement 1A ) . We verified that the selected Pou5f3-target genes require Pou5f3 for their expression and that other genes do not ( Figure 5—figure supplement 1B ) , and analyzed the effect of a reduction in Pou5f3 levels on the timing of transcription of target genes ( Figure 5A ) . Consistent with our model , a reduction in the amount of Pou5f3 delayed the onset of transcriptional activation: transcripts were detected in the middle of 1K stage in embryos injected with control morpholino , while in the embryos injected with Pou5f3 morpholino , the genes started to be transcribed at high stage ( Figure 5B and Figure 5—figure supplement 1C ) . We analyzed embryos at mid 1K in this experiment , because the delay that we observe is weaker than with the histone cocktail . Comparison of gene expression levels in control morpholino and Pou5f3 morpholino-injected embryos at mid 1K ( when transcripts can first be detected in control morpholino-injected embryos ) revealed that the level of induction is significantly reduced upon injection of Pou5f3 morpholino ( Figure 5B and Figure 5—figure supplement 1C , bar graphs ) . Performing similar experiments for two additional transcription factors ( Sox19b and FoxH1 ) revealed that this effect is general , and not specific to Pou5f3 ( Figure 5—figure supplement 1D–G ) . Staging by morphology was corroborated by cell counting with absolute time between the analyzed stages being constant ( Figure 5—figure supplement 1H ) . Together , these data show that a decrease in the level of transcription factors in the embryo delays the onset of transcription of target genes . Next , we analyzed the effect of increasing the level of Pou5f3 on the transcription of the selected Pou5f3 target genes ( Figure 5A ) . We co-injected mRNA coding for Sox19b because it has been shown that Pou5f3 and Sox19b often co-occupy their target genes ( Chen et al . , 2014; Leichsenring et al . , 2013; Onichtchouk et al . , 2010 ) . Injecting mRNA encoding these transcription factors resulted in overexpression of both proteins and the expected phenotypes for Pou5f3 overexpression ( Figure 5—figure supplement 2A ) ( Belting et al . , 2011 ) . Although Pou5f3 was required for the expression of all genes we selected ( Figure 5—figure supplement 1B ) , Pou5f3 and Sox19b were only sufficient to increase the expression level of apoeb and dusp6 at high stage ( Figure 5—figure supplement 2B ) . In agreement with this observation , overexpression of Pou5f3 and Sox19b resulted in premature expression of apoeb and dusp6: transcripts were detected at early 1K stage in embryos injected with pouf53 and sox19b , whereas in the embryos injected with control mRNA , transcripts could be detected only at high stage ( Figure 5C ) . Comparison of gene expression levels in control , and pouf53 and sox19b mRNA-injected embryos at early 1K stage ( one time-point prior to when genes are first induced in uninjected embryos ) revealed that the level of expression is increased significantly upon injection of pou5f3 and sox19b mRNA for apoeb and dusp6 ( Figure 5C , bar graphs ) . Such an effect on the timing of transcription was not observed for the other genes ( Figure 5—figure supplement 2C ) . Staging by morphology was corroborated by cell counting , with absolute time between the analyzed stages being constant ( Figure 5—figure supplement 2D ) . These experiments show that an increase in the level of Pou5f3 and Sox19b in the embryo can cause premature transcription . Taken together , our results show that changing the concentration of endogenous transcription factors can affect the onset of transcription . This is in agreement with our model in which the relative levels of histones and transcription factors determine the onset of transcription . If transcription factors and histones compete for DNA binding , it would be predicted that transcription factor binding is sensitive to histone levels . To directly test competition at the chromatin level , we determined whether the binding of the transcriptional machinery is affected by histone levels . We analyzed the binding of Pou5f3 to its predicted target sites upstream of apoeb and dusp6 by chromatin immunoprecipitation ( ChIP ) and identified co-precipitated DNA fragments by qPCR ( Figure 5D ) . In embryos that were injected with mRNA encoding both Pou5f3 and Sox19b , we found that the binding of Pou5f3 was readily detected at early 1K stage ( Figure 5D , white bars ) . When the HC was co-injected , binding of the transcription factor was reduced ( Figure 5D , gray bars ) . We expect , but did not test , that nucleosome density is concordantly increased at these binding sites . A control region in genomic DNA did not show any binding of Pou5f3 ( Figure 5D ) . Taken together , this shows that the binding of an endogenous transcription factor is sensitive to the amount of histones present in the embryo . Our results support a model in which transcription is regulated by the relative levels of histones and transcription factors . Endogenous gene regulation , however , is intrinsically complex , with multiple transcription factors providing input on the same gene , and often there is limited information on the number and strength of activator-binding sites . Because this might have affected the results we obtained with endogenous transcription factors and genes ( Figure 5 ) , we decided to take advantage of a heterologous system to confirm our results . The integrated inducible transgene TRE:GFP ( Figure 6A ) contains seven binding sites for tTA–VP16 as well as a CMV promoter and is strictly dependent on tTA–VP16 for its expression ( data not shown ) . tTA-VP16 was tagged with HA and a protein product was detected at the 1K stage following injection of mRNA ( Figure 6—figure supplement 1A ) . Injection of 5 pg of mRNA encoding the heterologous transcription factor tTA-VP16 resulted in the detection of transcripts at high stage , in accordance with the onset of zygotic transcription of endogenous genes ( Figure 6B ) . Next , we analyzed the transcriptional activity of this transgene upon injection of 300 pg of mRNA encoding tTA-VP16 and we observed that transcripts could be detected at early 1K ( Figure 6B ) . Comparison of gene expression levels at early 1K stage ( one stage prior to when genes are first induced in embryos injected with 5 pg of mRNA ) in embryos injected with 5 and 300 pg of mRNA revealed that the number of transcripts is increased significantly upon increasing the level of transcription factor ( Figure 6B , bar graph ) . Next , we tested whether an increase in histone levels would negate the effect of high levels of transcription factor . As predicted by the competition model , the increase in transcriptional activity that is observed upon the injection of 300 pg of tTA-VP16 mRNA is lost when the histone cocktail is co-injected ( Figure 6C ) . Finally , we determined whether the binding of tTA is affected by histone levels . We analyzed the binding of tTA-VP16 to the TRE sites in the transgene by ChIP-qPCR ( Figure 6A ) . We found that upon injecting 300 pg of tTA-VP16 mRNA , the binding of the transcription factor was readily detected at early 1K stage ( Figure 6D and Figure 6—figure supplement 1B ) . As expected , binding of the transcription factor was significantly reduced when the HC was co-injected . Control regions within the transgene and in genomic DNA did not show any binding of tTA-VP16 ( Figure 6D and Figure 6—figure supplement 1B ) . This shows that the binding of a heterologous transcription factor is sensitive to the amount of histones present in the embryo . These results are in agreement with the results obtained with the endogenous transcription factors ( Figure 5 ) . Taken together , our data provide direct evidence for a model in which competition between histones and transcription factors determines the onset of transcription . 10 . 7554/eLife . 23326 . 019Figure 6 . Direct experimental evidence for the competition model using a heterologous transgene . ( A ) Schematic representation of the experimental procedure and TRE:GFP transgene . tTA-VP16 and/or histone levels were increased by injecting mRNA or HC into the cell or yolk , respectively , of 1-cell transgenic embryos . The TRE element contains seven binding sites for tTA-VP16 and is joined to a CMV promoter . qPCR and ChIP-qPCR analysis was carried out at stages around genome activation . Orange crosses represent the timing of stages used for the analysis . ( B ) Expression of gfp was analyzed by qPCR at 512-cell , early 1K and high stage in embryos injected with 5 or 300 pg tTA-VP16 mRNA . Bar graphs focus on the early 1K stage . Error bars represent SEM ( n ≥ 4 ) . p<0 . 05 ( two-tailed Student’s t-test , compared to 5 pg tTA-VP16 mRNA ) . ( C ) Expression of gfp was analyzed by qPCR at early 1K stage in embryos injected with 5 pg , 300 pg tTA-VP16 mRNA and 300 pg tTA-VP16 mRNA plus histone cocktail . Error bars represent SEM ( n = 4 ) . p<0 . 05 ( Ordinary one-way ANOVA ) . ( D ) Binding of tTA-VP16 to the TRE element and control regions was analyzed by ChIP-qPCR at the early 1K stage in embryos injected with 300 pg tTA-VP16 mRNA or 300 pg tTA-VP16 mRNA plus histone cocktail . Enrichment of pulled-down fragments was normalized to input . Primer sets used for ChIP-qPCR analysis are indicated by arrows in panel A . A control region on the transgene was used in addition to a genome control region on chromosome 23 . Error bars represent SEM ( n = 3 ) . p<0 . 05 ( two-tailed Student’s t-test ratio paired , compared to 300 pg tTA-VP16 mRNA-injected embryos ) . In B and C , mRNA levels are normalized to the expression of eif4g2α . DOI: http://dx . doi . org/10 . 7554/eLife . 23326 . 01910 . 7554/eLife . 23326 . 020Figure 6—figure supplement 1 . Direct experimental evidence for the competition model using a heterologous transgene . ( A ) Western blot using an HA antibody shows the protein level at 1K stage after injection of either 5 pg or 300 pg tTA-VP16 mRNA . Tubulin was used to control for equal loading . Blot shown is a representative example ( n = 2 ) . ( B ) Binding of tTA-VP16 to the TRE element was analyzed by ChIP-qPCR at the early 1K stage in embryos injected with 300 pg tTA-VP16 or 300 pg tTA-VP16 mRNA plus histone cocktail . Enrichment of pulled-down fragments was normalized to input . All primers used for ChIP-qPCR are indicated using arrows . Error bars represent SEM ( n = 3 ) . p<0 . 05 ( two-tailed Student’s t-test ratio paired , compared to 300 pg tTA-VP16 mRNA plus histone cocktail ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23326 . 020 Our observation that histone levels affect the time of transcription onset in zebrafish embryos is in agreement with previous studies that showed a role of histones in the regulation of transcription in early Xenopus embryos and extracts ( Almouzni and Wolffe , 1995; Amodeo et al . , 2015 ) . However , our finding that histones are neither completely depleted from the soluble fraction , nor generally depleted from chromatin , argues against a model in which a global loss of nucleosome density on chromatin causes the onset of transcription ( Amodeo et al . , 2015 ) . Our work does not exclude the importance of other factors , such as the linker histone H1 ( Pérez-Montero et al . , 2013 ) , the embryonic form of which is stably present during zebrafish genome activation , but it establishes that core histones themselves function as actual repressors of transcription . Our discovery that all core histones are required to regulate the onset of transcription suggests that the nucleosome is important for the repressor function of histones . In a previous study , premature transcription of injected plasmids caused by an excess of non-specific DNA was negated by the addition of the four core histones , but histones were not tested separately and it was not clear whether one or more histones were required for the observed effect on transcription ( Almouzni and Wolffe , 1995 ) . This left open the possibility that single histones could repress transcriptional activity in the embryo , for example by binding to a transcription factor and preventing it from binding to DNA . However , our observation that in the embryo all core histones are important for the regulation of transcription would then require all histones to independently take part in this mode of repression . Because this is a very unlikely scenario , we propose that repression takes place close to DNA , where histones are assembled into a histone octamer to form the nucleosome . We propose that genome activation follows a decrease in the concentration of non-DNA bound histones in the nucleus . One possible way to explain the reduction in nuclear histone concentration is the exponential increase in DNA content during the cleavage stages of zebrafish development . Because histones have a high affinity for DNA , an increase in the amount of DNA would titrate out non-DNA-bound histones . Several experiments have indeed shown that changes in DNA content can affect the time of transcription in Drosophila , zebrafish and Xenopus ( Dekens et al . , 2003; Lu et al . , 2009; Newport and Kirschner , 1982a; Prioleau et al . , 1994 ) . Our data suggest that these effects were the result of reducing the concentration of non-DNA-bound histones . In zebrafish embryos , the amount of histones is so large that the increase in DNA content leading up to genome activation may contribute only moderately to a decrease in the concentration of non-DNA-bound histones in the nucleus . Another possible explanation for the decrease in nuclear histone concentration is the marked increase in the ratio of nuclear over cytoplasmic volume during the cleavage stages ( Figure 4—figure supplement 1A ) . We suggest that this may limit the capacity of the nucleus to concentrate histones . The process of nuclear transport has been investigated in great detail ( Kim and Elbaum , 2013a , 2013b; Kopito and Elbaum , 2007 , 2009 ) , and we can assume that the nuclear envelope can create a certain fold difference in concentrations between the nucleus and cytoplasm . During the initial stages of zebrafish development , the nucleus occupies only a very small fraction of the cell volume ( 1 . 1% at 128-cell stage , Figure 4—figure supplement 1A ) . As a result , when the nucleus concentrates histones up to the maximum fold difference between cytoplasm and nucleus , the cytoplasmic histone concentration is hardly altered . Later in development , when approaching the onset of zygotic transcription , the nucleus takes up a notably larger part of the total cell volume ( 7 . 1% at high stage , Figure 4—figure supplement 1A ) . Now , when histones are imported into the nucleus , the concentration of histones in the cytoplasm noticeably decreases . The nuclear envelope is still able to create approximately the same fold difference of concentrations , but due to the reduced cytoplasmic concentration , the achieved nuclear concentration is not as high as during the initial stages . Thus , in this scenario , the nuclear histone concentration decreases due to the increasing relative nuclear size , which alters the distribution of histones among cellular compartments . It may be expected that the relative increase in nuclear size affects the nuclear concentration of transcription factors as well , and it remains to be seen how the concentration of histones and transcription factors change with respect to each other in order to activate transcription . Experiments recently performed in Xenopus showed that changing the size of the nucleus affects the timing of transcription ( Jevtić and Levy , 2015 ) , providing further evidence for a role of nuclear size in regulating the onset of transcription . Our finding that the onset of transcription depends on the concentration of histones , but also on the concentration of transcription factors , is consistent with previous studies that suggested an important role for transcriptional activators in the temporal regulation of zygotic transcription ( Almouzni and Wolffe , 1995; Prioleau et al . , 1994; Veenstra et al . , 1999 ) . Because transcription can be induced prior to the onset of genome activation , both by adding DNA ( Dekens et al . , 2003; Lu et al . , 2009; Newport and Kirschner , 1982a ) or removing histones ( this study ) , we suggest that transcription factors required for the onset of transcription are in principle present prior to genome activation . To shift the balance from repression to activation , the relative concentrations of histones and transcription factors need to be changed in favor of transcription factors . This would explain previous observations in Xenopus , where the addition of TBP ( in combination with adding DNA ) or GAL4-VP16 resulted in premature transcription ( Almouzni and Wolffe , 1995; Veenstra et al . , 1999 ) . Based on our findings , those experiments would have shifted the balance in favor of transcriptional activity , similar to the effect observed when we increased transcription factor levels or decreased histone levels . Our model in which histones and transcription factors dynamically compete for DNA binding to regulate transcription in the embryo is consistent with the notion that most transcription factors cannot bind DNA when it is wrapped around a nucleosome and thus compete with nucleosomes for DNA access ( Almouzni and Wolffe , 1993; Almouzni et al . , 1990; Hayes and Wolffe , 1992; Miller and Widom , 2003; Mirny , 2010; Ramachandran and Henikoff , 2016; Raveh-Sadka et al . , 2012; Schild-Poulter et al . , 1996; Svaren et al . , 1994 ) . Initially , there is a large excess of histones stockpiled in the embryo and transcription starts when the concentration of non-DNA-bound histones in the nucleus drops and the transcriptional machinery gains access to DNA . In contrast to a competition model that was previously proposed ( Prioleau et al . , 1994 , 1995 ) , transcription factors do not need to be pre-bound to DNA in order to compete with histones , but rather , they dynamically compete with histones for DNA binding . Our experiments did not address when competition takes place during the cell cycle . Competition for DNA binding might either occur immediately following replication , on temporarily naked DNA , or following chromatin assembly . Recent studies assessing the nucleosome landscape following replication have revealed that replication-coupled nucleosome assembly initially outcompetes transcription factors for binding to DNA but that chromatin remodeling and phasing of nucleosomes by remodelers and transcription factors occurs rapidly thereafter ( Fennessy and Owen-Hughes , 2016; Ramachandran and Henikoff , 2016; Vasseur et al . , 2016 ) . Future experiments will determine the details of competition in the early embryo , with its rapid cell cycles and large pool of soluble histones . To gain further insight in the molecular details of competition that lead up to genome activation , it will be important to determine which factors compete with histones for DNA binding . In theory , the binding of all factors that require access to DNA could be affected by histone levels , suggesting that competition might take place at many levels of transcription regulation: the formation of higher order chromatin structure , chromatin remodeling , the binding of transcription factors , and the assembly of the basal transcription complex . Our results show that the transcription factors that have been identified to regulate many genes during genome activation in zebrafish ( Pou5f3 and Sox19b ) ( Lee et al . , 2013; Leichsenring et al . , 2013 ) , as well as FoxH1 and the heterologous transcription factor tTA-VP16 , compete with histones for binding to DNA ( Figures 5 and 6 ) . In this context , it is interesting to note that the transcription factors that have been identified to play a role in genome activation have either been suggested to be pioneer factors ( Lee et al . , 2013; Leichsenring et al . , 2013 ) , or there is indication for such a role because of their homology with mammalian pioneer factors ( Lee et al . , 2013; Leichsenring et al . , 2013; Soufi et al . , 2012 ) . Pioneer factors are able to interact with DNA that is nucleosome bound ( Zaret and Carroll , 2011 ) . In the context of the competition model , it will be interesting to see whether these factors also have pioneering activity in the early embryo , and how this affects their role in activating transcription in the embryo . The applicability of the competition model might extend well beyond the onset of zygotic transcription in zebrafish . First , given the excess of histones in a large number of species including Drosophila , Xenopus , and zebrafish ( Adamson and Woodland , 1974; Li et al . , 2012; Marzluff and Duronio , 2002; Osley , 1991; Vastenhouw et al . , 2010; Woodland and Adamson , 1977 ) , it is likely that histone levels play a role in the timing of zygotic transcription across these species . As discussed , in Xenopus embryos there is indeed evidence for a role of histone levels in regulating transcriptional activity ( Almouzni and Wolffe , 1995; Amodeo et al . , 2015 ) . Additional experiments will be required to determine whether the competition model we propose applies to these and other species . Second , the onset of zygotic transcription in the embryo takes place in the context of the mid-blastula transition and is accompanied by a lengthening of the cell cycle and changes in chromatin structure . Although it had previously been suggested that the rapid cell cycles lacking G1 and G2 phases might interfere with productive transcription during early developmental stages ( Collart et al . , 2013; Edgar and Schubiger , 1986; Kimelman et al . , 1987 ) , it was recently shown that the lengthening of the cell cycle might be a direct consequence of the onset of transcription in Drosophila embryos ( Blythe and Wieschaus , 2015b ) . This would suggest that what regulates the onset of zygotic transcription might also dictate the lengthening of the cell cycle . Finally , post-translational modification of histones often requires a chromatin-modifying enzyme to bind to DNA , much like transcription factors . Thus , competition is likely to affect the de novo modification of histones as well , explaining why many histone modifications are only observed around the onset of zygotic transcription in their temporal profile ( Lindeman et al . , 2011; Vastenhouw et al . , 2010 ) . Importantly , we observe an effect on the timing of transcription by adding unmodified histones . This suggests that post-translational modifications of histones are either downstream of the timing of transcriptional activation , or the enzymes that modify histones are not limiting in the embryo . The competition model can explain why genome activation is gene specific ( Aanes et al . , 2011; Collart et al . , 2014; Harvey et al . , 2013; Heyn et al . , 2014; Lott et al . , 2011; Owens et al . , 2016; Pauli et al . , 2012; Sandler and Stathopoulos , 2016; Tan et al . , 2013 ) , and even why the first zygotic transcripts can be detected several cell cycles before the stage that is traditionally defined as the time point of ZGA ( De Renzis et al . , 2007; Heyn et al . , 2014; Skirkanich et al . , 2011; Yang et al . , 2002 ) . The sensitivity of genes for a given histone concentration logically depends on their enhancers and the affinity and concentration of the transcription factors that bind to them . In this context , it is interesting to note that many genes that are activated during zebrafish genome activation respond to the same set of transcription factors , which are also the most highly translated transcription factors before genome activation ( Lee et al . , 2013; Leichsenring et al . , 2013 ) . Conversely , the affinity of transcription factors and the number of transcription-factor-binding sites might provide a mechanism to explain why some genes overcome repression earlier than others ( Heyn et al . , 2014 ) . Indeed , in Drosophila , it has been shown that the number of transcription-factor-binding sites as well as the level of transcription factors can affect the timing of gene expression ( Foo et al . , 2014; Harrison et al . , 2010 ) . Recent literature suggests that histone levels might play a role in the regulation of transcription during developmental transitions other than genome activation . In contrast to the situation in early embryos , where histone and DNA levels do not scale , it was generally believed that in somatic cells , histone production is tightly coupled to replication ( Nurse , 1983 ) . Recently , however , histone levels have been shown to change during ageing and differentiation ( Feser et al . , 2010; Hu et al . , 2014; Karnavas et al . , 2014; O'Sullivan et al . , 2010 ) suggesting that histone levels might not be as tightly coupled to DNA replication as previously thought . Moreover , histone chaperones were identified as inhibitors of reprogramming ( Cheloufi et al . , 2015; Ishiuchi et al . , 2015 ) and it was proposed that the lack of histone chaperones facilitates transcription factor binding ( Cheloufi et al . , 2015 ) . Taken together , these studies might suggest that the availability of histones could play a role in the regulation of transcription during differentiation and reprogramming . We have shown that the onset of transcription is regulated by a dynamic competition for DNA binding between histones and transcription factors . This suggests that the relative levels of histones and transcription factors in the nucleus determine the time at which transcription begins in the embryo . Future studies will be required to improve our understanding of the molecular mechanism of competition , the regulation of repressor and activator concentrations in the nucleus , and the role of competition during other developmental transitions . Zebrafish were maintained and raised under standard conditions . Wild-type ( TLAB ) ( WT-TL RRID:ZIRC_ZL86 , WT-AB RRID:ZIRC_ZL1 ) and transgenic embryos were dechorionated immediately upon fertilization , synchronized and allowed to develop to the desired stage at 28°C . Stage was determined by morphology and corroborated by cell counting . In terms of absolute time , the time between collected stages around ZGA was consistent between all conditions within an experiment and for all experiments . Histone cocktail and BSA ( A9418; Sigma , St . Louis , MO ) were injected into the yolk at the 1-cell stage at 22 ng per embryo . Pou5f3 anti-sense morpholino was injected at 6 ng per embryo , together with 1 ng of p53 morpholino ( Langheinrich et al . , 2002 ) . Sox19b anti-sense morpholino ( Okuda et al . , 2010 ) was injected at 2 ng per embryo and FoxH1 anti-sense morpholino ( Pei et al . , 2007 ) was injected at 4 ng per embryo , together with 1 ng of p53 morpholino . Dead-end ( Weidinger et al . , 2003 ) or control morpholino were injected as a control at the same concentration . Morpholino sequences can be found in Table 1 . α amanitin ( A2263; Sigma ) was injected at the 1-cell stage at a concentration of 0 . 2 ng per embryo . 2 . 8 mg/ml rhodamine-dextran ( D3307; Molecular Probes , Eugene , OR ) was used as an injection marker for the HC and BSA experiments . For all other injections , 0 . 1% Phenol red ( P0290; Sigma ) was injected . Bright-field images of whole embryos were acquired on a Leica M165 C dissecting scope equipped with a Leica MC170 HD camera ( Leica , Wetzlar , Germany ) . 10 . 7554/eLife . 23326 . 021Table 1 . List of morpholinos used . DOI: http://dx . doi . org/10 . 7554/eLife . 23326 . 021MorpholinosTargetSequenceCompanyReferencep535’-GCGCCATTGCTTTGCAAGAATTGGeneToolsLangheinrich et al . ( 2002 ) Pou5f35′-CGCTCTCTCCGTCATCTTTCCGCTAGeneToolsBurgess et al . ( 2002 ) Sox19b5'-ACGAGCGAGCCTAATCAGGTCAAACGeneToolsOkuda et al . ( 2010 ) Foxh15'-TGCTTTGTCATGCTGATGTAGTGGGGeneToolsPei et al . ( 2007 ) Dead-end5'-GCTGGGCATCCATGTCTCCGACCATGeneToolsWeidinger et al . ( 2003 ) Ctrl MO5'-CCTCTTACCTCAGTTACAATTTATAGeneToolsGeneTools , LLC mRNA was synthesized using the Ambion mMESSAGE mMACHINE SP6 Transcription Kit ( AM1430; ThermoFisher Scientific , Waltham , MA ) . Human PTX3 cDNA was cloned into a pCS2+ vector with a C-terminal RFP . ptx3-rfp and rfp mRNA were injected into the cell at the 1-cell stage at a concentration of 300 pg per embryo . Zebrafish Pou5f3 and Sox19b cDNA were cloned into a pCS2+ vector containing 2xHA sequences . For gene expression experiments , pou5f3-2xHA and sox19b-2xHA mRNA were each injected into the cell of 1-cell embryos at 300 pg per embryo . mRNA encoding cytoplasmic gfp was injected as a control at 600 pg per embryo . For ChIP-qPCR experiments , pou5f3-2xHA and sox19b-mEos2 mRNA were each injected into the cell of 1 cell embryos at 150 pg per embryo . A subsequent injection of either histone cocktail or mock ( histone buffer ) into the yolk was carried out . Human H4 cDNA was cloned into a pCS2+ vector with C-terminal sfGFP ( 50550; addgene , Cambridge , MA ) ( Olson et al . , 2014 ) . mRNA encoding H4-sfGFP was injected into the cell of 1-cell embryos at 240 pg per embryo . tTA-VP16 DNA was cloned into a pCS2+ vector containing 2xHA sequences . mRNA encoding tTA-VP16-2xHA was injected into the cell of 1-cell Tg ( TRE:GFP ) embryos either at 5 pg or 300 pg per embryo . The combination injection of 300 pg tTA-VP16-2xHA mRNA and histone cocktail , involved two subsequent injections into the cell of 1-cell embryos and yolk , respectively . The 300 pg tTA-VP16-2xHA mRNA only injections also received a secondary mock injection into the yolk . Twenty-five embryos per developmental stage were snap frozen in liquid nitrogen . RNeasy Mini Kit ( 74104; Qiagen , Venlo , the Netherlands ) was used to extract RNA . For Tg ( TRE:GFP ) embryos , contaminating DNA was removed from RNA preparations using the DNA-free Kit ( AM1906; ThermoFisher Scientific ) . mRNA was converted to cDNA using the iScript cDNA Synthesis Kit ( 1708891; Bio-Rad Laboratories , Hercules , CA ) . SYBR green ( AB-1158 . ; ThermoFisher Scientific ) with Rox ( R1371; ThermoFisher Scientific; 100 nM ) was used as the qPCR master mix . Primers were used at a final concentration of 500 nM and sequences can be found in Table 2 . Two or three technical replicates were performed for each sample . Ct values were normalized to the maternally loaded gene eif4g2a or input in ChIP-qPCR analysis . Relative mRNA expression levels were calculated via 1/ ( 2∧ ( gene-eif4g2a ) ) . Fold difference was calculated by dividing the relative mRNA expression level value of the test sample over control . 10 . 7554/eLife . 23326 . 022Table 2 . List of primers used . Location of primer sets with respect to transcription start-sites are indicated in brackets . DOI: http://dx . doi . org/10 . 7554/eLife . 23326 . 022Primer listGenePrimerseif4g2α5’-GAGATGTATGCCACTGATGAT5’-GCGCAGTAACATTCCTTTAGmxtx25’-ACTGACTGCATTGCTCAA5’-ACCATACCTGAATACGTGATTfam212aa5’-GCAAATGAGTATCTAAAACTGCT5’-CATCATATAGCGCATCTGGTnnr5’-GAGACATACCACAGGTGAAGC5’-CCGCTCTGGTCTGTTGCvox5’-TTATTCGTCGGGTTATGAGAG5’-AACCAAGTTCTGATCTGTGTsox19a5’-GAGGATGGACAGCTACGG5’-CTATAGGACATGGGGTTGTAGgrhl35’-AGACGAGCAGAGAGTCCT5’-TTGCTGTAATGCTCGATGATGapoeb5’-GCAGAGAGCTTGACACACTAA5’-TGCATTCTGCTCCATCATGGdusp65'-AGCCATCAGCTTTATTGATGAG5'-CAAAGTCCAAGAGTTGACCCklf175’-ATAGTTCGGGACTGGAAAGTTG 5’- TGAGGTGTTGTCGTTGTCAGirx75’-TGGCACACATTAGCAATTCC5’-GCATGATCTTCTCGCCTTTGklf2b5’-GCTCTGGGAGGATAGATGGA5’-CTCGGAGTGGGAGATGAACflh5'-CACTGAAGCTCAGGTTAAAGTC5'-ACAATCTGGGGAAAATCATGGwnt115'-CAGACAGGTGCTTATGGACT5'-CATCTCTCGGGGCACAAGgadd45bb5'-CAACTCATGAATGTGGATCCAG5'-ATGCAGTGAAGGTCTCTTGGGFP5’-GCACCATCTTCTTCAAGGAC5’-TTGTCGGCCATGATATAGACPou5f3 binding site ( −2270 apoeb ) 5'-TAAAGTGAGCAAATGTATGGCC5'-TTTGTTGATTAAATCGCTTGTGAPou5f3 binding site ( −3095 dusp6 ) 5'-CATATGTTAAGCGGGGTGAAAC5'-ATCCTGTCTCCTGTGTCATTTGTRE binding site ( −222 ) 5'-TCTTGATAGAGAGGCTGCAAAT5'-TCGAGATGGGCCCTTGATATRE binding site ( 13 ) 5'-TCGTATAGGGATAACAGGGTAATG5'-TACACGCCTACCTCGACCTRE binding site ( 217 ) 5'-GTACGGTGGGAGGCCTATAT5'-CTTCTATGGAGGTCAAAACAGCTransgene control5'-CTCTACAAATGTGGTATGGCTG 5'-ATTACCCTGTTATCCCTAAGGCGenomic control5'-CCATCATATTCACATCTTGCAAG5'-GTTCGTATGAACCGGAAGC Embryos were fixed in 4% formaldehyde in Danieau’s solution at 4°C overnight . The next day , embryos were washed with Danieau’s solution and then permeabilized with 0 . 2% Triton X in Danieau’s solution for 30 min . Subsequently , embryos were incubated for 10 min in DAPI ( 1 µg/ml ) and then washed several times with Danieau’s solution . Embryos were placed in an inverted agarose holder and covered with Danieau’s solution for imaging . An upright Zeiss LSM 780 NLO microscope equipped with a coherent Chameleon Vision II infrared laser was used for two-photon excitation ( Carl Zeiss AG , Oberkochen , Germany ) . DAPI was excited with 780 nm and detected using a non-descanned GaAsP detector ( BIG-Module ) with BP450/60 or SP485 . Samples were imaged with either a Zeiss W Plan-Apochromat 20 × 1 . 0 or 40 × 1 . 0 dipping objective . Images were acquired using a four tile scan of multiple z-sections ( 3–3 . 5 µm steps ) . Tiles were stitched with the ZEN software ( RRID:SCR_013672 , Zeiss ) . Images were imported into the Imaris software ( RRID:SCR_007370 , Bitplane , Belfast , Northern Ireland ) and the spot tool was used to calculate cell number . Embryos were manually deyolked at the desired stage and snap frozen in liquid nitrogen . For all proteins , equal numbers of embryos were analyzed for each developmental stage ( H4 and H2A [n = 10] , all other proteins [n = 5] ) . Samples were boiled with SDS loading buffer at 98°C , run on 4–12% polyacrylamide NuPAGE Bis-Tris gels ( NP0321BOX; ThermoFisher Scientific ) and blotted onto a nitrocellulose membrane ( 10600002; GE Life Sciences , Chicago , IL ) . Primary antibodies were incubated at RT for 1 hr or overnight at 4°C and secondary antibodies were incubated at RT for 45 min . Primary and secondary antibodies used are listed in Tables 3 and 4 , respectively . Membranes were analyzed on an Odyssey Infrared Imaging System ( LI-COR , Lincoln , NE ) or via chemiluminescent detection ( GE Life Sciences ) and X-ray film ( GE Life Sciences ) . Tubulin was examined visually on all blots as a loading control . 10 . 7554/eLife . 23326 . 023Table 3 . List of primary antibodies used . DOI: http://dx . doi . org/10 . 7554/eLife . 23326 . 023Primary antibodiesTarget/NameCompanyCompany codeRRID[WB][IF][IP]H3Abcamab1791AB_3026131:10 , 000H4Abcamab10158AB_2968881:10001:300H2AAbcamab18255AB_4702651:1000H2BAbcamab1790AB_3026121:3000α-tubulinSigmaT6074AB_4775821:20 , 000RFPAbcamab152123AB_2637080ExcessPTX3Cosmo BioPPZ1724AB_19622801:15 , 000RNA Pol IIBioLegendMMS-126RAB_100136651:1000HAAbcamab9110AB_3070191:5000ExcessIgG from rabbit serumSigmaI5006AB_1163659ExcessWB , Western blotting; IF , immunofluorescence; IP , immunoprecipitation . 10 . 7554/eLife . 23326 . 024Table 4 . List of secondary antibodies used . DOI: http://dx . doi . org/10 . 7554/eLife . 23326 . 024Secondary antibodiesNameCompanyCompany codeRRID[WB][IF]Alexa 488 goat anti-mouse IgG H&LThermoFisherA-11029AB_1384041:1000Alexa 594 goat anti-rabbit IgG H&LThermoFisherA-11037AB_25340951:500IRDye 800CW donkey anti-rabbit IgG H&LLI-CORP/N 926–32213AB_6218481:20 , 000IRDye 800CW donkey anti-mouse IgG H&LLI-CORP/N 926–32212AB_6218471:20 , 000Peroxidase AffiniPure goat anti-rabbit IgG H&LJackson ImmunoResearch111-035-144AB_23073911:20 , 000Peroxidase AffiniPure rabbit anti-mouse IgG H&LJackson ImmunoResearch315-035-003AB_23400611:20 , 000WB , Western blotting; IF , immunofluorescence . We selected five proteotypic peptides ( Worboys et al . , 2014 ) for each of the four histones: H3 , H4 , H2A and H2B . The peptides do not discriminate between known histone variants for H3 and H2A . A chimeric gene encoding these peptides ( Beynon et al . , 2005 ) in addition to reference peptides from BSA and Glycjogen Phosphorylase B ( PhosB ) ( five each ) , and flanked by Strep- and His-tags , was chemically synthesized ( Gene Art , ThermoFisher Scientific ) . This gene was expressed in a Lys , Arg dual-auxotroph E . coli strain ( BL21DE3pRARE ) that was grown in media complemented with 13C15N-Arg and 13C-Lys ( Silantes , Munich , Germany ) . In a separate LC-MS/MS experiment , we established that the full-length chimeric protein was correctly expressed and the rate of incorporation of isotopically labeled amino acids was ca . 99% . The gel band corresponding to the chimeric protein was co-digested with the gel slab containing the histones from the samples of interest ( Shevchenko et al . , 2006 ) and with the band containing the exactly known amount of the reference protein ( BSA ) . The recovered tryptic peptides were analysed by nanoLC-MS/MS on a LTQ Orbitrap Velos coupled with Dionex Ultimate 3000 nano-HPLC system ( ThermoFisher Scientific ) . Three biological replicates were analyzed for each sample , with two technical replicates for each sample . The peptides were separated using C18 reversed phase column ( Acclaim PepMap 100 ) over a linear gradient from 0 to 55% solvent B in a mixture of solvents A and B , delivered in 120 min ( Solvent A 0 . 1% FA , Solvent B 60% ACN + 0 . 1% FA ) . The identification of peptides was performed using Mascot v2 . 2 . 04 ( Matrix Science , London , United Kingdom ) against a custom-made database composed of sequences from all histones , BSA , PhosB , affinity tag and the sequences of common contaminants such as human keratins and porcine trypsin . The raw abundances of extracted ion chromatograms ( XIC ) peaks of peptide precursors were reported by Progenesis LC-MS v4 . 1 software ( Nonlinear Dynamics , Newcastle , United Kingdom ) . First the chimeric protein was quantified by comparing the abundances of BSA peptides comprised in its sequence with the corresponding peptides obtained by co-digestion of a known amount of BSA protein standard . In turn , the molar content of target zebrafish histones was inferred from the content of the chimera protein and the ratio of relative abundances of XIC peaks of precursor ions of matching pairs of labeled ( originating from chimera ) and unlabeled ( originating from histones ) peptides . Note that all peptides were recovered from the same in-gel digest and quantified at the same LC-MS/MS run . Absolute histone amounts were measured using mass spectrometry . The number of histones bound to a diploid zebrafish genome was calculated as 31 , 324 , 994 per genome for each histone . To arrive at this calculation ( Figure 4—source data 1 ) , we have previously shown that the average distance between the centers of neighboring nucleosomes in the zebrafish embryo around genome activation is 187 base pairs ( Zhang et al . , 2014 ) . The size of a zebrafish genome is 1 . 46 Gb ( GRCz10 ) . This was multiplied by two to reach the diploid genome size which was then divided by the nucleosome repeat length to arrive at the number of nucleosomes per genome . As each histone is represented twice in a nucleosome , this number was multiplied by two to arrive at 3 . 13 × 107 copies of each histone that are required to wrap one zebrafish genome into chromatin ( see Figure 4—source data 1 ) . To arrive at numbers of histones in ‘genomes worth of histones’ , the actual number of histones was divided by the amount of histones required to wrap one diploid genome . For the excess of histones per cell calculation ( Figure 4D ) , the level of H2B ( Figure 4—source data 1 ) and the cell numbers in Figure 1—source data 1 were used . Because we have shown that all four core histones contribute to the repressor effect , we used the level of H2B for these calculations , as H2B is the lowest abundant histone and therefore may be limiting for the formation of nucleosomes . We subtracted the number of histones that are assumed to be bound to DNA , which amounts to one to two genomes worth of histones assuming replication ( we used the average ) . The total concentration of non-DNA bound histones was calculated by dividing the total amount of non-DNA-bound histones by the volume of the animal cap ( Figure 4—figure supplement 1A ) . Recombinant histones were of human origin and produced in E . Coli ( NEB , Ipswich , MA; 1 mg/ml: H3 . 1 M2503S , H4 M2504S , H2A M2502S , H2B M2505S ) . We used human histones because histones are highly conserved between species and these are readily available . To remove DTT , H3 . 1 was dialyzed in histone buffer ( 300 mM NaCl , 1 mM EDTA , 20 mM NaPO4 , pH 7 . 0 at 25°C ) using a Slide-A-Lyzer MINI dialysis device , 7K MWCO ( ThermoFisher Scientific ) at RT for 30 min or at 4°C overnight . Stoichiometric amounts of all four core histones were combined , spun for 5 min at 6600 rcf on a bench top centrifuge , supernatant was removed , and recovery of histones was measured via quantitative Western blot analysis and calculated using a standard . On average 5756 genomes worth of histone were injected with an error of ±388 ( n = 3 ) . This method involves assigning a unique color-coded barcode to transcripts of interest for single-molecule imaging and counting . The number of times the unique barcode is detected , is used as a readout of the expression level or number of ‘counts’ for the gene of interest . We developed a custom-made probe set of zygotically expressed genes as well as control genes ( Figure 2 , Figure 2—source data 1 ) . Probe-sets were hybridized to 100 ng of mRNA extracted from a batch of 25 embryos using the RNeasy Mini Kit and processed following the manufacturer’s recommendations ( NanoString Technologies , Seattle , WA ) ( Kulkarni , 2011 ) . More information about the analysis can be found in the legend of Figure 2—figure supplement 2 . At the desired stage , 65–100 embryos were manually deyolked and snap frozen in a cell lysis buffer ( CLB: 10 mM HEPES pH 7 . 9 , 10 mM KCl , 1 . 5 mM MgCl2 , 0 . 34 M sucrose , 10% glycerol , 0 . 1% NP-40 , 1x protease inhibitor ( Roche , Basel , Switzerland ) ) ( Méndez and Stillman , 2000 ) . Thawed embryos were shaken at 4°C for 5 min , then placed on ice and flicked intermittently for 5 min . Samples were spun in a bench top centrifuge at 1700 rcf for 5 min . Supernatant was removed and the pellet was washed with CLB . After another spin , the pellet was washed with a nuclear lysis buffer ( 3 mM EDTA , 0 . 2 mM EGTA ) . The sample was spun down and resuspended with high-salt solubilization buffer ( 50 mM Tris-HCL pH 8 . 0 , 2 . 5 M NaCl , 0 . 05% NP-40 , 1x protease inhibitor ) ( Shechter et al . , 2009 ) . The sample was vortexed for 2 min and placed on a rotator at RT for 10 min . The complete sample was then used in Western blot analysis . Per IP , 500 staged embryos were deyolked as previously described ( Link et al . , 2006 ) . Cells were immediately resuspended in cell lysis buffer ( 10 mM Tris-HCl at pH 7 . 5 , 10 mM NaCl , 0 . 5% NP-40 , 1x protease inhibitor ( Roche ) ) , and lysed for 15 min on ice . Nuclei were pelleted by centrifugation and the supernatant was collected and rotated overnight at 4°C with 25 mL of protein G magnetic Dynabeads ( Invitrogen , Carlsbad , CA ) that had been pre-bound to an excess amount of antibody . Bound complexes were washed six times with RIPA buffer ( 50 mM HEPES at pH 7 . 6 , 1 mM EDTA , 0 . 7% DOC , 1% Igepal , 0 . 5 M LiCl , 1x protease inhibitor ) followed by 10 min of boiling in SDS loading buffer . Beads from the sample were subsequently removed by centrifugation and Western blotting was used for further analysis . Histone H4 was incubated with Cyanine5 NHS ester ( 10:1 molar ratio ) ( Lumiprobe , Hannover , Germany ) rotating overnight at 4°C . The next day , the solution was dialyzed in histone buffer for 30 min at RT . ~1 ng was injected into embryos of the Tg ( h2afz:h2afz-GFP ) transgenic fish line ( Pauls et al . , 2001 ) and embryos were imaged live on an upright LSM 510 META microscope equipped with a Zeiss W Plan-Apochromat 40 × 1 . 0 dipping objective . GFP was excited at 488 nm , detected with a PMT using BP527 . 5/545 and a pinhole size of 72 µm . Cy5 was excited at 633 nm , detected with the META detector using BP649-756 and a pinhole size of 96 µm . Images are 512*512 pixels , pixel size is 0 . 22 µm and were acquired with eight-bit mode . We determined the concentration of non-DNA bound histones in the nucleus as follows . First , we obtained volumetric data from live embryos , in which H4-sfGFP fusion protein was translated from injected mRNA to label animal cap and cell nuclei , at low- and high-intensity levels , respectively ( see ‘Live embryo tracking of nuclei and animal cap volumes’ and ‘Automated image analysis’ below ) . Imaging live embryos prevented volume alterations due to fixation , permeabilization , and wash steps in immunofluorescence . Next , we determined relative histone distributions in cytoplasm and nucleus by immunofluorescence detection of endogenous histone H4 , thus avoiding potential offsets or sub-cellular redistribution of the endogenous histone pool due to the addition of labeled fusion protein ( see ‘Immunofluorescence’ and ‘Automated image analysis’ below ) . Then , we combined volumetric and nuclear-over-cytoplasmic intensity ratio data to allocate the total amount of histone H4 per embryo , as measured by mass spectrometry ( Figure 4—source data 1 ) , to the cytoplasmic and the nuclear sub-compartment ( see ‘Calculation of non-DNA-bound nuclear histone concentration’ below ) . Lastly , aiming to calculate the concentration of only non-DNA-bound histones , the histones bound on chromatin in a given nucleus were subtracted from the total nuclear concentration of histones . To monitor the volumes of the animal cap and individual nuclei as well as nuclear import dynamics , histone H4 and PCNA were imaged in whole live embryos at a time resolution of 2 min or faster ( see Figure 4—source data 2 ) . H4 was introduced as a fusion with sfGFP by mRNA injection . PCNA was monitored using offspring of transgenic fish with PCNA-RFP ( Tg ( bactin:RFP-pcna ) [Strzyz et al . , 2015] ) . Embryos were mounted in glass capillaries with 1% low-melting agarose ( Invitrogen ) dissolved in 0 . 3x Danieau’s solution and imaged with a Zeiss Z . 1 lightsheet microscope using a 10x water dipping objective ( NA 0 . 5 ) for acquisition , a lightsheet thickness below 5 µm , and dual side illumination ( Icha et al . , 2016 ) . The microscopy chamber was filled with 0 . 3x Danieau’s and kept at 28 . 5°C . Optical sectioning was 1 or 1 . 5 µm , time resolution was 2 min or faster for the acquisition of a whole 3D stack . A time series of wild-type TLAB embryos covering 64-cell to sphere stages was collected , immunostained following a protocol optimized for full transparency and penetration of antibody , and imaged using a Zeiss Z . 1 lightsheet microscope . Wild-type TLAB embryos were transferred at a given stage by transfer from 0 . 3x Danieau’s into 2% formaldehyde in 0 . 3x Danieau’s with 0 . 2% Tween-20 and left to fix overnight at 4°C . On the next day , embryos were washed three times for 10 min in PBST ( Dulbecco’s PBS with 0 . 1% Tween-20 ) , then further permeabilized by washing twice in double-distilled water followed by 5 min waiting at room temperature , and then blocked with 4% BSA in PBST with 1% DMSO for at least 30 min . Primary antibodies against histone H4 and RNA polymerase II were diluted in 2% BSA in PBST with 1% DMSO and applied for incubation at 4°C for at least 48 hr . Embryos were washed three times for 10 min in PBST . Secondary antibodies were diluted in 2% BSA in PBST with 1% DMSO and applied overnight or longer at 4°C . Embryos were then washed three times for at least 10 min in PBST and stored at 4°C until imaging . Mounting for imaging was done in glass capillaries using 2% low-melting agarose dissolved in Dulbecco’s PBS . 3D stacks were acquired using a 20x water dipping objective ( NA 1 . 0 ) for acquisition , a lightsheet thickness below 5 µm , and dual side illumination . The microscopy chamber was filled with Dulbecco’s PBS . Optical sectioning was 1 µm or less . Microscopy data were analyzed with a custom MatLab code using the Open Microscopy Environment bioformats plugin for stack reading . Nuclei were segmented using iterative thresholding for individual nuclei to compensate for differing intensities across the sample . 3D segments representing nuclei were dilated in two steps , giving a once- and a twice-extended shell around any given nucleus ( Stasevich et al . , 2014 ) . The once-extended shell was removed from the twice-extended shell , along with any other nuclei that happened to be covered by the twice-extended shell . The resulting 3D segment thus covered cytoplasm in the vicinity of a given nucleus . The nucleus and the cytoplasm 3D segments were then used as masks to extract the mean intensity of a given cell’s nucleus and cytoplasm . The animal cap was segmented in 3D based a single , global threshold determined from maximum intensity z-projections using Otsu’s method . ( For code , see Hilbert L . 2016 GitHub . https://github . com/lhilbert/NCRatio_Analysis . a7a5849 ) . For live-imaging data , individual nuclei were tracked across consecutive time frames based on minimal centroid distances . The volume fraction of the animal cap taken up by nuclei was calculated from the sum of volumes of all nuclei detected in a given stage , at their individual times of maximal extension in the respective cell cycle . ( For code , see Hilbert L . 2016 GitHub . https://github . com/lhilbert/NucCyto_Ratio_TimeLapse . 55ed0fc ) . For immunofluorescence data , nuclei were segmented based on the Pol II signal , which exhibited strong nuclear localization during interphase for all stages . Nuclear and cytoplasmic intensities for both Pol II and H4 were then extracted based on the Pol II segmentation as described above . To remove nuclei that were not in interphase or suffered signal degradation due to excessive spherical aberration or out-of-focus light , only nuclei with a nuclear-over-cytoplasmic intensity ration of greater than two were included in the analysis . Intensity ratios were strongly affected by background staining , so that H4 intensity values were corrected by subtraction of background levels before calculating ratios . Background levels were obtained from control embryos incubated with secondary , but not primary antibodies , which were imaged in the same session and with the same settings as the fully stained samples . To obtain nuclear histone concentration values , one considers that the total number of histones must correspond to the contributions from all cells’ cytoplasm and nuclei , Htotal=[Hnuclear]×Vnucleussum+[Hcytoplasm]×Vcytoplasmsum , where Htotal , Hnuclear , Hcytoplasm are the total , the nuclear , and the cytoplasmic concentration of endogenous histone H4 , respectively . Vnucleussum and Vcytoplasmsum are the summed volumes of all cells’ nuclei and cytoplasm , respectively . Dividing both sides by the total animal cap volume , Vtotal , one findsHtotalVtotal=[Hnuclear]×v+[Hcytoplasm]× ( 1−v ) , where v=Vnucleussum/Vtotal is the fraction of the total animal cap volume taken up by nuclei ( also corresponds to the average fraction of cell volume occupied by the cell nucleus ) . Considering the N/C intensity ratio , R , to represent the concentration ratio , R≈[ Hnuclear ]/[Hcytoplasm] , one can solve for [Hnuclear] , [Hnuclear]=HtotalVtotal×R1+v ( R−1 ) . Realizing that this measured nuclear concentration results from non-DNA-bound histones as much as chromatin bound histones , one needs to subtract the concentration of chromatin bound histones to arrive at the non-DNA-bound histone H4 concentration , [ Hfree ]=[ Hnucleus ]−[ Hbound ]=[ Hnucleus ]−gVnucleussingle . g quantifies the number of complete , histone wrapped genomes ( in units of genomes worth ) being present in the volume of an individual nucleus , Vnucleussingle . Dropping the single superscript for ease of notation , the final expression is[Hfree]=HtotalVtotal×R1+v ( R−1 ) −gVnucleus . We measured all variables except g on the right hand side using mass spectrometry ( [Htotal] ) , lightsheet imaging of whole live embryos injected with mRNA for H4-sfGFP ( Vtotal , v , Vnucleus , see above ) , or immunofluorescence of endogenous histone H4 ( R ) ( see Figure 4—figure supplement 1B ) . g was assigned a value of 1 . 5 genomes worth , to fall between 1 ( before replication of the genome ) and 2 ( complete replication of the genome ) , under the assumption of full occupation of the DNA by histones . Per IP , ~550 staged embryos were fixed at RT for 15 min in 1 . 85% formaldehyde . The fixative was quenched with 125 mM glycine and rotation at RT for 5 min . Embryos were then rinsed 3x in ice cold PBS ( Accugene , Willowbrook , IL ) , resuspended in cell lysis buffer ( same as co-IP ) and lysed for 15 min on ice . Nuclei were pelleted by centrifugation , resuspended in nuclear lysis buffer ( 50 mM Tris-HCl at pH 7 . 5 , 10 mM EDTA , 1% SDS , 1x protease inhibitor ) and lysed for 10 min on ice . Two volumes of IP dilution buffer ( 16 . 7 mM Tris-HCl at pH 7 . 5 , 167 mM NaCl , 1 . 2 mM EDTA , 0 . 01% SDS , 1x protease inhibitor ) was added and the sample was sonicated to produce DNA fragments of between 200 and 300 bp ( for Pou5f3-2xHA ChIP ) or 400 and 500 bp ( for tTA-VP16-2xHA ChIP ) as determined using a bioanalyzer . 0 . 8% Triton X was added to the chromatin , which was then centrifuged to remove residual cellular debris . A sample was saved for input and the rest was divided over 25 mL of protein G magnetic Dynabeads that had been pre-bound to an excess amount of either HA antibody or IgG control antibody ( Table 3 ) . These were rotated overnight at 4°C . Bound complexes were washed six times with RIPA buffer followed by TBS . Elution buffer ( 50 mM NaHCO3 , 1% SDS ) was added to the beads , which were then vortexed and incubated for 15 min at RT on a rotator . Elutant was collected after centrifugation at 13 , 200 rcf , and beads were subjected to a repetition of the elution step . The same volume of elution buffer was added to the input sample , and 300 mM NaCl was added to all samples to reverse crosslink at 65°C overnight . Three volumes of 100% ethanol was added and samples were incubated for 1 hr at −80°C . Samples were spun at 13 , 200 rcf at 4°C for 10 min followed by supernatant removal and air drying . 100 µL water was added and samples were shaken at RT for 5 hr . A PCR purification kit ( Qiagen ) was used before qPCR analysis . A minimum of three biological replicates were used for each experiment , with most experiments having four or more biological replicates ( see figure legends for sample size ) .	The DNA in a fertilized egg contains all the information required to form an animal’s body . In order for the animal to develop properly , particular genes encoded in the DNA are only active at specific times . The DNA is wrapped around proteins called histones , which allows the DNA to be tightly packed inside the cell . However , histones can block other proteins called transcription factors from binding to the DNA to activate the genes . Young embryos initially develop with all of their genes switched off , relying on the nutrients and other molecules provided by their mother . After some time , the embryo starts to switch on its own genes to take control of its own development , but it was not clear how this happens . Joseph et al . investigated how genes are activated in zebrafish embryos , which are often used as models to study how animals develop . The experiments show that competition between histones and transcription factors for binding to DNA controls when genes are switched on . In young fish embryos , there are so many histones present that transcription factors have no opportunity to bind to DNA . Over time , however , the numbers of histones decrease , allowing transcription factors to bind to DNA and switch on genes . Histones and transcription factors regulate the activity of genes throughout the life of the animal . Therefore , competition between these two types of protein may also control gene activity in other situations . A better understanding of how gene activity is controlled could allow researchers to more easily grow different types of cell in the laboratory or to reprogram specific cells in the body . As such , these new findings may aid the development of therapies to regenerate organs or tissues that have been damaged by injury or disease .	[ "Abstract", "Introduction", "Results", "Discussion", "M" ]	[ "chromosomes", "and", "gene", "expression", "developmental", "biology" ]	2017	Competition between histone and transcription factor binding regulates the onset of transcription in zebrafish embryos
Maintenance of connective tissue integrity is fundamental to sustain function , requiring protein turnover to repair damaged tissue . However , connective tissue proteome dynamics remain largely undefined , as do differences in turnover rates of individual proteins in the collagen and glycoprotein phases of connective tissue extracellular matrix ( ECM ) . Here , we investigate proteome dynamics in the collagen and glycoprotein phases of connective tissues by exploiting the spatially distinct fascicular ( collagen-rich ) and interfascicular ( glycoprotein-rich ) ECM phases of tendon . Using isotope labelling , mass spectrometry and bioinformatics , we calculate turnover rates of individual proteins within rat Achilles tendon and its ECM phases . Our results demonstrate complex proteome dynamics in tendon , with ~1000 fold differences in protein turnover rates , and overall faster protein turnover within the glycoprotein-rich interfascicular matrix compared to the collagen-rich fascicular matrix . These data provide insights into the complexity of proteome dynamics in tendon , likely required to maintain tissue homeostasis . Maintaining the structural and mechanical integrity of tissues in the musculoskeletal system , and other connective tissues , is fundamental to sustain tissue homeostasis and healthy function , requiring protein synthesis and degradation to repair and/or replace damaged tissue before damage accumulates and leads to injury ( Humphrey et al . , 2014 ) . However , while the composition and structure of connective tissues is well defined ( Scott , 1983 ) , relatively little is known regarding proteome dynamics in connective tissues , particularly at the level of the individual constituent proteins . Connective tissues consist of fibrous proteins ( predominantly collagen ) embedded in a glycoprotein-rich matrix ( Scott , 1983 ) , and variation in the organisation of both phases give rise to tissues with distinct structural and mechanical properties ( Culav et al . , 1999 ) . Research in a variety of tissues including skin , tendon and cartilage indicates a relatively long half-life of collagens , with more rapid turnover of glycoproteins and other non-collagenous proteins ( Thorpe et al . , 2010; Maroudas et al . , 1998; Sivan et al . , 2006; Sivan et al . , 2008; Verzijl et al . , 2000 ) . Indeed , several studies have reported negligible turnover of collagen , the major component of tendon and other connective tissues , within an individual’s lifetime ( Thorpe et al . , 2010; Heinemeier et al . , 2013; Heinemeier et al . , 2016 ) . However other studies have measured relatively rapid collagen synthesis in tendon , both at basal levels and in response to exercise ( Miller et al . , 2005 ) , and also identified soluble collagen with a much shorter half-life compared to the majority of collagen in human articular cartilage ( Hsueh et al . , 2019 ) ; the location ( s ) of these more labile proteins remain to be identified . When taken together , these contradictory findings indicate that proteome dynamics in connective tissues is complex , and that the glycoprotein-rich phase may be replenished more rapidly than the collagen-rich phase . However , the turnover rate of individual proteins within the different phases of the extracellular matrix ( ECM ) , and the potential contribution of differential regulation of protein turnover to maintenance of tissue homeostasis remain undefined . Tendon provides an ideal model in which to separately interrogate protein turnover in these ECM phases , as it consists of highly aligned , collagen-rich fascicular matrix ( FM ) , interspersed by a less dense glycoprotein-rich phase , termed the interfascicular matrix ( IFM , also referred to as the endotenon ) ( Kastelic et al . , 1978 ) . While the FM is predominantly composed of type I collagen , glycoproteins are also present in this region at low abundance ( Thorpe et al . , 2016b ) . Similarly , the IFM contains small amounts of a variety of collagens ( Södersten et al . , 2013; Thorpe et al . , 2016b ) . Due to the highly aligned structure of tendon tissue , it is possible to separate FM and IFM for individual analysis using laser capture microdissection ( Thorpe et al . , 2016b; Zamboulis et al . , 2018 ) . Indeed , these , and other studies , have shown greater expression of markers of ECM degradation , as well as increased neo-peptide levels , a proxy measure of protein turnover , within the IFM ( Spiesz et al . , 2015; Thorpe et al . , 2015a; Thorpe et al . , 2016b ) , suggesting this region is more prone to microdamage , likely due to the high shear environment that occurs due to interfascicular sliding ( Thorpe et al . , 2012; Thorpe et al . , 2015b ) . Taken together , these findings suggest that there is greater turnover of ECM proteins localised to the IFM than within the FM to repair local microdamage and maintain tendon homeostasis . Due to a lack of available methodology , it has not been possible until recently to study differential rate of turnover at the individual protein level . However , novel bioinformatics software developments , in combination with in vivo isotope labelling , and mass spectrometry technologies , now provide the capacity to determine the turnover rates of individual proteins within a sample ( Kim et al . , 2012; Lam et al . , 2014; Lau et al . , 2016 ) . The aim of this study is therefore to apply this technique to tendon , firstly to establish the proteome-wide turnover rate of tendon , and secondly to combine this approach with laser capture microdissection to test the hypothesis that the proteome of the IFM is more dynamic than in the FM , with faster turnover of individual proteins . More rapid remodelling of the IFM would provide a mechanism by which shear-induced microdamage to this region could be repaired , preventing damage accumulation and subsequent injury . In rats labelled with deuterium over a period of 127 days ( Figure 1a ) , 2H enrichment of serum occurred rapidly and in a similar manner to that reported previously in rodents ( Kim et al . , 2012 ) , reaching a plateau of 5 . 6% by day 4 , and remained constant throughout the study ( Figure 1b ) . The enrichment curve , which was empirically derived from the GC-MS measurements at the sampled time points , defines two parameters: deuterium enrichment rate ( kp ) = 0 . 7913 and plateau ( pss ) = 0 . 0558; these values were used for subsequent kinetic curve fitting to calculate peptide turnover rate constants ( k ) . 190 proteins with ≥2 unique peptides were identified in whole tendon digests and protein interactions are shown in Figure 2 . Of these proteins , 72 were classified as ECM or ECM-related proteins by MatrisomeDB ( Hynes and Naba , 2012 ) . In samples collected by laser capture microdissection of tendon cryosections , 266 proteins with ≥2 unique peptides were identified in the IFM , 79 of which were ECM or ECM-related proteins ( Figure 3 ) . In the FM , 116 proteins were identified , of which 71 were ECM or ECM-related proteins ( Figure 4 ) . Protein interactions for each tendon component demonstrate a complex and highly interconnected proteome in tendon and its ECM phases . Peptide turnover rates were calculated using ProTurn ( v2 . 1 . 05; available at http://proturn . heartproteome . org; Lam et al . , 2014; Lau et al . , 2016; Lau et al . , 2018; Wang et al . , 2014 ) , which automatically calculates turnover rate constants for all peptides that pass the selection criteria using non-steady state curve fitting ( Figure 1 ) . To assure data quality , we used a stringent cut-off , passing only peptides that are identified at 1% false discovery rate ( FDR ) and quantified at four or more time points . In total , 455 peptides , relating to 41 proteins , passed the ProTurn selection criteria and were used to calculate protein half-life in whole tendon samples . The relative abundance of the unlabelled monoisotopic peak ( M0 ) , plotted as a function of time for selected decorin and collagen peptides are shown in Figure 5a&b . Non-steady state curve fitting was performed by ProTurn to calculate turnover rate constants ( k ) for each peptide , and resultant fractional synthesis curves in Figure 5c&d demonstrate much faster turnover of decorin peptides compared to collagen types 1 and 3 . k values for all peptides identified , and corresponding protein half-lives are shown in Figure 6 . As expected , the smallest k values related to collagenous proteins , with corresponding half-lives of 330 to 1086 days . By contrast , the protein identified with the fastest turnover was the glycoprotein clusterin , with a half-life of 1 . 4 days . The half-life of proteoglycans ranged from 21 days for decorin to 72 days for lumican . With the exception of collagens , which all exhibited low turnover rates , there was no clear relationship between protein class and rate of turnover ( Figure 6c ) . 246 peptides , relating to 20 proteins , and 121 peptides , relating to 12 proteins , passed the ProTurn selection criteria in the FM and IFM respectively . 55 peptides were present both in the IFM and FM , and k values were significantly greater for these peptides in the IFM compared to the FM ( median: 0 . 018 vs . 0 . 010; p<0 . 0001 ) , demonstrating an overall faster rate of protein turnover in the IFM . 39 peptides relating to collagen type I were identified in both the IFM and FM , with significantly greater k values in the IFM ( p<0 . 0001; Figure 7a ) . The turnover rate constants and resultant half-lives for proteins identified in each tendon phase are shown in Figure 4b . Turnover rate constants for Col1a1 and Col1a2 were significantly higher in the IFM compared to the FM , but there were no significant differences in rate constants for Col3a1 between tendon phases . Due to ProTurn software identifying a low number of proteins in laser captured samples , particularly in the IFM , turnover rate constants in tendon phases were calculated manually for a number of proteins of interest ( Figure 8 ) . Rate constants of turnover of decorin peptides were significantly higher in the IFM compared to the FM ( Figure 8 ) . It was not possible to assess differences in turnover of other proteins due to a low number of peptides identified at sufficient time points to allow accurate curve fitting ( Figure 8c ) . This is the first study to determine the turnover rate of individual proteins in different connective tissue phases , demonstrating complex proteome dynamics in tendon , with an overall faster turnover of proteins in the glycoprotein-rich IFM phase . Greater capacity for turnover of this phase of the tendon matrix may be indicative of a mechanism that reduces the accumulation of damage caused by the high shear environment within this region , although this remains to be directly determined . The techniques used here provide a powerful approach to interrogate alterations in connective tissue protein turnover with ageing and/or disease which are likely to influence injury risk . - Female Wistar rats ( n = 24 , 12 week old , weight: 270 ± 18 g , range: 228–299 g , specific pathogen free , Charles River Company , UK ) were randomly housed in polypropylene cages in groups of 3 , subjected to 12 hr light/dark cycle with room temperature at 21 ± 2°C and fed ad libitum with a maintenance diet ( Special Diet Services , Witham UK ) . All procedures complied with the Animals ( Scientific Procedures ) Act 1986 , were approved by the local ethics committee at the Royal Veterinary College , were performed under project licence PB78F43EE and are reported according to the ARRIVE guidelines ( Kilkenny et al . , 2010 ) . All rats , with the exception of the controls ( n = 3 ) received two intraperitoneal injections of 99% [2H]2O ( 10 ml/kg; CK isotopes Ltd ) spaced 4 hr apart , and were then allowed free access to 8% [2H]2O ( v/v ) in drinking water for the duration of the study to maintain steady-state labelling ( Kim et al . , 2012 ) . All rats were acclimatized for 1 week prior to commencement of the study , and monitored daily and weighed weekly throughout the study; no adverse effects were observed . Rats in the control group were sacrificed on day 0 . Rats in the isotope labelled groups were sacrificed on day 1 , 3 , 6 , 15 , 31 , 63 and 127 ( n = 3 per time point ) . Rats were culled at the same time of day ( 10 am ) at each time point , to minimise any effect from diurnal variations . Blood was collected from each rat immediately post-mortem , and allowed to clot at room temperature . Blood samples were centrifuged at 1500 g for 10 min , and serum collected and stored at −20°C prior to analysis . Both Achilles tendons were harvested within 2 hr of death . The left Achilles was snap frozen in n-hexane cooled on dry ice for proteomic analysis of the whole tissue , and the right Achilles was embedded in optimal cutting temperature compound and snap frozen in n-hexane cooled on dry ice for isolation of fascicular and interfascicular matrices ( Figure 1a ) . 2H labelling in serum was measured via gas chromatography-mass spectrometry ( GC-MS ) after exchange with acetone and extraction with chloroform as described previously ( McCabe et al . , 2006 ) . 30 µl extract was analysed using GC-MS ( Waters GCT mass spectrometer , Agilent J and W DB-17MS column ( 30m × 0 . 25 mm x 0 . 25 µm ) ) . The carrier gas was helium ( flow rate: 0 . 8 ml/min ) . The column temperature gradient was as follows: 60°C initial , 20 °C/min increase to 100°C , 1 min hold , then 50 °C/min increase to 220°C . The injection volume was 1 µl and injector temperature was 220°C . The mass spectrometer operated in positive ion electron ionisation mode , the source temperature was 180°C and the range scanned was m/z 40–600 ( scan time: 0 . 9 s ) . Mass spectral intensities for m/z values 58 and 59 were produced by combining the mass spectra in the acetone peak to create an averaged spectrum ( Waters MassLynx ) . Serum 2H enrichment was calculated by comparison to a standard curve , and first order curve fitting ( GraphPad Prism 8 ) used to calculate the rate constant ( kp ) and plateau ( pss ) of deuterium enrichment ( Figure 1b ) . Left Achilles tendons were thawed , and their surface scraped with a scalpel , followed by PBS washes , to remove the epitenon and any residual non-tendinous tissue . Tendons were finely chopped , flash frozen in liquid nitrogen and pulverised in a dismembrator ( Sartorius , mikro-dismembrator U , 1800rpm , 2 mins . ) . Proteins were extracted using previously optimised methodologies ( Ashraf Kharaz et al . , 2017 ) . Briefly , following deglycosylation in chondroitinase ABC , proteins were extracted in guanidine hydrocholoride ( GuHCl ) as described previously ( Kharaz et al . , 2016 ) . Protein content was determined using a Pierce protein assay according to the manufacturer’s instructions . Samples were stored at −80°C prior to preparation for LC-MS/MS by centrifugal filtration . Filter units ( Vivacon 500; 10 000 MWCO ) were rinsed with 1% ( v/v ) formic acid and a balance volume of buffer ( 4M GuHCl/50 mM ammonium bicarbonate ( ambic ) ) was added . A volume equivalent to 50 µg protein was added to each filter and the samples vortexed gently . Filters were centrifuged ( 15 min , 12500 rpm , 20°C ) , washed with GuHCl buffer and centrifuged again . Proteins were reduced by DTT incubation ( 100 µl of 8 mM in 4M GuHCl , 15 min , 56°C ) and then centrifuged ( 10 min , 12500 rpm ) . DTT was removed by washing twice with buffer as described above . Proteins were alkylated with 100 μl 50 mM iodoacetamide in 4M GuHCl solution , vortexed and incubated in the dark ( 20 min . , room temperature ) . Iodoacetamide was removed by washing twice with GuHCl buffer as described above . Buffer was exchanged to ambic by 3 washes with 50 mM ambic , centrifuging after each wash . 1 µg trypsin in 50 mM ambic ( 40 µl ) was added and proteins digested overnight at 37°C with mixing . Flow through was collected after centrifugation and addition of 40 µl 50 mM ambic . The flow though was combined and acidified using trifluoroacetic acid ( 10% ( v/v ) ) , and diluted 20-fold in 0 . 1% ( v/v ) TFA/3% ( v/v ) acetonitrile for LC-MS/MS analysis . Longitudinal cryosections ( 15 µm ) from the right Achilles tendons ( five time points , n = 2–3/time point ) were adhered to membrane slides ( PEN , Leica ) and stored at −80°C . Sections were prepared for laser capture as described previously ( Thorpe et al . , 2016b ) and approximately 1 mm2 of FM and IFM from each sample were collected into 50 µl molecular biology grade H2O ( Leica LMD7000; Figure 1a ) . Samples were immediately frozen and stored at −80°C . Prior to digestion , samples were centrifuged immediately after removal from −80°C storage . Volume was adjusted to 80 µl by adding 50 mM ambic , and Rapigest SF ( 1% ( v/v ) in 25 mM ambic , Waters , UK ) was added to a final volume of 0 . 1% ( v/v ) . Samples were mixed ( room temperature , 30 min ) then incubated at 60°C ( 1 hr , 450 rpm ) . Samples were centrifuged ( 17200 g , 10 min ) , vortexed , then incubated at 80°C ( 10 min ) . Proteins were reduced by incubating with DTT ( 5 μL , 60 mM in 25 mM ambic , 10 min , 60°C ) then alkylated by adding 5 μl 178 mM iodoacetamide in 25 mM ambic ( 30 min . incubation , room temperature , in dark ) . Trypsin ( Promega Gold sequencing grade ) was diluted in 25 mM ambic and added at an enzyme to protein ratio of 1:50 ( based on estimated protein amount from tissue volume collected ) . Digests were incubated overnight at 37°C with an enzyme top-up after 3 . 5 hr . Rapigest was hydrolysed with TFA ( 0 . 5 μl , 37°C , 45 min . ) . Digests were centrifuged ( 17200 g , 30 min . ) , aspirated into low-bind tubes and de-salted on stage-tips as previously described ( Thorpe et al . , 2016b ) . Data-dependent LC-MS/MS analyses were conducted on a QExactive quadrupole-Orbitrap mass spectrometer ( Williamson et al . , 2016; Michalski et al . , 2011 ) coupled to a Dionex Ultimate 3000 RSLC nano-liquid chromatograph ( Thermo Fisher , UK ) . Digest was loaded onto a trapping column ( Acclaim PepMap 100 C18 , 75 µm x 2 cm , 3 µm packing material , 100 Å ) using a loading buffer of 0 . 1% ( v/v ) TFA , 2% ( v/v ) acetonitrile in water for 7 min ( flow rate: 9 µl/min ) . The trapping column was set in-line with an analytical column ( EASY-Spray PepMap RSLC C18 , 75 µm x 50 cm , 2 µm packing material , 100 Å ) and peptides eluted using a linear gradient of 96 . 2% A ( 0 . 1% ( v/v ) formic acid ) :3 . 8% B ( 0 . 1% ( v/v ) formic acid in water:acetonitrile [80:20] ( v/v ) ) to 50% A:50% B over 30 min ( flow rate: 300 nl/min ) , followed by 1% A:99% B for 5 min and re-equilibration of the column to starting conditions . The column was maintained at 40°C , and the effluent introduced directly into the integrated nano-electrospray ionisation source operating in positive ion mode . The mass spectrometer was operated in DDA mode with survey scans between m/z 300–2000 acquired at a mass resolution of 70 , 000 ( FWHM ) at m/z 200 . The maximum injection time was 250 ms , and the automatic gain control was set to 1e6 . The 10 most intense precursor ions with charges states of between 2+ and 4+ were selected for MS/MS with an isolation window of 2 m/z units . The maximum injection time was 100 ms , and the automatic gain control was set to 1e5 . Peptides were fragmented by higher-energy collisional dissociation using a normalised collision energy of 30% . Dynamic exclusion of m/z values to prevent repeated fragmentation of the same peptide was used ( exclusion time: 20 s ) . Proteins were identified from RAW data files , with trypsin specified as the protease , with one missed cleavage allowed and with fixed modifications of carbamidomethyl cysteine and variable modifications of oxidation of methionine and proline ( Peaks Studio v8 . 5 , Bioinformatics Solutions , Waterloo , Canada ) . Searches were performed against the UniProt Rattus Norvegicus database ( www . uniprot . org/proteomes ) , with an FDR of 1% , ≥2 unique peptides per protein and a confidence score >20 . Protein network analysis was performed using the Search Tool for Retrieval of Interacting Genes/Proteins ( STRING ) , v11 . 0 ( Szklarczyk et al . , 2019 ) , and proteins were further classified using MatrisomeDB ( Hynes and Naba , 2012 ) and the Protein ANalysis THrough Evolutionary Relationships ( PANTHER ) Classification System ( Mi et al . , 2013 ) . The proteomic data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository ( Perez-Riverol et al . , 2019 ) with the data set identifier PXD015928 and 10 . 6019/PXD015928 . Peptides at each time point were identified using ProLuCID ( Xu et al . , 2015 ) searching against a reverse-decoyed protein sequence database ( UniProt Rattus Norvigicus , reviewed , accessed 23/04/2018 ) . Fixed modifications of carbamidomethyl cysteine and ≤3 variable modifications of oxidation of methionine and proline were allowed . Tryptic , semi-tryptic , and non-tryptic peptides within a 15-ppm mass window surrounding the candidate precursor mass were searched . Protein identification was performed using DTASelect ( v . 2 . 1 . 4 ) ( Cociorva et al . , 2006 ) , with ≤1% global peptide FDR and ≥2 unique peptides per protein . Protein turnover rates and resulting half-lives were calculated using custom software ( ProTurn v2 . 1 . 05; available at: http://proturn . heartproteome . org ) as previously described and validated ( Lam et al . , 2014; Lau et al . , 2016; Lau et al . , 2018; Wang et al . , 2014 ) . Briefly , RAW files were converted into mzML format ( ProteoWizard , v3 . 0 . 11781 ) ( Adusumilli and Mallick , 2017 ) for input into ProTurn , along with DTAselect-filter text files for protein identification . ProTurn parameters were as follows: area-under-curve integration width: 60 p . p . m . , extracted ion chromatograph smoothing: Savitzky-Golay filter ( Savitzky and Golay , 1964 ) over seven data points . Non-steady state curve fitting was used to account for the initial delay in uptake of 2H , using a first-order kinetic curve to approximate the equilibration of 2H2O in body water . Values of kp and pss were inputted from the resultant curve fitting of 2H enrichment in serum samples . To control against false positive identifications , only peptides that were explicitly identified in ≥4 data points were accepted for the calculation of protein turnover . The ‘Allow Peptide Modification’ option was selected to include any identified post-translationally modified peptides in kinetic curve-fitting , and peptide isotopomer time series were included if R2 ≥0 . 8 or standard error of estimate ( S . E ) ≤0 . 05 ( Lam et al . , 2014 ) . Protein-level turnover rate is reported as the median and median absolute deviation of the turnover rate constants of each accepted unique constituent peptide , from which protein half-life can be calculated assuming a first order reaction ( Lau et al . , 2018 ) . For some proteins of interest , particularly in the IFM , automated calculation of turnover rates was unsuccessful , due to unsuccessful automated identification of isotopomer peaks . In these cases , turnover rate constants were calculated manually , using isotopomer peak height to calculate the relative abundance of M0 as a function of time ( Thermo Xcalibur v2 . 2 ) , and fitting first order kinetic curves to estimate k ( GraphPad Prism v8 . 0 . 2 ) . As the labelling pattern of each peptide is dependent on the number of available labelling sites ( N ) , unlabelled relative abundance of the 0th isotopomer ( a ) , as well as the plateau level of enrichment of deuterium ( pss ) ; these values were used to calculate the plateau values of the 0th isotopomer ( A0 ) which occurs when the peptide is fully labelled . N was calculated from the literature ( Lam et al . , 2014; Commerford et al . , 1983 ) . a was calculated from the peptide sequence and natural abundance of heavy isotopes of carbon , nitrogen , oxygen and sulphur ( Lam et al . , 2014 ) . pss was calculated from the serum enrichment as detailed above . The following equation was then used to calculate the plateau of A0:A0 , PLATEA=a ( 1−pss ) N Previous work demonstrates this accurately predicts the plateau measured experimentally ( Lam et al . , 2014 ) , therefore , for curve fitting calculations the plateau was constrained to the calculated value for each peptide . To assess the degree of differences in curve fitting performed by automated and manual approaches , k values were calculated using both approaches for 10 peptides . Resulting k values differed by an average of 23% . This method does not allow for direct comparison between manual and automated calculations of k , due to differences in curve fitting method , but does allow differences in peptide k values between IFM and FM to be assessed , when both are calculated manually . When ≥ 3 peptides were identified corresponding to a particular protein , statistical differences in turnover rate constants in tendon phases were assessed using paired t-tests , Wilcoxon matched pairs tests , or Mann-Whitney tests , with p<0 . 05 ( GraphPad Prism , v8 . 2 ) . The statistical test chosen was dependent on whether data were normally distributed , which was assessed with Kolmogorov-Smirnov tests , and whether the same peptides were identified in tendon phases , which allowed for paired statistical tests .	Muscles are anchored to bones through specialized tissues called tendons . Made of bundles of fibers ( or fascicles ) linked together by an ‘interfascicular’ matrix , healthy tendons are required for organisms to move properly . Yet , these structures are constantly exposed to damage: the interfascicular matrix , in particular , is highly susceptible to injury as it allows the fascicles to slide on each other . One way to avoid damage could be for the body to continually replace proteins in tendons before they become too impaired . However , the way proteins are renewed in these structures is currently not well understood – indeed , it has long been assumed that almost no protein turnover occurs in tendons . In particular , it is unknown whether proteins in the interfascicular matrix have a higher turn over than those in the fascicles . To investigate , Choi , Simpson et al . fed rats on water carrying a molecular label that becomes integrated into new proteins . Analysis of individual proteins from the rats’ tendons showed great variation in protein turnover , with some replaced every few days and others only over several years . This suggests that protein turnover is actually an important part of tendon health . In particular , the results show that turnover is higher in the interfascicular matrix , where damage is expected to be more likely . Protein turnover also plays a part in conditions such as cancer , heart disease and kidney disease . Using approaches like the one developed by Choi , Simpson et al . could help to understand how individual proteins are renewed in a range of diseases , and how to design new treatments .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2020	Heterogeneity of proteome dynamics between connective tissue phases of adult tendon
Platelet-neutrophil interactions are important for innate immunity , but also contribute to the pathogenesis of deep vein thrombosis , myocardial infarction and stroke . Here we report that , under flow , von Willebrand factor/glycoprotein Ibα-dependent platelet ‘priming’ induces integrin αIIbβ3 activation that , in turn , mediates neutrophil and T-cell binding . Binding of platelet αIIbβ3 to SLC44A2 on neutrophils leads to mechanosensitive-dependent production of highly prothrombotic neutrophil extracellular traps . A polymorphism in SLC44A2 ( rs2288904-A ) present in 22% of the population causes an R154Q substitution in an extracellular loop of SLC44A2 that is protective against venous thrombosis results in severely impaired binding to both activated αIIbβ3 and VWF-primed platelets . This was confirmed using neutrophils homozygous for the SLC44A2 R154Q polymorphism . Taken together , these data reveal a previously unreported mode of platelet-neutrophil crosstalk , mechanosensitive NET production , and provide mechanistic insight into the protective effect of the SLC44A2 rs2288904-A polymorphism in venous thrombosis . To fulfil their hemostatic function , platelets must be recruited to sites of vessel damage . This process is highly dependent upon von Willebrand factor ( VWF ) . Upon vessel injury , exposed subendothelial collagen binds plasma VWF via its A3 domain ( Cruz et al . , 1995 ) . Elevated shear , or turbulent/disturbed flow , then unravels tethered VWF and exposes its A1 domain , facilitating specific capture of platelets via glycoprotein ( GP ) Ibα . As well as capturing platelets under flow , the A1-GPIbα interaction also induces shear-dependent signaling events ( Bryckaert et al . , 2015 ) . For this , GPIbα first binds the A1 domain of immobilized VWF ( Zhang et al . , 2015 ) . Rheological forces then cause unfolding of the GPIbα mechanosensitive domain that translates the mechanical stimulus into a signal within the platelet ( Ju et al . , 2016; Zhang et al . , 2015 ) . This leads to release of intracellular Ca2+ stores and activation of the platelet integrin , αIIbβ3 ( Gardiner et al . , 2010 ) . VWF-mediated signaling transduces a mild signal . Consequently , these signaling events are often considered redundant within hemostasis as platelets respond more dramatically to other agonists present at sites of vessel injury ( e . g . collagen , thrombin , ADP , thromboxane A2 ) ( Jackson et al . , 2003; Senis et al . , 2014 ) . Full platelet activation involves release of α- and δ-granules , presentation of new cell surface proteins , activation of cell surface integrins and alterations in the membrane phospholipid composition . The extent of platelet activation is dependent upon both the concentration , and identity , of the agonist ( s ) to which the platelets are exposed , which is dictated by the location of the platelets relative to the damaged vessel . For example , platelets in the core of a hemostatic plug/thrombus are exposed to higher concentrations of agonists and are more highly activated ( i . e . P-selectin-positive procoagulant platelets ) than those in the surrounding shell ( P-selectin-negative ) ( de Witt et al . , 2014; Shen et al . , 2017; Stalker et al . , 2013; Welsh et al . , 2014 ) . Thus , platelets exhibit a ‘tunable’ activation response determined by agonist availability . Aside from hemostasis , platelets also have important roles as immune cells by aiding in targeting of bacteria by leukocytes ( Gaertner et al . , 2017; Kolaczkowska et al . , 2015; Sreeramkumar et al . , 2014; Wong et al . , 2013 ) . Platelet-leukocyte interactions also influence the development of inflammatory cardiovascular conditions . In deep vein thrombosis ( DVT ) , VWF-dependent platelet recruitment , platelet-neutrophil interactions and the production of highly thrombotic neutrophil extracellular traps ( NETs ) all contribute to the development of a pathological thrombus ( Brill et al . , 2011; Brill et al . , 2012; Fuchs et al . , 2012a; Schulz et al . , 2013; von Brühl et al . , 2012 ) . Although the precise sequence of events still remains unclear , it appears that during the early stages of DVT , VWF-bound platelets acquire the ability to interact with leukocytes ( von Brühl et al . , 2012 ) . Exactly how this is mediated given the lack of vessel damage is unclear . It also remains to be determined precisely how platelet-tethered neutrophils undergo NETosis in DVT in the absence of an infectious agent . Known direct platelet-leukocyte interactions involve either P-selectin or CD40L on the surface of platelets binding to P-selectin glycoligand-1 ( PSGL-1 ) and CD40 , respectively , on leukocytes ( Lievens et al . , 2010; Mayadas et al . , 1993; Palabrica et al . , 1992 ) . As platelets must be potently activated to facilitate P-selectin/CD40L exposure , such interactions unlikely mediate the early platelet-leukocyte interactions that occur in the murine DVT model . Consistent with this , lack of platelet P-selectin has no effect upon either leukocyte recruitment or thrombus formation in murine DVT ( von Brühl et al . , 2012 ) . Leukocytes can also indirectly interact with platelets through Mac-1 ( integrin αMβ2 ) , which can associate with activated αIIbβ3 via fibrinogen ( Weber and Springer , 1997 ) , or directly via GPIbα ( Simon et al . , 2000 ) . Interactions are also possible through lymphocyte function-associated antigen 1 ( LFA-1/integrin αLβ2 ) that can bind intercellular adhesion molecule 2 ( ICAM-2 ) on platelets ( Damle et al . , 1992; Diacovo et al . , 1994 ) . In both instances though , leukocyte activation is necessary to activate Mac-1 or LFA-1 integrins before interactions can occur . Although it is often assumed that only activated platelets bind leukocytes , recent studies have revealed that platelets captured under flow by VWF released from activated endothelial cells can recruit leukocytes ( Doddapattar et al . , 2018; Zheng et al . , 2015 ) . If VWF-GPIbα-dependent signaling is capable of promoting leukocyte binding , this may be highly relevant to the non-hemostatic platelet functions ( particularly when other agonists are not available/abundant ) , but may also provide major mechanistic insights into the early recruitment of leukocytes during the initiation of DVT . Genome wide association studies ( GWAS ) on venous thromboembolism ( VTE ) have identified a panel of genes ( ABO , F2 , F5 , F11 , FGG , PROCR ) with well-described influences upon coagulation and thrombotic risk , as well as those with well-established causative links ( e . g . PROS , PROC , SERPINC1 ) ( Germain et al . , 2015; Germain et al . , 2011; Rosendaal and Reitsma , 2009 ) . This is consistent with the efficacy of therapeutic targeting of coagulation to protect against DVT with anticoagulants ( Chan et al . , 2016 ) . However , although the use of anticoagulants is effective , dosing and efficacy are limited by the increase in the risk of bleeding in treated individuals ( Chan et al . , 2016; Schulman et al . , 2014; Schulman et al . , 2009; Schulman et al . , 2013; Wolberg et al . , 2015 ) . Therefore , alternative targets that inhibit DVT disease processes , but that do not modify bleeding risk may provide new adjunctive therapies to further protect against the development or recurrence of DVT . GWAS studies have also identified additional risk loci for VTE , but with no known role in coagulation ( Apipongrat et al . , 2019; Germain et al . , 2015; Hinds et al . , 2016 ) . This provides encouragement that alternative therapeutic targets may exist with the potential to modify the disease process without affecting bleeding risk . These loci include SLC44A2 and TSPAN15 genes ( Apipongrat et al . , 2019; Germain et al . , 2015; Hinds et al . , 2016 ) . Despite the identification of these loci , the function of these cell surface proteins with respect to their involvement in the pathogenesis of venous thrombosis remains unclear . Using microfluidic flow channels to enable analysis of the phenotypic effects of VWF-GPIbα signaling under flow , we confirm the rapid activation of the platelet integrin , αIIbβ3 . Activated αIIbβ3 is capable of binding directly to neutrophils via a direct interaction with SLC44A2 . Under flow , this interaction transduces a signal into neutrophils capable of driving NETosis . A single-nucleotide polymorphism ( SNP; rs2288904-A ) in SLC44A2 ( minor allele frequency 0 . 22 ) that is protective against VTE ( Germain et al . , 2015 ) encodes a R154Q substitution in the first extracellular loop of the receptor that markedly reduces neutrophil-platelet binding via activated αIIbβ3 . These results provide a functional explanation for the protective effects of the rs2288904-A SNP and highlight the potential of SLC44A2 as an adjunctive therapeutic target in DVT ( Constantinescu-Bercu et al . , 2020 ) . Platelets bound to either FL-VWF , A1 or A1* formed small aggregates after about 2 min ( Figure 2ai ) due to activation of the platelet integrin , αIIbβ3 , and its binding to plasma fibrinogen . Consistent with this , when plasma-free blood ( i . e . RBCs , leukocytes and platelets resuspended in plasma-free buffer ) was used , platelets remained as a uniform monolayer , and did not form microaggregates ( Figure 2aii ) . Similarly , when activated αIIbβ3 was blocked in whole blood with eptifibatide or GR144053 , aggregation was also inhibited ( Figure 2aiii and iv ) . Irrespective of the surface ( VWF , A1 or A1* ) , platelet aggregation was markedly reduced if plasma-free blood was used , or if αIIbβ3 was blocked ( Figure 2b–e ) . These results demonstrate that the A1-GPIbα interaction leads to activation of αIIbβ3 , which is consistent with previous reports ( Goto et al . , 1995; Kasirer-Friede et al . , 2004 ) . In support of this , fluorescent fibrinogen bound to platelets tethered via FL-VWF , but not to platelets captured to channel surfaces using an anti-PECAM-1 antibody ( Figure 2—figure supplement 1a ) . To investigate the effect of A1-GPIbα-dependent signaling , platelets were preloaded with the Ca2+-sensitive fluorophore , Fluo-4 AM . Platelets bound to A1* under flow exhibited repeated transient increases in fluorescence , corresponding to Ca2+ release from platelet intracellular stores in response to A1-GPIbα binding under flow ( Video 2; Kasirer-Friede et al . , 2004; Mu et al . , 2010 ) . Despite intracellular Ca2+ release , this did not lead to appreciable P-selectin exposure ( i . e . α-granule release ) ( Figure 2—figure supplement 1b ) . Intraplatelet Ca2+ release was not detected when platelets were captured under flow using an anti-PECAM1 antibody . We therefore propose that flow-dependent VWF-GPIbα signaling ‘primes’ , rather than activates , platelets . This ‘priming’ is characterized by activation of αIIbβ3 , but minimal α-granule release , and represents part of the tunable response of platelets . To explore the influence of platelet ‘priming’ upon their ability to interact with leukocytes , platelets were captured and ‘primed’ on VWF for 3 min at 1000 s−1 . Thereafter , leukocytes in whole blood ( also labeled with DiOC6 ) that were perfused at 50 s−1 , rolled on the platelet-covered surface ( Video 3 ) . Leukocytes did not interact with platelets captured via an anti-PECAM-1 antibody ( Figure 3a ) , demonstrating the dependency on prior A1-GPIbα-mediated platelet ‘priming’ . As VWF-‘primed’ platelets present activated αIIbβ3 , we hypothesized that ‘outside-in’ integrin signaling might be important for platelet-leukocyte interactions to occur ( Durrant et al . , 2017 ) . Contrary to this , we observed a significant ( ~2 fold ) increase in the number of leukocytes interacting with the VWF-bound platelets in plasma-free conditions ( Figure 3b ) . Moreover , addition of purified fibrinogen to plasma-free blood to 50% normal plasma concentration significantly reduced platelet-leukocyte interactions ( Figure 3b ) suggesting that leukocytes and fibrinogen compete for binding ‘primed’ platelets . Blocking αIIbβ3 ( Figure 3a–b & Figure 3—figure supplement 1a ) also significantly decreased platelet-leukocyte interactions irrespective of whether platelets were captured on FL-VWF or A1* , or whether experiments were performed in whole blood or plasma-free blood ( Figure 3a–c ) . To explore the role of activated αIIbβ3 in binding leukocytes , platelets were captured onto anti-PECAM-1 coated channels and an anti-β3 antibody ( ligand-induced binding site – LIBS ) that induces activation of αIIbβ3 applied ( Du et al . , 1993 ) . Antibody-mediated activation of αIIbβ3 caused a significant increase in the number of leukocytes binding in a manner that could be blocked with GR144053 ( Figure 3d ) . The best characterized platelet-leukocyte interaction is mediated by P-selectin on activated platelets binding to PSGL-1 on leukocytes ( Vandendries et al . , 2004 ) . Although we detected little/no P-selectin on the surface of VWF-‘primed’ platelets , this did not formally exclude a role for P-selectin in leukocyte adhesion . Therefore , we first established the efficacy of P-selectin blockade through the marked reduction of leukocyte binding to collagen captured/activated platelets ( Figure 3—figure supplement 1b–c ) . However , blockade of P-selectin on FL-VWF-bound platelets from whole blood or plasma-free blood had no effect upon the number of leukocytes interacting with the platelet surface , suggesting that the recruitment of leukocytes is independent of P-selectin ( Figure 3e ) . Leukocytes rolled faster over platelet surfaces after blocking P-selectin in plasma-free blood ( Figure 3f and Video 3 ) or whole blood ( Figure 3—figure supplement 1d ) , suggesting that whereas leukocyte capture is highly dependent on activated αIIbβ3 ( and not P-selectin ) , once recruited , small amounts of P-selectin on the platelet surface may slow leukocyte rolling . To more specifically test the leukocyte interaction with activated αIIbβ3 ( and to exclude other platelet receptors ) , purified αIIbβ3 was covalently coupled to microchannels and , thereafter , activated with Mn2+ ( Litvinov et al . , 2005 ) . Isolated peripheral blood mononuclear cells ( PBMCs ) and polymorphonuclear cells ( PMNs ) were perfused through αIIbβ3-coated channels at 50 s−1 . Cells from both PBMCs and PMNs ( Figure 4ai–ii & and b ) directly attached to the activated αIIbβ3 surface . This binding was significantly diminished ( >70% ) by adding either purified fibrinogen or eptifibatide to the PMNs ( Figure 4aiv-v and b ) . We also captured purified αIIbβ3 to flow channel surfaces using the activating anti-β3 ( LIBS ) antibody . Leukocytes were again efficiently captured to this surface in a manner that could be inhibited ( ~70% ) by blocking αIIbβ3 ( Figure 4b ) . Activated ( rather than resting ) leukocytes can interact with platelets via Mac-1 ( αMβ2 ) ( either directly through GPIbα or via fibrinogen bridge with activated αIIbβ3 ) or LFA-1 ( αLβ2 ) via ICAM-2 ( Damle et al . , 1992; Diacovo et al . , 1994; Simon et al . , 2000; Weber and Springer , 1997 ) . However , blocking β2 suggested no role for either of these activated integrins in leukocyte binding to VWF-‘primed’ platelets ( Figure 4c ) . In summary , we show leukocytes bind αIIbβ3 directly dependent upon its RGD-binding groove , but in a manner that is independent of Mac-1 or LFA-1 . We found no evidence of either CD14+ monocytes or CD19+ B-cells in PBMCs interacting with activated αIIbβ3 . T-cells were the only cell type amongst the PBMCs capable of binding activated αIIbβ3 or VWF-‘primed’ platelets ( Figure 4d–e ) . Using isolated PMNs , we found that cells stained with anti-CD16 bound to activated αIIbβ3-coated channels and also to VWF-‘primed’ platelets ( Figure 4f and g ) . Based on multi-lobulated segmented nuclear morphology ( Figure 4f ) , these cells were indicative of CD16+ neutrophils . Neutrophils scanned the platelet- or αIIbβ3-coated surfaces ( Figure 4g and Video 4 ) suggesting that the binding of neutrophils to αIIbβ3 under flow may itself initiate signaling events within neutrophils . In line with this , PMNs bound to VWF-‘primed’ platelets ( Figure 5a ) or activated αIIbβ3 ( Figure 5b ) surfaces exhibited similar intracellular Ca2+ release ( Video 5 ) that reached a maximum after 200–300 s ( Figure 5c–d ) . Platelets assist in the targeting of intravascular bacterial pathogens through stimulation of the release of NETs ( Brinkmann et al . , 2004; Gaertner et al . , 2017; Wong et al . , 2013; Yeaman , 2014 ) . However , the physiological agonists or mechanisms that drive NETosis are not fully resolved ( Nauseef and Kubes , 2016 ) . We therefore examined whether the binding of neutrophils to αIIbβ3 might induce NETosis . Isolated PMNs were perfused over either activated αIIbβ3 or anti-CD16 ( negative control ) at 50 s−1 for 10 min and NETosis was subsequently analyzed under static conditions ( Figure 6a and Video 6 ) . Nuclear decondensation was evident after ~60 min , and Sytox Green fluorescence , indicative of cell permeability that precedes NETosis , was detected from ~85 min . Nuclear decondensation , increased cell permeability and positive staining with a cell impermeable DNA fluorophore do not specifically identify NETosis . Therefore , an anti-citrullinated histone H3 antibody was perfused through the channels after 90 min to more specifically identify NETs . The introduction of flow at this point caused the DNA to form extended mesh-like NETs that were stained positively by Hoechst and the anti-citrullinated histone H3 antibody ( Figure 6b ) . Very similar results were obtained with PMN bound to either αIIbβ3 ( captured by the activating anti-β3 antibody ) , or to platelets ‘primed’ by A1* or FL-VWF , which all exhibited a similar proportion of neutrophils undergoing NETosis within the 2 hr timeframe , suggesting that interaction with αIIbβ3 alone can promote NETosis and that any platelet releasate present does not appreciably augment this process under these conditions . On activated αIIbβ3 , 69% ± 14% of neutrophils through the entire channel formed NETs after 2 hr , compared to minimal NETosis events ( 8 ± 8% ) when neutrophils were captured by anti-CD16 ( Figure 6c ) . When neutrophils were captured on αIIbβ3 in the absence of flow , neutrophils attached , but NETosis was significantly reduced by fourfold , with only 17% of neutrophils exhibiting signs of NETosis ( Figure 6c ) . This suggested that the signaling mechanism from the platelet to the neutrophil is mechano-sensitive and does not require other platelet receptors or releasate components . Neutrophils captured by an anti-CD16 antibody and stimulated with phorbol 12-myristate 13-acetate ( PMA ) for 2 hr led to 100 ± 0 . 5% of neutrophils releasing NETs ( Figure 6d ) . PMA-induced NETosis was not significantly inhibited in the presence of TMB-8 ( an antagonist of intracellular Ca2+ release; 90 ± 9% ) , but was effectively inhibited in by DPI ( NADPH oxidase inhibitor; 16 ± 13% ) , similar to previous reports ( Gupta et al . , 2014 ) . NETosis of neutrophils captured by αIIbβ3 under flow was significantly inhibited by TMB-8 ( 17 ± 9% ) and DPI ( 25 ± 10% ) ( Figure 6e ) . This highlights the dependency of both intracellular Ca2+ release and NADPH oxidase signaling pathways in NETosis in response to binding αIIbβ3 under flow . Our data point to the presence of a specific receptor on the surface of neutrophils ( and T-cells ) that is not present on B cells or monocytes and that is capable of binding to activated αIIbβ3 and transducing a signal into the cell . To identify this leukocyte counter-receptor , we analyzed RNA sequencing data from different leukocyte populations , selecting genes that are expressed at higher levels in neutrophils ( or in CD4+ T-cells ) than in monocytes ( Adams et al . , 2012; Grassi et al . , 2019; Figure 7 ) . We further limited the candidate search by selecting those genes that code for transmembrane proteins . Using this approach , we identified 93 candidate genes . Of these , 33 genes were excluded as they are primarily associated with intracellular membranes . An additional 16 genes were also excluded due to the presence of short extracellular regions/domains ( <30 a . a . ) that would unlikely be capable of facilitating interactions with an extracellular binding partner ( Figure 7 ) . We then analyzed proteomic data to verify the preferential expression of the remaining candidates in neutrophils as opposed to monocytes ( Rieckmann et al . , 2017 ) . These data suggested that the protein product of 14 of the remaining genes appeared to be detected in higher abundance in monocytes , which we used as a further exclusion criterion ( Figure 7—figure supplement 1a ) . From the remaining 30 candidate genes , the SLC44A2 gene was selected for validation due to its recent identification as a risk locus for both DVT and stroke ( Germain et al . , 2015; Hinds et al . , 2016 ) , both of which are pathologies associated with described contributions of platelet-leukocyte interactions . SLC44A2 is a cell surface receptor with 10 membrane-spanning domains and five extracellular loops of 178a . a . , 38a . a . , 72a . a . , 38a . a . and 18a . a . in length , respectively ( Nair et al . , 2016 ) . We sourced antibodies against SLC44A2 that specifically recognize amino acid sequences within the first and second extracellular loops . Published proteomic profiling confirmed the preferential expression of SLC44A2 in neutrophils ( Figure 7—figure supplement 1a; Rieckmann et al . , 2017 ) . Western blotting of isolated granulocyte lysates revealed two bands representing SLC44A2 ( glycosylated and nascent/non-glycosylated SLC44A2 ) ( Figure 7—figure supplement 1b ) . Perfusing human neutrophils over immobilized activated αIIbβ3 in the presence of the first anti-SLC44A2 antibody ( anti-SLC44A2 #1 ) that recognizes the second extracellular loop revealed a dose-dependent blockade of neutrophil binding when compared to no antibody or control rabbit IgG ( Figure 8a–b ) . A second anti-SLC44A2 antibody ( anti-SLC44A2 #2 ) that recognizes the first extracellular loop region of SLC44A2 confirmed these findings ( Figure 8a ) . The anti-SLC44A2 #2 almost completely blocked neutrophil binding to activated αIIbβ3 suggesting that this antibody more effectively blocks the neutrophil binding to the integrin than anti-SLC44A2 #1 . This may suggest that the first and longest extracellular loop is involved in interaction with an extracellular ligand . Based on these results , we transfected HEK293T cells with an expression vector for human SLC44A2 fused to turbo green fluorescent protein ( tGFP ) at the intracellular C-terminus . Transfected cells were perfused through activated αIIbβ3 coated channels and cell binding was quantified . Transfected cells bound to these surfaces in a manner that could be blocked by GR144053 ( that blocks αIIbβ3 ) or by the anti-SLC44A2 #1 antibody ( Figure 8c ) . The SNP in SLC44A2 identified by GWAS studies that is protective against VTE and stroke ( rs2288904-A ) causes a missense mutation ( R154Q ) in the first 178a . a . extracellular loop of SLC44A2 ( Germain et al . , 2015 ) . Based on this , we hypothesized that this substitution might exert a functional influence upon the ability of SLC44A2 to interact with αIIbβ3 . Consistent with this hypothesis , HEK293T cells transfected with the SLC44A2 ( R154Q ) -tGFP expression vector exhibited reduced ability to interact with immobilized αIIbβ3 ( Figure 8c ) . Western blot analysis of transfected HEK293T cells revealed that expression of the SLC44A2 ( R154 ) -tGFP and SLC44A2 ( Q154 ) -tGFP was similar and that the tGFP remained uniformly associated with the fusion protein ( Figure 8c inset ) . To further explore the potential interaction between SLC44A2 and activated αIIbβ3 on platelets , we first captured and ‘primed’ platelets over VWF-coated surfaces and , thereafter , perfused SLC44A2-tGFP-transfected HEK293T cells . Again , these cells bound to VWF-‘primed’ platelets in a manner that could be blocked completely with GR144053 ( to block αIIbβ3 ) or the anti-SLC44A2 #1 antibody ( Figure 8d–e ) . Consistent with the previous results , HEK293T cells transfected with SLC44A2 ( R154Q ) exhibited markedly reduced binding to VWF-‘primed’ platelets ( Figure 8d–e ) . A previous report suggested that SLC44A2 might bind directly to VWF ( Bayat et al . , 2015 ) . However , when SLC44A2-tGFP-transfected HEK293T cells were perfused of VWF surfaces , in the absence of platelets , no binding was detected ( Figure 8d ) . Similarly , isolated neutrophils also failed to interact directly with VWF-coated surfaces , demonstrating the absolute dependence of platelets in facilitating cell capture under flow . The rs2288904-A SNP in SLC44A2 has a minor allele frequency of 0 . 22 and is protective against VTE ( Germain et al . , 2015 ) . It is therefore the common/wild-type allele , rs2288904-G , that is the risk allele for VTE with an odds ratio of 1 . 2–1 . 3 . The frequency of individuals homozygous for the protective rs2288904-A allele amongst VTE cases is 30–50% lower than in healthy controls , which perhaps provides a better indication to its protective phenotype . Given its prevalence , we genotyped a group of healthy volunteers to identify individuals homozygous for the major allele ( rs2288904-G/G ) , SLC44A2 ( R154/R154 ) , and for the protective allele ( rs2288904-A/A ) , SLC44A2 ( Q154/Q154 ) ( Figure 7—figure supplement 1c ) . SLC44A2 ( R154/R154 ) neutrophils interacted with VWF-‘primed’ platelets as before ( Figure 8f and Video 7 ) . Consistent with the previous blocking experiments , this binding was partially blocked with anti-SLC44A2 #1 , and almost completely blocked by anti-SLC44A2 #2 ( Figure 8f and Video 7 ) . Furthermore , and consistent with the transfection studies , neutrophils homozygous for the protective allele , SLC44A2 ( Q154/Q154 ) , exhibited markedly reduced ( ~75% ) binding to VWF-‘primed’ platelets ( Figure 8f and Video 7 ) demonstrating a functional consequence of the rs2288904-A polymorphism on this neutrophil-platelet interaction . Consistent with this , neutrophils heterozygous form the polymorphism SLC44A2 ( R154/Q154 ) exhibited a trend toward intermediate binding to platelets ( Figure 8f ) , although this did not reach statistical significance when compared to the SLC44A2 ( R154/R154 ) ( p=0 . 07 ) or SLC44A2 ( Q154/Q154 ) ( p=0 . 69 ) genotypes . Although the ability of platelet GPIbα binding to VWF to mediate intraplatelet signaling events has been known for many years , the role that this signaling fulfils remains poorly understood ( Goto et al . , 1995 ) . We demonstrate that under flow GPIbα-A1 binding ‘primes’ , rather than activates , platelets , based on the rapid activation of αIIbβ3 , but the lack of appreciable surface P-selectin exposure ( Figure 2 and Figure 2—figure supplement 1b ) . Some studies have reported that GPIbα-VWF-mediated signaling can induce modest α-granule release . However , the use of static conditions and processing of platelets may explain those observations . Despite this , when compared to other platelet agonists , degranulation and P-selectin exposure induced by VWF binding are both very low ( de Witt et al . , 2014; Deng et al . , 2016 ) . That platelet binding to VWF under flow ‘primes’ , rather than activates , platelets is consistent with in vivo observations . At sites of vessel damage , VWF is important for platelet accumulation through all layers of the hemostatic plug ( Joglekar et al . , 2013; Lei et al . , 2014; Verhenne et al . , 2015 ) . All platelets within a thrombus/hemostatic plug likely form interactions with VWF . Despite this , it is only the platelets in the ‘core’ of the thrombus that become P-selectin-positive , procoagulant platelets , whereas the more loosely bound platelets that form the surrounding ‘shell’ remain essentially P-selectin-negative ( Welsh et al . , 2014 ) . If VWF-binding alone were sufficient to fully activate platelets , the differential platelet characteristics of the ‘core’ and ‘shell’ would not be observed . Although it is frequently implied that VWF is only important for platelet capture under high shear conditions , murine models of venous thrombosis with no collagen exposure have repeatedly revealed an important role for VWF-mediated platelet accumulation ( Bergmeier et al . , 2008; Brill et al . , 2011; Chauhan et al . , 2007 ) . Platelet binding to VWF occurs most efficiently at arterial shear rates , but still occurs under lower linear venous shear ( Miyata and Ruggeri , 1999; Yago et al . , 2008; Zheng et al . , 2015 ) . However , linear channels do not mimic the distorted and branched paths of the vascular system that cause more disturbed flow patterns , particularly around valves . Using channels with changing geometry under lower shear conditions , we and others have noted that VWF captures platelets appreciably more efficiently in areas of disturbed flow ( Zheng et al . , 2015 ) . Indeed , at venous flow rates through bifurcated channels ( Figure 9a ) , we detected platelet capture on FL-VWF with concomitant ‘priming’ and leukocyte binding ( Figure 9b ) . This was appreciably augmented at bifurcation points where disturbed flow exists . Consistent with our earlier findings , leukocyte binding was almost completely inhibited when αIIbβ3 was blocked ( Figure 9c ) . This implies that VWF can function in platelet recruitment within the venous system , particularly in areas of turbulence ( e . g . branch sites , valves ) , which are frequently the nidus for thrombus formation in DVT . In venous thrombosis , the thrombus generally forms over the intact endothelium , in the absence of vessel damage . This poses the question of how VWF might contribute to DVT if subendothelial collagen is not exposed . It is likely that this reflects the function of newly secreted ultra-large VWF released from endothelial cells . Under low disturbed flow , released ultra-large VWF may tangle to form strings/cables over the surface of the endothelium . Tangled VWF strings/cables are appreciably more resistant to ADAMTS13 proteolysis than VWF that is simply unraveled . In the murine stenosis model of DVT , complete VWF-deficiency prevents platelet binding over the endothelium ( Bergmeier et al . , 2008; Brill et al . , 2011; Chauhan et al . , 2007 ) . Similarly , blocking GPIbα binding to VWF also completely blocks platelet accumulation and thrombus formation in the stenosis model of DVT . Thus , when platelets bind to VWF under flow in such settings , platelets may become ‘primed’ facilitating both aggregation and neutrophil binding through activated αIIbβ3 , but without activating them into procoagulant platelets . Our study reveals , that T-cells and neutrophils can bind directly to activated αIIbβ3 on platelets or that has been coupled to microchannel surfaces ( Figure 3a–d & Figure 4a–b ) . In both cases , the interaction can be inhibited by eptifibatide and GR144053 suggesting that both cell types may share the same receptor , in a manner that is dependent upon the RGD binding groove of activated αIIbβ3 . Previous studies have identified roles for β2 integrins , Mac-1 ( αMβ2 ) and LFA-1 ( αLβ2 ) , on leukocytes in mediating interactions with platelets . It should be recognized that the interactions of these molecules are dependent upon the integrins first being activated ( and therefore also the cell ) , which is not the case in our system and is in contrast to previous studies implicating Mac-1 and LFA-1 . However , we provide evidence that Mac-1 ( αMβ2 ) and LFA-1 ( αLβ2 ) are not involved by: 1 ) Leukocytes do not bind to ‘unprimed’ platelets captured by anti-PECAM-1 . As Mac-1 and LFA-1 bind to GPIbα and ICAM-2 , respectively , both of which are constitutively presented on the platelet surface ( Kuijper et al . , 1998; Simon et al . , 2000 ) , if Mac-1 and LFA-1 were the receptors involved binding would have been observed in these experiments ( Figure 3a–b ) . 2 ) Mac-1 on leukocytes can bind indirectly to activated αIIbβ3 via a fibrinogen bridge ( Weber and Springer , 1997 ) . However , we demonstrate that fibrinogen competes for leukocyte binding to bind to activated αIIbβ3 ( Figure 3b–c & Figure 4b ) . If fibrinogen were required , removal of fibrinogen from our perfusion system would have diminished the number of leukocyte interactions with primed platelets/αIIbβ3 if Mac-1 were involved . 3 ) Antibody-mediated blocking of β2 integrins did not reduce VWF-‘primed’ platelet-leukocyte interactions ( Figure 4c ) . 4 ) Only neutrophils and T cells interact with the ‘primed’ platelets , whereas Mac-1 and LFA-1 are also highly expressed in monocytes which do not bind ( Figure 4d–f ) . We also excluded a role for P-selectin-PSGL-1 for the platelet-leukocyte interaction that we observe as; 1 ) we detected little/no P-selectin on VWF-‘primed’ platelets ( Figure 2—figure supplement 1 ) , suggestive of minimal degranulation occurring; this also provides indirect evidence for the lack of CD40L on the platelet surface . 2 ) P-selectin blockade had no effect upon the number of leukocytes binding ( Figure 3e ) , and 3 ) only T-cells and neutrophils bind VWF-‘primed’ platelets ( Figure 4 ) - given that all leukocytes express PSGL-1 , ( Laszik et al . , 1996 ) and CD40 , if the capture of leukocytes were entirely P-selectin or CD40L-mediated , such cell-type selectivity would not be observed . We did , however , measure an influence of P-selectin upon the rolling speed of leukocytes over VWF-primed platelets ( Figure 3f and Figure 3—figure supplement 1d ) . This suggests that although low levels of P-selectin present on the platelet surface is insufficient to facilitate leukocyte capture , it may synergize to slow rolling of leukocytes that are first captured by αIIbβ3 . There are several studies that provide support for P-selectin-independent interactions of neutrophils and T-cells with platelets . Guidotti et al demonstrated the interaction of T-cells with small intrasinusoidal platelet aggregates in the liver during hepatotropic viral infections was independent of both P-selectin and CD40L in platelets ( Guidotti et al . , 2015 ) . Using a murine model of peritonitis , Petri et al demonstrated that neutrophil recruitment and extravasation was highly dependent upon VWF , GPIbα , and platelets , but largely independent of P-selectin ( Petri et al . , 2010 ) . Two further studies also corroborate the contention that VWF/GPIbα-bound platelets are capable of promoting neutrophil recruitment/extravasation in murine models of ischemia/reperfusion via P-selectin-independent mechanisms ( Gandhi et al . , 2012; Khan et al . , 2012 ) . These studies support the idea that both VWF and platelets can function beyond hemostasis to fulfil a role in leukocyte recruitment at sites of inflammation . As T-cells and neutrophils ( and not B-cells or monocytes ) can bind platelets via activated αIIbβ3 , this suggests that a specific receptor exists on these cells that is absent on B-cells or monocytes . Using transcriptomic and proteomic data , we identified 30 transmembrane candidates that were preferentially expressed in neutrophils ( or T-cells ) over monocytes . From this list , SLC44A2 stood out due to its recent identification as a risk locus for both VTE and stroke , but with as yet unknown functional association with these pathologies ( Apipongrat et al . , 2019; Germain et al . , 2015; Hinds et al . , 2016 ) . As platelet-leukocyte interactions are involved in both of these thrombotic disorders , we hypothesized that SLC44A2 functions as the neutrophil counter receptor for activated αIIbβ3 . The cellular function of SLC44A2 is not well-defined . It contains 10 transmembrane domains with five extracellular loops . The intracellular N-terminal tail contains several putative phosphorylation sites of unknown functional significance . As well as neutrophils , SLC44A2 expression has also been reported in endothelial cells and platelets . However , proteomic data suggest that levels in neutrophils are >300 fold greater in neutrophils that platelets ( Rieckmann et al . , 2017 ) . We provide several lines of evidence to support the direct interaction between SLC44A2 and activated αIIbβ3 . 1 ) Two different anti-SLC44A2 antibodies that recognize extracellular loops of the receptor blocked the binding of neutrophils to both VWF-‘primed’ platelets and to activated αIIbβ3 . 2 ) Recombinant expression of SLC44A2 in HEK293T cells imparted the ability of these cells to bind both VWF-primed platelets and activated αIIbβ3 under flow in a manner that can be blocked by either GR144053 or by anti-SLC44A2 antibodies . 3 ) Introduction of the rs2288904-A SNP in SLC44A2 that is protective against VTE resulted in markedly reduced binding of transfected HEK293T cells to both VWF-‘primed’ platelets and activated αIIbβ3 . 4 ) Neutrophils homozygous for the rs2288904-A/A SNP exhibit significantly reduced binding to VWF-‘primed’ platelets . Although NET production is an established mechanism through which neutrophils control pathogens ( Brinkmann et al . , 2004 ) , many questions remain as to how NETosis is regulated ( Nauseef and Kubes , 2016 ) . Binding of platelets to Kupffer cells in the liver of mice following infection with B . cereus or S . aureus is mediated by VWF ( Wong et al . , 2013 ) . This binding augments the recruitment of neutrophils , NET production and the control of infection ( Kolaczkowska et al . , 2015 ) . Mice lacking VWF or GPIbα do not form these aggregates and so have diminished neutrophil recruitment and , therefore , decreased survival ( Wong et al . , 2013 ) . How NETosis is initiated following platelet binding remains uncertain . Alone , lipopolysaccharide ( LPS ) is not a potent activator of NETosis ( Clark et al . , 2007 ) . However , LPS-stimulated platelets , which bind of fibrinogen ( i . e . αIIbβ3 is activated ) and also robustly activate NETosis independent of P-selectin ( Clark et al . , 2007; Looney et al . , 2009; Lopes Pires et al . , 2017; McDonald et al . , 2012 ) . We detected rapid release of intracellular Ca2+ ( within minutes ) in bound neutrophils ( Figure 5 and Video 5 ) that preceded the release of NETs after 80–90 min ( Figure 6 ) . This process was dependent upon neutrophils being captured under flow , suggesting that signal transduction through binding of αIIbβ3 to SLC44A2 may be mechanosensitive , which is consistent with the recent report suggesting a major influence of shear upon NETosis in the presence of platelets ( Yu et al . , 2018 ) . As NETosis can be induced following binding to purified αIIbβ3 under flow alone , this suggests that this process does not require a component of the platelet releasate ( e . g . high mobility group box 1 , platelet factor 4 , RANTES and thromboxane A2 ) which have been reported to be capable of driving NETosis ( Carestia et al . , 2016 ) . Naturally , if platelet degranulation occurs ( in response to other agonists ) , these components may have the potential to further augment this process . We propose a model in which platelets have a tunable response that can distinguish their roles in hemostasis and immune cell activation ( Figure 10 ) . The ‘priming’ of platelets by binding to VWF under flow ( in the absence of other platelet agonists ) may assist in the targeting of leukocytes to resolve pathogens or mediate vascular inflammatory response . The activated αIIbβ3 integrin can then mediate neutrophil recruitment through binding to SLC44A2 ( Figure 10 ) . We do not exclude a supporting role for P-selectin in maintaining T-cell/neutrophil recruitment , but this is not essential for initiating recruitment . Under flow SLC44A2 transduces a mechanical stimulus capable of promoting NETosis via a pathway involving synergy between NADPH oxidase and Ca2+ signaling ( Figure 10 ) . Homeostatically , this may be beneficial for immune responses . However , during chronic infection or vascular inflammation NET production may promote intravascular thrombosis . It remains to be determined whether the rs2288904-A SNP in SLC44A2 that is protective against VTE might also ( negatively ) influence the host response to infection , or certain routes of infection that are more reliant upon platelet immune cell function . This should determine the net selective pressure upon the SNP . However , as VTE is a comparatively recent selective pathology ( in evolutionary terms ) and rates of VTE are typically low during the most common child-bearing years , this positive selective pressure on rs2288904-A SNP may not strong . Indeed , the prevalence of this SNP in different global populations suggests that the SNP has not originated recently . Moreover , if the host response to infection is , at least in certain settings , also impaired in individuals carrying the rs2288904-A SNP , this would be predicted to slow the rate of positive selection . More work is needed to understand role of rs2288904-A SNP in VTE and infection risk and penetrance of SNP within large genomic datasets . This study identifies activated αIIbβ3 as a receptor and agonist for neutrophils through SLC44A2 . This provides a previously uncharacterized mechanism of how platelet-neutrophil cross-talk is manifest in innate immunity; it also provides an explanation for how VWF and platelet-dependent neutrophil recruitment and NETosis may occur in thrombotic disorders such as DVT ( Laridan et al . , 2019 ) , but also thrombotic microangiopathies like thrombotic thrombocytopenic purpura ( Fuchs et al . , 2012b ) . Identification of SLC44A2 as the counter-receptor for activated αIIbβ3 in conjunction with the prior identification of the protective rs2288904-A SNP in SLC44A2 that impairs the binding of neutrophils to platelets highlights SLC44A2 as a potential therapeutic target . Recent data reveal that SLC44A2-deficient mice exhibit normal hemostatic responses ( Tilburg et al . , 2018 ) but are protected against development of venous thrombosis ( Tilburg et al . , 2020 ) , provide further encouragement for this strategy . The coding sequence for the human VWF A1 domain ( Glu1264 to Leu1469 ) was cloned into the pMT-puro vector , containing a C-terminal V5 and polyhistidine tag . The Y1271C/C1272R mutations were introduced by PCR into the A1 domain ( A1* ) ( Blenner et al . , 2014 ) . All vectors were verified by sequencing . S2 insect cells stably expressing either VWF A1-V5-His or VWF A1-V5-His were selected using puromycin ( Life Technologies ) . Cells were cultured under sterile conditions at 28°C in Schneider’sDrosophilamedium ( Lonza ) , supplemented with 10% heat-inactivated fetal bovine serum ( FBS ) , 50 μg/ml penicillin and 50 U/ml streptomycin . Cells were grown in suspension in 2L conical flasks to a density of 2 × 106 cells/ml . Expression of VWF A1 or A1 was induced by addition of 500 μM CuSO4 for 5–7 days , at 28°C and 110 rpm . Conditioned media from S2 cells were tested using the Venor GeM mycoplasma detection kit ( Sigma ) and confirmed to be mycoplasma free . Conditioned media were harvested , cleared by centrifugation , concentrated by tangential flow filtration and dialyzed against 20 mM Tris ( pH 7 . 8 ) 500 mM NaCl . VWF A1 or A1* were purified by a two-step purification method using a Ni2+-HiTrap column followed by a heparin-Sepharose column ( GE Healthcare ) and elution with 20 mM Tris , 600 mM NaCl . VWF A1 and A1* were dialyzed in phosphate-buffered saline ( PBS ) . A1 and A1* concentrations were determined by absorbance at 280 nm . Proteins were analyzed by SDS-PAGE under reducing and non-reducing , and by western blotting using anti-His ( RRID:AB_298652 ) or anti-VWF ( RRID:AB_2315602 ) antibodies . Full length , multimeric VWF was isolated from Haemate P by gel filtration and quantified by a specific VWF ELISA , as previously described ( O'Donnell et al . , 2005 ) . Fresh blood was collected from consented healthy volunteers in 40 µM PPACK ( for whole blood experiments ) , 3 . 13% citrate ( for leukocyte isolation ) or 85 mM sodium citrate , 65 mM citric acid , 111 mM D ( + ) glucose , pH 4 . 5 ( 1x ACD , for plasma-free blood preparation ) . For reconstituted plasma-free blood , red blood cells ( RBCs ) and leukocytes were pelleted and washed twice . Separately , platelets were washed twice in 1x HEPES-Tyrode ( HT ) buffer containing 0 . 35% BSA , 75mU apyrase and 100 nM prostaglandin E1 ( Sigma ) . RBCs , leukocytes and platelets were resuspended in 1x HT buffer supplemented with 0 . 35% BSA . In some experiments , 1 . 3 mg/ml purified fibrinogen ( Haem Tech ) was added . For Ca2+ assays , PRP was incubated with 5 µM Fluo-4 AM ( Thermo Fisher Scientific ) for 30 min at 37°C prior to washing , and plasma-free blood was recalcified with 1 mM CaCl2 ( final concentration ) immediately prior to flow experiments . PMNs and PBMCs separated using Histopaque1077 and Histopaque1119 were resuspended in 1x HT , supplemented with 1 . 5 mM CaCl2 . For Ca2+ assays , PMNs were preloaded with 1 µM Fluo-4 AM for 30 min at 37°C , before washing . This study was approved by the Imperial College Research Ethics Committee ( approval reference 19IC5523 ) , and informed consent and consent to publish was obtained from all healthy volunteers . VenaFluoro8+ microchips ( Cellix ) were coated directly with 2 μM VWF in PBS overnight at 4°C in a humidified chamber . Coated channels were blocked for 1 hr with 1x HEPES-Tyrode ( HT ) buffer containing 1% bovine serum albumin ( BSA ) . For the isolated VWF A1 and A1* domains , NTA PEGylated microchips ( Cellix ) were used to capture the A1 or A1* via their His tags ( Tischer et al . , 2014 ) . Channels were stripped with EDTA before application of Co2+ and washing with 20 mM HEPES , 150 mM NaCl , pH 7 . 4 ( HBS ) . To each channel , 20 μl of 3 . 75 μM VWF A1 or A1* were applied at room temperature for 20 min in a humidified chamber . Channels were then incubated with H2O2 for 30 min to oxidize Co2+ to Co3+ , which stabilizes the binding of His-tagged A1/A1* ( Wegner et al . , 2016 ) . To NHS-microchannels ( Cellix ) , 2 . 6 μM purified αIIbβ3 ( ERL ) , 0 . 25 mg/ml PECAM-1 ( RRID:AB_314328 ) , anti-β3/LIBS2 antibody ( RRID:AB_10806476 ) , anti-CD16 ( RRID:AB_467129 ) or 0 . 25 mg/ml BSA were covalently attached by amine-coupling according to manufacturer’s instructions . For directly coated αIIbβ3 channels , the surface was washed with HBS containing 1 mM MnCl2 , 0 . 1 mM CaCl2 following coating . Mn2+ was maintained in all subsequent buffers to cause αIIbβ3 to favor its open , ligand binding conformation , as previously reported ( Litvinov et al . , 2005 ) . To anti-β3/LIBS2 antibody-coated channels , αIIbβ3 ( ERL ) was perfused over the surface to facilitate both capture and activation of αIIbβ3 on the surface . Whole blood or plasma-free blood was perfused through channels coated with either FL-VWF , A1 , A1* or anti-PECAM-1 at shear rates of 500–1500 s−1 for 3 . 5 min , followed by 50 s−1 for 15 min using a Mirus Evo Nanopump and Venaflux64 software ( Cellix ) . In separate experiments , 2 . 4 μM eptifibatide ( Sigma ) , 2 μM GR144053 ( Tocris ) , or 50 μg/ml anti-P-selectin blocking antibody ( RRID:AB_395908 ) were supplemented to whole blood or plasma-free blood . DiOC6 ( 2 . 5 μM; Invitrogen ) was used to label platelets and leukocytes . Cells were monitored in real-time using an inverted fluorescent microscope ( Zeiss ) or a SP5 confocal microscope ( Leica ) . Leukocytes and platelets were distinguished by their larger size . For presentation and counting purposes , leukocytes were pseudo-colored to distinguish them . In some experiments , antibodies that recognize the second extracellular loop of SLC44A2 , rabbit anti-SLC44A2 #1 ( RRID:AB_2827953 ) or the first extracellular loop , rabbit anti-SLC44A2 #2 ( RRID:AB_2827954 ) ( 0–20 μg ml−1 ) to block SLC44A2 were compared to non-immune rabbit IgG ( Abcam; 20 μg ml−1 ) to explore the influence of SLC44A2 on neutrophils to bind to either VWF-‘primed’ platelets or isolated/activated αIIbβ3 . Isolated PMNs and PBMCs were perfused through channels coated either directly or indirectly with αIIbβ3 , or BSA at 50 s−1 for 15 min . Antibodies specific to the different types of leukocytes were added to isolated leukocytes , that is anti-CD16 ( RRID:AB_2016663 ) conjugated to allophycocyanin ( APC ) to identify neutrophils , anti-CD14-APC ( RRID:AB_314190 ) for monocytes , anti-CD3-APC ( RRID:AB_314066 ) for T-cells and anti-CD19-APC ( RRID:AB_314242 ) for B-cells . To visualize NETosis , neutrophils were labeled with 8 μM Hoechst dye ( cell permeable ) and 1 μM Sytox Green ( cell impermeable ) and monitored for 2 hr . As indicated , isolated PMNs were preincubated with 20 μM TMB-8 ( Ca2+ antagonist and protein kinase C inhibitor; Sigma ) , for 15 min , or 30 μM DPI ( NADPH oxidase inhibitor; Sigma ) for 30 min at 37°C prior to NETosis assays . In some experiments , neutrophils were captured on microchannels coated with anti-CD16 and stimulated with 160 nM PMA prior to analysis of NETosis in the presence and absence of inhibitors . To confirm the presence of NETs , neutrophils that were captured by activated αIIbβ3 and fixed with 4% paraformaldehyde after 2 hr . Fixed neutrophils were permeabilized with 0 . 1% Triton X-100 in PBS for 10 min , blocked with 3% BSA in PBS and , thereafter , incubated with rabbit polyclonal anti-citrullinated H3 ( RRID:AB_304752 , 10 μg/ml ) overnight at 4°C ( Martinod et al . , 2013 ) . Neutrophils were incubated with a goat anti-rabbit secondary antibody conjugated with Alexa647 ( Abcam , 1:500 ) and with the Hoechst dye ( 8 μM ) for 2 hr , washed and then visualized by confocal microscopy . Quantitation of platelet rolling , aggregation and intracellular Ca2+ release was achieved using SlideBook 6 . 0 software ( RRID:SCR_014300 ) . The number of leukocytes rolling/attaching per minute at 50 s−1 was derived by counting the number of cells in one field of view over a period of 13 min . NETosis was quantified by determining the proportion of all neutrophils in the microchannel that had undergone NETosis after 2 hr . RNA sequencing data from different leukocytes were obtained from the BLUEPRINT consortium ( Grassi et al . , 2019 ) . For this , neutrophils and monocytes were isolated from peripheral blood . PBMCs were separated by gradient centrifugation ( Percoll 1 . 078 g/ml ) whilst neutrophils were isolated by CD16 positive selection ( Miltenyi ) from the pellet , after red blood cell lysis . PBMCs were further separated to obtain a monocyte-rich layer using a second gradient ( Percoll 1 . 066 g/ml ) and monocytes further purified by CD14-positive selection ( Miltenyi ) after CD16 depletion . For neutrophils and monocytes , gene expression was tested also on Illumina HT12v4 arrays ( accession E-MTAB-1573 at arrayexpress ) . The purification of naive B lymphocytes , naive CD4 lymphocytes , naive CD8 lymphocytes used in this study has been extensively described . Regulatory CD4 lymphocytes ( T regs ) were isolated by flow activated cytometry using the following surface markers combinations: CD3+ CD4+ CD25+ CD127low . Cell type purity was assessed by flow cytometry and morphological analysis . RNA was extracted using TRIzol according to manufacturer’s instructions , quantified using a Qubit RNA HS kit ( Thermofisher ) and quality controlled using a Bioanalyzer RNA pico kit ( Agilent ) . For all cell types , libraries were prepared using a TruSeq Stranded Total RNA Kit with Ribo-Zero Gold ( Illumina ) using 200 ng of RNA . Trim Galore ( v0 . 3 . 7 ) ( http://www . bioinformatics . babraham . ac . uk/projects/trim_galore/ ) with parameters ‘-q 15 s 3 --length 30 -e 0 . 05’ was used to trim PCR and sequencing adapters . Trimmed reads were aligned to the Ensembl v75 human transcriptome with Bowtie 1 . 0 . 1 using the parameters ‘-a --best --strata -S -m 100 -X 500 --chunkmbs 256 --nofw --fr’ . MMSEQ ( v1 . 0 . 10 ) was used with default parameters to quantify and normalize gene expression . Differential gene expression analyses were performed: mature neutrophils ( n = 7 ) vs monocytes ( n = 5 ) and CD4-positive/αβ T cells ( n = 8 ) vs monocytes ( n = 5 ) . Regulatory T cells ( Treg , n = 1 ) and native B cells ( n = 1 ) , are included in the heatmap , for comparison but were not used in differential gene expression analysis due to the low number of biological replicates . We selected genes that were expressed significantly higher in neutrophils than in monocytes , and also those that were significantly higher in CD4-positive/αβ T cells than in monocytes . Their intersection identified 750 genes ( 598 of which protein coding ) . From these 598 genes , we selected the 93 genes that contained the Uniprot annotation of ‘INTRAMEMBRANE DOMAIN’ or ‘TRANSMEM DOMAIN’ . The effective log2 ( FPKM+1 ) data were presented in the heatmap . Further selection involved discarding those transmembrane proteins that are not present on the extracellular membrane , or primarily associated with intracellular membranes as determined by Uniprot annotation . Proteins that ( where known ) had extracellular regions of <30 amino acids , as determined in Uniprot , that might be less likely capable of mediating specific ligand binding were also excluded . Finally , analysis of proteomic data from the ImmProt ( http://immprot . org ) resource was used to verify higher levels of protein of each selected gene in neutrophils than in monocytes . The mammalian expression vector , pCMV6-Entry containing the human SLC44A2 cDNA C-terminally fused to tGFP was purchased from OriGene . To introduce the rs2288904 SNP encoding a R154Q substitution , site-directed mutagenesis was performed using the R154Q ‘top’ and ‘bot’ primers ( see Key Resources Table ) . Successful introduction of the SNP was confirmed by sequencing . HEK293T cells ( RRID:CVCL_0063 ) were cultured as adherent layers , in humidified incubators at 37°C , 5% CO2 , in minimum essential media ( Sigma ) supplemented with 10% FBS , 1 U/ml Penicillin 0 . 1 mg/ml Streptomycin , 1% non-essential amino acids ( Sigma ) and 2 mM L-glutamine . Conditioned media from HEK293T cells were tested using the Venor GeM mycoplasma detection kit ( Sigma ) and confirmed to be mycoplasma free . Cells were authenticated by out-sourced short tandem repeat analysis ( NorthGene ) of genomic DNA extracted from HEK293T cells using PureLink Genomic DNA kit ( Invitrogen ) and confirmed to be HEK293T cells . HEK293T cells were seeded in 6-well plates 24 hr prior to transfection and transfected using Lipofectamine 2000 ( Invitrogen ) . Transfection efficiency was estimated visually by fluorescent microscopy and quantified using flow cytometry . In all cases , transfection efficiency was >75% . Cells were harvested 24 hr post-transfection with Tryplex ( Life Tech ) to obtain a single-cell suspension . Cells were washed with complete medium and cells resuspended in serum-free OptiMEM ( Life Tech ) until use . Transfected HEK293T cell lysates were harvested for Western blot analysis using an anti-tGFP monoclonal antibody ( RRID:AB_2622256 ) at 0 . 3 μg ml−1 to verify expression of the fusion proteins , and the consistent and uniform fusion of tGFP to SLC44A2 variants . Microchannels were coated with FL-VWF or αIIbβ3 ( coupled via the anti-β3/LIBS2 antibody ) . Thereafter , unlabeled plasma-free blood was perfused over the FL-VWF coated channels at high shear for 3 . 5 min to capture a layer of ‘primed’ platelets . Platelet coverage was monitored in bright-field . Channels were subsequently washed with 1xHT buffer to remove the blood and SLC44A2-tGFP or SLC44A2 ( R154Q ) -tGFP transfected HEK293T cells were perfused at low shear ( 25 s−1 ) for 10 min . Transfected HEK293T cells were also perfused through FL-VWF coated channels ( in the absence of platelets for 30 min at 25 s−1 ) to examine any direct interaction with VWF . Transfected HEK293T cells were also perfused through αIIbβ3 ( coupled via the anti-β3/LIBS2 antibody ) channels at 25 s−1 for 10 min . Binding of fluorescent HEK293T cells was quantified by counting the number of cells attached after 10 min across the whole channel and then expressing this as the mean number of cells/field of view . In separate experiments , the ability of GR144053 to block αIIbβ3 , or antibodies that recognize the second or first extracellular loop of SLC44A2 , respectively , rabbit anti-SLC44A2 #1 ( RRID:AB_2827953 ) or rabbit anti-SLC44A2 #2 ( RRID:AB_2827954 ) ( 0–20 μg ml−1 ) to block SLC44A2 were compared to non-immune rabbit IgG ( Abcam; 20 μg ml−1 ) . To identify individuals homozygous for the SLC44A2 rs2288904-A SNP or for the common/wild-type allele rs2288904-G , 25 μl blood was taken by pin prick from healthy volunteers that provided written informed consent that was approved by the Imperial College Research Ethics Committee ( approval reference 19IC5523 ) . Genomic DNA was extracted using PureLink Genomic DNA kit ( Invitrogen ) . DNA yield was quantified by NanoDrop . Genomic DNA from each volunteer was used as a template to PCR amplify a 410 base pair fragment of the SLC44A2 gene spanning the SNP site using SLC44A2 ‘top’ and ‘bot’ primers ( see Key Resources Table ) . After amplification , samples were separated by agarose gel electrophoresis and the 410 bands excised , purified using the Gel Extraction kit ( Qiagen ) and sequenced using the ‘top’ PCR primer . PMN isolated from genotyped individuals were subsequently used to examine their ability to bind both activated αIIbβ3 ( captured using the LIBS2 , anti-β3 antibody ) and VWF-‘primed’ platelets , as described above . Statistical analysis was performed using Prism 6 . 0 software ( RRID:SCR_002798 ) . Differences between data/samples was analyzed using unpaired two-tailed Student’s t-test or Mann-Whitney , as appropriate and as indicated in figure legends . Data are presented as mean ± standard deviation , or median ±95% confidence interval . The number of individual experiments performed ( n ) is given in each legend . Values of p<0 . 05 were considered statistically significant .	Platelets in our blood form clots over sites of injury to stop us from bleeding . Blood clots can also occur in places where they are not needed , such as deep veins in our legs or other regions of the body . Developing such clots – also known as deep vein thrombosis ( or DVT for short ) – is one of the most common cardiovascular diseases and a major cause of death . Although certain inherited factors have been linked to DVT , the underlying mechanisms of the disease remain poorly understood . In addition to platelets , the pathological ( or dangerous ) clots that cause DVT also contain immune cells called neutrophils which fight off bacterial infections . Platelets are recruited to the wall of the vein by a protein called “von Willebrand Factor” ( or VWF for short ) . However , it remained unclear how these recruited platelets interact with neutrophils and whether this promotes the onset of DVT . To answer this question , Constantinescu-Bercu et al . used a device that mimics the flow of blood to study how human platelets change when they are exposed to VWF . This revealed that VWF ‘primes’ the platelets to interact with neutrophils via a protein called integrin αIIbβ3 . Further experiments showed that integrin αIIbβ3 binds to a protein on the surface of neutrophils called SLC44A2 . Once the neutrophils interacted with the ‘primed’ platelets , they started making traps which increased the size of the blood clot by capturing other blood cells and proteins . Finally , Constantinescu-Bercu et al . studied a genetic variant of the SLC44A2 protein which is found in 22% of people and is associated with a lower risk of developing DVT . This genetic mutation caused SLC44A2 to interact with ‘primed’ platelets more weakly , which may explain why people with this genetic variant are protected from getting DVT . These findings suggest that blocking the interaction between ‘primed’ platelets and neutrophils could reduce the risk of DVT . Although current treatments for DVT can prevent patients from forming dangerous blood clots , they can also cause severe bleeding . Since neutrophils are not crucial for normal blood clots to form at the site of injury , drugs targeting SLC44A2 could inhibit inappropriate clotting without causing excess bleeding .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2020	Activated αIIbβ3 on platelets mediates flow-dependent NETosis via SLC44A2
Information about nutrient availability is assessed via largely unknown mechanisms to drive developmental decisions , including the choice of Caenorhabditis elegans larvae to enter into the reproductive cycle or the dauer stage . In this study , we show that CMK-1 CaMKI regulates the dauer decision as a function of feeding state . CMK-1 acts cell-autonomously in the ASI , and non cell-autonomously in the AWC , sensory neurons to regulate expression of the growth promoting daf-7 TGF-β and daf-28 insulin-like peptide ( ILP ) genes , respectively . Feeding state regulates dynamic subcellular localization of CMK-1 , and CMK-1-dependent expression of anti-dauer ILP genes , in AWC . A food-regulated balance between anti-dauer ILP signals from AWC and pro-dauer signals regulates neuroendocrine signaling and dauer entry; disruption of this balance in cmk-1 mutants drives inappropriate dauer formation under well-fed conditions . These results identify mechanisms by which nutrient information is integrated in a small neuronal network to modulate neuroendocrine signaling and developmental plasticity . Discrete alternate phenotypes arising from a single genotype in response to varying environmental cues is referred to as polyphenism ( Michener , 1961; Mayr , 1963; Stearns , 1989 ) . Well-described polyphenic traits include the exhibition of wings on locusts , caste hierarchy in social insects , and environmental sex determination in reptiles ( Nijhout , 2003; Beldade et al . , 2011; Simpson et al . , 2011 ) . In well-studied cases as in insects , it has been shown that animals integrate sensory cues during specific developmental stages to promote the expression of alternate phenotypic traits via regulation of endocrine and neuromodulatory signaling ( Simpson et al . , 2011; Watanabe et al . , 2014 ) . Environmental cues that trigger developmental plasticity include pheromones , temperature , mechanical stimuli , and food ( Nijhout , 2003; Simpson et al . , 2011 ) . In particular , nutrient availability and quality during development is a major regulator of polyphenism in many species ( Wheeler , 1986; Greene , 1989; Pfennig , 1992; Bento et al . , 2010 ) . Although the extent and adaptive value of polyphenism has been extensively discussed ( Pfennig et al . , 2010; Moczek et al . , 2011 ) , the underlying molecular and neuronal mechanisms that allow animals to sense and integrate signals from food and feeding-state signals in the context of other cues to regulate phenotypic plasticity are not fully understood . Caenorhabditis elegans exhibits polyphenism in response to environmental cues sensed during a critical period in development . Shortly following hatching , C . elegans larvae assess crowding in their environment via concentrations of a complex mixture of small molecules called ascarosides ( collectively referred to as dauer pheromone ) produced by conspecifics ( Golden and Riddle , 1982 , 1984c; Jeong et al . , 2005; Butcher et al . , 2007; Edison , 2009; Ludewig and Schroeder , 2013 ) . High concentrations of one or more of these chemicals promote entry of larvae into the alternate stress-resistant and long-lived dauer developmental stage , whereas under uncrowded conditions , larvae continue in the reproductive cycle ( Cassada and Russell , 1975 ) ( Figure 1A ) . Although pheromone is the instructive cue for dauer entry , additional cues , such as temperature and food availability , also regulate this binary decision ( Golden and Riddle , 1984a , 1984b , 1984c; Ailion and Thomas , 2000 ) ( Figure 1A ) . Thus , high ( low ) concentrations of food or low ( high ) temperature can efficiently inhibit ( promote ) pheromone-induced dauer formation , allowing animals to assess and integrate diverse sensory cues in order to make a robust developmental choice . 10 . 7554/eLife . 10110 . 003Figure 1 . CMK-1 acts in the AWC and ASI/AWA neurons to inhibit dauer formation in fed animals . ( A ) Simplified model of sensory inputs modulating TGF-β and insulin signaling in the regulation of the dauer decision . See text for details . ( B ) Quantification of dauer formation in wild-type animals in the presence of 6 μM ascr#3 and the indicated amounts of heat-killed ( blue circles ) or live ( red circles ) OP50 bacteria . Each filled circle represents one assay; n > 65 animals per assay , three independent experiments . Line represents best fit to the data . and * indicate different from values using 160 μg of heat-killed bacteria at p < 0 . 01 and 0 . 001 , respectively ( ANOVA and Games-Howell post-hoc test ) . ( C ) Dauers formed by strains of the indicated genotypes in the presence of 6 μM ascr#3 and 80 μg live OP50 . Each data point is the average of ≥3 independent experiments of >65 animals each . Errors are SEM . * , , and * indicate different from wild-type at p < 0 . 05 , 0 . 01 , and 0 . 001 , respectively ( ANOVA and Games-Howell post-hoc test ) . ( D ) Dauer formation in cmk-1 mutants grown with 6 μM ascr#3 and the indicated amounts of live OP50 . Each data point is the average of ≥3 independent experiments of >65 animals each . Errors are SEM . and * indicate different from corresponding values using 80 μg OP50 at p < 0 . 01 and 0 . 001 , respectively ( Student's t-test ) . ( E ) Dauers formed by strains of the indicated genotypes grown on plates containing 6 μM ascr#3 and 80 μg live OP50 . Promoters used to drive wild-type cmk-1 cDNA expression were: cmk-1p—cmk-1 upstream regulatory sequences; ASK—sra-9p; AFD—ttx-1p; ASJ—trx-1p; ASI/AWA—gpa-4p; AWC—ceh-36Δp . Each data point is the average of ≥3 independent experiments of >65 animals each . For transgenic strains , data are averaged from 1–4 independent lines each . Errors are SEM . * , , and * indicate different from wild-type at p < 0 . 05 , 0 . 01 , and 0 . 001 , respectively; ### indicates different from cmk-1 ( oy21 ) at p < 0 . 001 ( ANOVA and Games-Howell post-hoc test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10110 . 00310 . 7554/eLife . 10110 . 004Figure 1—source data 1 . Dauer assay data for individual trials in Figure 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 10110 . 00410 . 7554/eLife . 10110 . 005Figure 1—source data 2 . Dauer assay data for individual trials in Figure 1—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 10110 . 00510 . 7554/eLife . 10110 . 006Figure 1—figure supplement 1 . CMK-1 inhibits dauer formation in fed animals . ( A ) Dauer formation in wild-type and cmk-1 mutants on 80 μg live OP50 in the absence of exogenous pheromone . n > 65 animals per assay; at least 12 independent trials . Errors are SEM . ( B ) Dauers formed by daf-22 ( m130 ) animals in the presence of 6 μM ascr#3 and the indicated amounts of heat-killed ( blue circles ) or live ( red circles ) OP50 bacteria . Each filled circle represents one assay; n > 65 animals per assay , three independent experiments . and * indicate different from values using 160 μg of heat-killed bacteria at p < 0 . 01 and 0 . 001 , respectively ( ANOVA and Games-Howell post-hoc test ) . ( C , D ) Dauers formed by wild-type or cmk-1 mutants on 80 μg live OP50 and the indicated amounts of ascr#2 or icas#9 . * indicates different from wild-type at p < 0 . 001 ( ANOVA and Games-Howell post-hoc test ) . n > 65 animals each , three independent experiments . Errors are SEM . ( E ) Egg-laying is sensitive to bacterial food in wild-type and cmk-1 ( oy21 ) mutants . Shown is the average number of eggs laid per hour by adult animals in the presence or absence of live OP50 . * indicates different between the indicated values at p < 0 . 001 ( Student's t-test ) . n > 30 animals for each condition . Errors are SEM . ( F ) Dauers formed by wild-type or cmk-1 ( oy21 ) mutants on 80 μg of live OP50 and 6 μM ascr#3 at 20°C . For each assay: n > 65 animals each , three independent experiments . Errors are SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 10110 . 00610 . 7554/eLife . 10110 . 007Figure 1—figure supplement 2 . cmk-1p::gfp is expressed broadly in multiple neurons . Shown is the expression pattern of cmk-1p::gfp in an L1 larva ( left panel: GFP; right panel: DIC ) . Note GFP expression in multiple neurons in the head and tail . Scale bar: 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 10110 . 007 Decades of investigation have shown that environmental stimuli detected by sensory neurons modulate neuroendocrine signaling to regulate the choice of larval developmental trajectory in C . elegans ( Fielenbach and Antebi , 2008 ) . High pheromone concentrations , low food abundance and high temperature cues downregulate expression of the daf-7 TGF-β ligand and several insulin-like peptide ( ILP ) genes in subsets of ciliated sensory neurons in the head amphid organs ( Ren et al . , 1996; Schackwitz et al . , 1996; Li et al . , 2003; Cornils et al . , 2011; Entchev et al . , 2015 ) ( Figure 1A ) . Downregulated TGF-β and insulin signaling in turn decrease biosynthesis of dafachronic acid steroid hormones by neuronal and non-neuronal endocrine cells ( Fielenbach and Antebi , 2008 ) . In the absence of these steroid hormones , the DAF-12 nuclear hormone receptor promotes dauer entry , whereas in the ligand-bound form , DAF-12 promotes reproductive development ( Antebi et al . , 1998 , 2000; Ludewig et al . , 2004 ) . Ciliated chemosensory neurons required to sense a subset of ascarosides for the regulation of dauer entry have been identified ( Schackwitz et al . , 1996; Kim et al . , 2009; McGrath et al . , 2011; Park et al . , 2012 ) . However , little is known about how food is sensed , and how food signals are integrated with pheromone signals at the level of endocrine gene expression to influence the dauer decision . Here , we identify the CMK-1 calcium/calmodulin-dependent protein kinase I ( CaMKI ) as a key player in the regulation of dauer formation as a function of feeding state . Expression of the daf-7 TGF-β and daf-28 ILP genes are downregulated in well-fed cmk-1 mutants , and we find that CMK-1 acts cell-autonomously in the ASI sensory neurons , and non cell-autonomously in the AWC sensory neurons , to regulate the expression of daf-7 and daf-28 , respectively . We show that the subcellular localization of CMK-1 in AWC is feeding state-dependent , and that CMK-1 promotes expression of anti-dauer ILP genes in AWC . Our results indicate that a balance of CMK-1-regulated anti-dauer signals from AWC , as well as pro-dauer signals , regulates dauer entry as a function of feeding state , and that this balance is disrupted in cmk-1 mutants to inappropriately promote dauer entry under well-fed conditions . We also find that basal activity levels in AWC are enhanced upon prolonged starvation in wild-type animals and in well-fed cmk-1 mutants , and that increased activity acts in parallel with CMK-1 to antagonize dauer formation . Together , these results identify CMK-1 CaMKI as a key molecule that encodes information about nutrient availability within a sensory neuron network to regulate neuroendocrine signaling and a polyphenic developmental choice . To verify that pheromone-induced dauer formation in wild-type animals is suppressed by bacterial food , we quantified dauers formed in the presence of pheromone and different concentrations of non-replicative ( heat-killed ) as well as replicative ( live ) bacteria . Although no dauers were observed in the absence of added pheromone ( Figure 1—figure supplement 1A; Figure 1—source data 2 ) , >80% of wild-type larvae entered into the dauer stage on plates containing 6 μM ascr#3 ( also referred to as asc-ΔC9 or C9 ) and l60 μg heat-killed OP50 bacteria ( Figure 1B; Figure 1—source data 1 ) . Dauer formation decreased upon increasing the amount of heat-killed bacteria and was fully suppressed by only 80 μg of live bacteria ( Figure 1B ) . We could not reliably quantify the effects of lower concentrations of live bacteria since as reported previously , animals arrest development postembryonically when food becomes limiting ( Hong et al . , 1998; Fukuyama et al . , 2006; Baugh , 2013 ) . Food also inhibited dauer formation in daf-22 mutants that are unable to produce most ascarosides ( Golden and Riddle , 1985; Butcher et al . , 2009 ) ( Figure 1—figure supplement 1B ) , indicating that under these conditions , dauer formation is not increased simply due to enhanced endogenous pheromone signaling . These observations confirm that food cues modulate pheromone-induced dauer formation . To begin to explore the mechanisms by which food signals are integrated with pheromone cues to regulate dauer entry , we focused on genes previously implicated in different aspects of nutrient sensing and/or metabolism in C . elegans . These include the aak-1 and aak-2 AMP-activated protein kinases ( Apfeld et al . , 2004; Narbonne and Roy , 2009; Cunningham et al . , 2012 ) , the crh-1 CREB transcription factor and the cmk-1 CaMKI kinase ( Kimura et al . , 2002; Suo et al . , 2006; Suo and Ishiura , 2013 ) , the egl-4 cGMP-dependent protein kinase ( Daniels et al . , 2000; You et al . , 2008 ) , the eat-4 glutamate transporter ( Avery , 1993; Hills et al . , 2004 ) , the tph-1 tryptophan hydroxylase and cat-2 tyrosine hydroxylase enzymes ( Sawin et al . , 2000; Hills et al . , 2004; Suo et al . , 2009; Ezcurra et al . , 2011; Entchev et al . , 2015 ) , and the skn-1 , hlh-30 and mxl-3 transcription factors ( Paek et al . , 2012; O'Rourke and Ruvkun , 2013 ) ( Figure 1C ) . Specifically , we reasoned that mutations in genes essential for food signal integration would result in inappropriate entry into the dauer stage in the presence of pheromone and plentiful food but would not lead to constitutive dauer formation ( dauer-constitutive or Daf-c ) ( Hu , 2007 ) . Mutations in the cmk-1 CaMKI gene fulfilled both these criteria . cmk-1 ( oy20 ) missense , as well as cmk-1 ( oy21 ) putative null , mutants consistently formed dauers in the presence of 80 μg live bacteria and 6 μM ascr#3—conditions that fully suppressed dauer formation in wild-type animals ( Figure 1C ) . Moreover , few dauers were observed in the absence of added pheromone ( Figure 1—figure supplement 1A ) , indicating that cmk-1 mutants do not form dauers constitutively . cmk-1 mutants also formed dauers in the presence of live food and ascr#2 ( also referred to as asc-C6-MK or C6 ) and icas#9 ( also referred to as IC-asc-C5 or C5 ) ( Figure 1—figure supplement 1C , D ) , indicating that the response was not specific to a particular ascaroside . We could not reliably examine the effects of heat-killed bacteria in this assay since cmk-1 mutants exited the dauer stage prematurely under these conditions ( not shown ) , possibly due to additional metabolic defects that we have not explored further in this study . Increasing the amount of live food to 320 μg decreased but did not fully suppress dauer formation in cmk-1 mutants ( Figure 1D ) , implying that these animals are able to respond to food , but exhibit a shifted threshold of response to feeding . Consistent with the notion that cmk-1 mutants retain the ability to respond to food cues , egg-laying was modulated by bacterial food in both wild-type and cmk-1 ( oy21 ) adult animals ( Figure 1—figure supplement 1E ) . In addition to food and pheromone , temperature also regulates dauer formation ( Golden and Riddle , 1984b ) , and we and others previously showed that cmk-1 mutants exhibit altered thermosensory behaviors ( Satterlee et al . , 2004; Schild et al . , 2014; Yu et al . , 2014 ) . cmk-1 ( oy21 ) mutants retained the ability to respond to temperature in the context of dauer formation , since dauer formation was fully suppressed at a lower temperature of 20°C in the presence of pheromone ( Figure 1—figure supplement 1F ) . We infer that cmk-1 mutants are defective in correctly integrating food signals into the dauer decision pathway . cmk-1 is expressed broadly in multiple sensory and non-sensory neuron types in C . elegans ( Kimura et al . , 2002; Satterlee et al . , 2004 ) ( Figure 1—figure supplement 2 ) . We first verified that expression of a cmk-1 cDNA under its endogenous regulatory sequences rescues the dauer formation phenotype in the presence of live bacteria and exogenous ascr#3 ( Figure 1E ) . We next performed cell-specific rescue experiments to identify the site ( s ) of CMK-1 function in the regulation of dauer entry under these conditions . Expression in the ASK pheromone-sensing , or the AFD thermosensory , neurons did not affect the dauer formation phenotype of cmk-1 mutants ( Figure 1E ) , suggesting , but not proving , that CMK-1 does not act simply by modulating pheromone or temperature sensitivity . However , expression of wild-type cmk-1 sequences in the ASI/AWA or AWC sensory neurons resulted in partial , and nearly complete , suppression of dauer formation , respectively . Expression of cmk-1 in both ASI/AWA and AWC suppressed dauer formation to the same extent as expression in AWC alone ( Figure 1E ) . No rescue was observed upon expression in the ASJ sensory neurons which have also been previously implicated in the regulation of dauer formation ( Bargmann and Horvitz , 1991; Schackwitz et al . , 1996 ) ( Figure 1E ) . Thus , CMK-1 acts primarily in AWC but also in ASI/AWA to regulate dauer formation . The DAF-7 TGF-β and ILP neuroendocrine signaling pathways act in parallel to regulate dauer formation ( Fielenbach and Antebi , 2008 ) . We asked whether CMK-1 acts in either or both these pathways to regulate dauer formation . daf-7 TGF-β null mutants are strongly Daf-c , and this phenotype is suppressed upon loss of DAF-3 SMAD or DAF-5 transcription factor function ( Golden and Riddle , 1984c; Vowels and Thomas , 1992; Thomas et al . , 1993 ) . Despite the presence of multiple ILPs , the C . elegans genome encodes a single insulin receptor encoded by the daf-2 gene . Although loss of function of single ILP genes such as daf-28 results in only weak effects on dauer formation , likely due to redundancy ( Li et al . , 2003; Cornils et al . , 2011; Ritter et al . , 2013; Hung et al . , 2014 ) , daf-2 insulin receptor mutants are Daf-c ( Thomas et al . , 1993; Gottlieb and Ruvkun , 1994; Gems et al . , 1998 ) . The Daf-c phenotype of daf-2 mutants is suppressed by loss of daf-16 FOXO transcription factor function ( Riddle et al . , 1981; Gottlieb and Ruvkun , 1994 ) . To determine whether CMK-1 reports food information to either , or both , the TGF-β and insulin pathways , we examined whether daf-3 , daf-5 , or daf-16 mutations suppress the dauer formation phenotype of cmk-1 mutants . We found that mutations in daf-3 and daf-5 partly suppressed the dauer formation defects of cmk-1 mutants , whereas loss of daf-16 function fully suppressed this phenotype ( Figure 2A; Figure 2—source data 1 ) . These results suggest that CMK-1 influences both the TGF-β and insulin pathways to regulate dauer formation . 10 . 7554/eLife . 10110 . 008Figure 2 . CMK-1 acts cell-autonomously to regulate daf-7 TGF-β expression in ASI . ( A ) Dauers formed by the indicated strains on 80 μg live OP50 and 6 μM ascr#3 . Alleles used were: cmk-1 ( oy21 ) , daf-3 ( mgDf90 ) , daf-5 ( e1385 ) , and daf-16 ( mgDf50 ) . Shown are the averages of ≥3 independent experiments with >65 animals each . Errors are SEM . ( B ) Representative images of daf-7p::gfp expression in L1 larvae of wild-type or cmk-1 ( oy21 ) animals under the indicated conditions . Schematic at top indicates the position of ASI cell body ( lateral view ) ; boxed region is shown in panels below . Occasional weak expression is observed in the ADL neurons . Animals were grown with plentiful live OP50 or starved for at least 6 hr in the absence or presence of 1 unit of crude pheromone ( see ‘Materials and methods’ ) . White arrowheads indicate cell bodies of ASI . Numbers in bottom left hand corners indicate the percentage of examined larvae that exhibit the shown phenotype; n > 50 each; three independent experiments . Lateral view; scale bar: 10 μm . ( C ) Scatter plot of fluorescence intensity of daf-7p::gfp expression in ASI in wild-type or cmk-1 ( oy21 ) mutants . Median is indicated by a red horizontal line . Animals were grown on ample live OP50 in the absence of exogenous pheromone . Each dot is the fluorescence intensity of a single neuron in a given experiment; n > 60 neurons total each , at least three independent experiments . Promoters driving wild-type cmk-1 cDNA were: ASI—srg-47p; AWC—ceh-36Δp . ( D ) Dauers formed by shown strains on 80 μg live OP50 and 6 μM ascr#3 . The srg-47 promoter was used to drive expression of wild-type daf-7 cDNA in ASI . Shown are the averages of ≥3 independent experiments with >65 animals each . Errors are SEM . Unless indicated otherwise , * and *** indicate different from wild-type at p < 0 . 05 and p < 0 . 001 , respectively , # , ## , and ### indicate different from cmk-1 at p < 0 . 05 , 0 . 01 , and 0 . 001 , respectively . ( ANOVA and Games-Howell post-hoc test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10110 . 00810 . 7554/eLife . 10110 . 009Figure 2—source data 1 . Dauer assay data for individual trials in Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 10110 . 009 Previous work has shown that food and pheromone cues regulate the expression of both daf-7 TGF-β and the daf-28 ILP genes to modulate entry into the dauer stage . daf-7 expression in ASI , and daf-28 expression in both ASJ and ASI , are downregulated upon starvation or upon exposure to high pheromone concentrations ( Ren et al . , 1996; Schackwitz et al . , 1996; Li et al . , 2003; Entchev et al . , 2015 ) . Since CMK-1 acts in both the TGF-β and insulin pathways to regulate dauer formation , we asked whether the inability of food to fully suppress dauer formation in cmk-1 mutants is in part due to defects in regulation of expression of one or both ligands in cmk-1 mutants . Since the cmk-1 mutant dauer phenotype is suppressed by downstream mutations in the TGF-β pathway ( Figure 2A ) , we first asked whether CMK-1 acts cell-autonomously in ASI to regulate expression of the daf-7 TGF-β ligand . As reported previously , a daf-7p::gfp reporter gene was expressed strongly and specifically in the ASI neurons of fed wild-type L1 larvae grown under the conditions of low endogenous pheromone concentrations ( Ren et al . , 1996; Schackwitz et al . , 1996; Entchev et al . , 2015 ) ( Figure 2B , C ) . This reporter has been validated to reflect expression of the endogenous daf-7 gene ( Ren et al . , 1996 ) ; the expression level of this gene is one of the components encoding food levels and is variable at all examined food concentrations ( Meisel et al . , 2014; Entchev et al . , 2015 ) . Expression of daf-7p::gfp was strongly decreased upon starvation , or in the presence of crude pheromone ( Figure 2B ) . We found that unlike in wild-type animals , expression of daf-7p::gfp was reduced , but not abolished , in cmk-1 mutant larvae grown on live bacteria and no exogenous pheromone ( Figure 2B , C ) . The strongly reduced expression under fed conditions in cmk-1 larvae precluded us from determining whether starvation or addition of pheromone resulted in a further decrease in daf-7p::gfp expression levels in this mutant background ( Figure 2B ) . Expression of cmk-1 in ASI , but not in AWC , rescued the daf-7p::gfp expression defects of cmk-1 mutant larvae ( Figure 2C ) , indicating that CMK-1 acts cell-autonomously in ASI to regulate daf-7 expression . We next asked whether CMK-1 acts in parallel to the TGF-β pathway to also regulate ILP gene expression . A daf-28p::gfp reporter is expressed strongly in both ASI and ASJ neurons in fed wild-type larvae ( Li et al . , 2003 ) ( Figure 3A , B ) . Under our conditions , addition of crude pheromone decreased expression primarily in ASI , whereas starvation resulted in decreased expression in both ASI and ASJ ( Figure 3A ) . Addition of pheromone to starved wild-type larvae did not appear to further decrease expression in either cell type ( Figure 3A ) . Expression of daf-28p::gfp was unaffected by temperature ( Figure 3—figure supplement 1 ) . Interestingly , we observed that in fed cmk-1 mutant larvae grown on live bacteria without exogenous pheromone , daf-28p::gfp expression was strongly decreased in ASJ and more weakly affected in ASI ( Figure 3A , B ) . As in wild-type larvae , addition of pheromone or starvation decreased daf-28p::gfp expression in ASI; we were unable to detect whether expression in ASJ was further reduced under these conditions in cmk-1 mutants ( Figure 3A ) . The daf-28p::gfp expression defect of cmk-1 mutants in ASJ was partly but significantly rescued upon expression of wild-type cmk-1 sequences in AWC but not in ASI/AWA or ASJ ( Figure 3B ) . These results indicate that while pheromone regulates daf-28 expression primarily in ASI , food-dependent regulation of daf-28 is observed in both ASI and ASJ . Moreover , we find that CMK-1 acts non cell-autonomously in AWC to regulate daf-28 ILP gene expression specifically in ASJ under well-fed conditions . 10 . 7554/eLife . 10110 . 010Figure 3 . CMK-1 acts non cell-autonomously in AWC to regulate daf-28 insulin-like peptide ( ILP ) gene expression in ASJ . ( A ) Representative images of daf-28p::gfp expression in L1 larvae of wild-type or cmk-1 ( oy21 ) animals under the indicated conditions . Schematic of worm head indicating positions of the ASI and ASJ sensory neuron cell bodies is shown at top; boxed region is shown in panels below . Animals were grown with plentiful live OP50 or starved for at least 6 hr in the absence or presence of 1 unit of crude pheromone ( see ‘Materials and methods’ ) . White arrowheads and arrows indicate cell bodies of ASI and ASJ , respectively . Yellow arrowheads indicate expression in an ectopic cell observed in ∼14% of wild-type and ∼50% of cmk-1 mutants under all conditions . Numbers in bottom left hand corners indicate the percentage of examined larvae that exhibit the shown expression patterns; n > 50 each; three independent experiments . Lateral view; scale bar: 10 μm . ( B ) Scatter plot of fluorescence intensity of daf-28p::gfp expression in ASJ in wild-type or cmk-1 ( oy21 ) mutants . Median is indicated by red horizontal line . Animals were grown on ample live OP50 in the absence of exogenous pheromone . Each dot is the fluorescence intensity of a single neuron in a given experiment; n > 60 neurons total each , at least three independent experiments . For transgenic strains , a representative line was selected from experiments shown in Figure 1E and crossed into the reporter strains . Promoters driving wild-type cmk-1 cDNA were: ASI/AWA—gpa-4p; AWC—ceh-36Δp; ASJ—trx-1p . ( C ) Dauers formed by shown strains on 80 μg live OP50 and 6 μM ascr#3 . Promoters used to drive expression of wild-type daf-28 cDNA were: ASI—srg-47p and ASJ—trx-1p . Shown are the averages of ≥3 independent experiments with >65 animals each . ( D ) Dauers formed by the indicated strains on 80 μg OP50 at 20°C in the absence of exogenous pheromone . Alleles used were: cmk-1 ( oy21 ) and daf-28 ( tm2308 ) . Shown are the averages of ≥3 independent experiments with >65 animals each . Errors are SEM . Unless indicated otherwise , * and * indicate different from wild-type at p < 0 . 01 and 0 . 001 , respectively; # and ### indicate different from cmk-1 at p < 0 . 05 and 0 . 001 , respectively; &&& indicates different from daf-28 at p < 0 . 001 ( ANOVA and Games-Howell post-hoc test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10110 . 01010 . 7554/eLife . 10110 . 011Figure 3—source data 1 . Dauer assay data for individual trials in Figure 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 10110 . 01110 . 7554/eLife . 10110 . 012Figure 3—figure supplement 1 . daf-28 expression is not affected by cultivation temperature . ( Top ) Schematic of worm head indicating positions of the ASI and ASJ sensory neuron cell bodies . Boxed area is shown in images below . ( Below ) Representative images of daf-28p::gfp in developmentally synchronized L1 animals , grown on live OP50 and in the absence of exogenous pheromone , at the indicated temperatures . White arrowheads and arrows indicate cell bodies of ASI and ASJ , respectively . Lateral view; scale bar: 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 10110 . 012 If reduced expression of daf-7 and daf-28 in cmk-1 mutants is causal to their dauer formation phenotypes , we would predict that increased expression of daf-7 or daf-28 would rescue the dauer formation defects of cmk-1 mutants . We found that constitutive expression of daf-7 and daf-28 in ASI and ASJ , respectively , rescued the dauer formation phenotype of cmk-1 mutants ( Figures 2D , 3C; Figure 3—source data 1 ) . daf-28 expression in ASI failed to rescue ( Figure 3C ) indicating that neuron-specific release properties or spatial diffusion constraints may require DAF-28 expression in ASJ to rescue the cmk-1 phenotype ( Cornils et al . , 2011; Chen et al . , 2013 ) . Taken together , these results indicate that CMK-1 may relay food information into the dauer regulatory pathway by acting cell-autonomously to regulate daf-7 TGF-β expression in ASI , and non cell-autonomously in AWC to regulate daf-28 ILP expression in ASJ . We further tested the regulatory relationship between CMK-1 and the parallel TGF-β and DAF-28 ILP signaling pathways by performing genetic epistasis experiments . If CMK-1 acts in both the TGF-β and DAF-28 pathways , we predicted that cmk-1; daf-28 double mutants would exhibit increased dauer formation defects compared to either single mutant alone , in part due to reduced daf-7 expression in cmk-1 mutants . As shown in Figure 3D , we found that a significantly larger percentage of cmk-1; daf-28 double mutant animals entered into the dauer stage as compared to cmk-1 or daf-28 null mutants alone in the presence of live food and no added pheromone . Taken together , these results confirm that CMK-1 regulates both TGF-β and insulin signaling to modulate dauer formation . The AWC neurons have not previously been implicated in dauer formation . We further explored the role of CMK-1 in AWC in the regulation of dauer formation as a function of feeding state . We and others previously showed that CMK-1 shuttles between the nucleus and the cytoplasm in thermosensory neurons based on growth temperature ( Schild et al . , 2014; Yu et al . , 2014 ) , and regulates the expression of thermotransduction genes ( Yu et al . , 2014 ) . We asked whether CMK-1 subcellular localization in AWC is similarly affected by feeding status . In fed L1 larvae , a functional CMK-1::GFP fusion protein was present mostly uniformly throughout the soma of the AWC neurons ( Figure 4A , B ) . Food withdrawal for 30 min resulted in a transient nuclear enrichment of CMK-1 that persisted for approximately 1 hr ( Figure 4A , B ) . However , after prolonged starvation , CMK-1::GFP was enriched in the cytoplasm of AWC ( Figure 4A , B ) . In contrast , food did not affect subcellular localization of CMK-1::GFP in AFD ( Figure 4A ) , and we previously showed that temperature does not affect CMK-1 localization in AWC ( Yu et al . , 2014 ) . Overexpression of a constitutively nuclear-enriched CMK-1::GFP protein in AWC strongly rescued the dauer formation phenotype of cmk-1 mutants ( Figure 4C; Figure 4—source data 1 ) , whereas a constitutively cytoplasmically enriched protein rescued more weakly ( Figure 4C ) . The subcellular localization of these proteins in AWC was unaffected by genotype or feeding state ( Figure 4—figure supplement 1 ) . However , we note that neither the nuclear-enriched nor the cytoplasmically enriched CMK-1::GFP fusion proteins are fully excluded from the cytoplasmic or nuclear compartment , respectively , possibly due to overexpression ( Figure 4—figure supplement 1 ) . These results suggest that feeding state regulates CMK-1 subcellular localization in AWC , and that CMK-1 activity in the nuclei of AWC may be important to encode food information in the dauer regulatory pathway . 10 . 7554/eLife . 10110 . 013Figure 4 . Subcellular localization of CMK-1 in AWC is regulated by feeding state . ( A , B ) Representative images ( A ) and quantification ( B ) of subcellular localization of CMK-1::GFP in AWC neurons following removal from food for the indicated times . Representative images of CMK-1::GFP localization in AFD are also shown in A . Scale bar: 5 μm ( A ) . n > 75 AWC neurons each ( B ) . * indicates different from distribution at 0 min at p < 0 . 001 ( χ2 test ) . ( C ) Dauers formed by the indicated strains on 80 μg live OP50 and 6 μM ascr#3 . CMK-1::NLS::GFP and CMK-1::NES::GFP were expressed in AWC under the ceh-36Δ promoter . Shown are the averages of ≥3 independent experiments with >65 animals each . For transgenic strains , data are averaged from two independent lines each . Errors are SEM . and ** indicate different from wild-type at p < 0 . 05 and 0 . 001 , respectively; ### indicates different from cmk-1 ( oy21 ) at p < 0 . 001 ( ANOVA and Games-Howell post-hoc test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10110 . 01310 . 7554/eLife . 10110 . 014Figure 4—source data 1 . Dauer assay data for individual trials in Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 10110 . 01410 . 7554/eLife . 10110 . 015Figure 4—figure supplement 1 . Localization of CMK-1::NLS::GFP and CMK-1::NES::GFP in AWC . Shown is the localization of NLS::GFP or NES::GFP tagged CMK-1 protein in AWC in wild-type ( A ) and cmk-1 ( oy21 ) mutants ( B ) in the indicated conditions . Scale bar: 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 10110 . 015 We next investigated the nature of the CMK-1-regulated signal in AWC that may transmit food information to the dauer regulatory pathway . The transient nuclear localization of CMK-1 in AWC upon food withdrawal suggests that CMK-1 regulates gene expression in AWC as a function of feeding state . A link between nutrient availability and insulin signaling is now well established in many organisms ( Erion and Sehgal , 2013; Riera and Dillin , 2015 ) , and in C . elegans , the expression of several ILP genes has been shown to be regulated by food availability ( Li et al . , 2003; Cornils et al . , 2011; Ritter et al . , 2013; Chen and Baugh , 2014 ) . Together with the hours-long timescale of integration of sensory cues for the regulation of dauer formation ( Golden and Riddle , 1984c; Schaedel et al . , 2012 ) , and the fact that AWC may signal feeding state to regulate downstream hormonal signaling , we hypothesized that CMK-1 may regulate the expression of ILP genes in AWC as a function of feeding state . Of the subset of 40 predicted ILP genes ( Pierce et al . , 2001; Li and Kim , 2010 ) ( www . wormbase . org ) whose expression is reported to be regulated by nutrient availability in C . elegans , only ins-26 and ins-35 are known to be expressed in AWC ( as well as in additional cells ) ( Chen and Baugh , 2014 ) . However , whether their expression in AWC is regulated by food has not been previously determined . We asked whether ins-26 and ins-35 expression in AWC is regulated by food and CMK-1 . Under fed conditions , ins-26p::yfp was expressed in ASI , ASE , and a subset of AWC neurons in wild-type L1 larvae; expression in AWC , and to a lesser extent in ASI , was reduced upon starvation ( Figure 5A , B ) . Relative to wild-type , ins-26p::yfp expression was decreased in ASE and AWC in cmk-1 mutants grown with plentiful food , whereas expression in ASI decreased only upon starvation ( Figure 5A , B ) . Similarly , ins-35p::yfp was largely expressed in ASE and AWC in fed wild-type L1 larvae; expression in both neurons was significantly decreased upon starvation , and in fed and starved cmk-1 mutants ( Figure 5A , B ) . Weak expression of ins-35 was also observed in the ASK neurons , particularly in cmk-1 mutants , although expression in this neuron type was not further modulated by starvation ( Figure 5B ) . Expression of a constitutively nuclear-enriched CMK-1 protein in AWC neurons increased ins-35 expression in cmk-1 mutants in both fed and starved conditions , to a slightly greater extent than a constitutively cytoplasmically enriched protein ( Figure 5B ) . These observations suggest that the expression of ins-26 and ins-35 in AWC is regulated by food availability , and that their expression pattern in AWC in fed cmk-1 mutants resembles that of starved wild-type animals . Moreover , nucleocytoplasmic shuttling of CMK-1 may be important for feeding state-dependent regulation of ins-35 gene expression . 10 . 7554/eLife . 10110 . 016Figure 5 . CMK-1 maintains a balance of anti- and pro-dauer signals from AWC as a function of feeding state . ( A ) ( Top ) Schematic of worm head indicating positions of sensory neuron soma . Boxed area is shown in images below . ( Below ) Representative images of ins-26p::yfp and ins-35p::yfp expression in fed wild-type and cmk-1 ( oy21 ) mutants . White and yellow arrowheads indicate ASI and ASE , respectively; white arrow indicates AWC; yellow asterisk marks expression in the intestine . The location of ASK is indicated by a red arrowhead; fluorescence in ASK is weak and not visible at this exposure in shown images . Lateral view; scale bar: 10 μm . ( B ) Quantification of expression in each neuron type in fed and starved ( >6 hr ) conditions . Solid and hatched bars indicate strong and weak expression , respectively , in each cell . n > 50 animals each; three independent experiments . ( C ) Dauers formed by shown strains on 160 μg heat-killed OP50 and 60 nM ascr#2 ( left ) , and 80 μg live OP50 and 6 μM ascr#3 + 600 nM ascr#5 ( right ) . Alleles used were: ins-26 ( tm1983 ) , ins-35 ( ok3297 ) and ins-32 ( tm6109 ) . n > 4 assays of 65 animals each; at least three independent experiments . ( D ) Dauers formed by shown strains on 6 μM ascr#3 and 80 μg live OP50 . ins-26 and ins-35 cDNAs were expressed in ASE , ASJ , and AWC under che-1 , trx-1 , and ceh-36∆ regulatory sequences , respectively . At least two independent lines were analyzed for each transgenic strain . Shown are the averages of at least three independent experiments with >65 animals each . ( E ) Scatter plot of fluorescence intensity of daf-28p::gfp expression in ASJ in the indicated genetic backgrounds . ins-26 and ins-35::SL2::mCherry cDNAs were expressed in AWC under the odr-1 promoter . Only animals expressing mCherry in AWC were scored . Median is indicated by red horizontal line . Each dot is the fluorescence intensity of a single neuron in a given experiment . n > 60 neurons total each , four independent experiments . ( F ) Quantification of dauer formation in AWC-ablated animals in the presence of 6 μM ascr#3 and the indicated amounts of heat-killed ( blue circles ) or live ( red circles ) OP50 . Each filled circle represents one assay; n > 65 animals per assay , five independent experiments . Line represents best fit to the data . Dashed line indicates the curve for wild-type animals from Figure 1B shown for comparison . * , , and * indicate different from values using 160 μg of heat-killed bacteria at p < 0 . 05 , 0 . 01 , and 0 . 001 , respectively . ( G ) Dauers formed by the indicated strains on 80 μg live OP50 and 6 μM ascr#3 . For each assay: n > 65 animals; four independent experiments . Errors are SEM . Except where indicated , * , , and * indicate different from wild-type at p < 0 . 05 , 0 . 01 , and 0 . 001 , respectively; # , ## , and ### indicate different from cmk-1 ( oy21 ) at p < 0 . 05 , 0 . 01 , and 0 . 001 , respectively ( ANOVA and Games-Howell post-hoc test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10110 . 01610 . 7554/eLife . 10110 . 017Figure 5—source data 1 . Dauer assay data for individual trials in Figure 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 10110 . 01710 . 7554/eLife . 10110 . 018Figure 5—source data 2 . Dauer assay data for individua reliably quantify the effects of lower concentrations of live bactl trials in Figure 5—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 10110 . 01810 . 7554/eLife . 10110 . 019Figure 5—figure supplement 1 . Scatter plot of fluorescence intensity of daf-28p::gfp expression in ASJ . Median is indicated by red horizontal line . Animals were grown on live OP50 in the absence of exogenous pheromone . Each dot is the fluorescence intensity of a single neuron in a given experiment; n > 50 neurons each , at least three independent experiments . ** indicates different between indicated values at p < 0 . 01 ( Student's t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10110 . 01910 . 7554/eLife . 10110 . 020Figure 5—figure supplement 2 . CMK-1 regulates a BLI-4-dependent pro-dauer signal from AWC . Dauers formed by the indicated strains on 6 μM ascr#3 and 80 μg live OP50 . Alleles used were: cmk-1 ( oy21 ) and bli-4 ( e937 ) . bli-4 and egl-3 sense and antisense ( SAS ) constructs were expressed in AWC under the indicated promoters in cmk-1 mutants . Numbers shown are from two transgenic lines each with the exception of odr-1p::egl-3 ( SAS ) . * and * indicate different from wild-type at p < 0 . 05 and 0 . 001 , respectively; # , ## , and ### indicate different from cmk-1 at p < 0 . 05 , 0 . 01 , and 0 . 001 , respectively ( ANOVA and Games-Howell post-hoc test ) . n > 65 animals each; ≥3 independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 10110 . 020 Reduced expression of ins-26 and ins-35 in AWC in starved wild-type or fed cmk-1 mutants suggest that these peptides are candidates for an anti-dauer or pro-growth signal from AWC . If this were the case , we would predict that loss of function of these genes would enhance dauer formation in wild-type but not in cmk-1 mutants . Indeed , we found that ins-26; ins-35 double mutants formed more dauers on heat-killed bacteria in the presence of pheromone than wild-type animals ( Figure 5C; Figure 5—source data 1 ) . Dauer formation in cmk-1 mutants on live food was not significantly enhanced upon loss of either ins-26 or ins-35 ( Figure 5D ) . Since live vs heat-killed food provide different sensory inputs that might confound comparisons between wild-type and cmk-1 mutants , we established conditions under which we could force a small fraction of wild-type larvae to enter into the dauer stage on live food ( see ‘Materials and methods’ ) . ins-26; ins-35 double mutants also exhibited enhanced dauer formation on live food ( Figure 5C ) , further suggesting that these ILPs may comprise an anti-dauer signal . To confirm that reduced expression of these ILP genes in AWC in cmk-1 mutants is partly causal to their increased dauer formation phenotype , we asked whether restoration of ILP gene expression specifically in AWC was sufficient to rescue the dauer formation phenotype of cmk-1 mutants . As shown in Figure 5D , we found that overexpression of ins-26 and ins-35 specifically in AWC , but not in ASE or ASJ , rescued the dauer formation defect of cmk-1 mutants . Overexpression of ins-26 and ins-35 in AWC in cmk-1 mutants was also sufficient to partly restore daf-28p::gfp expression in ASJ ( Figure 5E ) . Together , these results suggest that food-dependent regulation of ins-26 and ins-35 in AWC may comprise an anti-dauer signal , and that inappropriate downregulation of these genes in AWC in cmk-1 mutants may in part cause the increased dauer formation phenotype in these animals under fed conditions . If AWC were in part required to transmit information about food availability to limit dauer formation , we would predict that ablation of AWC would result in increased dauer formation in wild-type animals , and that dauer formation in these animals would be less sensitive to regulation by food . Indeed , we found that AWC-ablated animals formed more dauers overall on heat-killed food than wild-type animals , and moreover , the ability of heat-killed bacteria to modulate dauer formation was reduced ( Figure 5F ) . Since the ins-26 and ins-35 anti-dauer signals are downregulated in cmk-1 mutants , we expected that ablation of AWC would not further increase dauer formation in cmk-1 mutants . Unexpectedly , we found that ablation of AWC instead suppressed dauer formation in cmk-1 mutants grown on live food ( Figure 5G ) . One interpretation of this observation is that AWC also sends a pro-dauer signal in cmk-1 mutants . Consistent with this notion , ablation of AWC significantly rescued daf-28p::gfp expression in ASJ in cmk-1 ( oy21 ) animals ( Figure 5—figure supplement 1 ) . We considered the possibility that this pro-dauer signal could be present aberrantly only in cmk-1 mutants . Alternatively , wild-type AWC neurons could also send a pro-dauer signal , but this signal may be masked by the different food conditions ( heat-killed vs live food ) employed to examine dauer formation in wild-type and cmk-1 mutants , respectively . To address this issue , we asked whether ablation of AWC alters dauer formation of wild-type animals on live food . Indeed , we found that ablation of AWC also decreased dauer formation in wild-type animals grown on live food ( Figure 5C ) , and increased daf-28p::gfp expression in ASJ ( Figure 5—figure supplement 1 ) suggesting that AWC also sends a pro-dauer signal in wild-type animals that is revealed under specific conditions . We next investigated the identity of this pro-dauer signal . The BLI-4 and EGL-3 proprotein convertases regulate the processing of different subsets of ILP precursors ( Leinwand and Chalasani , 2013; Hung et al . , 2014 ) . Knocking down bli-4 , but not egl-3 in the AWC neurons suppressed dauer formation in cmk-1 mutants ( Figure 5—figure supplement 2; Figure 5—source data 2 ) , suggesting that a BLI-4-dependent insulin signal ( s ) from AWC promotes dauer formation in cmk-1 mutants . Of the nineteen predicted BLI-4 targets ( Leinwand and Chalasani , 2013 ) , only the ins-32 ILP gene was previously reported to be expressed in AWC in adult hermaphrodites ( Takayama et al . , 2010 ) . We found that ins-32 mutants formed fewer dauers on heat-killed food , as well as live food , and pheromone in an otherwise wild-type background ( Figure 5C ) . Loss of ins-32 also suppressed the increased dauer formation phenotype of cmk-1 mutants on live food and pheromone ( Figure 5D ) . We could not detect ins-32 expression in AWC in L1 larvae although we noted that expression of ins-32 reporter gene was weak and dynamic ( SJN and PS , unpublished ) , raising the possibility that ins-32 is expressed in AWC only transiently during the period of dauer signal integration in early postembryonic development ( Golden and Riddle , 1984c; Schaedel et al . , 2012 ) . Together , these observations suggest that different insulin peptides comprise the food-dependent anti- and pro-dauer signals from AWC , and that the balance between expression and/or release of these antagonistic signals is disrupted in cmk-1 mutants . CaMKs such as CaMKI and CaMKIV play critical roles in regulating neuronal activity-dependent processes , including activity-dependent gene expression ( Wayman et al . , 2008 , 2011; Cohen et al . , 2015 ) . In turn , changes in gene expression feed back to regulate neuronal state and properties . We asked whether AWC activity is altered upon starvation or in cmk-1 mutants . It has previously been shown that the AWC neurons in adult animals are silenced in the presence of food or food-related odorants and are activated upon food/odor withdrawal ( Chalasani et al . , 2007 ) . Consistent with this observation , the AWC neurons of well-fed wild-type L2 larvae expressing the GCaMP calcium indicator ( Nakai et al . , 2001 ) exhibited few , if any somatic calcium transients ( Figure 6A , B ) . However , we found that wild-type AWC neurons became hyperactive following prolonged food deprivation . Both the frequency and amplitude of spontaneous calcium transients increased upon starvation for 2 hr , although no effects were observed after 1 hr ( Figure 6A , B ) . Interestingly , the AWC neurons in well-fed cmk-1 mutant larvae exhibited a high frequency and amplitude of somatic calcium transients similar to the activity pattern observed in starved wild-type larvae ( Figure 6A , B ) ; activity in AWC in cmk-1 larvae was not further increased by starvation ( Figure 6A , B ) . Overexpression of wild-type cmk-1 sequences in AWC suppressed basal GCaMP expression ( data not shown ) , implying that AWC may be silenced under these conditions , and precluded examination of cell-specific effects of CMK-1 on AWC activity . 10 . 7554/eLife . 10110 . 021Figure 6 . The AWC neurons exhibit increased basal activity in fed cmk-1 , and starved wild-type animals . ( A ) Heat maps showing the fluorescence intensity ( ∆F/F0 ) in the soma of AWC neurons in fed or starved wild-type and cmk-1 ( oy21 ) L2 larvae expressing GCaMP 3 . 0 in AWC under the ceh-36∆ promoter . Animals were cultured at 20°C , starved for 0 , 1 , or 2 hr and imaged at 20°C . Each horizontal line shows calcium dynamics in a single AWC neuron; n = 20 neurons each . ( B ) Box-and-whisker plots quantifying total duration of calcium responses >5% above baseline for each genotype and condition shown in A . Median is indicated by a red line . Tops and bottoms of boxes indicate the 75th and 25th percentiles , respectively; whiskers represent fifth and 95th percentiles . Outliers are indicated by + signs . indicates different from wild-type at 0 hr at p < 0 . 01 ( Kruskal–Wallis test ) . n . s . —not significant . ( C ) Average calcium responses in AWC neurons of fed and starved animals expressing GCaMP 3 . 0 under the ceh-36∆ promoter in the presence , or upon removal of the odorant isoamyl alcohol ( IAA ) , diluted to 10−3 ( blue ) or 10−4 ( red ) . Error bars are SEM and are represented by light gray shading . n ≥ 10 for each genotype and condition shown . ( D ) Scatter plot of the peak fluorescence amplitudes of individual neuron responses following odorant removal for the indicated conditions and genotypes . Horizontal black bars represent the median . * represents different between the indicated values at p < 0 . 05 ( Student's t-test ) . ( E ) Dauer formation in the presence of 80 μg live OP50 and 6 μM ascr#3 by non-transgenic and transgenic animals expressing the Drosophila histidine-gated chloride channel 1 ( HisCl1 ) in the presence ( black ) or absence ( gray ) of 10 mM histidine . Error bars are SEM . n . s . —not significant; * and *** indicate different between indicated values at p < 0 . 05 and 0 . 001 , respectively ( Student's t-test ) . ( F ) Dauer formation in the presence of 80 μg live OP50 and 6 μM ascr#3 by non-transgenic and transgenic animals expressing the constitutively active potassium channel TWK-18 ( gf ) . Error bars are SEM . * and indicate different between indicated values at p < 0 . 05 and 0 . 01 , respectively ( Student's t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10110 . 02110 . 7554/eLife . 10110 . 022Figure 6—source data 1 . Dauer assay data for individual trials in Figure 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 10110 . 02210 . 7554/eLife . 10110 . 023Figure 6—figure supplement 1 . AWC neurons exhibit increased responses to odorant removal in fed cmk-1 and starved wild-type animals . ( Left ) Average calcium responses in AWC neurons of fed ( blue ) or starved ( red ) animals expressing GCaMP3 . 0 under the ceh-36Δ promoter in the presence , or upon removal of , the odorant benzaldehyde ( Bz ) . Error bars are SEM and are represented by light gray shading . n ≥ 10 for each genotype and condition shown . ( Right ) Scatter plot of the peak fluorescence amplitudes of individual neuron responses following odorant removal for the indicated conditions and genotypes . Horizontal black bars are the median . indicates different between indicated values at p < 0 . 01 ( Student's t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10110 . 023 Fed wild-type animals must be transiently starved for the AWC neurons to respond to odor/food removal following a brief exposure to the stimulus ( S Chalasani , personal communication ) . Confirming this observation , we found that removal of either of the attractive odorants isoamyl alcohol or benzaldehyde following a 1 min exposure failed to elicit a response in the AWC neurons of wild-type adult animals transferred directly from food , but led to responses in animals starved on a food-free plate ( Figure 6C , D , Figure 6—figure supplement 1 ) . Since the activity state of AWC in fed cmk-1 mutants resembles that of starved wild-type animals , we asked whether AWC neurons in cmk-1 mutant adults are able to respond to odor removal without prior starvation . Indeed , the AWC neurons in cmk-1 adult animals responded robustly to odorant removal regardless of their feeding state ( Figure 6C , D , Figure 6—figure supplement 1 ) . Together , these observations imply that the basal activity state of the AWC neurons in fed cmk-1 mutants is similar to the activity state of these neurons in wild-type animals starved for extended periods . Importantly , these results also indicate that AWC retains the ability to respond to food-associated odors in cmk-1 mutants . We next explored the relationship between increased AWC activity in cmk-1 mutants and their enhanced dauer phenotype . To address this issue , we inhibited activity in AWC neurons and examined the effect of this inhibition on dauer formation . Addition of exogenous histamine has been shown to silence neuronal activity in C . elegans neurons expressing the Drosophila histamine-gated chloride ( HisCl1 ) channel ( Pokala et al . , 2014 ) . Expression of an activated potassium channel ( twk-18 ( gf ) ) has also been shown to hyperpolarize C . elegans neurons ( Kunkel et al . , 2000; Kawano et al . , 2011; Zhang and Zhang , 2012 ) . We found that growth on 10 mM histamine increased dauer formation in transgenic wild-type as well as in cmk-1 mutants expressing HisCl1 in AWC but had no effect on non-transgenic animals ( Figure 6E; Figure 6—source data 1 ) . Similarly , dauer formation was enhanced in both cmk-1 ( oy21 ) , and to a weaker extent in wild-type animals expressing twk-18 ( gf ) in AWC ( Figure 6F ) . These results suggests that activity in AWC antagonizes dauer formation in both wild-type and cmk-1 mutants , and acts in parallel with CMK-1 to regulate dauer formation ( Figure 7; see ‘Discussion’ ) . 10 . 7554/eLife . 10110 . 024Figure 7 . Model for the role of AWC and CMK-1 in the regulation of the dauer decision as a function of feeding state . CMK-1 acts in AWC to drive expression of the anti-dauer ins-26 and ins-35 ILP genes . CMK-1 may also regulate expression of the ins-32 ( or other ) pro-dauer signal in AWC; alternatively , the pro-dauer signal could be present tonically regardless of environmental state . Under fed conditions , CMK-1-regulated anti-dauer signals from AWC predominate and drive expression of the daf-28 ILP gene in ASJ . CMK-1 also acts cell-autonomously to regulate expression of daf-7 TGF-β in ASI . Together , daf-7 TGF-β and daf-28 ILP signals promote reproductive development . When starved , anti-dauer signals from AWC are downregulated resulting in decreased expression of daf-28 in ASJ . daf-7 and daf-28 expression in ASI are also downregulated upon starvation . The shifted balance towards pro-dauer signals promotes dauer formation in the presence of pheromone . In cmk-1 mutants , loss of the anti-dauer signals from AWC downregulates daf-28 expression in ASJ under fed conditions . daf-7 expression in ASI is also downregulated in cmk-1 mutants . Consequently , the inappropriate predominance of pro-dauer signals in cmk-1 mutants promotes dauer formation in the presence of plentiful food and pheromone . Feeding conditions also modulate AWC basal neuronal activity; increased activity upon starvation or in cmk-1 mutants may limit dauer formation via a parallel pathway . DOI: http://dx . doi . org/10 . 7554/eLife . 10110 . 024 Environmental signals such as food availability are encoded in the level of expression of the daf-7 TGF-β , and daf-28 and other ILP genes , to regulate the binary decision between entry into the reproductive cycle or the dauer stage ( Ren et al . , 1996; Schackwitz et al . , 1996; Li et al . , 2003; Entchev et al . , 2015 ) . We have identified AWC as key neurons that translate food information into changes in daf-28 ILP gene expression to regulate dauer formation . Our results suggest a model in which the AWC neurons transmit both anti- and pro-dauer signals; the balance between these signals is determined by food inputs and feeding state to regulate a critical developmental decision in the nematode life cycle . We show that the anti-dauer signals are comprised in part of the AWC-expressed ins-26 and ins-35 ILP genes , whereas the ins-32 ILP gene may constitute a component of the pro-dauer signal ( Figure 7 ) . Expression of ins-26 and ins-35 in AWC is downregulated under starvation conditions that promote dauer formation , and loss of both genes enhances dauer formation ( Figure 7 ) . Consistent with a role for these ILPs in antagonizing dauer formation , ins-26 and ins-35 mutations were reported to enhance dauer formation in sensitized backgrounds , and ins-35 was proposed to represent a genetic ‘hub’ regulating dauer entry ( Fernandes de Abreu et al . , 2014 ) . Conversely , loss of ins-32 suppresses dauer formation , although it remains possible that ins-32 acts in cell types other than AWC to promote dauer formation . The complex balance between pro- and anti-dauer signals as a function of environmental conditions is highlighted in the effects of AWC ablation on dauer formation . Ablation of AWC enhances dauer formation on heat-killed food but suppresses dauer formation on live food . Bacterial food is a complex cue consisting of many volatile and non-volatile compounds ( Orth et al . , 2011 ) that are sensed by multiple neuron types . Live and heat-killed bacteria provide different signals which may be sensed and integrated by distinct networks in the dauer decision . Thus , integration of different food signals from diverse sensory neurons likely differentially influences dauer formation on different food types . Indeed , ASI detects food directly to regulate daf-7 expression ( Ren et al . , 1996; Schackwitz et al . , 1996; Gallagher et al . , 2013 ) , and recent work has shown that complex patterns and shapes of expression of daf-7 and the tph-1 tryptophan hydroxylase genes across multiple neuron types represents a ‘neural code’ for food abundance in the context of regulation of adult lifespan ( Entchev et al . , 2015 ) . Given the consequences of phenotypic plasticity on fitness ( Nijhout , 2003; Avery , 2014 ) , the precise tuning of anti- and pro-dauer signals in multiple neurons , including the AWC neurons , is critical to allow animals to make the optimal developmental decision in response to specific environmental conditions . We have shown that the CMK-1 CaMKI plays a critical role in regulating the balance between anti- and pro-dauer signals from AWC . Our results suggest that in cmk-1 mutants , the balance between these signals is decoupled from feeding state , resulting in deregulated dauer formation under conditions that suppress dauer formation in wild-type animals . This hypothesis is based on several lines of evidence . First , cmk-1 mutants form dauers inappropriately on food conditions that fully suppress dauer formation in wild-type animals . Second , the expression patterns of daf-7 in ASI and daf-28 in ASJ in fed cmk-1 mutants are similar to those in starved wild-type animals , indicating that the expression of these key dauer-regulatory hormones does not reflect food abundance accurately in the absence of CMK-1 . Third , expression of ins-26 and ins-35 ILP genes in AWC in fed cmk-1 mutants resembles those in starved wild-type animals , further supporting the notion that food signals and gene expression patterns are uncoupled in cmk-1 mutants ( Figure 7 ) . Overexpression of ins-26 and ins-35 from AWC , and loss of ins-32 suppress the enhanced dauer formation phenotype of cmk-1 mutants indicating that the altered balance between anti-dauer and pro-dauer signals together with decreased daf-7 expression in ASI , sensitizes the threshold for dauer formation in cmk-1 mutants . Consistent with the notion that this balance is disproportionately shifted towards increased pro-dauer signals from AWC in cmk-1 mutants , ablation of AWC suppresses increased dauer formation and partially restores daf-28p::gfp expression in this background . It is important to note , however , that the expression pattern changes in fed cmk-1 mutants are not identical to those in starved wild-type animals , indicating that additional mechanisms translate food abundance into changes in gene expression patterns . How might CMK-1 link food stimuli to dauer signals from AWC ? We report that the subcellular localization of CMK-1 in AWC is highly dynamic and feeding state-dependent . Starvation results in transient nuclear localization of CMK-1 , followed by cytoplasmic enrichment upon prolonged food deprivation . Although we do not yet know the nature of the signal that triggers the initial nuclear translocation , neuronal activity has been shown in multiple contexts to regulate subcellular localization of both mammalian and C . elegans CaMKI ( Eto et al . , 1999; Ueda et al . , 1999; Sakagami et al . , 2005; Schild et al . , 2014; Yu et al . , 2014 ) . Since the AWC neurons must be transiently starved to exhibit responses to odorant removal , one possibility is that this odor removal-induced depolarization promotes CMK-1 nuclear translocation . Consistent with a well-described role of CaMKs in mediating activity-dependent gene expression ( Flavell and Greenberg , 2008; Wayman et al . , 2011; West and Greenberg , 2011; Bengtson and Bading , 2012 ) , nuclear localized CMK-1 may then regulate the transcription of anti- and pro-dauer signals . We show that constitutive nuclear localization of CMK-1 is sufficient to suppress dauer formation in the cmk-1 background and increase ins-35 expression under both fed and starved conditions . An intriguing possibility is that CMK-1 initially upregulates anti-dauer signals such as ins-26 and ins-35 upon transient food removal . As food deprivation persists , CMK-1 moves into the cytoplasm resulting in downregulation of anti-dauer signals to promote dauer formation . CMK-1 may also upregulate pro-dauer signals from AWC . Alternatively , pro-dauer signals may be expressed and/or released tonically , and CMK-1-mediated regulation of anti-dauer signals may be sufficient to promote or inhibit dauer formation . C . elegans larvae integrate food signals over an extended period of time to allow accurate assessment of their past , current , and future environment ( Schaedel et al . , 2012; Avery , 2014 ) . We speculate that not only food abundance but also temporal variability in food availability must be encoded in the balance of anti- and pro-dauer signals . Such mechanisms would allow larvae to ignore relatively transient fluctuations in environmental cues and ensure that irreversible commitment to the dauer stage occurs only under persistently adverse conditions . Basal activity in AWC is increased upon prolonged starvation in wild-type animals , and in fed cmk-1 mutants . In wild-type animals , increased AWC activity upon prolonged starvation may be a consequence of feedback from internal state . Similar modulation of peripheral sensory neuron responses as a function of feeding state has been suggested to underlie state-dependent plasticity in sensory behaviors in other organisms ( Jyotaki et al . , 2010; Palouzier-Paulignan et al . , 2012; Sengupta , 2013; Pool and Scott , 2014 ) . In starved wild-type animals , or in the absence of CMK-1 function , this state-dependent feedback may increase AWC activity , and this increased activity in turn , may antagonize dauer formation by regulating an as yet uncharacterized pathway that acts in parallel to the CMK-1 regulated pathway in AWC . Why would animals need to continue to antagonize dauer formation even following prolonged starvation ? Pre-dauer L2d larvae integrate environmental cues from 16 hr–33 hr post hatching and make an irreversible commitment to dauer entry only after 33 hr ( Schaedel et al . , 2012 ) . Given this long timeframe of sensory signal integration , and that entry into the dauer stage is a bet-hedging strategy that maximizes fitness under uncertain environmental conditions ( DeWitt and Scheiner , 2004; Avery , 2014 ) , the availability of multiple dauer-regulatory pathways that integrate environmental cues on different timescales may be critical to correctly promote or suppress the dauer decision . A goal for the future will be to define how activity in AWC modulates dauer formation , to identify the circuit mechanisms by which AWC activity is altered as a function of starvation , and to correlate temporal changes in AWC activity with commitment to the dauer stage . Although dauer formation and other forms of polyphenism are the extreme examples of phenotypic plasticity , environmental cues experienced during defined sensitive or critical periods during development also underlie phenotypic variation in mammals ( Gluckman et al . , 2007 ) . Our results describe how C . elegans integrates and translates environmental cues into hormonal signaling to regulate the dauer decision . We expect that continued investigation of the molecular and neuronal regulation of phenotypic plasticity by sensory cues in different species will lead to insights into the general principles underlying these critical developmental decisions , as well as provide information about the mechanisms of sensory integration that direct the choice of the appropriate developmental pathway . C . elegans was maintained on nematode growth medium ( NGM ) agar plates at 20°C , with Escherichia coli OP50 as a food source . Strains were constructed using standard genetic procedures . The presence of mutations was confirmed by PCR-based amplification and/or sequencing . A complete list of strains used in this study is shown in Supplementary file 1 . Promoter sequences and cDNAs were amplified from genomic DNA or a cDNA library , respectively , from a population of mixed-stage wild-type animals . cDNA sequences were verified by sequencing . The promoters used in this study are as follows: cmk-1p ( 3 . 1 kb ) , sra-9p ( ASK , 2 . 9 kb ) , ttx-1p ( AFD , 2 . 7 kb ) , trx-1p ( ASJ , 1 . 0 kb ) , gpa-4p ( ASI/AWA , 2 . 8 kb ) , ceh-36Δp ( AWC , 0 . 6 kb ) , srg-47p ( ASI , 1 . 0 kb ) , ins-26p ( see Chen and Baugh , 2014 ) , ins-35p ( see Chen and Baugh , 2014 ) , che-1p ( ASE , 0 . 7 kb ) , odr-1p ( AWC [strong] , AWB [weak] , 2 . 4 kb ) and odr-3p ( AWC [strong] , AWA , AWB , ASH , ADF [weak] , 2 . 7 kb ) . Sense and antisense constructs for RNAi were generated by amplifying exons 2–8 ( 1 . 5 kb; bli-4 ) and 6–8 ( 568 bp; egl-3 ) from cDNAs , and cloning into vectors containing cell-specific promoter sequences . Linearized vectors containing either sense or antisense sequences were injected at 100 ng/μl each . The twk-18 ( gf ) allele ( Kunkel et al . , 2000; Kawano et al . , 2011; Zhang and Zhang , 2012 ) and the Drosophila HisCl1 channel cDNA ( Pokala et al . , 2014 ) fused via SL2 to an mCherry reporter were expressed under ceh-36Δp regulatory sequences . ceh-36Δp::HisCl1 and ceh-36Δp::twk-18 ( gf ) were injected at a concentration of 50 ng/μl . Dauer assays were performed essentially as described ( Neal et al . , 2013 ) , using the indicated food sources . Briefly , young adult worms were allowed to lay 65–85 eggs on an assay plate ( Neal et al . , 2013 ) containing either ethanol ( control ) or pheromone , and a defined amount of bacteria . Animals were grown at 25°C unless indicated otherwise . Assays using heat-killed food or live food were examined for dauer and non-dauer larvae approximately 84 hr or 66 hr , respectively , after the midpoint of the egg lay . In order to induce wild-type animals to form dauers on assay plates containing 80 μg live food , a mixture of 6 μM ascr#3 + 600 nM ascr#5 was used . For experiments involving transgenic animals expressing the Drosophila HisCl1 channel , 30 μl of 1 M histidine ( Sigma-Aldrich , St . Louis , MO ) was mixed with either pheromone or ethanol and was added to the assay plates and overlaid with molten assay agar such that the final concentration was 10 mM . At least three independent trials were conducted for each condition with two technical replicates each . Statistical analyses of dauer data were performed in SPSS ( IBM , Armonk , NY ) , as described in the figure legends . Strains expressing fluorescent reporters were growth-synchronized by hypochlorite treatment , and embryos were allowed to develop for 20–24 hr at 20°C to the end of the L1 larval stage on OP50 , in the presence or absence of crude pheromone ( Golden and Riddle , 1982; Zhang et al . , 2013 ) . Crude pheromone plates were prepared by spreading 20 μl of 1:4 crude pheromone ( ∼1 unit , defined as the amount required for forming ∼33% dauers on heat-killed OP50 at 25°C ) on the agar surface and allowing it to dry completely prior to seeding with bacteria . For starvation conditions , worms were washed from growth plates in M9 buffer and were transferred to assay plates for 4–6 hr . Prior to imaging , worms were collected by centrifugation , transferred to a 2% agarose pad on a microscope slide , and immobilized using 10 mM levamisole ( Sigma-Aldrich ) . Animals were visualized on a Zeiss Axio Imager . M2 microscope using either a 40× ( NA 1 . 3 ) or 63× ( NA 1 . 4 ) oil objective , and images were captured using a Hamamatsu Orca camera . Neurons were identified by position relative to the subset of neurons filled with DiI ( Sigma-Aldrich ) . For quantification of fluorescence intensity , images were acquired from a single focal plane . The exposure time for each fluorescent reporter was adjusted in the wild-type background to ensure that pixel intensity in the cell of interest was in the linear range . Pixel intensities for the soma ( daf-7p::gfp reporter ) or the nucleus ( daf-28p::gfp reporter ) were measured in ImageJ ( NIH ) by calculating the mean pixel intensity for the entire region of interest . All measurements within a single experiment were normalized to the median wild-type expression value ( set at 1 ) to account for variation in conditions across trials . All imaging and subsequent quantification was performed blind to the genotype . For shown representative images , z-stacks ( 0 . 5 μm per slice ) were acquired through the head of the animal , and a sub-stack containing all GFP-expressing cells was rotated and maximally projected in ImageJ . All images within a panel were collected using the same exposure times . Adjustments to levels and/or contrast for optimal viewing were applied equally to images within each panel . Images used for quantification were not processed . Imaging of spontaneous calcium dynamics in AWC was performed essentially as previously described ( Biron et al . , 2008 ) . Briefly , L2 larvae were glued to an NGM agar pad on a cover glass , bathed in M9 , and mounted under a second cover glass for imaging . The edge of the sandwiched cover glasses was sealed with a mixture of paraffin wax ( Fisher Scientific , Pittsburgh , PA ) , and Vaseline , and the sample was transferred to a slide placed on a Peltier device on the microscope stage . The elapsed time from removal of the animal from the incubator to initiation of imaging was <3 min . The temperature was maintained at 20°C via temperature-regulated feedback using LabView ( National Instruments , Austin , TX ) and measured using a T-type thermocouple . Individual animals were imaged for 90 s at a rate of 2 Hz . Images were captured using MetaMorph ( Molecular Devices , Sunnyvale , CA ) and a Hamamatsu Orca digital camera . Data were analyzed using custom scripts in MATLAB ( The Mathworks , Natick , MA ) ( Source code 1–4 ) . A neuronal response was defined as the percent change of the relative fluorescence of the neuron from its baseline fluorescence level after background subtraction . A fluorescence change of >5% in AWC was considered a response . The duration of calcium events was calculated as the sum of all events in each animal , and averaged for each genotype and condition . Odor-evoked imaging was performed as described previously ( Jang et al . , 2012; Ryan et al . , 2014 ) using custom microfluidics devices . Imaging was conducted on an Olympus BX52WI or Carl Zeiss Axio Observer A1 microscope equipped with a 40X oil objective and a Hamamatsu Orca CCD or a Zeiss Axiocam 506 mono camera . Animals were exposed to a 1 min pulse of diluted odorant in S-basal . Recording was performed at 4 Hz during the last 10 s of the pulse and 50 s following removal of the stimulus . Starved worms were transferred to an assay plate with no food before imaging . Recorded image stacks were aligned using the StackReg plugin ( Thevenaz et al . , 1998 ) for ImageJ ( rigid body option ) and were cropped to a region containing the AWC cell body . Relative changes in fluorescence , following background subtraction , were calculated using custom MATLAB scripts ( Source code 5 , 6 ) . Individual traces were normalized to their average baseline value for the five seconds prior to odorant removal .	Living organisms have the remarkable ability to adapt to changes in their external environment . For example , when conditions are favorable , the larvae of the tiny roundworm C . elegans rapidly mature into adults and reproduce . However , when faced with starvation , over-crowding or other adverse conditions , they can stop growing and enter a type of stasis called the dauer stage , which enables them to survive in harsh conditions for extended periods of time . The worms enter the dauer stage if they detect high levels of a pheromone mixture that is produced by other worms—which indicates that the local population is over-crowded . However , temperature , food availability , and other environmental cues also influence this decision . A protein called TGF-β and other proteins called insulin-like peptides are produced by a group of sensory neurons in the worm's head . These proteins usually promote the growth of the worms by increasing the production of particular steroid hormones . However , high levels of the pheromone mixture , an inadequate supply of food and other adverse conditions decrease the expression of the genes that encode these proteins , which allows the worm to enter the dauer state . It is not clear how the worm senses food , nor how this is integrated with the information provided by the pheromones to influence this decision . To address these questions , Neal et al . studied a variety of mutant worms that lacked proteins involved in different aspects of food sensing . The experiments show that worms missing a protein called CaMKI enter the dauer state even under conditions in which food is plentiful and normal worms continue to grow . CaMKI inhibits entry into the dauer stage by increasing the expression of the genes that encode TGF-β and the insulin-like peptides in sensory neurons in response to food . Neal et al . 's findings reveal how CaMKI enables information about food availability to be integrated with other environmental cues to influence whether young worms enter the dauer state . Understanding how food sensing is linked to changes in hormone levels will help us appreciate why and how the availability of food has complex effects on animal biology and behavior .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "neuroscience" ]	2015	Feeding state-dependent regulation of developmental plasticity via CaMKI and neuroendocrine signaling
EAG-like ( ELK ) voltage-gated potassium channels are abundantly expressed in the brain . These channels exhibit a behavior called voltage-dependent potentiation ( VDP ) , which appears to be a specialization to dampen the hyperexitability of neurons . VDP manifests as a potentiation of current amplitude , hyperpolarizing shift in voltage sensitivity , and slowing of deactivation in response to a depolarizing prepulse . Here we show that VDP of D . rerio ELK channels involves the structural interaction between the intracellular N-terminal eag domain and C-terminal CNBHD . Combining transition metal ion FRET , patch-clamp fluorometry , and incorporation of a fluorescent noncanonical amino acid , we show that there is a rearrangement in the eag domain-CNBHD interaction with the kinetics , voltage-dependence , and ATP-dependence of VDP . We propose that the activation of ELK channels involves a slow open-state dependent rearrangement of the direct interaction between the eag domain and CNBHD , which stabilizes the opening of the channel . Ion channels in the KCNH family ( EAG , ERG and ELK ) are voltage-gated potassium channels important for nervous system function , cardiac physiology , and cancer biology ( Warmke and Ganetzky , 1994; Ganetzky et al . , 1999; Pardo et al . , 1999; Morais-Cabral and Robertson , 2015 ) ( Figure 1—figure supplement 1 ) . ERG channels ( Kv11 ) constitute the fast delayed rectifier in cardiomyocytes and are partly responsible for the repolarization of the cardiac action potential ( Sanguinetti et al . , 1995; Trudeau et al . , 1995 ) . EAG channels ( Kv10 ) and ELK channels ( Kv12 ) are abundantly and almost exclusively expressed in the brain where they also regulate electrical excitability , though their precise physiological function is not well understood ( Shi et al . , 1998; Warmke and Ganetzky , 1994; Zou et al . , 2003; Saganich et al . , 2001; Martin et al . , 2008 ) . Genetic deletion of ELK channels was shown to cause hippocampal hyperexcitability and epilepsy in mice ( Zhang et al . , 2010 ) . In addition , EAG channels are also abundantly expressed in many forms of cancer ( Camacho , 2006; Pardo and StuhmerStühmer , 2014 ) . Like other voltage-gated potassium channels , KCNH channels are composed of four subunits around a centrally located pore , where each subunit contains six transmembrane segments and an intracellular N-terminal and C-terminal region ( Figure 1A ) . Although the KCNH channels contain a cyclic nucleotide-binding homology domain ( CNBHD ) in the C-terminal region , the channels do not bind and are not directly regulated by cyclic nucleotides , including cAMP and cGMP ( Brelidze et al . , 2009; Robertson et al . , 1996 ) . Instead , the analogous cyclic nucleotide-binding pocket of the CNBHD is occupied by an ‘intrinsic ligand’ from a short sequence at the C-terminal end of the CNBHD ( Marques-Carvalho et al . , 2012; Brelidze et al . , 2012 ) . This intrinsic ligand regulates KCNH channel function ( Marques-Carvalho et al . , 2012; Brelidze et al . , 2012; Zhao et al . , 2017 ) and explains , in part , why KCNH channels are not regulated by cyclic nucleotides . Another important structural feature of KCNH channels is the interaction between the N-terminal eag domain ( PAS domain and PAS cap ) and C-terminal CNBHD ( Gianulis et al . , 2013; Gustina and Trudeau , 2009; Haitin et al . , 2013; Whicher and MacKinnon , 2016 ) ( Figure 1A ) . This interaction has been demonstrated to be critical for the proper function of KCNH channels . Mutations in KCNH channels that impair this eag domain-CNBHD interaction lead to alterations in channel trafficking and gating , which are thought to underlie some forms of Long QT Syndrome and cancer ( Curran et al . , 1995; Gustina and Trudeau , 2009 ) . 10 . 7554/eLife . 26355 . 003Figure 1 . VDP of zELK channels . ( A ) Homology model of the structure of the zELK channel illustrating the intersubunit eag domain-CNBHD interaction ( side-view parallel to the plasma membrane ) , based on the cryo-EM structure of the rEAG1 channel ( PDB code: 5K7L ) ( Whicher and MacKinnon , 2016 ) . Red arrow highlights the direct eag domain-CNBHD interaction . ( B ) Representative current-voltage ( I–V ) recordings of zELK channels in the cell-attached configuration using the voltage protocol on the left . The red trace is the double-exponential fitting of the current elicited by a +120 mV voltage pulse ( τ1 = 4 ms and τ2 = 206 ms ) . ( C ) Representative conductance-voltage ( G–V ) curves of zELK channels in the cell-attached configuration without and with a +60 mV prepulse . The dashed curve is the same data as the black solid curve but normalized to the amplitude of the red curve . ( Right ) Summary of the V1/2 of the G-V curves from multiple patches ( n = 4–14 ) . ( D ) zELK currents ( right ) elicited by a voltage protocol with increasing durations of +60 mV pulse ( left ) . The deactivation time constants for the red traces are 8 . 5 and 22 . 9 ms respectively ) . ( E ) A 6-state kinetic model for the VDP of KCNH channels . ( F–H ) Simulated data based on the 6-state model for zELK channels using QuB software ( State University of New York at Buffalo ) using the same protocols as panels B–D respectively . Voltage-dependent rate constants are given by k ( V ) = k0 exp ( k1V ) , where V is voltage , k0 is the rate at 0 mV , and k1 the voltage dependence of the rate . For the forward rate constant α of the voltage-dependent transition: k0 = 80 s−1 and k1 = 0 . 025 mV−1; for the reverse rate constant β of the voltage-dependent transition: k0 = 600 s−1 and k1 = −0 . 025 mV−1 . For the rate constant of the VDP transition step: ε = 35 s−1 . For the other transitions illustrated: γ = 60 s−1 , d = 200 s−1 , ο = 5000 s−1 , κ = 70 s−1 . DOI: http://dx . doi . org/10 . 7554/eLife . 26355 . 00310 . 7554/eLife . 26355 . 004Figure 1—figure supplement 1 . Dendrogram of KCNH channel family . Phylogenetic tree illustrating the evolutionary relationships of EAG , ERG and ELK channels in the KCNH channel family . DOI: http://dx . doi . org/10 . 7554/eLife . 26355 . 004 One behavior shared by ERG and ELK channels is mode shift or hysteresis ( Li et al . , 2015; Tan et al . , 2012; Goodchild et al . , 2015 ) . This electrical property is characterized by a shift in the voltage dependence of activation to more hyperpolarized voltages in response to a depolarizing prepulse . In ERG channels , this mode shift is thought to be responsible for the slowing of deactivation that contributes to the repolarization of the cardiac action potential ( Sanguinetti et al . , 1995; Trudeau et al . , 1995 ) . This phenomenon , also called prepulse facilitation , has been found in other types of ion channels including N-type and P/Q-type calcium channels and HCN channels ( Hoshi et al . , 1984; Hoshi and Smith , 1987; Bean , 1989; Männikkö et al . , 2005; Elinder et al . , 2006 ) . We refer to this behavior in KCNH channels as voltage-dependent potentiation ( VDP ) . In this paper , we studied the structural mechanism underlying the VDP in ELK channels . Using deletions , mutations , and chimeric constructs we show that VDP involves the interaction between the eag domain and CNBHD . To measure the distance between positions in the eag domain and CNBHD , we used transition metal ion FRET ( tmFRET ) ( Taraska et al . , 2009a ) together with incorporation of a fluorescent noncanonical amino acid ( Chatterjee et al . , 2013 ) . By simultaneously measuring channel current and tmFRET using patch-clamp fluorometry ( PCF ) ( Zheng and Zagotta , 2003 ) , we showed that the distance between the eag domain and CNBHD decreases with the time course , voltage-dependence , and ATP-dependence of VDP . These results indicate that VDP in ELK channels involves a slow rearrangement of the interaction between the eag domain and the CNBHD that is coupled to channel opening . For this study , we used a vertebrate ELK channel from zebrafish ( zELK ) which exhibits robust expression in heterologous expression systems ( Figure 1—figure supplement 1 ) . Previously , the basic electrophysiological properties of zELK channels were shown to be similar to the mammalian orthologs , and the structure of the C-linker/CNBD of the channel was solved by X-ray diffraction ( Brelidze et al . , 2012 ) . zELK channels were expressed in Xenopus oocytes and activated by depolarizing voltage steps from −120 mV to +120 mV in the cell-attached patch-clamp configuration ( Figure 1B ) . As for other KCNH channels , zELK is a K+-selective channel activated by membrane depolarization with prominent inward tail currents seen with high concentrations of potassium in the recording electrode . VDP of zELK channels manifests in three ways . The first is that the activation of zELK channels at depolarizing voltages exhibits prominent double exponential kinetics , with a fast ( ~4 ms ) and a slow ( ~200 ms ) component ( Figure 1B ) . This suggests that prolonged depolarization is causing the channel to transition to a second more stable open conformation . The second manifestation of VDP is hysteresis in the steady-state conductance-voltage ( G-V ) curve . We applied a 500 ms depolarizing prepulse to +60 mV before a family of voltage steps . The depolarizing prepulse caused about a −60 mV shift in the G-V curve to more hyperpolarized voltages as well as a dramatic increase in the peak tail-current amplitude ( Figure 1C ) . The third manifestation of VDP is apparent by varying the duration of depolarizing voltages ( +60 mV ) and monitoring the amplitude and time course of the tail current at −100 mV . The tail current amplitude increased 3 . 5 ± 0 . 3 fold ( n = 11 ) for depolarizing voltage pulses of longer durations , with a time course that closely matched the slow component of the activation kinetics ( Figure 1D ) . More interestingly , the time constant of the tail current increased from 8 . 4 ± 0 . 4 ms to 16 . 9 ± 1 . 5 ms with longer pulses ( n = 12 ) ( Figure 1D ) . This suggests that channels closed more slowly from a potentiated open state . These three manifestations of VDP can be recapitulated in a simple kinetic scheme ( Figure 1E ) . In this scheme the voltage-dependent activation of the channel was modeled as a single voltage-dependent transition followed by a voltage-independent closed-to-open transition . VDP was then modeled as a slow transition to a potentiated mode with an unaltered voltage-dependent transition but a more favorable closed-to-open transition . As required by thermodynamics , the mode shift is more favorable from the open state than from the closed states , and is therefore coupled to activation . This coupling produces a voltage-dependence to the mode shift , and therefore VDP . While clearly oversimplified , this simple gating scheme could quantitatively account for the double exponential activation ( Figure 1F ) , the shift in the G-V curve with depolarizing prepulses ( Figure 1G ) , and the slowing of the tail currents with longer depolarizing pulses ( Figure 1H ) . In the rest of the paper , we determine the molecular mechanism that underlies the mode shift that produces VDP . zELK channels exhibit a dramatic run-up in activity after patch excision . Similar to VDP , this run-up manifested as an increase in tail current amplitude ( a 2 . 2 ± 0 . 3 fold increase compared to the cell-attached configuration , measured after +120 mV depolarization , n = 7 ) ( Figure 2A ) and a significant shift of the G-V curve of channel activation to more hyperpolarized voltages ( V1/2 = −52 . 5 ± 3 . 5 mV , n = 6 ) ( Figure 2C ) . The shift of the G-V curve happened gradually after excision and reached steady-state after about 20 mins ( Figure 2B ) . Interestingly , the VDP was almost completely eliminated in excised patches , with no further shift in the G-V curve with depolarizing prepulses ( Figure 2C ) , and no slowing of the tail currents with longer depolarizing pulses ( Figure 2D , E ) . This suggests that patch excision shifted the channels into the potentiated mode even without a depolarizing prepulse . Patch excision also revealed a prominent voltage-dependent inactivation in zELK channels , particularly at very depolarized voltages ( >+60 mV ) ( Figure 2A ) . 10 . 7554/eLife . 26355 . 005Figure 2 . Run-up of zELK channels and loss of VDP after patch excision . ( A ) Representative I-V recordings of zELK channels immediately after excision ( left ) and 20 mins after excision ( middle ) in the inside-out configuration , as well as after patch cramming ( right ) using the same voltage protocol illustrated in Figure 1B . ( B ) Time course of the V1/2 change of the G-V curve of zELK channels after patch excision; patch-cramming restored the V1/2 to that before patch excision ( n = 3–6 ) . ( C ) Representative G-V curves of zELK channels in the cell-attached configuration ( black ) , in the inside-out configuration with a −100 mV prepulse ( blue ) , and with a +60 mV prepulse ( red ) ( see the legend in panel G ) . ( D ) zELK currents elicited by a voltage protocol with increasing durations of +60 mV pulse in the inside-out configuration ( same protocol illustrated in Figure 1D ) . The deactivation time constants for the red traces are 10 . 3 and 13 . 7 ms respectively ) . ( E ) Plot of the time constants of deactivation versus the duration of the +60 mV pulse for zELK channels in inside-out patches with and without ATP/Mg2+ ( n = 4 ) . The corresponding data for the cell-attached configuration is shown for comparison . ( F ) I-V recordings of zELK channels in inside-out patches with ATP/Mg2+ in the bath solution showing no run-up after patch excision . ( G ) G-V curves of zELK channels in the same conditions as panel C with the addition of 2 mM ATP/Mg2+ to the bath solution . ( H ) zELK current elicited using the same voltage protocol in Figure 1D , with 2 mM ATP/Mg2+ in the bath solution . The deactivation time constants for the red traces are 7 . 2 and 14 . 3 ms , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 26355 . 00510 . 7554/eLife . 26355 . 006Figure 2—figure supplement 1 . Run-up of zELK channels is not prevented by reducing agents or ATP without Mg2+ . ( A ) Representative I-V recordings showing 5 mM DTT was not able to prevent the run-up of zELK channels after patch excision . ( B ) Representative I-V recordings showing 1 mM ATP alone without Mg2+ failed to prevent the run-up of zELK channels after patch excision . DOI: http://dx . doi . org/10 . 7554/eLife . 26355 . 006 Patch cramming , i . e . inserting the excised patch back into the intracellular regions of the oocyte , was able to completely restore the cell-attached channel behavior within five mins , indicating some cytosolic factors were lost in the inside-out configuration ( Figure 2A and B ) . Reducing reagent DTT in the bath did not prevent the run-up , suggesting disulfide bonding was not involved in this run-up ( Figure 2—figure supplement 1A ) . Previously , it was proposed that run-up of human ELK1 channels was mediated by PI ( 4 , 5 ) P2 hydrolysis ( Li et al . , 2015 ) . We found that supplementation of the bath solution with 2 mM ATP/Mg2+ was able to prevent the run-up and maintain the VDP observed in the cell-attached configuration ( Figure 2F , G and H ) . ATP alone without Mg2+ was not able to prevent the run-up ( Figure 2—figure supplement 1B ) . These results suggest that , consistent with previous findings in human ELK1 ( Li et al . , 2015 ) , PI ( 4 , 5 ) P2 is also required for VDP in zELK , and hydrolysis of PI ( 4 , 5 ) P2 after patch excision leaves the channel in a potentiated state . To determine the role of the intracellular eag domain and CNBHD in VDP , we made mutations in these domains and tested for VDP . We found the VDP was almost completely abolished in zELK channels with the eag domain deleted ( zELK Δeag ) ( Figure 3A ) . In the cell-attached configuration; the average change in V1/2 ( ΔV1/2 ) with a +60 mV prepulse was −12 . 7 ± 1 . 4 mV ( n = 7 ) ( Figure 3B ) compared to −54 . 0 ± 2 . 0 mV ( n = 14 ) in wild-type zELK . In addition , zELK Δeag did not show an increase in the time constant of deactivation with longer depolarizing pulses ( Figure 3C ) as seen in the wild-type channel . 10 . 7554/eLife . 26355 . 007Figure 3 . Structural perturbations of the eag domain and CNBHD impair VDP . ( A ) I-V recordings of zELK Δeag channels in the cell-attached configuration ( τ = 5 . 9 ms for the faster deactivation highlighted in red ) . ( B ) Representative G-V curves for zELK Δeag channels without or with a +60 mV prepulse . The dashed trace illustrates the G-V curve of the wild-type channel after a +60 mV prepulse . ( C ) Plot of the time constant for deactivation vs . the duration of the +60 mV pulse for zELK Δeag channels . ( D ) Ribbon structure of the eag domain/CNBHD complex of mEAG1 channels ( PDB code: 4LLO ) ( Haitin et al . , 2013 ) , highlighting a salt bridge between the eag domain and CNBHD formed by R57 and D681 in the analogous positions of zELK . ( E ) and ( F ) Summary of effects of salt-bridge mutations on V1/2 ( E ) and VDP measured by ΔΔG ( prepulse ) ( F ) , ( n = 4–5 ) . p<0 . 05 . ( G ) Representative I-V recordings of mEAG1 channels showing that the kinetics of activation has only one component ( τ = 7 . 7 ms for the red trace ) . The fit is applied to the exponential activation following small sigmoidal delay . ( H ) G-V curves of mEAG1 channels in the same conditions as panel B . ( I ) G-V curves of a zELK-mEAG1 chimera containing the N- and C-terminal intracellular domains from zELK and transmembrane ( S1–S6 ) domain from mEAG1 . The dashed curve is the same data as the black solid curve but normalized to the amplitude of the red curve . DOI: http://dx . doi . org/10 . 7554/eLife . 26355 . 007 We next mutated an intersubunit salt bridge between the eag domain and the CNBHD predicted based on structures of the EAG1 channel ( Figure 3D ) ( Haitin et al . , 2013; Whicher and MacKinnon , 2016 ) . In zELK , the homologous positions of the salt-bridging residues are R57 in the eag domain and D681 in the CNBHD . We found that charge reversal mutations ( zELK-R57D or D681R ) that would disrupt the salt bridge not only shifted the initial V1/2 , but also attenuated the VDP ( Figure 3E and F ) . When we made the charge-swapping mutations ( zELK-R57D , D681R ) by combining the individual reversal mutations , we partially rescued the channel behavior to that of the wild-type channels ( Figure 3E and F ) . These results indicate that the eag domain and CNBHD are interacting via the salt bridge and this interaction is supporting the VDP . To test the hypothesis that the intracellular eag domain-CNBHD interaction is sufficient to confer VDP to the channel , we engineered a chimeric channel with the S1 to S6 transmembrane domains of mEAG1 ( amino acids: 209–503 ) and N- ( amino acids: 1–217 ) and C- ( amino acids: 544–914 ) terminal regions from zELK . Wild-type mouse EAG1 channels do not have VDP ( Figure 3G , H ) . However , the mEAG1-zELK chimera exhibited prominent VDP ( Figure 3I ) . The average ΔV1/2 with the +60 mV prepulse was −39 . 1 ± 2 . 6 mV ( n = 9 ) for the mEAG1-zELK chimera in the cell-attached configuration . These results suggest that the eag domain-CNBHD complex of zELK is sufficient to confer VDP on mEAG1 . To determine if there is a rearrangement of the eag domain-CNBHD interaction during VDP , we used transition metal ion FRET ( tmFRET ) combined with patch-clamp fluorometry . tmFRET measures the FRET between a donor fluorophore and a nonfluorescent transition metal ion acceptor ( Latt et al . , 1972; Horrocks et al . , 1975; Richmond et al . , 2000; Sandtner et al . , 2007; Taraska et al . , 2009a , 2009b ) . The efficiency of tmFRET is steeply dependent on the distance between the donor fluorophore and the acceptor metal ion and can be directly measured from the percent quenching of the donor’s fluorescence upon addition of the metal ion . Compared to traditional FRET , tmFRET measures much shorter distances ( 10–20 Å ) and has less orientation dependence , making it ideal for measuring intramolecular distances in proteins ( Taraska et al . , 2009a ) . As the donor fluorophore for tmFRET , we used the fluorescent noncanonical amino acid L-Anap ( Figure 4A ) . L-Anap was site-specifically incorporated into zELK channels using the amber ( TAG ) stop-codon suppression strategy ( Chatterjee et al . , 2013; Kalstrup and Blunck , 2013; Aman et al . , 2016; Sakata et al . , 2016; Zagotta et al . , 2016 ) . As the tmFRET acceptor , we used Co2+ coordinated by a dihistidine pair engineered into an α helix in zELK ( Figure 4A ) . The emission spectrum of L-Anap overlaps with the absorption spectrum of Co2+-dihistidine , predicting a distance for 50% FRET efficiency ( R0 ) of 12 Å ( Figure 5—figure supplement 1 ) ( Zagotta et al . , 2016; Aman et al . , 2016 ) . 10 . 7554/eLife . 26355 . 008Figure 4 . Strategy of combining tmFRET , patch-clamp fluorometry and a fluorescent noncanonical amino acid L-Anap to study conformational changes of zELK channels . ( A ) Ribbon diagram of eag domain-CNBHD complex illustrating the strategy of using tmFRET between an noncanonical amino acid L-Anap ( structure shown on the right ) and Co2+ chelated by dihistidines to measure interdomain ( intramolecular ) distances . ( B ) Cartoon illustrating the zELK channel construct with L-Anap site , dihistidine site , and C-terminal YFP . ( C ) Representative patch-clamp fluorometry images showing the specific incorporation of L-Anap into zELK channels using the amber stop-codon suppression strategy . ( D ) and ( E ) L-Anap fluorescence in patches correlated with zELK channel current ( D ) or YFP fluorescence ( E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26355 . 008 When combined with patch-clamp fluorometry ( Zheng and Zagotta , 2003 ) , tmFRET is able to detect the distance change between protein domains with simultaneous electrophysiological measurements while controlling the membrane voltage and intracellular solution . L-Anap was incorporated into the zELK eag domain by mutating the codon for amino acid 51 in the A helix to the amber stop codon TAG ( Figure 4A , B ) . We also fused a YFP at the C-terminal end of the zELK channels as a fluorescent reporter to confirm the successful expression of the full-length channel ( Figure 4B ) . Xenopus oocytes were then injected with the zELK-E51TAG-YFP mRNA , L-Anap , and a plasmid pANAP coding for the orthogonal amber suppressor tRNA/aminoacyl-tRNA synthetase ( aaRS ) pair for L-Anap ( Chatterjee et al . , 2013; Kalstrup and Blunck , 2013; Aman et al . , 2016; Sakata et al . , 2016; Zagotta et al . , 2016 ) . Only patches from oocytes injected with all three components exhibited Anap fluorescence ( Figure 4C ) . The linear correlation between the Anap fluorescence and the YFP fluorescence and current in the patches indicates that virtually all of the Anap fluorescence was coming from L-Anap incorporated within the functional channel and not from any nonspecific background fluorescence ( Figure 4D and E ) . Indeed , the negative controls in the absence of channels or with wild-type zELK channels ( no TAG mutation ) produced negligible Anap fluorescence ( Figure 4C ) . With E51Anap located within the A helix of the eag domain , a dihistidine ( K729H , E733H ) was introduced to the C helix of the CNBHD ( Figure 4A and B ) . The tmFRET efficiency between L-Anap and Co2+ bound to the dihistidine was measured by the degree of quenching of Anap fluorescence by Co2+ . With an increasing concentration of Co2+ applied to the intracellular face of excised patches held at −100 mV , the Anap fluorescence decreased monotonically ( Figure 5A ) . The Co2+-mediated quenching was reversed by applying 10 mM EDTA to chelate the divalent cations ( Figure 5—figure supplement 2 ) . This quenching of Anap fluorescence by Co2+ is indicative of FRET between the L-Anap and the bound Co2+ . The channel without the dihistidine produced only a small amount of quenching up to 1 mM Co2+ ( Figure 5A ) and was used to correct the FRET efficiency for any quenching that was not due to Co2+ binding to the dihistidine ( see tmFRET efficiency calculation in the Materials and methods ) ( Figure 5B ) . The apparent tmFRET efficiency increased with increasing Co2+ concentration and was well described by a Langmuir isotherm with an apparent affinity of around 70 µM , and a maximal FRET efficiency of 0 . 71 ± 0 . 05 ( n = 4 ) . These results suggest that Co2+ binding to the dihistidine in the CNBHD is in close proximity to L-Anap in the eag domain , as predicted from the X-ray crystal and cryoEM structures ( Haitin et al . , 2013; Whicher and MacKinnon , 2016 ) . 10 . 7554/eLife . 26355 . 009Figure 5 . Measuring the ATP/Mg2+-dependent and voltage-dependent change in the distance between the eag domain and CNBHD of zELK channels using tmFRET . ( A ) Quenching of Anap fluorescence measured using PCF by different concentrations of Co2+ with or without dihistidines and in the absence and presence of 2 mM ATP/Mg2+ at the resting holding voltage of −100 mV . ( B ) tmFRET efficiency as a function of Co2+ concentration in the absence and presence of ATP/Mg2+ as described . The smooth curves are fits of the Langmuir isotherm , Apparent FRETeff . = FRETeff [Co2+] / ( K1/2 + [Co2+] ) , with the following parameters: FRETeff = 0 . 46 , K1/2 = 48 . 0 µM with ATP ( red ) and FRETeff = 0 . 71 , K1/2 = 66 . 1 µM without ATP ( green ) . For the control construct zELK-E51Anap without the dihistidine , the quenching data with and without ATP/Mg2+ were merged since ATP/Mg2+ did not produce any significant difference . ( C ) Inside-out patch-clamp recordings of zELK E51Anap , K729H-E733H channels exhibiting a ATP/Mg2+-dependent slow component of activation typical of VDP in wild-type zELK channels . ( D ) Representative G-V curves of zELK-E51Anap , K729H-E733H channels exhibiting prepulse-dependent shift in the voltage-dependence of activation typical of VDP in wild-type zELK channels ( different patch from panel C ) . ( E ) Representative PCF images showing Anap fluorescence decreased when the membrane voltage was stepped from −100 mV to +120 mV for zELK-E51Anap , K729H-E733H channels in the presence of 1 mM Co2+ and ATP/Mg2+ in the bath . ( F ) Summary data showing the Anap fluorescence decrease by depolarization was abolished without the dihistidines or when the VDP disappeared in the absence of ATP/Mg2+ ( n = 5 ) , p<0 . 05 . The fluorescence was measured using a bandpass filter for Anap emission . ( G ) Spectral images of L-Anap emission at −100 mV and +120 mV . ( H ) Emission spectra from the spectral images shown in panel G . DOI: http://dx . doi . org/10 . 7554/eLife . 26355 . 00910 . 7554/eLife . 26355 . 010Figure 5—figure supplement 1 . tmFRET between L-Anap and transition metal ions . ( A ) Spectra of L-Anap excitation and emission and absorbance ( measured as extinction coefficient ) of cobalt and copper coordinated by a dihistidine pair . ( B ) Förster distance of the L-Anap/Co2+-dihistidine FRET pair ( labeled with the dashed line ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26355 . 01010 . 7554/eLife . 26355 . 011Figure 5—figure supplement 2 . Reverse of Co2+ quenching by EDTA . ( A ) Representative PCF images of Anap fluorescence from zELK-E51Anap , K729H-E733H channels , for the control , 1 mM Co2+ and the subsequent application of 10 mM EDTA . ( B ) Quantification of the Anap fluorescence of the conditions in panel A for zELK-E51Anap , K729H-E733H channels . Background fluorescence was subtracted . DOI: http://dx . doi . org/10 . 7554/eLife . 26355 . 011 We showed above that zELK channels in patches excised in the standard saline solution are potentiated and lose VDP , while channels excised in the presence of ATP/Mg2+ maintain VDP ( Figure 2 ) . zELK-E51Anap , K729H-E733H channels behaved similarly to wild-type zELK channels and exhibited VDP in ATP/Mg2+ ( Figure 5C ) . In the presence of ATP/Mg2+ , zELK-E51 , K729H-E733H channels exhibited a shift in the G-V curve ( ΔV1/2 ) of −43 . 5 ± 2 . 7 mV with depolarizing prepulses , only slightly less than the wild-type channel in the cell-attached configuration ( Figure 5D ) . We next measured the tmFRET efficiency between sites in the eag domain and CNBHD in the absence and presence of ATP/Mg2+ ( Figure 5A and B ) . FRET efficiency was determined at a saturating concentration of 1 mM Co2+ where the dihistidine sites are expected to be completely bound by Co2+ ( see tmFRET efficiency calculation in the Materials and methods ) . We found that , with ATP/Mg2+ added to the bath , the FRET efficiency at −100 mV decreased to 0 . 46 ± 0 . 08 ( n = 4 ) compared to 0 . 71 ± 0 . 05 without 2 mM ATP/Mg2+ added ( Figure 5B ) . Using the Förster equation and an R0 of 12 Å , this corresponds to a distance change of 2 . 0 Å . These results suggest that , at hyperpolarizing voltages , these two sites within the eag domain and CNBHD are further apart when the channel is not potentiated and closer together when the channel is potentiated . We next measured the voltage-dependence of the conformational change between the eag domain and the CNBHD . With patch-clamp fluorometry , we found the steady-state Anap fluorescence intensity with 1 mM Co2+ decreased at +120 mV compared to −100 mV ( Figure 5E and F ) . This decrease was not present without dihistidine or in the absence of ATP/Mg2+ ( Figure 5F ) . L-Anap is an environmentally sensitive fluorophore whose emission spectrum shifts to shorter wavelengths in more hydrophobic environments ( Chatterjee et al . , 2013 ) . To determine if the decreased fluorescence was associated with a change of the environment of L-Anap , we measured the emission spectra of Anap fluorescence in patches at −100 mV and +120 mV . We found the wavelength of the L-Anap peak emission was not significantly shifted despite the reduction in the intensity of peak emission produced by the +120 mV voltage pulse in the presence of 1 mM Co2+ ( Figure 5G and H ) . Together with the absence of a fluorescence change without a dihistidine ( Figure 5F ) , these results are consistent with a FRET mechanism for Co2+ quenching and not a change in environment ( Figure 5G and H ) . These results indicate that , similar to VDP , membrane depolarization causes a ATP/Mg2+-dependent and voltage-dependent rearrangement between the eag domain and CNBHD . This suggests that the rearrangement between the eag domain and the CNBHD is associated with VDP . To further test that the rearrangement between the eag domain and CNBHD is associated with VDP , we measured the kinetics of the change in tmFRET and compared it with kinetics of the development and the recovery of VDP . Using patch-clamp fluorometry , we simultaneously measured the time course of the development of VDP and the time course of the domain rearrangement in zELK-E51 , K729H-E733H channels . Fluorescent images were captured every 100 ms with a 50 ms exposure time . In the presence of ATP/Mg2+ and 1 mM Co2+ , the Anap fluorescence decreased with a +60 mV depolarization and recovered after repolarization to −100 mV ( Figure 6A ) . The time constant , 261 ± 56 ms , for the decrease in Anap fluorescence was not statistically different from the time constant of approximately 308 ± 26 ms for the slow component of channel activation associated with the development of VDP ( Figure 6A and B , also Figure 1D ) . In the absence of ATP/Mg2+ , the channel activated quickly with the +60 mV depolarization without a slow component , and the concurrent Anap fluorescence was unchanged by the voltage step ( Figure 6A ) . 10 . 7554/eLife . 26355 . 012Figure 6 . Voltage-dependent rearrangement of the eag domain-CNBHD interaction for zELK channels . ( A ) Kinetic measurement of Anap fluorescence during a +60 mV depolarization pulse with simultaneous current recordings , in the presence or absence of ATP/Mg2+ . ( B ) and ( C ) Comparison of the kinetics of the development ( B ) and the recovery ( C ) of tmFRET and VDP ( n = 4–6 ) . ( D ) Cartoon illustrating that the VDP of zELK channels involves a rearrangement of the direct interaction between the eag domain and CNBHD . The intersubunit interaction of the eag domain and CNBHD of diagonal subunits is illustrated at hyperpolarizing and depolarizing voltages showing a rearrangement of this interaction accompanies the VDP . Other changes such as pore opening and movement of the S4 are also important for channel activation . The yellow box indicates the intrinsic ligand . PI ( 4 , 5 ) P2 in the inner leaflet of the plasma membrane is shown to highlight its potential role in regulating VDP . DOI: http://dx . doi . org/10 . 7554/eLife . 26355 . 01210 . 7554/eLife . 26355 . 013Figure 6—figure supplement 1 . Measuring the recovery of VDP . ( A ) Recovery of VDP measured by a voltage protocol with hyperpolarizing recovery pulses of various durations . Representative control current traces without the +60 mV prepulse are shown , and representative current traces with +60 mV prepulse and subsequent 0 , 100 and 300 ms recovery pulses are shown . ( B ) Relationship between the duration of the recovery pulse and V1/2 of zELK channel activation after a +60 mV prepulse ( n = 6 ) . Theτ of the exponential fit in red is 136 ms . ( C ) Relationship between the duration of the recovery pulse and the peak tail-current amplitudes of zELK channels after a +60 mV prepulse ( n = 6 ) . Theτ of the exponential fit in red is 293 ms . DOI: http://dx . doi . org/10 . 7554/eLife . 26355 . 013 The time course of the recovery of tmFRET also closely matched the time course of the recovery of VDP . The time constant for recovery of the Anap fluorescence at −100 mV was 137 . 5 ± 16 ms ( Figure 6A and C ) . To measure the recovery rate of VDP , we used a new voltage protocol applying a −100 mV recovery pulse of variable durations after a +60 mV prepulse ( Figure 6—figure supplement 1A ) . With an increased duration of the −100 mV pulse , the VDP gradually disappeared; a 500 ms recovery pulse shifted the V1/2 back to the control value without the +60 mV prepulse ( Figure 6—figure supplement 1 ) . The recovery of the peak tail-current amplitude happened in a similar but somewhat slower time frame compared to the V1/2 recovery ( Figure 6—figure supplement 1B and C ) . The time constant of the V1/2 recovery of wild-type channels was not significantly different from the time constant of the recovery of Anap fluorescence ( p>0 . 05 , Figure 6C ) . Overall these tmFRET experiments demonstrate that there is a rearrangement between the eag domain and CNBHD that exhibits the same ATP/Mg2+-dependence , voltage-dependence , and kinetics as VDP . Combined with our finding that VDP is altered or eliminated by mutations of the eag domain and CNBHD ( Figure 3 ) , these experiments suggest that the VDP is produced partially or fully by a slow open-state dependent rearrangement of the direct interaction between the eag domain and CNBHD , which stabilizes the opening of the channel . In this paper , we show zELK channels exhibit VDP that results from the channel undergoing a slow state-dependent transition to a mode with a more favorable opening transition . We then show that the VDP transition involves an interaction between the intracellular N-terminal eag domain and C-terminal CNBHD . Combining transition metal ion FRET , patch-clamp fluorometry , and incorporation of a fluorescent noncanonical amino acid , we show a rearrangement between the eag domain and CNBHD that exhibits the same ATP/Mg2+-dependence , voltage-dependence , and kinetics as VDP . We proposed that this rearrangement of the eag domain-CNBHD interaction is coupled to channel opening and underlies VDP in these channels ( Figure 6D ) . VDP of mammalian ELK and ERG channels appears to be an adaptation to dampen the hyperexitability of neurons and cardiac tissue . Previously , it has been shown that hELK1 is downregulated by PI ( 4 , 5 ) P2 ( Li et al . , 2015 ) . Similarly , we found that excision of the patch causes a run-up of the current that is prevented by the presence of ATP/Mg2+ . This downregulation by PI ( 4 , 5 ) P2 is unusual in ion channels which are generally upregulated by PI ( 4 , 5 ) P2 ( Hille et al . , 2015 ) . Furthermore , for both hELK1 ( Li et al . , 2015 ) and zELK , PI ( 4 , 5 ) P2 degradation leaves the channels in a potentiated mode that no longer undergoes VDP . These results suggest that VDP could result from a simple mechanism where PI ( 4 , 5 ) P2 binds with higher affinity to the closed state of the channel than the open state , and unbinds slowly upon depolarization . This mechanism could account for our ATP/Mg2+-dependence and voltage-dependence of VDP . It would suggest that PI ( 4 , 5 ) P2 regulation is linked to a rearrangement of the eag domain-CNBHD interaction . However , hERG channels are also thought to undergo VDP ( Tan et al . , 2012; Goodchild et al . , 2015 ) but are not appreciably regulated by PI ( 4 , 5 ) P2 ( Kruse and Hille , 2013 ) . The precise role of PI ( 4 , 5 ) P2 in VDP has yet to be fully understood . The interaction between the eag domain and CNBHD has been well studied in ERG and EAG channels . Direct interaction between the eag domain and CNBHD in hERG channels has been demonstrated by multiple approaches including FRET ( Gianulis et al . , 2013; Gustina and Trudeau , 2009 , 2013 ) . In hERG channels , the eag domain-CNBHD interaction is necessary to maintain the slow deactivation and normal inactivation of the channel ( Gianulis et al . , 2013; Gustina and Trudeau , 2009 , 2013 ) . Furthermore , the eag domain-CNBHD interface of ERG channels is altered in some forms of long QT syndrome and schizophrenia ( Chen et al . , 1999; Huffaker et al . , 2009 ) . In EAG channels , direct interaction between the EAG domain and CNBHD was demonstrated using X-ray crystallography and cryo-EM ( Haitin et al . , 2013; Whicher and MacKinnon , 2016 ) . Indeed , the eag domain-CNBHD complex has been shown to adopt two closely related but different conformations in X-ray crystallography ( Haitin et al . , 2013 ) . These two conformations of the complex predict a small change in the distances between the eag domain and CNBHD . Breaking the critical salt bridge between eag domain and CNBHD significantly altered activation gating in EAG channels . Moreover , mutations at this interface of EAG channels have being associated with cancer ( Haitin et al . , 2013 ) . For both ERG and EAG channels , the interaction has been shown to be intersubunit rather than intrasubunit ( Gianulis et al . , 2013; Whicher and MacKinnon , 2016 ) . It appears that intersubunit eag domain-CNBHD interactions are a general self-regulatory mechanism among all three subfamilies of KCNH channels , EAG , ERG , and ELK; though the forms of regulation are different for the individual channels . Our experiments with zELK suggest that the intracellular domains are responsible for the VDP . We hypothesize that , for KCNH channels , a rearrangement of the eag domain-CNBHD interaction is necessary for VDP . Moreover , we suggest that there is a rearrangement of the eag domain-CNBHD interaction coupled to the opening of the channel . Since the rearrangement is coupled to opening , it occurs preferentially at depolarized voltage and stabilizes channel opening . Indeed , thermodynamics dictates that any open state-dependent transition will stabilize channel opening . Therefore , while the molecular mechanism for VDP might be distinct for different channels , the presence of a slow state-dependent transition might be a general theme that underlies the VDP of all channels . The full length D . rerio zELK construct ( GI: 159570347 ) was synthesized ( Bio Basic , Amherst , NY ) and subcloned into a modified pcDNA3 vector that contained a C-terminal YFP , a T7 promoter and 3’ and 5’ untranslated regions of a Xenopus β-globin gene . Point mutations were made using Quickchange II XL Site-Directed Mutagenesis kit ( Agilent technologies , Santa Clara , CA ) . The chimeras and deletions were made using standard overlapping PCR followed by ligation using T4 ligase or Gibson Assembly ( New England Biolabs ) . The sequences of the DNA constructs were confirmed by fluorescence-based DNA sequencing ( Genewiz LLC , Seattle , WA ) . The RNA was synthesized in vitro using HiScribe T7 ARCA mRNA Kit ( New England Biolabs , Ipswich , MA ) or mMESSAGE mMACHINE T7 ULTRA Transcription Kit ( ThermoFisher , Waltham , MA ) from the linearized cDNA . mEAG1 was a gift from Dr . Gail Robertson ( University of Wisconsin-Madison , Madison , WI ) . Xenopus oocytes were prepared as previously described ( Varnum et al . , 1995 ) . The pANAP plasmid cDNA ( ~50 nL of 100 ng/ml ) containing the orthogonal tRNA/aminoacyl-tRNA synthetase specific to L-Anap ( Chatterjee et al . , 2013 ) was injected into the Xenopus oocyte nucleus . L-Anap ( ~50 nL of 1 mM free-acid form , AsisChem , Waltham , MA ) as well as channel mRNA were injected into the cytosolic regions of oocytes separately . 2 to 4 days after injection , currents were recorded in the cell-attached and inside-out configuration of the patch-clamp technique using an EPC-10 ( HEKA Elektronik , Germany ) or Axopatch 200B ( Axon Instruments , Union City , CA ) patch-clamp amplifier and PATCHMASTER software ( HEKA Elektronik ) . For oocyte patch-clamp recording , the standard bath and pipette saline solutions contained 130 mM KCl , 10 mM HEPES , 0 . 2 mM EDTA , pH 7 . 2 . For patch-clamp fluorometry , 0 . 5 mM niflumic acid was added to the bath solution and the perfusion solution to remove calcium-activated CI- currents . Different concentrations of Co2+ were added to the perfusion solution with EDTA eliminated . Borosilicate patch electrodes were made using a P97 micropipette puller ( Sutter Instrument , Novato , CA ) . The initial pipette resistance was 0 . 3–0 . 7 MΩ for oocyte recordings . Recordings were made at 22°C to 24°C . The channel conductance-voltage relationship ( G-V curve ) was measured from the instantaneous tail currants at −100 mV as a function of the voltage of the main pulse . It was fitted with a Boltzmann equation: I = Imin + ( Imax – Imin ) / ( 1 + exp[ ( V1/2 – V ) /Vs] ) where Imax is the maximum tail current at −100 mV , Imin is the minimum tail current after hyperpolarizing voltage steps , V is the membrane potential , V1/2 is the potential for half-maximal activation , and Vs is the slope factor . The change in Gibbs free energy of channel activation was calculated according to the following equation: ΔG = RTV1/2/Vs , where R is the gas constant , and T is temperature in kelvin . The VDP due to a prepulse was calculated using the following equation: ΔΔG ( prepulse ) = ΔG ( after prepulse ) - ΔG ( before prepulse ) . Patch-clamp fluorometry ( PCF ) was performed using a Nikon Eclipse TE2000-E microscope with a 60X water immersion objective ( N . A . =1 . 2 ) . Epifluorescent recording of L-Anap was performed with wide-field excitation using a Lambda LS Xenon Arc lamp ( Sutter Instruments ) , as well as a 376/30 nm excitation filter and 485/40 nm emission filter . YFP was excited with a 490/10 nm excitation filter and 535/30 nm emission filter . Images were collected with a 50 or 100 ms exposure using an Evolve 512 EMCCD camera ( Photometrics , Tucson , AZ ) and MetaMorph software ( Molecular Devices , Sunnyvale , CA ) . VC3-8xP series valve-controlled pressurized perfusion system ( Scientific Instruments , Farmingdale , NY ) was used to minimize electronic noise during PCF experiments . For spectral measurements , images were collected by a Cascade 512B intensified CCD camera ( Roper Scientific , Tucson , AZ ) attached to a spectrograph ( Acton research , Acton , MA ) on the output port of the microscope . Spectra were analyzed by measuring line-scans across the patch area . Spectra were background subtracted using a line-scan of the non-fluorescent region outside of patch . The tmFRET efficiency measured by the decrease in donor fluorescence upon addition of the metal acceptor can be affected by nonspecific decreases in fluorescence that do not involve FRET with the metal bound to the dihistidine motif . The FRET efficiency can be corrected for these nonspecific decreases in donor fluorescence using the fluorescence decrease for channels without the dihistidine . The precise form of the correction depends on the source for the nonspecific fluorescence decrease . If the decrease comes from a source that does not involve energy transfer , such as static quenching , bleaching , inner filter effect , or nonspecific loss of the fluorophore , then the FRET efficiency can be calculated using the following equation: ( 1 ) FRETeff=1−FHHFnoHH where FHH is the normalized fluorescence of channels with dihistidines and FnoHH is the normalized fluorescence of channels without dihistidines . In each case , the F values are the fluorescence measured at metal concentrations that saturate the binding sites normalized by the fluorescence in the absence of metal , e . g . ( 2 ) FHH=fl ( metal ) fl ( no metal ) If the nonspecific decrease in fluorescence is due to collisional quenching or FRET to a different metal ion bound to an endogenous metal binding site , then the mechanism of quenching involves an additional pathway for relaxation of the fluorophore from its excited state . Consider the following scheme for a fluorophore relaxing from the excited state . photon ↑kph nonspecific quenching←kno F* →kHHFRET to metal dihistidine where F* is the excited state of the fluorophore , kph is the rate constant for emission of a photon by the excited-state fluorophore , kHH is the rate constant for energy transfer to the metal bound to the dihistidine , and kno is the sum of the rate constants for nonspecific sources of energy transfer . FRETeff in terms of the rate constants is given by: ( 3 ) FRETeff=kHHkph+kHH Rearranging: ( 4 ) 1+kHHkph=11−FRETeff Furthermore , FnoHH in terms of the rate constants is given by: ( 5 ) FnoHH=kphkph+kno Rearranging: ( 6 ) knokph=1FnoHH−1 Finally , FHH in terms of the rate constants is given by: ( 7 ) FHH=kphkph+kHH+kno Rearranging: ( 8 ) 1FHH=1+kHHkph+knokph Substituting in Equations 4 and 6 into Equation 8 gives: ( 9 ) 1FHH=11−FRETeff+1FnoHH−1 And solving for FRETeff gives: ( 10 ) FRETeff=1− 11+1FHH−1FnoHH= FnoHH−FHHFHH∗FnoHH+FnoHH−FHH Both equations ( Equations 1 and 10 ) produce similar values of FRETeff . when the value of FnoHH is near one ( little decrease in donor fluorescence for channels without the dihistidine ) , as seen for the experiments in this paper . Since the small nonspecific fluorescence decrease in these experiments likely involved collisional quenching or FRET to a metal ion bound to an endogenous metal binding site , we used Equation 10 to calculate the tmFRET efficiency . The distance ( R ) between L-Anap and the metal ion was calculated using the Förster equation: R = R0 ( 1/ FRETeff−1 ) 1/6 , where R0 is the Förster distance for FRET between L-Anap and Co2+-dihistidine ( 12 Å ) ( Aman et al . , 2016 ) . All data were analyzed using IgoPro ( Wavemetrics , Lake Oswago , OR ) . Data parameters were expressed as mean ± SEM of n experiments . Statistical significance ( p<0 . 05 ) was determined by using Student’s t test .	In humans and other animals , electrical signals trigger the heart to beat and carry information around the brain and nervous system . Particular cells can generate these signals by regulating the flow of ions into and out of the cell via proteins called ion channels . These proteins sit in the membrane that surrounds the cell and will open or close in response to specific signals . For example , an ion channel in humans called hERG allows positively-charged potassium ions to flow out of a heart cell to help the cell return to its “resting” state after producing an electrical signal . Defects in hERG can alter the rhythm at which the heart beats , leading to a serious condition called Long QT syndrome . The human hERG channel is part of a family of related channels known as the KCNH channels . These channels are made of four protein subunits that assemble to form a pore that spans the cell membrane . When a cell is resting before producing an electrical signal , KCNH channels are generally closed . However , once an electrical signal starts , the flow of ions through other ion channels in the cell membrane changes an electrical property across the membrane known as the “voltage” . This change in voltage causes KCNH channels to open . Dai and Zagotta studied how a KCNH channel known as ELK from zebrafish responds to changes in membrane voltage . The experiments show that the manner in which ELK channels respond to the voltage is due to changes in how the subunits interact in the part of the channel that lies inside the cell . Further experiments using several new techniques reveal in much more detail how the shape of the channel alters as the voltage changes . These new techniques could also be used to observe how other KCNH channels in the heart and brain change shape in response to changes in voltage . This could lead to the design of new drugs to treat heart and neurological diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics", "neuroscience" ]	2017	Molecular mechanism of voltage-dependent potentiation of KCNH potassium channels
The field of tissue engineering entered a new era with the development of human pluripotent stem cells ( hPSCs ) , which are capable of unlimited expansion whilst retaining the potential to differentiate into all mature cell populations . However , these cells harbor significant risks , including tumor formation upon transplantation . One way to mitigate this risk is to develop expandable progenitor cell populations with restricted differentiation potential . Here , we used a cellular microarray technology to identify a defined and optimized culture condition that supports the derivation and propagation of a cell population with mesodermal properties . This cell population , referred to as intermediate mesodermal progenitor ( IMP ) cells , is capable of unlimited expansion , lacks tumor formation potential , and , upon appropriate stimulation , readily acquires properties of a sub-population of kidney cells . Interestingly , IMP cells fail to differentiate into other mesodermally-derived tissues , including blood and heart , suggesting that these cells are restricted to an intermediate mesodermal fate . Human pluripotent stem cells ( hPSCs; including human embryonic stem [hES] cells and human induced pluripotent stem [hiPS] cells ) have the potential to generate the various cell types of the adult body . With their capacity to expand indefinitely , hPSCs provide a potentially unlimited source of mature cell types that can be used for disease modeling , drug discovery , and regenerative medicine purposes . Current methods for generating these therapeutically relevant cell types follow a linear approach in which hPSCs are differentiated in small , discrete steps that mimic the sequence of events occurring during development . The initial step typically involves specification of hPSCs into one of the three embryonic germ layers—ectoderm ( EC ) , endoderm ( EN ) , or mesoderm . In the case of mesoderm , several protocols have been developed to derive mature tissues stemming from this lineage , including muscle , blood , and urogenital cells ( Kee and Reijo Pera , 2008; Ng et al . , 2008; Lian et al . , 2012; Taguchi et al . , 2014 ) . While these studies have demonstrated the potential of hPSC-derived mesodermal tissues for cell replacement therapies , these protocols result in the generation of heterogeneous cell populations , some with tumor forming potential , which limits their clinical utility . Additionally , because of the inefficiency of these established protocols , large numbers of input cells are necessary to generate cell types in the quantities necessary for clinical applications . Expansion of intermediate progenitor populations of differentiating hPSCs followed by subsequent differentiation is an alternative approach for generating highly enriched and well-defined cell populations required for cell-based therapies and disease modeling . For example , homogenous , expandable ectodermally- and endodermally-restricted progenitor populations have been generated from hPSCs ( Reubinoff et al . , 2001; Shin et al . , 2006; Chambers et al . , 2009; Cheng et al . , 2012 ) . However , similar methods to generate cell types restricted to the mesodermal lineage have yet to be developed . The cell microenvironment plays a critical role for regulating self-renewal and differentiation of many progenitor cell populations that exist within the developing and fully mature adult organism ( Moore and Lemischka , 2006; Jones and Wagers , 2008 ) . In this study , we used a multifactorial high-throughput screening technology ( Flaim et al . , 2005; Brafman et al . , 2012 ) to engineer in vitro microenvironments that allow for the homogenous expansion of a hPSC-derived mesodermally restricted progenitor population , which we refer to as mesodermal progenitors ( MPs ) . Gene expression analysis and differentiation assays indicated that these cell lack tumor forming potential and exhibit properties associated with intermediate mesoderm , an observation that led us to re-name MP cells as intermediate mesodermal progenitor ( IMP ) cells . Upon modulation of their culture conditions , IMPs readily generate cell types of the renal lineage . Interestingly , IMP cells fail to differentiate into other mesodermal lineages , such as blood and cardiac muscle . Therefore , IMP cells provide a useful tool to not only study the mechanisms that regulate human mesoderm development but also a homogenous , non-tumorigenic cell source for regenerative medicine purposes . Using a high-throughput screening platform previously developed in our laboratory referred to as arrayed cellular microenvironments ( ACMEs , [Brafman et al . , 2012] ) we sought to identify culture conditions to derive , maintain and expand a cell population with mesodermal properties from hPSCs ( including human embryonic and human induced pluripotent stem [hES and iPS] cells ) . To readily observe and detect acquisition of a mesodermal phenotype , we utilized the hES cell line H9/WA09 harboring the gene encoding green fluorescent protein ( GFP ) under the control of the Brachyury ( T ) promoter ( referred to as H9-T-GFP; [Kita-Matsuo et al . , 2009] ) . Brachyury , which is expressed early in embryonic development in the primitive streak , is transiently expressed as hPSCs exit the pluripotent state and differentiate into mesodermal ( ME ) lineages ( Rivera-Perez and Magnuson , 2005 ) . To induce mesodermal differentiation , we treated H9-T-GFP cells with the GSK3 inhibitor CHIR98014 ( CHR ) for 2 days , at which point cells uniformly expressed GFP ( Figure 1A ) and were seeded onto ACME slides printed with combinations of bioactive molecules . We performed two sequential screens to identify conditions that maintain GFP expression over a 3-day period: a first screen to identify an optimal substrate composed of extracellular matrix proteins ( ECMPs ) , and a second screen to identify growth factors ( GFs ) and small molecules ( SMs ) ( Figure 1A ) . The second screen was performed using the optimal substrate composition identified in the first screen . GFP expression for each condition was evaluated and quantified using a high content imaging system and software . 10 . 7554/eLife . 08413 . 003Figure 1 . Arrayed cellular microenvironment ( ACME ) screen identified conditions that maintain expression of the mesodermal reporter T-GFP . ( A ) Schematic of the ACME experimental design . Human ES cells carrying a green fluorescent protein ( GFP ) reporter under control of the BRY/T promoter were treated with CHIR98014 ( CHR ) . GFP positive ( T-GFP ) cells were seeded onto ACME slides printed with combinations of extracellular matrix proteins ( ECMPs ) , growth factors ( GF ) and small molecules ( SMs ) . A primary screen contained all possible combinations of ECMP Collagen I ( C1 ) , Collagen III ( C3 ) , Collagen IV ( C4 ) , Collagen V ( C5 ) , Fibronectin ( FN ) , Laminin ( LN ) , and Vitronectin ( VN ) . A second GF and SM screen contained all possible single , pairwise , and three-way combinations of Wnt3a ( WNT ) , CHIR98014 ( CHR ) , Rspondin ( RSP ) , Dkk-1 ( DKK ) , IWP-2 ( IWP ) , FGF-2 ( FGF ) , FGF-7 ( KGF ) , VEGF ( VGF ) , EGF ( EGF ) , SHH ( SHH ) , Activin ( ACT ) , Cyclopamine ( CYC ) , Dorsomorphin ( DSM ) , BMP4 ( BMP ) , SB4-31542 ( SB4 ) , and Noggin ( NOG ) . The second screen was performed on the optimal ECMP combination identified in the primary screen . 72 hr after seeding , GFP expression and DAPI staining were captured and analyzed using a high content imaging microscope . ( B ) Results of the primary ECMP screen . A heat map of average T-GFP intensity was generated showing the distribution across the data set . Representative clusters are magnified . The position of the Matrigel condition in the cluster is also indicated for reference . Rows represent different ECMP combinations . Columns 1–3 represent biological replicates for cell number ( Cell # ) or T-GFP ( GFP ) . Columns marked X¯ represent the average of the three biological replicates . ( C ) Representative images of ECMP conditions in the array format . Matrigel is shown in comparison to the hit condition C1 C3 C4 FN VN . Scalebar = 50 µm . ( D ) Results of the second GF and SM screen . A heat map of average T-GFP intensity was generated showing the distribution across the data set . Representative clusters are magnified . The position of the condition lacking GFs and SMs ( No Factor ) is also indicated for reference . Rows represent different GF and SM combinations . Columns 1–3 represent biological replicates for cell number ( Cell # ) or T-GFP ( GFP ) . Columns marked X¯ represent the average of the three biological replicates . ( E ) Representative images of GF and SM conditions in the array format . No GF or SM is shown in comparison to the hit condition CHR + FGF . Scalebar = 50 µm . Figure 1—figure supplement 1 provides a global main effects principal component analysis for all GF and SM used in this second screen . DOI: http://dx . doi . org/10 . 7554/eLife . 08413 . 00310 . 7554/eLife . 08413 . 004Figure 1—figure supplement 1 . Global main effects principal component analysis of GF and SM ACME screen demonstrates that WNT and FGF agonists exert positive effects on T-GFP expression . DOI: http://dx . doi . org/10 . 7554/eLife . 08413 . 004 In the first screen , all possible 128 combinations of 7 purified ECMPs ( Collagen 1 , 3 , 4 , 5 [C1 C3 C4 C5] , Fibronectin [FN] , Laminin [LN] , Vitronectin [VN] ) , were tested for their ability to support attachment and maintain GFP expression . Hit conditions were defined as those ECMP combinations that supported maximal cell numbers , as well as GFP expression . The distribution of total cell number and GFP signal intensity across conditions was summarized in a normalized , clustered heat map ( Figure 1B ) . Interestingly , several defined ECMP combinations increased total cell number relative to Matrigel , a commercially available extracellular matrix that is commonly used for growth of hPSCs and their derivatives . Further , several ECMP combinations maintained expression of GFP to a greater extent than Matrigel . Cells growing on one of these representative ‘hit’ conditions ( C1 C3 C4 FN VN ) is shown in Figure 1C . For the second GF and SM screen , we used one of the optimal matrix compositions ( C1 C3 C4 FN VN ) as a substrate to deposit combinations of up to three GF and SM , which are known to exert potent effects during early developmental processes . Certain factor combinations increased , while others decreased , cell number and GFP expression ( Figure 1D ) . Conditions with positive effects in this assay contained a Wnt agonist ( either Wnt3a [WNT] or CHR ) and a member of the FGF superfamily ( Figure 1D , E ) . Consistent with this observation , a global main effects principal component analysis of all GF and SM revealed that CHR , WNT , Rspondin and FGF exerted the most potent effects on GFP expression ( Figure 1—figure supplement 1 ) . To a lesser extent , the FGF family members VEGF ( VGF ) and KGF , also positively influenced GFP expression , whereas Wnt antagonists ( DKK1 and IWP2 ) negatively influenced GFP expression . We confirmed the ECMP hit conditions by scaling up the 10 top-performing matrix compositions shown in the heatmap of Figure 1B into traditional cell culture formats . Compared to Matrigel and a sub-optimal matrix ( C1 C4 C5 LN ) , 8 of the 10 ECMP hit conditions significantly increased the percentage of GFP positive cells ( Figure 2A ) . Importantly , in this scaled-up format , the optimal matrix identified in the primary screen ( C1 C3 C4 FN VN ) consistently led to higher cell numbers and GFP expression compared to the other top ECMP combinations , thus demonstrating the robustness of the ACME screening platform . 10 . 7554/eLife . 08413 . 005Figure 2 . Validation of high-throughput ACME screens . Scale up analysis of hits from the ACME screens . Human ES cells carrying a GFP reporter under control of the BRY/T promoter were treated with CHIR98014 ( CHR ) for 24 hr . After 48 hr , GFP positive ( T-GFP ) cells were cultured in multi-well plates for 72 hr to validate conditions from the ACME screens . ( A ) GFP+ cells were cultured in multi-well plates coated with 10 hit matrices from the primary ECMP screen as well as Matrigel and a sub-optimal matrix ( C1 C4 C5 LN ) . The optimal matrix ( C1 C3 C4 FN VN ) was defined as the condition that maintained the highest T-GFP expression and fostered the highest cell number . Statistical comparisons are made to the Matrigel condition . p < 0 . 05 , p < 0 . 005 . When p-values are not indicated with or ** , the statistical difference is not significant from the control . ( B ) GFP+ cells were cultured in multi-well plates coated with the optimal matrix ( C1 C3 C4 FN VN ) and various GF/SM combinations . Statistical comparisons are made to the conditions containing no GF/SM ( No Factor ) . p < 0 . 05 , p < 0 . 005 . When p-values are not indicated with or ** , the statistical difference is not significant from the control . ( C ) GFP+ cells were cultured in multi-well plates coated with the optimal matrix ( C1 C3 C4 FN VN ) and various concentrations of CHR and FGF2 ( FGF ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08413 . 005 We also plated cells in traditional cell culture format on the optimized matrix in the presence of individual soluble factors as well as the top 27 combinatorial hits from the GF-SM screen ( Figure 2B ) . This analysis confirmed that the combination of CHR and FGF yielded the highest level of GFP expression and cell number . Since bioactive molecules like CHR ( or Wnt ) and FGF often exhibit distinct effects at varying concentrations , we performed a dose response analysis to identify optimal CHR and FGF concentrations . The optimal CHR dose was 1 . 0 µM while the dose of FGF was less dynamic , with its effects saturating at 20 ng/ml FGF ( Figure 2C ) . The previous analysis was performed 3 days after plating cells in the optimized culture condition . We also examined to what extent this optimized culture condition could support long term growth and expansion of cells with mesodermal properties ( Figure 3A ) , which we preliminarily referred to as MP cells . In addition , we extended this analysis to include two additional cell lines , BJ-RiPS and HUES9 cells . When seeded at a density of 104 cells/cm2 , cells formed and grew in tight clusters ( Figure 3B ) . Cells with these morphological properties were expanded by serial passaging with approximate doubling rates of 60 . 2 ± 4 . 2 hr ( H9 = 55 . 4 hr , Figure 3C; Hues9 = 61 . 8 hr , RiPS = 63 . 4 hr , Figure 3—figure supplement 1 ) and expressed the proliferation marker Ki-67 ( Figure 3—figure supplement 2 ) . Cell counts taken at each passage revealed that 1 × 104 cells could theoretically be expanded to approximately 1 × 1012 cells over 10 passages ( Figure 3C and Figure 3—figure supplement 1 ) . These cells maintained 46 chromosomes ( Figure 3—figure supplement 3 ) , indicating that cultured cells did not acquire abnormal chromosome numbers commonly associated with late passage hPSCs . Reverse transcription quantitative PCR ( qPCR ) showed that expression of genes associated with pluripotency ( OCT4 , NANOG , SOX2 ) was rapidly lost during expansion ( Figure 3D; Figure 3—figure supplement 4 ) . This loss of pluripotency-associated properties was further confirmed by immunofluorescence ( IF ) staining ( OCT4 and NANOG , Figure 3—figure supplement 5 ) and flow cytometry ( TRA-1-81 and SSEA4 , Figure 3—figure supplement 6 ) . In contrast , genes associated with the mesoderm ( ME ) lineage ( MESP1 , MIXL1 , LHX1 ) were upregulated and maintained over 10 passages ( Figure 3E; Figure 3—figure supplement 7 ) . IF staining confirmed the presence of MIXL1 protein in these expanded cell cultures ( Figure 3F ) . Using flow cytometry , we furthermore showed that the expanded cells shared a cell surface signature of CD56+ CD326− ( Figure 3G; Figure 3—figure supplement 8 ) , previously defined for a multipotent mesoderm-committed cell population ( Evseenko et al . , 2010 ) . In addition , expression of the EN marker FOXA2 and the EC marker SOX1 was significantly reduced in these cells ( Figure 3H , I; Figure 3—figure supplements 9 , 10 ) . 10 . 7554/eLife . 08413 . 006Figure 3 . Characterization of mesodermal progenitor population . ( A ) Schematic showing derivation of mesoderm progenitor ( MP ) cells . Human ES cells were differentiated into mesoderm ( ME ) with CHIR98014 ( CHR ) and then replated onto the defined substrate C1 C3 C4 FN VN and cultured with CHR and FGF2 ( FGF ) for up to 20 passages ( p0 to p20 ) . ( B ) Representative images of MP cells derived from the hES cell line H9/WA09 at passage 1 and 10 in C1 C3 C4 FN VN with CHR and FGF . Scale bar = 50 µm . ( C ) Growth rate of MP cells derived from H9 T-GFP . Cell counts were taken at each passage . ( D ) Quantitative PCR ( qPCR ) analysis for expression of pluripotency markers OCT4 , NANOG , and SOX2 . Expression of these markers in MP cells at passages 1 , 5 and 10 is lower than in undifferentiated cells ( ES ) . Cells differentiated into ME , endoderm ( EN ) and ectoderm ( EC ) served as controls . All statistical comparisons are made to the ES sample . p < 0 . 05 , p < 0 . 005 . ( E ) qPCR analysis for expression of mesodermal markers MESP1 , MIXL1 , and LHX1 . Expression of these markers in MP cells at passages 1 , 5 and 10 is comparable to that observed in ME and higher than in ES , EN and EC . All statistical comparisons are made to the ME sample . p < 0 . 05 , *p < 0 . 005 . ( F ) MIXL1 immunofluorescence ( IF ) in MP cells . MP cells at passage 15 were fixed and stained with MIXL1-specific antibody . Number indicates percentage of MIXL1 expressing cells in the MP cell population . Standard deviation represents the variation between the fields of view used for counting ( n = 20 ) . Scale bar = 50 µm . ( G ) Flow cytometry analysis for CD56 ( NCAM ) and CD326 ( ECAM ) . Pluripotent cells ( ES , CD326+ CD56− ) are differentiated to ME cells ( CD326− CD56+ ) . MP cells at p10 exhibit a similar cell surface expression of these two markers as ME . ( H ) qPCR analysis for expression of the EN marker FOXA2 . Expression of FOXA2 is only detected in cells differentiated towards EN . All statistical comparisons are made to the ES sample . ( I ) qPCR analysis for expression of the EC marker SOX1 . Expression of SOX1 is only detected in cells differentiated towards EC . All statistical comparisons are made to the ES sample . ( J ) MP cells are non-tumorogenic . Nude mice were injected with H9-derived MP cells or H9 ES cells . Injected ES cells generated tumors while injected MP cells did not form any growth in 11/12 injections . Figure 3—figure supplement 1 through 10 provide additional analysis , including for two other hPSC lines ( BJ RiPS and HUES9 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08413 . 00610 . 7554/eLife . 08413 . 007Figure 3—figure supplement 1 . Growth rate of MP cells derived from Hues 9 or BJ RiPS . Cell counts were taken at each passage . DOI: http://dx . doi . org/10 . 7554/eLife . 08413 . 00710 . 7554/eLife . 08413 . 008Figure 3—figure supplement 2 . Flow cytometry analysis of Ki-67 in human ES , ME , and MP . MP cells were analyzed at passage 10 . DOI: http://dx . doi . org/10 . 7554/eLife . 08413 . 00810 . 7554/eLife . 08413 . 009Figure 3—figure supplement 3 . Karyotype of MP cells derived from the hES cell line H9/WA09 . Cytogenetic analysis was performed on two independent MP cell lines , one derived from H9 cells and grown to passage 10 ( p10 , top ) and one derived from H9 SOX17-GFP and grown to passage 15 ( p15 , bottom ) . For each line , twenty G-banded metaphase cells were analyzed . 19 cells of H9 MP cells demonstrated an apparently normal female karyotype while one cell demonstrated a non-clonal chromosome aberration ( 45 , XX , −20 ) , which is most likely an artifact of culture . None of the 20 cells of H9 SOX17-GFP MP cells exhibited chromosome aberrations . No abnormal cells with trisomy 12 and/or 17 were detected . DOI: http://dx . doi . org/10 . 7554/eLife . 08413 . 00910 . 7554/eLife . 08413 . 010Figure 3—figure supplement 4 . QPCR analysis for expression of pluripotency markers OCT4 , NANOG , and SOX2 . Expression of these markers in MP cells at passage 10 is lower than in undifferentiated cells ( Pluri ) . Cells differentiated into ME , EN and EC served as controls . All statistical comparisons are made to the ES sample . p < 0 . 05 , *p < 0 . 005 . DOI: http://dx . doi . org/10 . 7554/eLife . 08413 . 01010 . 7554/eLife . 08413 . 011Figure 3—figure supplement 5 . IF of Hues 9 ES and MP cells demonstrate that MP cells do not express OCT4 and NANOG proteins . Scale bar = 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 08413 . 01110 . 7554/eLife . 08413 . 012Figure 3—figure supplement 6 . Flow cytometry analysis of Hues 9 ES and MP ( p10 ) cells for Tra-1-81 and SSEA4 . MP cells do not express pluripotent cell surface markers . DOI: http://dx . doi . org/10 . 7554/eLife . 08413 . 01210 . 7554/eLife . 08413 . 013Figure 3—figure supplement 7 . QPCR analysis of MP cells derived from Hues 9 and BJ RiPS for expression of mesodermal markers MESP1 , MIXL1 , and LHX1 . Expression of these markers in MP cells at passages 1 , 5 and 10 is comparable to that observed in ME and higher than in ES , EN and EC . All statistical comparisons are made to the ME sample . p < 0 . 05 , *p < 0 . 005 . DOI: http://dx . doi . org/10 . 7554/eLife . 08413 . 01310 . 7554/eLife . 08413 . 014Figure 3—figure supplement 8 . Flow cytometry analysis for CD56 ( NCAM1 ) and CD326 ( EPCAM ) in undifferentiated RiPS cells as well as ME and MP ( p10 ) cells derived from RiPS cells . MP cells exhibit a similar cell surface expression of these two markers as ME . DOI: http://dx . doi . org/10 . 7554/eLife . 08413 . 01410 . 7554/eLife . 08413 . 015Figure 3—figure supplement 9 . QPCR analysis of MP cells derived from Hues 9 and BJ RiPS . Low expression of FOXA2 demonstrates that MP cells are not committed to the EN lineages . All statistical comparisons are made to the pluripotent ( ES or hPS ) sample . p < 0 . 05 , *p < 0 . 005 . DOI: http://dx . doi . org/10 . 7554/eLife . 08413 . 01510 . 7554/eLife . 08413 . 016Figure 3—figure supplement 10 . QPCR analysis of MP cells derived from Hues 9 and BJ RiPS . Low expression of SOX1 demonstrates that MP cells are not committed to the EC lineages . All statistical comparisons are made to the pluripotent ( ES or hPS ) sample . p < 0 . 05 , *p < 0 . 005 . DOI: http://dx . doi . org/10 . 7554/eLife . 08413 . 016 The apparent indefinite expansion of MP cells ( greater than 20 passages at the time of this submission ) raised the possibility that these cells , like undifferentiated hPSCs , harbored tumorigenic potential . Importantly , unlike hPSCs , MP cells did not produce tumors when injected into immune compromised mice ( Figure 3J ) . Among the 12 MP cell injections ( 2 injections per mouse ranging from 0 . 5 million to 1 million cells ) , only one site maintained a small lump ( ∼1 mm in diameter ) , which did not grow in size over 12 weeks . In contrast , all 6 hPSC injections ( 0 . 5 million cells per injection ) produced readily visible teratomas ( greater than 10 mm in diameter ) . Taken together , we have generated a non-tumorigenic progenitor population capable of nearly indefinite expansion potential with a mesodermal phenotype . From the ACME screens , we identified a defined matrix ( C1 C3 C4 FN VN ) and combination of soluble factors ( CHR + FGF ) that allow for the derivation and expansion of MP cells . We wanted to explore to what extent these defined conditions were critical for the derivation and expansion of MP cells . To this end , we first compared the effectiveness of our defined matrix relative to Matrigel and of CHR + FGF relative to no factors in deriving MP cells ( Figure 4A ) , as assayed by qPCR of mesodermal markers . Importantly , cells cultured in the absence of CHR and/or FGF failed to passage beyond one passage , indicating that these soluble factors are essential to the expansion of MP cells . Furthermore , although Matrigel with CHR and FGF yielded cells expressing the mesodermal markers MESP1 , MIXL1 , and LHX1 , our optimized matrix significantly increased their expression ( Figure 4B ) . By passage 3 , cells cultured in our optimized conditions expressed 1 . 5- to 2-fold greater levels of MESP1 , MIXL1 , and LHX1 compared to cells cultured on Matrigel ( Figure 4B ) . 10 . 7554/eLife . 08413 . 017Figure 4 . Optimized culture conditions are required to generate and maintain MP cells . ( A ) Human ES cells were treated with CHIR98014 ( CHR ) for 24 hr . After 48 hr , cells were cultured on either Matrigel or the optimal matrix ( C1 C3 C4 FN VN ) in the absence ( no factor ) or in the presence of the optimal GF/SM combination ( CHR + FGF ) . Only cells cultured with CHR + FGF could be serially passaged . ( B ) QPCR analysis for mesodermal markers MESP1 , MIXL1 , and LHX1 . Conditions containing no factor did not grow beyond passage 1 , while the CHIR + FGF samples represent expression at passage 3 . NF = no factor; C + F = CHR + FGF . Statistical comparisons are made to C1 C3 C4 FN VN with CHR + FGF condition . p < 0 . 05 , *p < 0 . 005 . ( C ) MP cells were expanded to p6 on the optimal ECMP ( C1 C3 C4 FN VN ) and GF/SM combination ( CHR + FGF ) . MP cells were then either transitioned to Matrigel or maintained on C1 C3 C4 FN VN in the absence or presence of CHR + FGF . ( D ) QPCR analysis for mesodermal markers MESP1 , MIXL1 , and LHX1 . Conditions containing no factor did not grow past p7 , while the CHR + FGF sample represents expression at p9 . All statistical comparisons are made to the C1 C3 C4 FN VN with CHR + FGF condition . p < 0 . 05 , **p < 0 . 005 . DOI: http://dx . doi . org/10 . 7554/eLife . 08413 . 017 Next , we compared the effectiveness of our defined matrix relative to Matrigel and of CHR + FGF relative to no factors in maintaining MP cells ( Figure 4C ) . For this analysis , MP cultures were grown in the optimized conditions ( C1 C3 C4 FN VN and CHR + FGF ) through passage 6 , at which point cultures were either passaged onto Matrigel or the defined matrix in the presence or absence of the soluble factors CHR and FGF . Again , the optimized culture condition produced a statistically significant difference in maintaining mesoderm marker expression compared to other conditions ( Figure 4D ) . Importantly , MP cultures without CHR and FGF failed to expand beyond the first passage . Taken together , these results indicate that the defined substrate C1 C3 C4 FN VN as well as CHR and FGF are required for optimal MP cell generation and maintenance . To further characterize the MP cell population derived and expanded under our defined culture conditions , we performed transcriptome analysis by RNA sequencing ( RNA-seq ) . For comparison , we analyzed the transcriptomes of undifferentiated hES cells , as well as of transient EC , EN , ME populations differentiated from hES cells . Cluster analysis of the RNA-seq data revealed that MP cells are more similar to ME cells than they are to EC , EN , and hES cells ( Figure 5A and Supplementary file 1A , B ) . Comparison of expressed genes in MP and ME cell populations confirmed a high degree of similarity , with a correlation coefficient of 0 . 9522 ( Figure 5B ) . Although this analysis revealed that MP cells are more similar to transient ME populations than they are to other cell populations examined , they are also distinct from ME cells . In contrast to ME cells , MP cells exhibit significantly lower levels of pluripotency regulators , including POU5F1 ( OCT4 ) and SOX2 . Several established early mesodermal markers ( T , MIXL ) were significantly elevated in ME cells relative to MP cells , suggesting that MP cells have progressed beyond this transient and early ME phenotype . 10 . 7554/eLife . 08413 . 018Figure 5 . Gene expression analysis reveals that MP cells have an intermediate mesodermal ( IM ) identity . RNA sequencing ( RNA-seq ) was used to analyze gene expression of MP cells . As a comparison , gene expression profiles were analyzed for hES ( ES ) cells and their differentiated progeny , ME , EN and EC . ( A ) MP cells resemble mesodermally differentiated cells . Hierarchical clustering analysis was performed for all genes with detectable expression ( reads per kilobase per million mapped reads [RPKM] values greater than 10 ) in one of the five cell populations . Supplementary file 1 provides the complete list of genes shared between MP and ME ( A ) and genes unique to MP ( B ) . The complete RNA-seq data set for MP cells is provided in Supplementary file 2 . ( B ) Correlation of gene expression profiles . Genes with expression values ( RPKM ) expression between 10 and 1500 were plotted for MP cells and ME . The correlation coefficient ( R ) for all expressed genes is 0 . 9522 . ( C ) Schematic depicting differentiation protocols from hES cells to IM and lateral plate mesoderm ( LM ) derivatives cardiomyocytes ( CMs ) and hematopoietic stem and progenitor ( HSP ) cells . ( D ) QPCR analysis of IM , CM , HSP , and MP cells revealed that MP cells have a similar expression profile as IM cells . ACT = Activin A , BMP = BMP4 , CHR = CHIR98014 , d = day , FGF = FGF2 , IWP = IWP-2 , RA = retinoic acid , VGF = VEGF . DOI: http://dx . doi . org/10 . 7554/eLife . 08413 . 018 During development as the ME germ layer matures , modulation of various signaling molecule pathways lead to its further specification into paraxial , intermediate , and lateral plate mesoderm ( PM , IM , and LM , respectively ) ( reviewed in Christ and Ordahl , 1995 ) . IM develops into cells of the urogenital system , whereas LM develops into tissues of the vascular system , including cardiomyocytes ( CMs ) and hematopoietic stem and progenitor ( HSP ) cells . Using established differentiation protocols ( Figure 5C ) , we examined expression by qPCR of several mesodermal markers in MP cells relative to IM , CM and HSP . Interestingly , MP cells most closely resembled the mesodermal gene expression profile of IM cells ( Figure 5D ) . In addition , we observed in the RNA-seq data that several IM markers ( CITED2 , EYA1 , GATA3 , LHX1 , SALL1 ) were expressed in MP cells ( Supplementary file 2 ) . Based on this gene expression analysis we speculated that MP cells are most closely related to cells of intermediate mesoderm and consequently renamed them from MP cells to intermediate mesodermal progenitor cells ( IMP ) . Based on the above findings , we hypothesized that the differentiation potential of IMP cells may be limited to cell types derived from IM , such as of the renal lineage . To test this hypothesis , we tested the ability of IMP cells to differentiate into various mesodermally derived tissues , including hematopoietic cells , CMs and renal progenitors . Using an established protocol for hematopoietic differentiation ( Figure 6A , adapted from Ng et al . ( 2008 ) ) , we successfully differentiated hES cells into cells expressing SOX17 , a marker of hemogenic endothelium , and CD34 and CD45 , two cell surface markers commonly used to monitor the presence of hematopoietic cell populations ( Figure 6B , C ) . In contrast , IMP cells derived from three independent hPSC lines and manipulated in a similar manner failed to express these markers at detectable levels ( Figure 6B , C ) , even when culture periods were extended beyond the standard protocol . 10 . 7554/eLife . 08413 . 019Figure 6 . IMP cells are unable to differentiate to cell types derived from lateral plate mesoderm ( LM ) . ( A ) Schematic of the hematopoietic differentiation protocol . Cells were differentiated in a step-wise manner using the indicated GFs and SMs from undifferentiated ES cells or from IMP cells to ME , endothelial cell ( ENC ) and subsequently to hematopoietic precursors ( HPs ) . Stage-specific marker genes and cell surface markers expressed during this differentiation process are indicated at the top . FGF = FGF2 , VGF = VEGF , SCF = Stem Cell Factor , BMP = BMP4 . ( B ) QPCR analysis of hES and MP cells differentiated towards HPs . Compared to hES cells , IMP cells do not differentiate towards HPs , as indicated by the absence of SOX17 expression . ( C ) Flow cytometry analysis of hES and IMP cells differentiated towards HPs for CD34 and CD45 . While hESC cells can differentiate into CD34+ CD35+ HPs , IMP cells fail to differentiate generate cells positive for CD34 and CD45 . ( D ) Schematic of the CM differentiation protocol . Cells were differentiated in a step-wise manner using the indicated GFs and SMs from undifferentiated ES cells or from IMP cells to ME , cardiac progenitor ( CP ) and subsequently to CM . Stage-specific marker genes expressed during this differentiation process are indicated at the top . CHR = CHIR98014 , IWP = IWP-2 . ( E ) QPCR analysis of MP cells differentiated towards CMs . Compared to hES cells , IMP cells do not differentiate towards CMs , as indicated by the absence of ISL1 and NKX2 . 5 expression . DOI: http://dx . doi . org/10 . 7554/eLife . 08413 . 019 Along similar lines , using an established protocol to derive CMs ( Figure 6D , adapted from Lian et al . ( 2012 ) ) , hES cells readily produced cardiac progenitors ( CPs ) and subsequently CMs , as monitored by expression of NKX2 . 5 and ISL1 ( Figure 6E ) . Cultures containing CMs exhibited the characteristic contractile activity associated with such cells . In contrast , IMP cells subjected to these same manipulations failed to express detectable levels of NKX2 . 5 and ISL1 ( Figure 6E ) , and never produced contractile activity . Furthermore , since CM differentiation from hES cells is enhanced by inhibition of Wnt signaling ( Willems et al . , 2011 ) , we reasoned that a prolonged withdrawal of CHR ( a potent Wnt agonist required to maintain IMP cells ) and addition of IWP ( a potent Wnt inhibitor ) may encourage IMP cells to enter the CM lineage . However , under no tested conditions were we able to promote CM differentiation from IMP cells . Taken together , IMP cells were unable to differentiate into cells with hematopoietic or cardiogenic properties , both derivatives of LM . Since the IMP cells described in this study failed to generate derivatives of LM , we reasoned that these cells may differentiate into cell populations derived from IM , such as kidney and gonads . To test this possibility , we employed a published protocol to differentiate hES cells into renal progenitors ( Figure 7A ) ( Taguchi et al . , 2014 ) . This protocol employed several GFs and SMs to promote the differentiation of hES cells to IM and subsequently metanephric mesenchyme ( MM ) . Importantly , IMP cells efficiently acquired gene expression signatures associated with IM and MM as monitored by qPCR ( Figure 7B ) . The gene expression profile of IMP-derived MM exhibited a striking similarity to that of fetal kidney cells . PAX2 and SIX2 were upregulated at day 14 of renal differentiation , indicating commitment to the kidney lineage ( Bush et al . , 2013 ) . Furthermore , immuno-fluorescence analysis demonstrated that a significant number of cells expressed IM and MM markers PAX2 , SALL1 , SIX2 , WT1 and CDH1 ( E-cadherin ) ( Figure 7C–E ) . These results suggested that IMP cells , as predicted by the gene expression profile , are restricted to IM and effectively differentiate into cells expressing genes associated with a renal phenotype . 10 . 7554/eLife . 08413 . 020Figure 7 . Differentiation of IMP cells into metanephric mesenchyme ( MM ) . ( A ) Schematic of the differentiation protocol . Cells were differentiated in a step-wise manner using the indicated GFs and SMs from undifferentiated ES cells or from IMP cells to IM and subsequently to MM . Stage-specific marker genes expressed during this differentiation process are indicated at the top . ACT = ActivinA , BMP = BMP4 , CHR = CHIR98014 , d = day , FGF = FGF2 , RA = retinoic acid . ( B ) Upon differentiation towards MM , cells expressed genes associated with kidney lineage . QPCR was performed on ES and IMP cells for the indicated genes at various time points . Fetal kidney RNA ( 11 gestation weeks ) was used as a control . The data is displayed as a heat map with black corresponding to minimal expression and red corresponding to maximal levels . ( C–E ) IF analysis of MP cell-derived MM . IMP cells were differentiated as depicted in panel A , fixed and stained for the indicated proteins and DNA ( DAPI ) . Numbers refer to percentages of cells expressing the protein of interest . Standard deviation represents the variation between the fields of view used for counting ( n = 20 ) . Scale bar = 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 08413 . 020 To further assess the ability of the IMP cells to generate cells with renal properties , we employed two rat explant assays that represent stringent measures of renal potential . In the first assay , we co-cultured IMP-derived MM cells with dissected embryonic rat spinal cords ( SCs ) , a tissue that produces potent nephrogenic inductive signals ( Figure 8A ) ( Kispert et al . , 1998; Osafune et al . , 2006; Gallegos et al . , 2012 ) . In this system , IMP-derived MM cells readily acquired expression of markers associated with renal cell types , including Lotus tetragonolobus lectin ( LTL ) , CDH1 , SALL1 and SIX2 ( Figure 8B ) . In contrast , undifferentiated hES cells failed to express of SIX2 ( Figure 8C ) , indicating that MM properties are required for efficient renal differentiation . Although IMP-derived MM cells expressed several markers associated with the renal lineage , they failed to generate tubule-like structures , including the nephron , suggesting that IMP cells differentiate effectively into a sub-population of kidney cells . These co-culture experiments demonstrate that IMP cells efficiently generate cell types with renal characteristics . 10 . 7554/eLife . 08413 . 021Figure 8 . Assessment of renal potential of IMP cells . ( A ) Schematic of spinal cord ( SC ) co-culture assay to assess renal differentiation potential of IMP cells . IMP cells were differentiated as depicted in Figure 7A and incubated in liquid–air interface cultures with rat embryonic SC explants . ( B ) Immuno-fluorescence analysis of markers expressed in renal progenitors . 4 days after co-cultures were established , cells were fixed and stained for the indicated proteins ( ECAD , SIX2 and SALL1 ) and for Lotus-tetragonolobus lectin ( LTL ) . The dashed line indicates the boundary between human cells and the SC explant . Scale bar = 100 µm . ( C ) Undifferentiated hES cells failed to express SIX2 when co-cultured with embryonic rat SCs . Scale bar = 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 08413 . 021 In a second assay , rat embryonic kidneys were dissociated to single cells and re-aggregated to form kidney-like organoids ( Unbekandt and Davies , 2010; Davies and Chang , 2014 ) . These aggregation experiments were performed in the presence of either IMP-derived MM cells ( Figure 9A ) or undifferentiated hES cells ( control ) , thereby assessing the renal potential of these cells . The contribution of human cells to the re-aggregated rat kidneys is readily detected by staining for the human specific nuclear antigen ( HuNu ) . In this assay , we consistently observed efficient incorporation of IMP-derived MM cells into the kidney organoids ( Figure 9B , Figure 9—figure supplement 1 ) . Interestingly , we primarily observed incorporation of these cells into the mesenchyme surrounding epithelial structures , which were visualized by staining with lectin Dolichos biflorus agglutinin ( DBA ) . Furthermore , incorporated human cells expressed FOXD1 , the expression of which is restricted to metanephric stromal mesenchyme ( Hatini et al . , 1996 ) and stained with LTL ( Figure 9C , Figure 9—figure supplement 2 ) . In contrast , undifferentiated hES cells failed to incorporate into these kidney organoids ( Figure 9D , Figure 9—figure supplement 3 ) and instead were found adjacent to the organoid structures ( Figure 9—figure supplement 3 , bottom row ) . Taken together , these co-culture experiments establish that IMP cells efficiently incorporated into the developing kidney . 10 . 7554/eLife . 08413 . 022Figure 9 . Incorporation of IMP cells into kidney mesenchyme . ( A ) Schematic of a re-aggregation assay to test renal potential . IMP cells were differentiated as depicted in Figure 7A and mixed with dissociated embryonic rat kidneys at a ratio of 7 . 5:92 . 5 and co-incubated for 4 days to form organoids in media-air interface co-culture . ( B ) Representative images of re-aggregated kidney organoids . IMP cells differentiated to MM are detected with the human specific nuclear antigen HuNu ( green ) . Human cells are clearly integrated into renal organoids and surround epithelial structures labeled with the lectin Dolichos biflorus agglutinin ( DBA ) ( red ) . Figure 9—figure supplement 1 provides additional images of MP cells incorporating into renal structures . Scale bar = 25 µm . ( C ) Representative images of re-aggregated kidney organoids . Renal organoids were labeled with DAPI ( blue ) to identify nuclei , HuNu ( green ) to identify human cells and with either FOXD1 antibody or LTL ( red ) . Two representative sets of images are shown to indicate co-localization of FOXD1 in HuNu positive cells . Scale bar = 25 µm . ( D ) Undifferentiated hES cells failed to integrate into renal organoids . Instead of MP cells , undifferentiated ES cells were mixed with dissociated embryonic rat kidneys . These cells failed to integrate into the renal organoid structures as indicated by the lack of HuNu staining . Figure 9—figure supplement 2 demonstrates that undifferentiated ES cells fail to incorporate into these structures . Scale bar = 25 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 08413 . 02210 . 7554/eLife . 08413 . 023Figure 9—figure supplement 1 . Additional assessment of renal potential of MP cells . Representative images of re-aggregated kidney organoids . MP cells differentiated to MM are detected with the human specific nuclear antigen HuNu ( green ) . Human cells are clearly integrated into renal organoids and surround epithelial structures labeled with the lectin DBA ( red ) . Scale bar = 25 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 08413 . 02310 . 7554/eLife . 08413 . 024Figure 9—figure supplement 2 . Staining controls relevant to Figure 9C . Samples were fixed and stained as in Figure 9C , except anti-HuNu antibody was excluded ( top row ) or anti-FOXD1 antibody was excluded ( bottom row ) . The same secondary antibodies were used to demonstrate that the HuNu and FOXD1 stains are dependent on primary antibodies and are not due to non-specific binding of secondary antibodies . DOI: http://dx . doi . org/10 . 7554/eLife . 08413 . 02410 . 7554/eLife . 08413 . 025Figure 9—figure supplement 3 . Additional assessment of renal potential of MP cells . Undifferentiated hES cells failed to integrate into renal organoids . Unlike MP cells , undifferentiated ES cells failed to integrate into the renal orgaoind structures as indicated by the lack of HuNu staining . The last row of images demonstrates ES cells are present in the culture but are not incorporated into the renal aggregates . Scale bar = 25 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 08413 . 025 In this study , we describe a novel progenitor cell population derived from hPSCs with the potential to differentiate into tissues of the IM lineage . By using the ACME screening technology , we were able to simultaneously define and optimize derivation and expansion conditions for these intermediate mesodermal progenitor ( IMP ) cells . Although it was our initial intention to produce a progenitor cell population with broad differentiation potential into all mesodermally-derived tissues , we made the surprising finding that the differentiation potential of these IMP cells was restricted to the IM lineage . Consequently , we were unable to coax IMP cells to differentiate into cell types derived from LM , such as blood and CMs . This exquisite lineage restriction was particularly surprising in light of the expression of multiple pan-mesodermal marker genes , such as LHX1 , MESP1 and MIXL1 . Given their ability to differentiate into cell types with gene expression patterns associated with renal lineages , we hypothesize that this IMP cell population is an in vitro counterpart to intermediate mesoderm . It will be interesting to investigate whether IMP cells are capable of differentiating into other derivatives of intermediate mesoderm , such as the Wolffian and Müllerian ducts of the developing reproductive system . Generation of expandable , lineage restricted progenitor cell populations offers several advantages over the use of undifferentiated hPSCs in tissue engineering approaches . First , differentiated cultures derived directly from hPSCs often harbor undifferentiated cells , which retain the potential to seed tumor growth . Such tumor-initiating potential is problematic when cells are intended for transplantation to repair or replace damaged tissue . Based on our sub-cutaneous injections into immune-compromised mice , IMP cells do not grow into teratomas , a defining property of undifferentiated pluripotent stem cells . Our gene expression analysis provides further evidence of this loss of pluripotency and hence of teratoma-seeding potential: IMP cells express nearly undetectable levels of pluripotency markers , such as POU5F1/OCT4 and SOX2 , both of which show residual expression in mesodermally differentiated hPSCs . Second , lineage-restricted progenitors require less elaborate manipulation to derive more mature cell populations . In the case of the IMP cells , early differentiation steps to usher cells into a mesodermal lineage are no longer needed , thereby truncating differentiation protocols to derive more mature cell populations . A third benefit for using expanded progenitor cells is that such cultures are often quite homogenous . In contrast , hPSC cultures instructed to differentiate into a specific lineage generally contain other cell types . Therefore , the yield of more mature cell types upon subsequent differentiation is higher when starting with a homogenous , lineage restricted cell population than when starting with undifferentiated hPSCs . The conditions that we developed for the culture and expansion of IMP cells are fully defined and free from animal-derived components , which will be important when cells are intended for therapeutic applications . Moreover , these optimized conditions are robust , as demonstrated by their ability to support derivation and expansion of IMP cells from two hES ( H9 and Hues9 ) and one hiPS ( RiPS ) cell lines . Additionally , IMP cells grown in these optimized conditions can be frozen and thawed without any detectable effect on proliferative capacity or differentiation potential . Finally , these optimized conditions allow for near unlimited expansion to quantities ( ∼1020 ) necessary for drug screening or regenerative medicine purposes ( Chen et al . , 2013 ) . Expandable lineage restricted cell populations have been developed for other lineages , including the neural and EN lineages . Several protocols have been described for the derivation of neural progenitor ( NP ) cells , which can proliferate extensively and differentiate into all the neural lineages and supporting cells ( neurons , astrocytes , and oligodendrocytes ) that compromise the central nervous system ( Reubinoff et al . , 2001; Shin et al . , 2006; Chambers et al . , 2009 ) . EN progenitor ( EP ) cells represent another example of lineage restricted progenitor cells ( Cheng et al . , 2012 ) . These cells retain the ability to differentiate into endodermally derived tissues , including liver and pancreas . Interestingly , differentiation into functional beta-cells is greatly improved when starting with EP cells compared to undifferentiated hPSCs . Although both EP and IMP cells exhibit restriction with respect to their developmental potency , IMP cells are more severely restricted as they fail to produce certain mesodermally-derived cell populations , such as blood and heart muscle . We currently do not understand the mechanism by which the culture conditions defined for the derivation and expansion of IMP cells lead to this highly restricted developmental potential . During embryogenesis , as the mesoderm emerges and migrates from the primitive streak , it is further specified into PM , LM , and IM . Interestingly , both FGF and WNT/β-catenin signaling regulate this ME cell specification , migration , and proliferation ( Ciruna and Rossant , 2001; Sweetman et al . , 2008; Aulehla and Pourquie , 2010 ) . Along similar lines , modulation of the certain signaling pathways , such as WNT , can further refine and specify the differentiation potential of hPSC-derived progenitors . For example , we previously showed that levels of WNT/β-catenin signaling instruct the positional identity of NPCs and , upon subsequent differentiation , of the resulting neuronal cell population ( Moya et al . , 2014 ) . Specifically , high levels of WNT signaling instructed NP cells to adopt a posterior fate , consistent with WNT's role in posterior patterning during development . In a separate study , the level of WNT activation achieved through GSK3-β inhibition was found to directly influence the ME subtype of differentiating hPSCs ( Mendjan et al . , 2014 ) . We speculate that continuous activation of the WNT and FGF signaling pathways is acting not only to stabilize the IMP cell state , but also to restrict its differentiation potential to cell types derived from the IM lineage . The development of lineage-restricted progenitors offers an opportunity to investigate mechanisms by which specific developmental stages can be paused . Recent studies to profile epigenetic changes during the differentiation of hPSCs to pancreatic beta cells indicate that specific chromosomal regions open during specific windows of differentiation , thereby conferring a certain developmental competence to sequentially acquire increased lineage restriction ( Wang et al . , 2015 ) . In the future , the intermediate mesodermally restricted cell population described here can provide a further window into the mechanisms by which developmental competence is established and maintained . Human ES cell lines H9 and Hues9 were obtained from WiCell and Harvard University , respectively . All experiments described in this study were approved by a Stem Cell Research Oversight Committee ( Protocol #100210ZX , PI Willert ) . The human induced pluripotent stem cell line BJ RiPS ( Warren et al . , 2010 ) was provided under a Material Transfer Agreement from Dr D Rossi ( Childrens Hospital Boston , MA , United States ) . The H9 line carrying GFP in the SOX17 locus ( Wang et al . , 2011 ) was provided under a Material Transfer Agreement from Dr Seung Kim ( Stanford School of Medicine ) . The following media were used: BJ RiPS and Hues 9 ES ( DMEM/F12 mixed , 20% ( vol/vol ) Knockout Serum Replacement , 1% ( vol/vol ) penicillin-streptomycin , 1% ( vol/vol ) nonessential amino acids , 2 mM L-glutamate , 0 . 1 mM β-mercaptoethanol and 10 ng/ml FGF2 ( PeproTech ) ) ; H9 ES ( DMEM/F12 supplemented with L-Ascorbic Acid , Selenium , Transferrin , NaHCO3 , Insulin , TGFβ1 , and FGF2 as described previously ( Chen et al . , 2011 ) ) . Fresh media was added daily to all cells . Every 5 days , colonies were enzymatically passaged with Accutase ( Thermo Fisher Scientific , Waltham , MA , United States ) and transferred to a Matrigel-coated culture dish . All media components are from Thermo Fisher Scientific unless indicated otherwise . For all experiments , hPSCs were used between passages 20 and 50 in this study . ACME slides were fabricated as previously described ( Brafman et al . , 2012 ) . Briefly , glass slides were cleaned , silanized , and then functionalized with a polyacrylamide gel layer . For ECMP arrays , stock solutions of ECMPs were suspended at 250 µg/ml in ECMP printing buffer ( 100 mM acetate , 5 mM EDTA , 20% [vol/vol] glycerol and 0 . 25% [vol/vol] Triton X-100 , pH 5 . 0 ) . ECMP solutions were mixed in all possible 128 combinations in a 384-well plate . For GF and SM arrays , stock solutions were suspended at 1 mg/ml in soluble factor printing buffer ( 100 mM acetate , 5 mM EDTA , 19% glycerol [vol/vol] and 0 . 25% [vol/vol] Triton X-100 , 10 mM trehalose dehydrate [Sigma] , 1% poly ( ethylene glycol ) , pH 5 ) . GF solutions were then mixed into 400 combinations representing all single , pairwise , and non-redundant three-way combinations possible in a 384-well plate . The following ECMPs , GFs , and SMs ( Product/Vendor/Catalog #/Concentration ) were used: Collagen I/Sigma–Aldrich ( St . Louis , MO , United States ) /C7774/250 µg/ml , Collagen III/Sigma–Aldrich/C4407/250 µg/ml , Collagen IV/Sigma–Aldrich/C7521/250 µg/ml , Collagen V/Sigma–Aldrich/C3657/250 µg/ml , Fibronectin/Sigma–Aldrich/F2518/250 µg/ml , Laminin/Sigma–Aldrich/L6274/250 µg/ml , Vitronectin/Sigma–Aldrich/V8379/250 µg/ml , Wnt3a/In House/100 ng/ml , R-Spondin/In House/100 ng/ml , CHIR98014/Selleck Chemicals ( Houston , TX , United States ) /S2745/50 ng/ml , Dkk-1/R&D Systems ( Minneapolis , MN , United States ) /5439-DK-010/50 ng/ml , IWP-2/Tocris ( United Kingdom ) /3533/50 ng/ml , FGF/Thermo Fisher Scientific/13256-029/40 ng/ml , KGF/Thermo Fisher Scientific/PHG0094/50 ng/ml , VEGF/R&D Systems/293-VE-010/50 ng/ml , EGF/R&D Systems/236-EG-01M/50 ng/ml , SHH/R&D Systems/464-SH-025/50 ng/ml , Cyclopamine/Tocris/1523/50 ng/ml , BMP4/R&D Systems/314-BP-010/50 ng/ml , Activin/R&D Systems/338-AC-010/50 ng/ml , Dorsomorphin/Sigma–Aldrich/P5499-5MG/50 ng/ml , SB 431542/Tocris/1614/50 ng/ml , Noggin/R&D Systems/6057-NG-025/50 ng/ml . The hit ECMP condition from the primary screen was used as a substrate to print the GFs and SMs in the second screen . 20 individual spots of each protein/GF/SM mixture , clustered into groups of five and printed in different quadrants of the slide , were deposited with a 450 µm pitch on the acrylamide gel pad using a SpotBot Personal Microarray Printer ( ArrayIt , Sunnyvale , CA , United States ) equipped with Stealth SMP 4 . 0 split pins . The pins were cleaned by sonication in 5% Micro Cleaning Solution ( ArrayIt ) and dH2O immediately before use . Between each sample in the source plate , the pins were dipped in a 50% DMSO and water solution , washed for 25 s with dH2O and dried . Slides were fixed with 4% PFA for 10 min at room temperature ( RT ) and washed with PBS . Slides were imaged using the CellInsight CX5 High Content Screening ( HCS ) Platform ( Thermo Fisher Scientific ) . The system was programmed to visit each spot on the array , perform autofocus , and acquire DAPI and FITC ( GFP ) . Cell counts and stain intensities were measured using Thermo Fisher Scientific HCS Studio 2 . 0 Software using the built-in object identification and cell intensity algorithms . Undifferentiated hPSCs were re-plated on Matrigel at a density of 3 × 103 cells/cm2 and cultured in ES cell culture medium for 4 days . To direct cells to the mesoderm lineage , the media was switched to serum free differentiation media ( consisting of RPMI 1640 , 1× B27 minus Insulin , and 1% [vol/vol] penicillin-streptomycin ) . Cells were treated with 10 µM CHIR-98014 ( CHR , Tocris ) for the first 24 hr and then allowed to recover for an additional 24 hr without CHR . Tissue culture plates were incubated with ECMP coating buffer ( PBS with 15 ng/ml Collagen I [C1] , 15 ng/ml Collagen III [C3] , 15 ng/ml Collagen IV [C4] , 50 ng/ml FN , 15 ng/ml VN ) overnight at 37° with volume sufficient to coat the surface area of the well . Mesoderm ( 48 hr ) cells were single-cell passaged with Accutase and replated onto C1 C3 C4 FN VN-coated plates at a density of 3 . 5 × 103 cells/cm2 in serum free differentiation media supplemented with 1 µM CHR and 20 ng/ml FGF . Media was also supplemented with 10 μM Y27632 ( Wako , Richmond , VA , United States ) for improve passaging efficiency . Optimal CHR concentration varied with cell line; Hues 9 MP cells propagated in colonies most efficiently at 0 . 25 µM while BJ RiPS cells did so at 0 . 05 µM . Manual picking of colonies in passage 1 improved MP/IMP expansion . Differentiated cells around colonies were scraped away before passaging . Half the media was changed the day after passaging and then full media changes were made every other day thereafter . For routine passaging , MP/IMP cell cultures reaching 85% confluency were dissociated using a 0 . 5 mM EDTA ( in Ca2+/Mg2+-free PBS , pH 8 . 0 ) at RT for 5 min . MP cells were removed from the plate via gentle washing with the EDTA solution . Using this method , MP cells were routinely passaged every 5–8 days . RNA was isolated using RNeasy Plus Micro Kit ( Qiagen , Germany ) reverse-transcribed with random primers and qScript cDNA Supermix ( Quanta , Gaithersburg , MD , United States ) . Before reverse transcription , 5 µg of RNA was digested by RNase-free DNase I ( Ambion/Thermo Fisher Scientific ) to remove genomic DNA . qPCR was carried out using a Real-Time PCR System ( Bio-Rad , Hercules , CA , United States ) and Taqman qPCR Mix with a 10-min gradient to 95°C followed by 40 cycles at 95°C for 15 s and 60°C for 1 min . The following Taqman ( Thermo Fisher Scientific ) gene expression assay primers ( Gene/ABI Assay # ) were used: 18s/Hs99999901_s1 , OCT4/Hs04260367_gH , NANOG/Hs04399610_g1 , SOX2/Hs01053049_s1 , FOXA2/Hs00232764_m1 , SOX1/Hs01057642_s1 , MESP1/Hs01001283_g1 , MIXL1/Hs00430824_g1 , LHX1/Hs00232144_m1 , PDGFRA/Hs00998018_m1 , PAX1/Hs01071293_g1 , TBX6/Hs00365539_m1 , TCF15/Hs00231821_m1 , MEOX1/Hs00244943_m1 , NKX2 . 5/Hs00231763_m1 , ISL1/Hs00158126_m1 , LMO2/Hs00153473_m1 , KDR/Hs00911700_m1 , PAX2/Hs01057416_m1 , EYA1/Hs00166804_m1 , SALL1/Hs01548765_m1 , OSR1/Hs01586544_m1 , LHX1/Hs00232144_m1 , WT1/Hs01103751_m1 , CITED2/Hs01897804_s1 , PECAM1/Hs00169777_m1 , HOXC9/Hs00396786_m1 , ITGA8/Hs00233321_m1 , PBX1/Hs00231228_m1 , HOXA10/Hs00172012_m1 , HOXA11/Hs00194149_m1 , GDNF/Hs01931883_s1 , FOXD1/Hs00270117_s1 , SIX2/Hs00232731_m1 , CDX2/Hs01078080_m1 , FGF5/Hs03676587_s1 . Gene expression was normalized to 18S rRNA levels . Delta Ct values were calculated as Cttarget−Ct18s . All experiments were performed with three technical replicates . Relative fold changes in gene expression were calculated using the 2−ΔΔCt method ( VanGuilder et al . , 2008 ) . The following antibodies were used ( Antibody/Vendor/Catalog #/Concentration ) : Rabbit anti-NANOG/Santa Cruz Biotechnology ( Dallas , TX , United States ) /SC-33759/1:50 , Rabbit anti-OCT4/Santa Cruz/SC-9081/1:50 , Mouse anti-MIXL1/R&D Systems/MAB2610/1:200 , Mouse anti-PAX2/Creative Diagnostics ( Shirley , NY , United States ) /DMABT-H14539/1:200 , Rabbit anti-SIX2/Abcam ( Cambridge , MA , United States ) /ab68908/1:200 , Rabbit anti-WT1/Santa Cruz Biotechnology/sc-192/1:200 , Rabbit anti-SALL1/Abcam/ab31526/1:200 , Mouse anti-E Cadherin/Abcam/ab1416/1:200 , Mouse anti-Human Nuclear Antigen/Abcam/ab191181/1:250 , Goat anti-FOXD1/Santa Cruz Biotechnology/sc-47585/1:200 , Rabbit anti-Ki67/Abcam/ab15580/1:250 , APC anti-human CD56 ( NCAM ) /BioLegend ( San Diego , CA , United States ) /318309/5 μl per test , PE anti-human CD326 ( EpCAM ) /BioLegend/324205/5 μl per test , Alexa-647 Mouse IgG2a Isotype Control/BD/558053/20 µl per test , PE Mouse IgG1 Isotype Control/BioLegend/400113/5 μl per test , PE Mouse IgG2a Isotype Control/BD Biosciences ( San Jose , CA , United States ) /561552/5 μl per test , Alexa 647 Donkey Anti-Goat/Thermo Fisher Scientific/A-21447/1:200 , Alexa 647 Donkey Anti-Rabbit/Thermo Fisher Scientific/A-31573/1:200 , Alexa 647 Donkey Anti-Mouse/Thermo Fisher Scientific/A-31571/1:200 , Alexa 546 Donkey Anti-Rabbit/Thermo Fisher Scientific/A-10040/1:200 , Alexa 546 Donkey Anti-Mouse/Thermo Fisher Scientific/A-10036/1:200 , Alexa 488 Streptavidin Conjugate/Thermo Fisher Scientific/S-11223/1:200 , Alexa 488 Donkey Anti-Rabbit/Thermo Fisher Scientific/A-21206/1:200 , Alexa 488 Donkey Anti-Mouse/Thermo Fisher Scientific/A-21202/1:200 . Cells were dissociated with Accutase ( Thermo Fisher Scientific ) at 37°C for 4 min and triturated using fine-tipped pipettes . For intracellular antibody staining , cells were fixed for 15 min with Cytofix ( BD Biosciences ) , washed twice with flow cytometry buffer ( PBS , 1 mM EDTA , and 0 . 5% FBS ) , permeabilized with Cytoperm ( BD Biosiences ) for 30 min on ice , and washed twice with flow cytometry buffer , and resuspended at a maximum concentration of 5 × 106 cells per 100 μl . Cells were incubated with primary antibodies on ice for 1 hr , washed twice with flow cytometry buffer . If necessary , cells were incubated with secondary antibodies on ice for 1 hr and then washed three times . After passing through a 40 μm cell strainer , cells were resuspended in flow cytometry buffer at a final density of 2 × 106 cells ml−1 . Propidium iodide ( Sigma ) was added at a final concentration of 50 mg ml−1 to exclude dead cells . Cells were analyzed on the FACS Fortessa ( Becton Dickinson ) . For each sample , at least three independent experiments were performed . Results were analyzed using FlowJo software . Monolayer cultures were gently washed with PBS prior to fixation . Cultures were fixed for 10 min at 4°C with fresh paraformaldehyde ( 4% [wt/vol] in PBS ) . For sectioning aggregates of cells in suspension , samples were fixed with 4% paraformaldehyde , embedded in optimal cutting temperature compound ( Tissue Tek ) and cryo-sectioned at 10-µm thickness before staining , Cells were blocked and permeabilized with 2% ( wt/vol ) BSA , 0 . 2% ( [vol/vol] in PBS ) Triton X for 30 min at RT . Cells were then washed twice with PBS . Primary antibodies were incubated overnight at 4°C and washed twice with PBS . Secondary antibodies were incubated for 1 hr at 37°C . Antibodies used are as listed above . Prior to imaging , samples were stained with DAPI for 10 min , washed and mounted in Vectashield ( Vector Laboratories , Burlingame , CA , United States ) , covered with coverslips , and sealed with nail polish . Images were taken using an Olympus FluoView1000 multi-photon confocal microscope . All IF analyses were repeated a minimum of three times and representative images are shown . Total RNA was isolated from cells , depleted of genomic DNA and rRNA and fragmented to ∼200 bp by RNase III . After ligating the Adaptor Mix , fragmented RNA was converted to the first strand cDNA by ArrayScript Reverse Transcriptase ( Ambion/Thermo Fisher Scientific ) , size selected ( 100–200 bp ) by gel electrophoresis , and amplified by PCR using adaptor-specific primers . Deep sequencing was performed on an Illumina ( San Diego , CA , United States ) Genome Analyzer II . Analysis of genome-wide expression data was performed as previously described ( Trapnell et al . , 2012 , 2013 ) . Briefly , raw reads were aligned to the reference human genome ( hg19 ) using TopHat . Cufflinks was used to assemble individual transcripts from the mapped reads . Cuffmerge was used to merge the assembled transcripts from the two biologically independent samples . Cuffdiff was used to calculate gene expression levels and test for the statistical significance of differences in gene expression . Reads per kilobase per million mapped reads were calculated for each gene and used as an estimate of expression levels . The full RNA-seq data set for the IMP cells is provided in Supplementary file 2 . hES or IMP-derived MM cells were cultured with mouse embryonic SC taken from E11 . 5 or E12 . 5 embryos at the air-fluid interface on a polycarbonate filter ( 0 . 8 mm; Whatman/Sigma-Aldrich ) fed with DMEM containing 10% fetal calf serum , as described previously ( Kispert et al . , 1998; Osafune et al . , 2006; Gallegos et al . , 2012; Martovetsky et al . , 2013 ) . The re-aggregation assay was performed as previously described ( Unbekandt and Davies , 2010; Davies and Chang , 2014 ) . To prepare the kidney tissue for recombination , embryonic kidneys from 12 . 5–13 . 5-dpc ( days post coitum ) mice were isolated and dissected free of surrounding tissues as previously described ( Gallegos et al . , 2012; Martovetsky et al . , 2013 ) . Briefly , embryonic kidneys were digested with trypsin at 37°C for 10 min and dissociated by manually pipetting . After the cells had been filtered through a 100 μm cell strainer , 4–10 × 105 embryonic kidney cells were recombined with 4% ( by number ) of hESC- or IMP-derived cells and then centrifuged at 400×g for 2 min to form a pellet . The pellet was allowed to aggregate by culturing in DMEM supplemented with 10% FBS overnight in a sterile PCR tube . The following day , the aggregate was transferred to the top of a Transwell polycarbonate filter ( 0 . 4 μm pore size ) . The filter was placed with the well of a 12-well dish to which DMEM supplemented with 10% FBS was added to bottom of the well . The aggregate was then cultured for 4 days at the air-fluid interface before fixation and analysis . The following lectins were used to stain organoid cultures ( Lectin/Vendor/Catalog #/Concentration ) : Biotinylated DBA/Vector Laboratories/B-1035/1:200 , LTL , Biotinylated-LTL/Vector Labs/B-1325/1:200 . An Alexa 647 Streptavidin Conjugate ( Thermo Fisher Scientific; S-21374 ) was used at 1:200 for detection of these lectins . For the subcutaneous injection , H9 hES or MP cells were dissociated , mixed with 250 µl Matrigel , and transplanted subcutaneously into the thigh and shoulder of nude mice . Each mouse received two injections of cells , one near the front legs and one near the hind legs . Teratoma formation was monitored over a period of 4–12 weeks . A total of 3 mice were injected with 0 . 5 × 106 hES cells per site . All six injection sites yielded teratomas of 10 mm or greater . Another 6 mice were injected with MP cells: 2 mice received 0 . 5 × 106 cells per injection , 2 mice received 0 . 75 × 106 cells per injection , and 2 mice received 1 . 0 × 106 cells per injection . Of the 12 injection sites , only one site maintained a small lump of 1 mm that did not grow in size . No MP cell injection yielded a growth of the size observed for hES cells . All animal work was approved by the institutional IACUC committee ( Protocol Number S06321 , PI Willert ) . Karyotype analysis was performed by Cell Line Genetics , Inc . , Madison , Wisconsin , United States . For each submitted MP cell line ( H9 at passage 10 and H9_SOX17-GFP at passage 15 [Wang et al . , 2011] , kindly provided by Dr Seung Kim , Stanford School of Medicine , Palo Alto , CA , United States chromosome numbers were determined for 20 cells using G-banded metaphase spreads . All averaged data are expressed ±standard error of the mean of three independent biological replicates unless otherwise stated . For comparisons of discrete data sets , unpaired Student's t-tests were performed to calculate p-values between experimental conditions and controls and a p-value <0 . 05 was considered statistically significant . For each ACME experiment , the ratio ( Ri ) of the log2 of the T-GFP signal and the DNA signal was calculated for each spot . From this a differentiation z-score was calculated for each spot ZDIF = ( Ri − μDIF ) /σDIF , where Ri was the ratio for the spot , μDIF was the average of the ratios for all spots on each array , and σDIF was the S . D . of the ratios for all spots on each array . Differentiation z-scores from replicate spots ( n = 5 per condition ) were averaged for each ECMP condition on the array . The replicate average z-scores were displayed in a heat map with rows corresponding to individual conditions and columns representing independent array experiments ( n = 5 for each replicate ) . For each array experiment , all columns were mean-centered and normalized to one unit S . D . The rows were clustered using Pearson correlations as a metric of similarity . All clustering was performed using Gene Cluster . The results were displayed using a color code with red and green representing an increase and decrease , respectively , relative to the global mean . All heat maps were created using Tree View . Global main effects principal component analysis was performed as previously described ( Box et al . , 2005 ) .	The development of ‘human pluripotent stem cells’ has the potential to revolutionize the future of medicine . This is because these cells can both replicate themselves indefinitely ( i . e . , they can self-renew ) and develop into any of the cell types found in the human body ( a process that is referred to as differentiation ) . These abilities mean that the cells could in theory be used to replace any tissues or organs that have been damaged by disease or injury . Unfortunately , transplanting stem cells that are capable of developing into any type of cell comes with the significant risk that these cells will form into a tumor . Once a cell has started to differentiate it can typically only go on to generate a restricted number of cell types . However , these differentiating cells also generally lose their ability to self-renew . Kumar et al . set out to challenge this fundamental property of differentiating cells . A high throughput-screening approach was used to test thousands of combinations of bioactive molecules ( i . e . , molecules that are known to affect living cells in different ways ) to identify some that could promote the self-renewal of cells with a restricted potential to differentiate . Kumar et al . found specific conditions that could cause a population of cells , which they referred to as ‘intermediate mesodermal progenitor cells’ ( or IMP cells for short ) , to self renew . These cells resemble those found in the middle layer of a very early human embryo , which typically go on to develop into only a subset of tissue types in the body—for example , muscle , kidneys and blood vessels , but not brain or lungs . Yet , when Kumar et al . stimulated the self-renewing IMP cells , these cells only differentiated into the cell types that make up the kidney and not any other types of cell . This tight restriction on the differentiation potential of these cells is highly important , because it means that these cells could greatly advance methods to generate kidney cells or even whole kidneys in the laboratory that are suitable for transplantation .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine" ]	2015	Generation of an expandable intermediate mesoderm restricted progenitor cell line from human pluripotent stem cells
We report a draft assembly of the genome of Hi5 cells from the lepidopteran insect pest , Trichoplusia ni , assigning 90 . 6% of bases to one of 28 chromosomes and predicting 14 , 037 protein-coding genes . Chemoreception and detoxification gene families reveal T . ni-specific gene expansions that may explain its widespread distribution and rapid adaptation to insecticides . Transcriptome and small RNA data from thorax , ovary , testis , and the germline-derived Hi5 cell line show distinct expression profiles for 295 microRNA- and >393 piRNA-producing loci , as well as 39 genes encoding small RNA pathway proteins . Nearly all of the W chromosome is devoted to piRNA production , and T . ni siRNAs are not 2´-O-methylated . To enable use of Hi5 cells as a model system , we have established genome editing and single-cell cloning protocols . The T . ni genome provides insights into pest control and allows Hi5 cells to become a new tool for studying small RNAs ex vivo . Lepidoptera ( moths and butterflies ) , one of the most species-rich orders of insects , comprises more than 170 , 000 known species ( Mallet , 2007; Chapman , 2009 ) , including many agricultural pests . One of the largest lepidopteran families , the Noctuidae diverged over 100 million years ago ( mya ) from the Bombycidae—best-known for the silkworm , Bombyx mori ( Rainford et al . , 2014 ) . The Noctuidae family member cabbage looper ( Trichoplusia ni ) is a widely distributed generalist pest that feeds on cruciferous crops such as broccoli , cabbage , and cauliflower ( Capinera , 2001 ) . T . ni has evolved resistance to the chemical insecticide Dichlorodiphenyltrichloroethane ( DDT; ( McEwen and Hervey , 1956 ) and the biological insecticide Bacillus thuringiensis toxin ( Janmaat and Myers , 2003 ) , rendering pest control increasingly difficult . A molecular understanding of insecticide resistance requires a high-quality T . ni genome and transcriptome . Hi5 cells derive from T . ni ovarian germ cells ( Granados et al . , 1986; 1994 ) . Hi5 cells are a mainstay of recombinant protein production using baculoviral vectors ( Wickham et al . , 1992 ) and hold promise for the commercial-scale production of recombinant adeno-associated virus for human gene therapy ( Kotin , 2011; van Oers et al . , 2015 ) . Hi5 cells produce abundant microRNAs ( miRNAs ) miRNAs , small interfering RNAs ( siRNAs ) , and PIWI-interacting RNAs ( Kawaoka et al . , 2009 ) ( piRNAs ) , making them one of just a few cell lines suitable for the study of all three types of animal small RNAs . The most diverse class of small RNAs , piRNAs protect the genome of animal reproductive cells by silencing transposons ( Saito et al . , 2006; Vagin et al . , 2006; Brennecke et al . , 2007; Houwing et al . , 2007; Aravin et al . , 2007; Kawaoka et al . , 2008 ) . The piRNA pathway has been extensively studied in the dipteran insect Drosophila melanogaster ( fruit fly ) , but no piRNA-producing , cultured cell lines exist for dipteran germline cells . T . ni Hi5 cells grow rapidly without added hemolymph ( Hink , 1970 ) , are readily transfected , and—unlike B . mori BmN4 cells ( Iwanaga et al . , 2014 ) , which also express germline piRNAs—remain homogeneously undifferentiated even after prolonged culture . In contrast to B . mori , no T . ni genome sequence is available , limiting the utility of Hi5 cells . To further understand this agricultural pest and its Hi5 cell line , we combined divers genomic sequencing data to assemble a chromosome-level , high-quality T . ni genome . Half the genome sequence resides in scaffolds > 14 . 2 megabases ( Mb ) , and >90% is assembled into 28 chromosome-length scaffolds . Automated gene prediction and subsequent manual curation , aided by extensive RNA-seq data , allowed us to examine gene orthology , gene families such as detoxification proteins , sex determination genes , and the miRNA , siRNA , and piRNA pathways . Our data allowed assembly of the gene-poor , repeat-rich W chromosome , which remarkably produces piRNAs across most of its length . To enable the use of cultured T . ni Hi5 cells as a novel insect model system , we established methods for efficient genome editing using the CRISPR/Cas9 system ( Ran et al . , 2013 ) as well as single-cell cloning . With these new tools , T . ni promises to become a powerful companion to flies to study gene expression , small RNA biogenesis and function , and mechanisms of insecticide resistance in vivo and in cultured cells . We combined Pacific Biosciences long reads and Illumina short reads ( Figure 1A , Table 1 , and Materials and methods ) to sequence genomic DNA from Hi5 cells and T . ni male and female pupae . The initial genome assembly from long reads ( 46 . 4 × coverage with reads >5 kb ) was polished using paired-end ( 172 . 7 × coverage ) and mate-pair reads ( 172 . 0 × coverage ) to generate 1976 contigs spanning 368 . 2 megabases ( Mb ) . Half of genomic bases reside in contigs > 621 . 9 kb ( N50 ) . Hi-C long-range scaffolding ( 186 . 5 × coverage ) produced 1031 scaffolds ( N50 = 14 . 2 Mb ) , with >90% of the sequences assembled into 28 major scaffolds . Karyotyping of metaphase Hi5 cells revealed that these cells have 112 ± 5 chromosomes ( Figure 1B , Figure 1—figure supplement 1 ) . Because lepidopteran cell lines are typically tetraploid ( Hink , 1972 ) , we conclude that the ~368 . 2 Mb T . ni genome comprises 28 chromosomes: 26 autosomes plus W and Z sex chromosomes ( see below ) . To evaluate the completeness of the assembled T . ni genome , we compared it to the Arthropoda data set of the Benchmark of Universal Single-Copy Orthologs ( Simão et al . , 2015 ) ( BUSCO v3 ) . The T . ni genome assembly captures 97 . 5% of these gene orthologs , more than either the silkworm ( 95 . 5% ) or monarch butterfly ( D . plexippus; 97 . 0% ) genomes ( Supplementary file 1A ) . All 79 ribosomal proteins conserved between mammals and D . melanogaster ( Yoshihama et al . , 2002; Marygold et al . , 2007 ) have orthologs in T . ni , further evidence of the completeness of the genome assembly ( Supplementary file 1B ) . Finally , a search for genes in the highly conserved nuclear oxidative phosphorylation ( OXPHOS ) pathway ( Porcelli et al . , 2007 ) uncovered T . ni orthologs for all known D . melanogaster OXPHOS genes ( Supplementary file 1C ) . The genomes of wild insect populations are typically highly heterogeneous , which poses a significant impediment to assembly ( Keeling et al . , 2013; You et al . , 2013 ) . We were unable to generate an isogenic T . ni strain by inbreeding . Therefore , our T . ni sequence reflects the genome of Hi5 cells , not cabbage looper itself . Hi5 cells presumably derive from a single immortalized , germline founder cell , which should reduce genomic variation among the cell line’s four sets of chromosomes . To test this supposition , we identified the sequence variants in the Hi5 genome . In total , we called variants at 165 , 370 genomic positions ( 0 . 0449% of the genome assembly ) , with 2710 in predicted coding regions ( 0 . 0132% of coding sequence ) , indicating that the genome of Hi5 cells is fairly homogenous . For the majority ( 88 . 8% ) of these genomic positions ( covering 0 . 0399% of the genome ) , only one copy of the chromosome has the variant allele while the other three chromosomal copies match the reference genome . We can make three conclusions . First , Hi5 cells originated from a single founder cell or a homogenous population of cells . Second , the founder cells were haploid . Third , most sequence variants were acquired after the original derivation of the line from T . ni eggs . We also assembled de novo T . ni genomes using paired-end DNA-seq data obtained from male and female pupae , but the resulting assemblies are fragmented ( scaffold N50 ≤ 2 . 4 kb , Supplementary file 1D ) , likely due to the limitations of short-insert libraries and the high levels of heterozygosity commonly observed for genomes of wild insect populations ( Keeling et al . , 2013; You et al . , 2013 ) . The animal genome contigs are highly concordant with the Hi5 genome , with ≤1 . 37% of animal contigs misassembled ( Supplementary file 1D ) . Although we cannot determine scaffold-level differences between the animal and Hi5 cells , at the contig-level the Hi5 genome assembly is representative of the T . ni animal genome . We annotated 14 , 034 protein-coding genes in the T . ni genome ( Supplementary file 1E ) , similar to other Lepidoptera ( Challis et al . , 2016 ) . Analysis of the homology of T . ni genes to genes in 20 species that span the four common insect orders ( Lepidoptera , Diptera , Coleoptera , Hymenoptera ) , non-insect arthropods , and mammals defines 30 , 448 orthology groups each containing orthologous proteins from two or more species ( Hirose and Manley , 1997 ) ; 9112 groups contain at least one T . ni gene . In all , 10 , 936 T . ni protein-coding genes are orthologous to at least one gene among the 20 reference species ( Figure 1C , Figure 1—figure supplement 2 ) . T . ni contains 2 , 287 Lepidoptera-specific orthology groups ( T . ni , B . mori , D . plexippus , and P . xylostella [diamondback moth] ) . Far fewer orthology groups are unique to Diptera ( 404 ) , Coleoptera ( 371 ) , or Hymenoptera ( 1344 ) , suggesting that the lepidopteran lifestyle requires more order-specific genes . The T . ni genome additionally contains 3098 orphan protein-coding genes for which we could detect no orthologous sequences in the 20 reference species . Of these orphan genes , 14 . 5% are present as two or more copies in the genome ( ‘in-paralogs’ ) , suggesting they evolved recently . Some of these in-paralogs may have arisen by gene duplication after the divergence of T . ni and B . mori ~111 mya ( Gaunt and Miles , 2002; Rota-Stabelli et al . , 2013; Wheat and Wahlberg , 2013; Rainford et al . , 2014 ) . The ability of insects to respond to light is crucial to their survival . Opsins , members of the G-protein-coupled receptor superfamily , play important roles in vision . Covalently bound to light-sensing chromophores , opsins absorb photons and activate the downstream visual transduction cascade ( Terakita , 2005 ) . The T . ni genome encodes ultraviolet , blue , and long-wavelength opsins . Thus , this nocturnal insect retains the full repertoire of insect opsins and has color vision ( Zimyanin et al . , 2008 ) ( Figure 1—figure supplement 3 ) . T . ni also encodes an ortholog of the non-visual Rh7 opsin , which is found in a variety of insects ( International Glossina Genome Initiative , 2014; Futahashi et al . , 2015 ) . In the D . melanogaster brain , Rh7 opsin participates in the entrainment of circadian rhythms by sensing violet light ( Ni et al . , 2017 ) . T . ni also encodes an ortholog of the vertebrate-like opsin , pterosin , which was first detected in the honeybee ( A . mellifera ) brain and is found widely among insects except for Drosophilid flies ( Velarde et al . , 2005 ) . Understanding the T . ni sex-determination pathway holds promise for engineering sterile animals for pest management . ZW and ZO chromosome systems determine sex in lepidopterans: males are ZZ and females are either ZW or ZO ( Traut et al . , 2007 ) . To determine which system T . ni uses and to identify which contigs belong to the sex chromosomes , we sequenced genomic DNA from male and female pupae and calculated the male:female coverage ratio for each contig . We found that 175 presumably Z-linked contigs ( 20 . 0 Mb ) had approximately twice the coverage in male compared to female DNA ( median male:female ratio = 1 . 92; Figure 2A , Figure 2—figure supplement 1A ) . Another 276 contigs ( 11 . 1 Mb ) had low coverage in males ( median male:female ratio = 0 . 111 ) , suggesting they are W-linked . We conclude that sex is determined in T . ni by a ZW system in which males are homogametic ( ZZ ) and females are heterogametic ( ZW ) . For some lepidopteran species , dosage compensation has been reported to equalize Z-linked transcript abundance between ZW females and ZZ males in the soma , while other species show higher expression of Z-linked genes in males ( Walters et al . , 2015; Gu et al . , 2017 ) . In the soma , T . ni compensates for Z chromosome dosage: transcripts from Z-linked genes are approximately equal in male and female thoraces ( Z ≈ ZZ , Figure 2B ) . In theory , somatic dosage compensation could reflect increased transcription of the single female Z chromosome , reduced transcription of both male Z chromosomes , or silencing of one of the two male Z chromosomes . To distinguish among these possibilities , we compared the abundance of Z-linked and autosomal transcripts ( Z/AA in female and ZZ/AA in male , Figure 2—figure supplement 1B and C ) . Z-linked transcripts in the male thorax are expressed at lower levels than autosomal transcripts , but not as low as half ( ZZ ≈ 70% AA ) . These data support a dosage compensation mechanism that decreases transcription from each Z chromosome in the T . ni male soma , but does not fully equalize Z-linked transcript levels between the sexes ( Z ≈ ZZ ≈ 70% AA ) . In contrast , T . ni lacks germline dosage compensation: in the ovary , Z-linked transcript abundance is half that of autosomal transcripts ( Z ≈ 50% AA ) , whereas in testis , Z-linked and autosomal transcripts have equal abundance ( ZZ ≈ AA ) . We conclude that T . ni , like B . mori ( Walters and Hardcastle , 2011 ) , Cydia pomonella ( Gu et al . , 2017 ) , and Heliconius butterflies ( Walters et al . , 2015 ) , compensates for Z chromosome dosage in the soma by reducing gene expression in males , but does not decrease Z-linked gene expression in germline tissues . Little is known about lepidopteran W chromosomes . The W chromosome is not included in the genome assembly of Manduca sexta ( Kanost et al . , 2016 ) or B . mori ( International Silkworm Genome Consortium , 2008 ) , and earlier efforts to assemble the silkworm W resulted in fragmented sequences containing transposons ( Abe et al . , 2005 , 2008; Kawaoka et al . , 2011 ) . The monarch genome scaffold continuity ( N50 = 0 . 207 Mb versus N50 = 14 . 2 Mb for T . ni; ( Zhan et al . , 2011 ) is insufficient to permit assembly of a W chromosome . Our genome assembly includes the 2 . 92 Mb T . ni W chromosome comprising 32 contigs ( contig N50 = 101 kb ) . In T . ni , W-linked contigs have higher repeat content , lower gene density , and lower transcriptional activity than autosomal or Z-linked contigs ( Figure 2B ) . Other lepidopteran W chromosomes are similarly enriched in repeats and depleted of genes ( Abe et al . , 2005; Fuková et al . , 2005; Traut et al . , 2007 ) . A search for T . ni genes that are homologous to insect sex determination pathway genes detected doublesex ( dsx ) , masculinizer ( masc ) , vitellogenin , transformer 2 , intersex , sex lethal , ovarian tumor , ovo , and sans fille . T . ni males produce a four-exon isoform of dsx , while females generate a six-exon dsx isoform ( Figure 2—figure supplement 1D ) . The Lepidoptera-specific gene masc encodes a CCCH zinc finger protein . masc is associated with the expression of the sex-specific isoforms of dsx in lepidopterans , including silkworm ( Katsuma et al . , 2015 ) . As in B . mori , T . ni masc lies next to the scap gene , supporting our annotation of T . ni masc . Lepidopteran masc genes are rapidly diverging and have low-sequence identity with one another ( 30 . 1% ) . Figure 2C shows the multiple sequence alignment of the CCCH zinc finger domain of Masc proteins from several lepidopteran species . Like many non-dipteran insects , T . ni has a single telomerase gene and telomeres containing TTAGG repeats ( Sahara et al . , 1999 ) . We found 40 ( TTAGG ) n stretches longer than 100 nt ( mean ± S . D . =600 ± 800 nt ) , nine at and 31 near contig boundaries ( Supplementary file 1F; distance between ( TTAGG ) n and contig boundary = 5000 ± 6000 nt for the 40 stretches ) , indicating that our assembly captures the sequences of many telomeres . More than half ( 59% ) of the sequences flanking the ( TTAGG ) n repeats are transposons , and ~49% of these belong to the non-long-terminal-repeat LINE/R1 family ( Supplementary file 1G ) . These telomeric and subtelomeric characteristics of T . ni resemble those of B . mori ( Fujiwara et al . , 2005 ) . Lepidopteran chromosomes generally lack a coherent , monocentric centromere and are instead holocentric or diffuse ( Labbé et al . , 2011 ) , and the silkworm , monarch butterfly , and diamondback moth genomes do not encode CenH3 , a protein associated with monocentric chromosomes . The T . ni genome similarly does not contain a gene for CenH3 , suggesting that its chromosomes are also holocentric . The T . ni genome is 35 . 6% GC , slightly less than B . mori ( 37 . 3% ) . The distributions of observed/expected CpG ratios in genes and across the genome ( Figure 2—figure supplement 2A ) reveal that T . ni is similar to other lepidopterans ( silkworm , monarch butterfly , diamondback moth ) and a coleopteran species ( red flour beetle , T . castaneum ) , but different from honeybee and fruit fly . The honeybee genome has a high CpG content in genes and exhibits a bimodal CpG distribution across the genome as a whole; the fruit fly genome is uniformly depleted of CpG dinucleotides . The differences in CpG patterns reflect the presence of both the DNMT1 and DNMT3 DNA methyltransferases in the honeybee , the absence of either in fruit fly , and the presence of only DNMT1 in T . ni , B . mori , D . plexippus , P . xylostella , and T . castaneum . Thus , like many other insects , the T . ni genome likely has low levels of DNA methylation ( Xiang et al . , 2010; Glastad et al . , 2011 ) . The T . ni genome contains 75 . 3 Mb of identifiable repeat elements ( 20 . 5% of the assembly ) , covering 458 repeat families ( Figure 2—figure supplement 2B , Supplementary file 1H ) . With this level of repeat content , T . ni fits well with the positive correlation between genome size and repeat content among lepidopteran genomes ( Figure 2—figure supplement 2C ) . The DNA transposon piggyBac was originally isolated from a T . ni cell line ( Fraser et al . , 1983 ) and transposes effectively in a variety of species ( Lobo et al . , 1999; Bonin and Mann , 2004; Wang et al . , 2008 ) . We identified 262 copies of piggyBac in the Hi5 cell genome assembly . The family divergence rate of piggyBac is ~0 . 17% , substantially lower than other transposon families in the genome ( Supplementary file 1I provides divergence rates for all transposon families ) . Among the individual piggyBac elements in the T . ni genome , 71 are specific to Hi5 cells . Compared to the 191 piggyBac insertions shared between T . ni and Hi5 cells ( divergence rate = 0 . 22% ) , the Hi5-cell-specific elements are more highly conserved ( divergence rate = 0 . 04% ) . We conclude that the piggyBac transposon entered the T . ni genome more recently than other transposons and , likely driven by the presence of many active piggyBac elements , expanded further during the immortalization of Hi5 cells in culture . miRNAs are ~22 nt non-coding RNAs that regulate mRNA stability and translation ( He and Hannon , 2004; Gao et al . , 2005 ) . In insects , miRNA targets function in metamorphosis , reproduction , diapause , and other pathways of insect physiology and development ( Lucas and Raikhel , 2013 ) . To characterize the T . ni miRNA pathway , we sequenced RNA and small RNA from ovary , testis , thorax , and Hi5 cells . Then , we manually identified miRNA biogenesis genes such as dcr-1 , pasha , drosha , and ago2 ( Supplementary file 2A ) and computationally predicted 295 miRNA genes ( Figure 3 , Supplementary file 3A and Supplementary file 4 ) , including 77 conserved , 31 Lepidoptera-specific , and 187 novel , T . ni-specific miRNAs . In thorax , 222 of 270 miRNAs had comparable abundance in males and females ( ≤2 fold difference or false discovery rate [FDR]≥0 . 1; Figure 3A ) . Of the 48 miRNAs having significantly different abundances in female and male thorax ( >2 fold difference and FDR < 0 . 1; Figure 3A ) , miR-1a , let-7 , and miR-278 were highly abundant ( >1000 parts per million [ppm] ) in either female or male thorax . miR-1a , a miRNA thought to be expressed in all animal muscle , was the most abundant miRNA in thorax in both sexes , but was 2 . 2-fold more abundant in males . miR-1 was previously shown to regulate muscle development in fruit flies ( Sokol and Ambros , 2005 ) and to increase when locusts transition from solitary to swarming ( Wei et al . , 2009 ) . T . ni let-7 , which has the same mature miRNA sequence as its D . melanogaster , C . elegans , and mammalian counterparts ( Lagos-Quintana et al . , 2001 ) was also more abundant in males , whereas miR-278 was 2 . 6-fold more abundant in females . let-7 may act in sex-specific pathways in metamorphosis ( Caygill and Johnston , 2008 ) , whereas miR-278 may play a sex-specific role in regulating energy homeostasis ( Teleman et al . , 2006 ) . A subset of less well-conserved miRNAs was also differentially expressed between male and female thorax . In general , poorly conserved miRNAs were less abundant: the median expression level for conserved miRNAs was 316 ppm , but only 161 ppm for Lepidoptera-specific and 4 . 22 ppm for T . ni-specific miRNAs . However , mir-2767 , a Lepidoptera-specific miRNA , and three T . ni-specific miRNAs ( mir-novel1 , mir-novel4 , mir-novel11 ) were both abundant ( >1000 ppm ) and differentially expressed in males and female thorax . We speculate that these recently evolved miRNAs may prove useful as targets for pest management . Ovary , testis , and Hi5 cells have distinct miRNA expression profiles . We analyzed the expression patterns of the 44 most abundant miRNAs ( Figure 3B and C ) , which explain 90% of miRNA reads in a tissue or cell line . Thirteen were expressed in ovaries , testes , and Hi5 cells . Of these 13 , 11 were significantly more abundant in testis , 5 in ovary , and 3 in Hi5 cells ( Figure 3B ) , suggesting that these miRNAs have important tissue- or cell-type-specific roles . miR-31 and miR-375 , highly expressed in T . ni testis , are both mammalian tumor suppressors ( Creighton et al . , 2010; Kinoshita et al . , 2012 ) . miR-989 , the most abundant miRNA in T . ni ovaries , plays an important role in border cell migration during Drosophila oogenesis ( Kugler et al . , 2013 ) . miR-10 , a miRNA in the Hox gene cluster , was preferentially expressed in Hi5 cells; its orthologs have been implicated in development and cancer ( Lund , 2010 ) , suggesting miR-10 played a role in the immortalization of the germline cells from which Hi5 cells derive . siRNAs , typically 20–22 nt long , regulate gene expression , defend against viral infection , and silence transposons ( Agrawal et al . , 2003; van Rij et al . , 2006; Sánchez-Vargas et al . , 2009; Tyler et al . , 2008; Tam et al . , 2008; Zambon et al . , 2006; Chung et al . , 2008; Okamura et al . , 2008b; Czech et al . , 2008; Okamura et al . , 2008b; Flynt et al . , 2009 ) . They are processed by Dicer from double-stranded RNAs or hairpins into short double-stranded fragments bearing two-nucleotide , overhanging 3′ ends , which are subsequently loaded into Argonaute proteins ( Bernstein et al . , 2001; Elbashir et al . , 2001; Siomi and Siomi , 2009 ) . siRNAs require extensive sequence complementarity to their targets to elicit Argonaute-catalyzed target cleavage . Endogenous siRNAs ( endo-siRNAs ) can derive from transposon RNAs , cis-natural antisense transcripts ( cis-NATs ) , and long hairpin RNAs ( Czech et al . , 2008; Ghildiyal et al . , 2008; Okamura et al . , 2008a; Chung et al . , 2008; Kawamura et al . , 2008; Okamura et al . , 2008a; Tam et al . , 2008; Watanabe et al . , 2008 ) ( hpRNAs ) . In T . ni ovary , testis , thorax , and Hi5 cells , 20 . 7–52 . 4% of siRNAs map to transposons , suggesting T . ni endogenous siRNAs suppress transposons in both the soma and the germline . Among the non-transposon siRNAs , <4 . 6% map to predicted hairpins , while 11 . 6–31 . 3% siRNAs map to cis-NATs ( Supplementary file 3B ) . Hi5 cells are latently infected with a positive-sense , bipartite alphanodavirus , TNCL virus ( Li et al . , 2007; Andrew Ball and Johnson , 1998 ) ( Tn5 Cell Line virus ) . We asked if TNCL virus RNA is present in our T . ni samples and whether the RNAi pathway provides antiviral defense via TNCL virus-derived siRNAs . We detected no viral RNA in the T . ni ovary , testis , or thorax transcriptome , but both TNCL virus RNA1 ( 5010 fragments per kilobase of transcript per million mapped reads [FPKM] ) and RNA2 ( 8280 FPKM ) were readily found in the Hi5 transcriptome ( Figure 4A ) . To test whether Hi5 cells mount an RNAi defense to TNCL virus infection , we mapped small RNA-seq reads that were not mappable to the T . ni genome to the two TNCL virus genomic segments . TNCL virus-mapping small RNAs showed a median length of 21 nt ( modal length = 20 nt; Figure 4A ) , typical for siRNAs , suggesting that the Hi5 RNAi pathway actively combats the virus . The TNCL virus-mapping small RNAs bear the two-nucleotide , 3′ overhanging ends that are the hallmark of siRNAs ( Figure 4B ) ( Elbashir et al . , 2001 ) . Moreover , the phased pattern of TNCL virus-mapping siRNAs suggests they are made one-after-another starting at the end of a dsRNA molecule: the distance between siRNA 5′ ends shows a periodicity of 20 nt , the length of a typical TNCL virus-mapping siRNA ( Figure 4C ) . In D . melanogaster , Dicer-2 processively produces siRNAs , using ATP energy to translocate along a dsRNA molecule ( Cenik et al . , 2011 ) . The phasing of anti-viral siRNAs in Hi5 cells suggests that T . ni Dicer-2 similarly generates multiple siRNAs from each molecule of dsRNA before dissociating . In addition to siRNAs , the TNCL-mapping small RNAs include some 23–32 nt RNAs . These are unlikely to be anti-viral piRNAs , because they lack the characteristic first-nucleotide uridine bias and show no significant ping-pong signal ( Z-score = −0 . 491 ) . We conclude that Hi5 cells do not use piRNAs for viral defense . The discovery that the 3′ ends of D . melanogaster siRNAs , but not miRNAs , are 2′-O-methylated ( Pélisson et al . , 2007 ) led to the idea that insects in general methylate both siRNAs and piRNAs . Resistance to oxidation by NaIO4 is the hallmark of 3′ terminal , 2′-O-methylation , and the enrichment of a small RNA in a high-throughput sequencing library prepared from NaIO4-treated RNA suggests 2′-O-methylation . Conversely , depletion of small RNAs , such as miRNAs , from such an oxidized RNA library is strong evidence for unmodified 2′ , 3′ vicinal hydroxyl groups . Surprisingly , TNCL virus-mapping siRNAs were 130-fold depleted from our oxidized small RNA-seq library ( 22 . 0 ppm ) compared to the unoxidized library ( 2870 ppm ) , suggesting that they are unmethylated . Sequencing of oxidized and unoxidized small RNA from T . ni ovary , testis , and thorax detected 20–22 nt peaks in unoxidized libraries; such peaks were absent from oxidized libraries ( Figure 4D ) , suggesting that T . ni genome-mapping , endogenous siRNAs also lack 2′-O-methylation . We conclude that both T . ni exo- and endo-siRNAs are not 2′-O-methyl modified . Are siRNAs unmethylated in other Lepidopteran species ? We sequenced oxidized and unoxidized small RNAs from two additional Lepidoptera: P . xylostella and B . mori . Like T . ni , siRNAs from these Lepidoptera were abundant in libraries prepared from unoxidized small RNA but depleted from oxidized libraries ( Figure 4—figure supplement 1A ) . The ratio of siRNAs in the oxidized library to siRNAs in the corresponding unoxidized library ( ox/unox ) provides a measure of siRNA 2′ , 3′ modification . For D . melanogaster siRNAs , the median ox/unox ratio was 1 . 00 , whereas the three Lepidoptera species had median ox/unox ratios between 0 . 17 and 0 . 22 ( Figure 4E ) , indicating their siRNAs were depleted from oxidized libraries and therefore bear unmodified 2′ , 3′ hydroxyl groups . We conclude that the last common ancestor of T . ni , B . mori , and P . xylostella , which diverged 170 mya , lacked the ability to 2′-O-methylate siRNA 3′ ends . We do not currently know whether the last common ancestor of Lepidoptera lost the capacity to methylate siRNAs or if some or all members of Diptera , the sister order of Lepidoptera , acquired this function , which is catalyzed by the piRNA-methylating enzyme Hen1 ( Saito et al . , 2007; Horwich et al . , 2007; Kirino and Mourelatos , 2007 ) . Terminal 2′ methylation of D . melanogaster siRNAs is thought to protect them from non-templated nucleotide addition ( tailing ) , 3′-to-5′ trimming , and wholesale degradation ( Ameres et al . , 2010 ) . Since T . ni siRNAs lack a 2′-O-methyl group at their 3′ ends , we first asked if we could observe frequent trimming by examining shorter TNCL-mapping siRNA ( 18–19 nt ) . These siRNAs account for 1 . 05% of all TNCL-mapping siRNAs . They did not possess the typical siRNA one-after-another pattern ( Z1 = −0 . 674 , p=0 . 500 ) , yet more than 97 . 5% of these were prefixes of longer , phased siRNAs , indicating that these were trimmed siRNAs . We conclude that TNCL siRNA trimming is rare in Hi5 cells . We next asked whether T . ni and other lepidopteran siRNAs have higher frequencies of tailing . Despite the lack of 2′-O-methylation , most TNCL virus siRNAs were not tailed: just 6 . 69% of all virus-mapping small RNA reads contained 3′ non-templated nucleotides ( Figure 4—figure supplement 1B ) . Among the 3′ non-templated nucleotides , the most frequent addition was one or more uridines ( 49 . 6% ) as observed previously for miRNAs and siRNAs in other animals ( Ameres et al . , 2010; Chou et al . , 2015 ) . Endogenous siRNA tailing frequencies for the lepidopterans T . ni ( 10 . 2% , ovary ) , B . mori ( 5 . 97% , eggs ) , and P . xylostella ( 8 . 58% , ovary ) were also similar to D . melanogaster ( 6 . 71% , ovary ) . We speculate that lepidopterans have other mechanisms to maintain siRNA stability or that trimming and tailing in lepidopterans are less efficient than in flies . siRNAs are non-randomly loaded into Argonaute proteins: the guide strand , the strand with the more weakly base paired 5′ end , is favored for loading ( Khvorova et al . , 2003; Schwarz et al . , 2003 ) ; the disfavored passenger strand is destroyed . Thus , loading skews the abundance of the two siRNA strands . To test if non-methylated siRNAs are loaded into Argonaute , we computationally paired single-stranded siRNAs that compose an siRNA duplex bearing two-nucleotide overhanging 3′ ends and calculated the relative abundance of the two siRNA strands . For TNCL-mapping siRNAs , 72 . 3% of siRNA duplexes had guide/passenger strand ratios ≥ 2 ( median = 3 . 90; mean = 10 . 2; Figure 4—figure supplement 2 ) . Among genome-mapping , 20–22 nt small RNAs 78 . 5% of duplexes had guide/passenger strand ratios ≥ 2 ( median 5 . 44; average 56 . 2 ) . We conclude that the majority of exogenous and endogenous siRNAs are loaded , presumably into Ago2 . In animals , piRNAs , ~23–32 nt long , protect the germline genome by suppressing the transcription or accumulation of transposon and repetitive RNA ( Girard et al . , 2006; Lau et al . , 2006; Vagin et al . , 2006; Brennecke et al . , 2007; Aravin et al . , 2007 ) . In D . melanogaster , dedicated transposon-rich loci ( piRNA clusters ) give rise to piRNA precursor transcripts , which are processed into piRNAs loaded into one of three PIWI proteins , Piwi , Aubergine ( Aub ) , or Argonaute3 ( Ago3 ) . Piwi acts in the nucleus to direct tri-methylation of histone H3 on lysine nine on transposon and repetitive genomic sequences ( Sienski et al . , 2012; Le Thomas et al . , 2014a , 2014b ) . In fly cytoplasm , piRNAs guide the Piwi paralog Aub to cleave transposon mRNAs . The mRNA cleavage products can then produce more piRNAs , which are loaded into Ago3 . In turn , these sense piRNAs direct Ago3 to cleave transcripts from piRNA clusters , generating additional piRNAs bound to Aub . The resulting ‘Ping-Pong’ feed-forward loop both amplifies piRNAs and represses transposon activity ( Brennecke et al . , 2007; Gunawardane et al . , 2007 ) . Finally , Ago3 cleavage not only produces Aub-bound piRNAs , but also initiates the production of Piwi-bound , phased piRNAs that diversify the piRNA pool ( Mohn et al . , 2015; Han et al . , 2015b ) . The T . ni genome contains a full repertoire of genes encoding piRNA pathway proteins ( Supplementary file 2B ) . These genes were expressed in both germline and somatic tissues , but were higher in ovary , testis , and Hi5 cells compared to thorax ( median ratios: ovary/thorax = 14 . 2 , testis/thorax = 2 . 9 , and Hi5/thorax = 4 . 9; Figure 5A ) . Expression of piRNA pathway genes in the Hi5 cell line suggests that it recapitulates the germline piRNA pathway . Although most T . ni piRNA pathway genes correspond directly to their D . melanogaster orthologs , T . ni encodes only two PIWI proteins , TnPiwi and TnAgo3 . The fly proteins Aub and Piwi are paralogs that arose from a single ancestral PIWI protein after the divergence of flies and mosquitos ( Lewis et al . , 2016 ) . We do not yet know whether TnPiwi functions more like Drosophila Aub or Piwi . In D . melanogaster , piRNA clusters—the genomic sources of most transposon-silencing germline piRNAs—are marked by the proteins Rhino , Cutoff , and Deadlock , which allow transcription of these heterochromatic loci ( Klattenhoff et al . , 2009; Pane et al . , 2011; Mohn et al . , 2014; Zhang et al . , 2014 ) . T . ni lacks detectable Rhino , Cutoff , and Deadlock orthologs . In fact , this trio of proteins is poorly conserved , and the mechanism by which they mark fly piRNA source loci may be unique to Drosophilids . In this regard , T . ni likely provides a more universal insect model for the mechanisms by which germ cells distinguish piRNA precursor RNAs from other protein-coding and non-coding transcripts . In both the germline and the soma , T . ni piRNAs originate from discrete genomic loci . To define these piRNA source loci , we employed an expectation-maximization algorithm that resolves piRNAs mapping to multiple genomic locations . Applying this method to multiple small RNA-seq datasets , we defined piRNA-producing loci comprising 10 . 7 Mb ( 348 clusters ) in ovary , 3 . 1 Mb ( 79 clusters ) in testis , 3 . 0 Mb ( 71 clusters ) in Hi5 cells , and 2 . 4 Mb ( 65 clusters ) in thorax ( Figure 5B ) . For each tissue or cell-type , these 393 clusters explain >70% of uniquely mapped piRNAs and >70% of all piRNAs when using expectation-maximization mapping . A core set of piRNA-producing loci comprising 1 . 5 Mb is active in both germline and somatic tissues . T . ni piRNA clusters vary substantially in size and expression level . In ovary , half the bases in piRNA clusters are in just 67 loci , with a median length of 53 kb . Among these , five span >200 kb , while the smallest is just 38 kb . The most productive piRNA source is a 264 kb locus on chromosome 13 ( Figure 5—figure supplement 1 ) ; 7 . 8% of uniquely mapped piRNAs—50 , 000 distinct piRNA sequences—reside in this locus . Collectively , the top 20 ovary piRNA loci explain half the uniquely mapped piRNAs , yet constitute only 0 . 7% of the genome . Globally , 61 . 9% of bases in piRNA clusters are repetitive , and 74 . 5% transposon-mapping piRNAs are antisense , suggesting that T . ni uses antisense piRNAs to suppress transposon transcripts . In the fly ovary germline , most piRNA clusters generate precursor RNAs from both DNA strands . These dual-strand clusters fuel the ‘Ping-Pong’ amplification cycle ( Brennecke et al . , 2007; Gunawardane et al . , 2007 ) . Other fly piRNA clusters , such as the paradigmatic flamenco gene ( Prud'homme et al . , 1995; Brennecke et al . , 2007; Pélisson et al . , 2007; Malone et al . , 2009; Goriaux et al . , 2014 ) are transcribed from one strand only and are organized to generate antisense piRNAs directly , without further Ping-Pong amplification ( Malone et al . , 2009 ) . These uni-strand clusters are the only sources of piRNAs in the follicle cells , somatic cells that support fly oocyte development and express only a single PIWI protein , Piwi ( Malone et al . , 2009 ) . The T . ni genome contains both dual- and uni-strand piRNA clusters . In ovary , 62 of 348 piRNA-producing loci are dual-strand ( Watson/Crick > 0 . 5 or Watson/Crick < 2 ) . These loci produce 35 . 9% of uniquely mapped piRNAs and 22 . 8% of all piRNAs; 71 . 6% of transposon-mapping piRNA reads from these loci are antisense . The remaining 286 uni-strand loci account for 54 . 8% of uniquely mapped piRNAs and 36 . 7% of all piRNAs . Most piRNAs ( 74 . 8% of reads ) from uni-strand clusters are antisense to transposons , the orientation required for repressing transposon mRNA accumulation . At least part of the piRNA antisense bias reflects positive selection for antisense insertions in uni-strand clusters: 57 . 1% of transposon insertions—79 . 7% of transposon-mapping nucleotides—are opposite the direction of piRNA precursor transcription , significantly different from dual-strand clusters , in which transposons are inserted randomly: 49 . 5% of transposon insertions in dual-strand clusters are in the antisense direction ( Figure 5—figure supplement 2A ) . For one 77 kb uni-strand cluster on chromosome 20 , 99 . 0% of piRNA reads ( 96% of piRNA sequences ) that can be uniquely assigned are from the Crick strand , while 67 . 6% of transposon insertions and 79 . 7% of transposon-mapping nucleotides at this locus lie on the Watson strand . The largest ovary cluster is a 462 kb W-linked region , consistent with our finding that the W chromosome is a major source of piRNAs ( Figure 5B and C and Figure 5—figure supplement 2B ) . Our data likely underestimates the length of this large piRNA cluster , as it is difficult to resolve reads mapping to its flanking regions: 70 . 8% of bases in the flanking regions do not permit piRNAs to map uniquely to the genome . In fact , 85 . 1% of the sequences between clusters on the W chromosome are not uniquely mappable . These gaps appear to reflect low mappability and not boundaries between discrete clusters . We propose that the W chromosome itself is a giant piRNA cluster . To further test this idea , we identified piRNA reads that uniquely map to one location among all contigs and measured their abundance per kilobase of the genome . W-linked contigs had a median piRNA abundance of 14 . 4 RPKM in ovaries , 379-fold higher than the median of all autosomal and Z-linked contigs , consistent with the view that almost the entire W chromosome produces piRNAs . In B . mori females , a plurality of piRNAs come from the W chromosome: ovary-enriched piRNAs often map to W-linked sequences , but not autosomes ( Kawaoka et al . , 2011 ) . Similarly , for T . ni , 27 . 2% of uniquely mapping ovary piRNAs derive from W-linked sequences , even though these contigs compose only 2 . 8% of the genome ( Figure 5C ) . The W chromosome may produce more piRNAs than our estimate , as the unassembled repetitive portions of the W chromosome likely also produce piRNAs . Thus , the entire W chromosome is a major source of piRNAs in T . ni ovaries ( Figure 5B ) . To our knowledge , the T . ni W chromosome is the first example of an entire chromosome devoted to piRNA production . To determine if there are W-linked regions devoid of piRNAs , we mapped all piRNAs to the W-linked contigs and found that 11 . 0% of the W-linked bases were not covered by any piRNAs , indicating at least part of the W chromosome does not produce any piRNAs . Next , we manually inspected 74 putative W-linked protein-coding genes and nine putative W-linked miRNAs . All nine W-linked miRNAs ( Figure 5B , Supplementary file 1J ) are T . ni-specific , and small RNAs mapping to these predicted miRNA loci showed significant ping-pong signature ( Z-score = 14 . 2 , p=1 . 81 × 10−45 ) , suggesting that these are likely piRNAs , not authentic miRNAs . For the putative protein-coding genes , we categorized them into orphan genes ( no homologs found ) , transposons ( good homology to transposons ) , uncharacterized/hypothetical proteins , and potential protein-coding genes with homology to the NCBI non-redundant protein sequences . We then asked whether piRNAs were produced from these genes ( Figure 5—figure supplement 2C ) . Among W-linked genes , those with transposon homology on average produced the most piRNAs ( 44 . 9 median ppm ) , whereas those with homology to annotated genes produced the fewest ( 9 . 81 median ppm ) . Some putative genes ( such as TNI001015 and TNI005339 ) produced no piRNAs at all . We conclude that although some W-linked loci do not produce piRNAs , nearly the entire W chromosome produces piRNAs . In contrast to the W chromosome , T . ni autosomes and the Z chromosome produce piRNAs from discrete loci—63 autosomal and 11 Z-linked contigs had piRNA levels > 10 rpkm . Few piRNAs are produced outside of these loci: for example , the median piRNA level across all autosomal and Z-linked contigs was ~0 in ovaries ( Figure 5—figure supplement 2B ) . In the T . ni germline , piRNA production from individual clusters varies widely , but the same five piRNA clusters produce the most piRNAs in ovary ( 34 . 9% of piRNAs ) , testis ( 49 . 3% ) , and Hi5 cells ( 44 . 0% ) , suggesting that they serve as master loci for germline transposon silencing . Other piRNA clusters show tissue-specific expression , with the W chromosome producing more piRNAs in ovary than in Hi5 cells , and three Z-linked clusters producing many more piRNAs in testis than in ovary ( 15 . 0–24 . 7 times more ) , even after accounting for the absence of dosage compensation in germline tissues ( Figure 6—figure supplement 1A ) . Hi5 cells are female , yet many piRNA-producing regions of the W chromosome that are active in the ovary produce few piRNAs in Hi5 cells ( Figure 6—figure supplement 1A ) . We do not know whether this reflects a reorganization of cluster expression upon Hi5 cell immortalization or if Hi5 cells correspond to a specific germ cell type that is underrepresented in whole ovaries . At least 40 loci produce piRNAs in Hi5 cells but not in ovaries . Comparison of DNA-seq data from T . ni and Hi5 identified 74 transposon insertions in 12 of the Hi5-specific piRNA clusters . Older transposons have more time to undergo sequence drift from the consensus sequence of the corresponding transposon family . The 74 Hi5-specific transposon insertions , which include both DNA and LTR transposons , had significantly lower divergence rates than those common to ovary and Hi5 cells ( Figure 6A ) , consistent with the idea that recent transposition events generated the novel piRNA clusters in Hi5 cells . We conclude that the Hi5-specific piRNA-producing loci are quite young , suggesting that T . ni and perhaps other lepidopterans can readily generate novel piRNA clusters . piRNA clusters active in thorax occupy ~0 . 57% of the genome and explain 86 . 8% of uniquely mapped somatic piRNAs in females and 89 . 5% in males . More than 90% of bases in clusters expressed in thorax are shared with clusters expressed in ovary ( Figure 6—figure supplement 1B ) . Such broadly expressed clusters explain 83 . 7% of uniquely mapping piRNAs in female thorax and 86 . 1% in male thorax . Thus , the majority of piRNAs in the T . ni soma come from clusters that are also active in the germline . In general , autosomal piRNA cluster expression is similar between female and male thorax , but 12 clusters are differentially expressed between male and female thorax . Of these , nine are W-linked clusters that produce significantly more piRNAs in female than in male thorax ( Figure 6B ) . In D . melanogaster , Rhino suppresses splicing of piRNA precursors transcribed from dual-strand piRNA clusters ( Mohn et al . , 2014; Zhang et al . , 2014 ) . Fly uni-strand piRNA clusters do not bind Rhino and behave like canonical RNA polymerase II transcribed genes ( Brennecke et al . , 2007; Goriaux et al . , 2014 ) . Although T . ni has no rhino ortholog , its piRNA precursor RNAs are rarely spliced as observed for clusters in flies . We identified splicing events in our RNA-seq data , requiring ≥10 reads that map across exon-exon junctions and a minimum splicing entropy of 2 to exclude PCR duplicates ( Graveley et al . , 2011 ) . This approach detected just 27 splice sites among all piRNA precursor transcripts from ovary , testis , thorax , and Hi5 piRNA clusters ( Figure 6C ) . Of these 27 splice sites , 19 fall in uni-strand piRNA clusters . We conclude that , as in flies , transcripts from T . ni dual-strand piRNAs clusters are rarely if ever spliced . Unlike flies ( Goriaux et al . , 2014 ) , RNA from T . ni uni-strand piRNA clusters also undergoes splicing infrequently . The absence of piRNA precursor splicing in dual-strand piRNA clusters could reflect an active suppression of the splicing machinery or a lack of splice sites . To distinguish between these two mechanisms , we predicted gene models for piRNA-producing loci , employing the same parameters used for protein-coding genes . For piRNA clusters , this approach generated 1332 gene models encoding polypeptides > 200 amino acids . These models comprise 2544 introns with consensus splicing signals ( Figure 6—figure supplement 1C ) . Notably , ~90% of these predicted gene models had high sequence similarity to transposon consensus sequences ( BLAST e-value <10–10 ) , indicating that many transposons in piRNA clusters have intact splice sites . We conclude that piRNA precursors contain splice sites , but their use is actively suppressed . To measure splicing efficiency , we calculated the ratio of spliced to unspliced reads for each predicted splice site in the piRNA clusters . High-confidence splice sites in protein-coding genes outside piRNA clusters served as a control . Compared to the control set of genes , splicing efficiency in piRNA loci was 9 . 67-fold lower in ovary , 2 . 41-fold lower in testis , 3 . 23-fold lower in thorax , and 17 . 0-fold lower in Hi5 cells ( Figure 6D ) , showing that T . ni piRNA precursor transcripts are rarely and inefficiently spliced . To test whether uni- and dual-strand piRNA cluster transcripts are differentially spliced in T . ni , we evaluated the experimentally supported splice sites from Hi5 , ovary , testis , and thorax collectively . Dual-strand cluster transcripts had 1 . 71-fold lower splicing efficiency compared to uni-strand clusters ( Figure 6D ) . Thus , T . ni suppresses splicing of dual- and uni-strand piRNA cluster transcripts by a mechanism distinct from the Rhino-dependent pathway in D . melanogaster . That this novel splicing suppression pathway is active in Hi5 cells should facilitate its molecular dissection . The study of arthropod piRNAs has been limited both by a lack of suitable cultured cell models and by the dominance of D . melanogaster as a piRNA model for arthropods generally . Although Vasa-positive D . melanogaster ovarian cells have been isolated and cultured ( Niki et al . , 2006 ) , no dipteran germ cell line is currently available . D . melanogaster somatic OSS , OSC and Kc167 cells produce piRNAs , but lack key features of the canonical germline pathway ( Lau et al . , 2009; Saito et al . , 2009; Vrettos et al . , 2017 ) . In addition to Hi5 cells , lepidopteran cell lines from Spodoptera frugiperda ( Sf9 ) and B . mori ( BmN4 ) produce germline piRNAs ( Kawaoka et al . , 2009 ) . The S . frugiperda genome remains a draft with 37 , 243 scaffolds and an N50 of 53 . 7 kb ( Kakumani et al . , 2014 ) . Currently , the BmN4 cell line is the only ex vivo model for invertebrate germline piRNA biogenesis and function . The B . mori genome sequence currently comprises 43 , 463 scaffolds with an N50 of 4 . 01 Mb ( International Silkworm Genome Consortium , 2008 ) . Unfortunately , BmN4 cells readily differentiate into two morphologically distinct cell types ( Iwanaga et al . , 2014 ) . Although genome editing with Cas9 has been demonstrated in BmN4 cells ( Zhu et al . , 2015 ) , no protocols for cloning individual , genome-modified BmN4 cells have been reported ( Mon et al . , 2004; Kawaoka et al . , 2009; Honda et al . , 2013 ) . In contrast , Hi5 cells are cultured using commercially available media , readily transfected , and , we report here , efficiently engineered with Cas9 and grown from single cells into clonal lines . The bacterial DNA nuclease Cas9 , targeted by a single guide RNA ( sgRNA ) , enables rapid and efficient genome editing in worms , flies , and mice , as well as in a variety of cultured animal cell lines ( Jinek et al . , 2012; Barrangou and Horvath , 2017; Komor et al . , 2017 ) . The site-specific double-strand DNA breaks catalyzed by Cas9 can be repaired by error-prone non-homologous end joining ( NHEJ ) , disrupting a protein-coding sequence or , when two sgRNAs are used , deleting a region of genomic DNA . Alternatively , homology-directed repair ( HDR ) using an exogenous DNA template allows the introduction of novel sequences , including fluorescent proteins or epitope tags , as well as point mutations in individual genes ( Cong et al . , 2013 ) . As a proof-of-concept , we used Cas9 and two sgRNAs to generate a deletion in the piRNA pathway gene TnPiwi . The two sgRNAs , whose target sites lie 881 bp apart ( Figure 7A ) , were transcribed in vitro , loaded into purified , recombinant Cas9 protein , and the resulting sgRNA/Cas9 ribonucleoprotein complexes ( RNPs ) transfected into Hi5 cells . PCR of genomic DNA isolated 48 hr later was used to detect alterations in the TnPiwi gene . A novel PCR product , ~900 bp smaller than the product amplified using DNA from control cells , indicated that the desired deletion had been created ( Figure 7B ) . Sanger sequencing of the PCR products confirmed deletion of 881–896 bp from the TnPiwi gene . The presence of indels—short deletions and non-templated nucleotide additions—at the deletion junction is consistent with a Cas9-mediated dsDNA break having been repaired by NHEJ ( Figure 7A ) . We note that these cells still contain at least one wild-type copy of TnPiwi . We have not yet obtained cells in which all four copies of TnPiwi are disrupted , perhaps because in the absence of Piwi , Hi5 cells are inviable . To test whether an exogenous donor DNA could facilitate the site-specific incorporation of protein tag sequences into Hi5 genome , we designed two sgRNAs with target sites ~ 90 bp apart , flanking the vasa start codon ( Figure 7C ) . As a donor , we used a single-stranded DNA ( ssDNA ) encoding EGFP and an HA epitope tag flanked by genomic sequences 787 bp upstream and 768 bp downstream of the vasa start codon ( Figure 7C ) . Cas9 and the two sgRNAs were cotransfected with the ssDNA donor , and , 1 week later , EGFP-positive cells were detected by fluorescence microscopy . PCR amplification of the targeted region using genomic DNA from EGFP-expressing cells confirmed integration of EGFP and the HA tag into the vasa gene ( Figure 7D ) . Sanger sequencing further confirmed integration of EGFP and the HA tag in-frame with the vasa open-reading frame ( Supplemental file 9 ) . To establish a clonal line from the EGFP-HA-tagged Vasa-expressing cells , individual EGFP-positive cells were isolated by FACS and cultured on selectively permeable filters above a feeder layer of wild-type Hi5 cells ( Figure 8A ) . Growth of the genome-modified single cells required live Hi5 feeder cells—conditioned media did not suffice—presumably because the feeder cells provide short-lived growth factors or other trophic molecules . Single EGFP-positive clones developed 1 month after seeding and could be further grown without feeder cells as a clonally derived cell line ( Figure 8B ) . In the germline of D . melanogaster and other species , components of the piRNA biogenesis pathway , including Vasa , Aub , Ago3 , and multiple Tudor-domain proteins , localize to a perinuclear structure called nuage ( Eddy , 1975; Findley et al . , 2003; Lim and Kai , 2007; Li et al . , 2009; Liu et al . , 2011a; Webster et al . , 2015 ) . Vasa , a germline-specific nuage component , is widely used as a marker for nuage . In BmN4 cells , transiently transfected Vasa localizes to a perinuclear structure resembling nuage ( Xiol et al . , 2012; Patil et al . , 2017 ) . To determine whether nuage-like structures are present in Hi5 cells , we examined Vasa localization in the Hi5 cells in which the endogenous vasa gene was engineered to fuse EGFP and an HA epitope tag to the Vasa amino-terminus . We used two different immunostaining strategies to detect the EGFP-HA-Vasa fusion protein: a mouse monoclonal anti-GFP antibody and a rabbit monoclonal anti-HA antibody . GFP and HA colocalized in a perinuclear structure , consistent with Vasa localizing to nuage in Hi5 cells ( Figure 8C ) . Using Hi5 cells , we have sequenced and assembled the genome of the cabbage looper , T . ni , a common and destructive agricultural pest that feeds on many plants of economic importance . Examination of the T . ni genome and transcriptome reveals the expansion of detoxification-related gene families ( Table 1 and Supplementary file 6 ) , many members of which are implicated in insecticide resistance and are potential targets of pest control . The T . ni genome should enable study of the genetic diversity and population structure of this generalist pest , which adapts to different environmental niches worldwide . Moreover , as the sister order of Diptera , Lepidoptera like T . ni provide a counterpoint for the well-studied insect model D . melanogaster . The use of Hi-C sequencing was an essential step in assembling the final 368 . 2 Mb T . ni genome into high-quality , chromosome-length scaffolds . The integration of long reads , short reads , and Hi-C provides a rapid and efficient paradigm for generating chromosome-level assemblies of other animal genomes . This strategy assembled the gene-poor , repeat-rich T . ni W chromosome , which is , to our knowledge , the first chromosome-level sequence of a lepidopteran W chromosome . Our analysis of autosomal , Z-linked , and W-linked transcripts provides insights into lepidopteran dosage compensation and sex determination . Our data show that T . ni compensates for Z chromosome dosage in the soma by reducing transcription of both Z homologs in males , but Z dosage is uncompensated in the germline . In addition to long RNAs , we characterized miRNAs , siRNAs , and piRNAs in T . ni gonads , soma , and cultured Hi5 cells . miRNAs are widely expressed in T . ni tissues , providing examples of germline-enriched and somatic miRNAs , as well as highly conserved , lepidopteran-specific , and novel T . ni miRNAs . Like flies , T . ni possess siRNAs that map to transposons , cis-NATs and hpRNAs . Unexpectedly , T . ni siRNAs—and likely all lepidopteran siRNAs—lack a 2′-O-methyl modification at their 3′ ends , unlike siRNAs in D . melanogaster . Consistent with siRNA production by a processive Dicer-2 enzyme , Hi5 cells produce phased siRNAs from the RNA genome of a latent alphanodavirus . The commonalities and differences between T . ni and D . melanogaster small RNA pathways will help identify both deeply conserved and rapidly evolving components . A major motivation for sequencing the T . ni genome was the establishment of a tractable cell culture model for studying small RNAs , especially piRNAs . We believe that our genome assembly and gene-editing protocols will enable the use of T . ni Hi5 cells to advance our understanding of how piRNA precursors are defined , made into piRNAs and act to silence transposons in the germline . Hi5 cells express essentially all known piRNA pathway genes except those specific to Drosophilids . Furthermore , T . ni Vasa localizes to a perinuclear , nuage-like structure in Hi5 cells , making them suitable for studying the assembly of the subcellular structures thought to organize piRNA biogenesis . We have defined genomic piRNA-producing loci in Hi5 cells , as well as in the soma , testis , and ovary . The most productive piRNA clusters are shared among ovary , testis , and Hi5 cells . In addition , Hi5 cells contain novel piRNA clusters not found in the moth itself , suggesting that the process of establishing new piRNA-producing loci can be recapitulated by experimental manipulation of Hi5 cells . As in D . melanogaster , splicing of T . ni piRNA precursor transcripts is efficiently suppressed , yet T . ni lacks paralogs of the proteins implicated in splicing suppression in flies . The ability to study the mechanisms by which piRNA clusters form and how precursor RNAs are transcribed , exported , and marked for piRNA production in T . ni promises to reveal both conserved and lepidopteran-specific features of this pathway . Notably , the W chromosome not only is a major piRNA source , but also produces piRNAs from almost its entirety . Future studies are needed to determine whether this is a common feature of W chromosomes in Lepidoptera and other insects . The establishment of procedures for genome editing and single-cell cloning of Hi5 cells , combined with the T . ni genome sequence , make this germ cell line a powerful tool to study RNA and protein function ex vivo . Our strategy combines transfection of pre-assembled Cas9/sgRNA complexes with single clone isolation using a selectable marker ( e . g . EGFP ) and feeder cells physically separated from the engineered cells . Compared with nucleic-acid-based delivery of Cas9 , transfection of Cas9 RNP minimizes the off-target mutations caused by prolonged Cas9 expression and eliminates the risk of integration of sgRNA or Cas9 sequences into the genome ( Lin et al . , 2014; Kim et al . , 2014 ) . Compared to plasmid donors ( Yu et al . , 2014; Ge et al . , 2016 ) , ssDNA homology donors similarly reduce the chance of introducing exogenous sequences at unintended genomic sites . Techniques for injecting the embryos of other lepidopteran species have already been established ( Wang et al . , 2013; Takasu et al . , 2014; Zhang et al . , 2015 ) . In principle , Cas9 RNP injected into cabbage looper embryos could be used to generate genetically modified T . ni strains both to explore lepidopteran biology and to implement novel strategies for safe and effective pest control . Hi5 cells ( ThermoFisher , Waltham , MA ) were cultured at 27°C in Express Five Serum Free Medium ( ThermoFisher ) following the manufacturer’s protocol . Thorax were dissected from four-day-old female or male T . ni pupa ( Benzon Research , Carlisle , PA ) . Cells or tissues were lysed in 2 × PK buffer ( 200 mM Tris-HCl [pH7 . 5] , 300 mM NaCl , 25 mM EDTA , 2% w/v SDS ) containing 200 μg/ml proteinase K at 65°C for 1 hr , extracted with phenol:chloroform:isoamyl alcohol ( 25:24:1; Sigma , St . Louis , MO ) , and genomic DNA collected by ethanol precipitation . The precipitate was dissolved in 10 mM Tris-HCl ( pH 8 . 0 ) , 0 . 1 mM EDTA , treated with 20 μg/ml RNase A at 37°C for 30 min , extracted with phenol:chloroform:isoamyl alcohol ( 25:24:1 ) , and collected by ethanol precipitation . DNA concentration was determined ( Qubit dsDNA HS Assay , ThermoFisher ) . Genomic DNA libraries were prepared from 1 µg genomic DNA ( Illumina TruSeq LT kit , NextSeq 500 , Illumina , San Diego , CA ) . Long-read genome sequencing with a 23 kb average insert range was constructed from 16 µg genomic DNA using the SMRTbell Template Prep Kit 1 . 0 SPv3 ( Pacific Biosciences , Menlo Park , CA ) according to manufacturer’s protocol . Sequence analysis was performed using P6/C4 chemistry , 240 min data collection per SMRTcell on an RS II instrument ( Pacific Biosciences ) . Mate pair libraries with 2 kb and 8 kb insert sizes were constructed ( Nextera Mate Pair Library Prep Kit , Illumina ) according to manufacturer’s protocol from 1 µg Hi5 cell genomic DNA . Libraries were sequenced to obtain 79 nt paired-end reads ( NextSeq500 , Illumina ) . Hi-C libraries were generated from Hi5 cells as described ( Belton et al . , 2012 ) , except that 50 million cells were used . Hi-C Libraries were sequenced using the NextSeq500 platform ( Illumina ) to obtain 79 nt , paired-end reads . Hi5 cells were first incubated in Express Five medium containing 1 µg/ml colcemid at 27°C for 8 hr ( Schneider , 1979 ) , then in 4 ml 0 . 075 M KCl for 30 min at 37°C , and fixed with freshly prepared methanol:acetic acid ( 3:1 , v/v ) precooled to −20°C . Mitotic chromosomes were spread , mounted by incubation in ProLong Gold Antifade Mountant with DAPI ( 4ʹ , 6ʹ-diamidino-2-phenylindole; ThermoFisher ) overnight in the dark , and imaged using a DMi8 fluorescence microscope equipped with an 63 × 1 . 40 N . A . oil immersion objective ( HCX PL APO CS2 , Leica Microsystems , Buffalo Grove , IL ) as described ( Matijasevic et al . , 2008 ) . Ovaries , testes , and thoraces were dissected from cabbage looper adults 24–48 hr after emerging . Total RNA ( 30 µg ) was isolated ( mirVana miRNA isolation kit , Ambion , Austin , TX ) and sequenced using the NextSeq500 platform ( Illumina ) to obtain 59 nt single-end reads as previously described ( Han et al . , 2015b ) . Adult ovaries , testes , or thoraces were dissected from cabbage looper adults 24 to 48 hr after emerging . Total RNA ( 3 µg ) was purified ( mirVana miRNA isolation kit , Ambion ) and sequenced as described ( Zhang et al . , 2012 ) using the NextSeq500 platform ( Illumina ) to obtain 79 nt , paired-end reads . Canu v1 . 3 ( Koren et al . , 2017 ) was used to assemble long reads into contigs , followed by Quiver ( github . com/PacificBiosciences/GenomicConsensus ) to polish the contigs using the same set of reads . Pilon ( Walker et al . , 2014 ) was used to further polish the assembly using Illumina paired-end reads . Finally , to assemble the genome into chromosome-length scaffolds , we joined the contigs using Hi-C reads and LACHESIS ( Burton et al . , 2013 ) . The mitochondrial genome was assembled separately using MITObim ( six iterations , D . melanogaster mitochondrial genome as bait; ( Hahn et al . , 2013 ) . To evaluate the quality of the genome assembly , we ran BUSCO v3 ( Simão et al . , 2015 ) using the arthropod profile and default parameters to identify universal single-copy orthologs . We further evaluated genome quality using conserved gene sets: OXPHOS and CRP genes . B . mori and D . melanogaster OXPHOS and CRP protein sequences were retrieved ( Marygold et al . , 2007; Porcelli et al . , 2007 ) and BLASTp was used to search for their T . ni homologs , which were further validated by querying using InterPro ( Jones et al . , 2014; Mitchell et al . , 2015 ) . We also assembled T . ni genomes from male and female animals respectively using SOAPdenovo2 ( kmer size 69; ( Luo et al . , 2012 ) . We then compared the animal genomes with the T . ni genome assembled from Hi5 cells using QUAST ( -m 500 ) ( Gurevich et al . , 2013 ) and the nucmer and mummerplot ( --layout --filter ) functions from MUMmer 3 . 23 ( Kurtz et al . , 2004 ) . To determine the genomic variants , we used HaplotypeCaller from GATK ( McKenna et al . , 2010; DePristo et al . , 2011; Van der Auwera et al . , 2013 ) ( -ploidy 4 -genotyping_mode DISCOVERY’ ) . To annotate the T . ni genome , we first masked repetitive sequences and then integrated multiple sources of evidence to predict gene models . We used RepeatModeler to define repeat consensus sequences and RepeatMasker ( -s -e ncbi ) to mask repetitive regions ( Smit et al . , 2017 ) . We used RNAmmer ( Lagesen et al . , 2007 ) to predict 8S , 18S , 28S rRNA genes , and Barrnap ( https://github . com/tseemann/barrnap ) to predict 5 . 8S rRNA genes . We used Augustus v3 . 2 . 2 ( Stanke et al . , 2006 ) and SNAP ( Korf , 2004 ) to computationally predicted gene models . Predicted gene models were compiled by running six iterations of MAKER ( Campbell et al . , 2014 ) , aided with homology evidence of well annotated genes ( UniProtKB/Swiss-Prot and Ensembl ) and of transcripts from related species ( B . mori ( Suetsugu et al . , 2013 ) and D . melanogaster ( Attrill et al . , 2016 ) . We used BLAST2GO ( Conesa et al . , 2005 ) to integrate results from BLAST , and InterPro ( Mitchell et al . , 2015 ) to assign GO terms to each gene . We used MITOS ( Bernt et al . , 2013 ) web server to predict mitochondrial genes and WebApollo ( Lee et al . , 2013 ) for manual curation of genes of interest . To characterize telomeres , we used ( TTAGG ) 200 ( Robertson and Gordon , 2006 ) as the query to search the T . ni genome using BLASTn with the option ‘-dust no’ and kept hits longer than 100 nt . The genomic coordinates of these hits were extended by 10 kb to obtain the subtelomeric region . To place genes into ortholog groups , we compared the predicted proteomes from 21 species ( Supplementary file 5 ) . Orthology assignment was determined using OrthoMCL ( Hirose and Manley , 1997 ) with default parameters . MUSCLE v3 . 8 . 31 ( Edgar , 2004 ) was used for strict 1:1:1 orthologs ( n = 381 ) to produce sequence alignments . Conserved blocks ( 66 , 044 amino acids in total ) of these alignments were extracted using Gblocks v0 . 91b ( Castresana , 2000 ) with default parameters , and fed into PhyML 3 . 0 ( Vastenhouw et al . , 2010 ) ( maximum likelihood , bootstrap value set to 1000 ) to calculate a phylogenetic tree . The human and mouse predicted proteomes were used as an outgroup to root the tree . The tree was viewed using FigTree ( http://tree . bio . ed . ac . uk/software/figtree/ ) and iTOL ( Shirayama et al . , 2012 ) . To identify sex-linked contigs , we mapped genomic sequence reads from males and females to the contigs . Reads with MAPQ scores ≥ 20 were used to calculate contig coverage , which was then normalized by the median coverage . The distribution of normalized contig coverage ratios ( male:female ratios , M:F ratios ) was manually checked to empirically determine the thresholds for Z-linked and W-linked contigs ( M:F ratio >1 . 5 for Z-linked contigs and M:F ratio <0 . 5 for W-linked contigs ) . Lepidopteran masc genes were obtained from Lepbase ( Challis et al . , 2016 ) . Z/AA ratio was calculated according to ( Gu et al . , 2017 ) . To curate genes related to detoxification and chemoreception , we obtained seed alignments from Pfam ( Finn et al . , 2016 ) and ran hmmbuild to build HMM profiles of cytochrome P450 ( P450 ) , amino- and carboxy-termini of glutathione-S-transferase ( GST ) , carboxylesterase ( COE ) , ATP-binding cassette transporter ( ABCs ) , olfactory receptor ( OR ) , gustatory receptor ( GR ) , ionotropic receptor ( IR ) , and odorant binding ( OBP ) proteins , ( Supplementary file 6 , 7 and 8 ) . We then used these HMM profiles to search for gene models in the predicted T . ni proteome ( hmmsearch , e-value cutoff: 1 × 10−5 ) . We also retrieved reference sequences of P450 , GST , COE , ABC , OR , GR , IR , OBP , and juvenile hormone pathway genes from the literature ( Hekmat-Scafe et al . , 2002; Bellés et al . , 2005; Wanner and Robertson , 2008; Yu et al . , 2008; Benton et al . , 2009; Gong et al . , 2009; Yu et al . , 2009; Croset et al . , 2010; Ai et al . , 2011; Liu et al . , 2011b; Dermauw and Van Leeuwen , 2014; Goodman and Granger , 2005; van Schooten et al . , 2016 ) . These were aligned to the T . ni genome using tBLASTx ( Altschul et al . , 1990 ) and Exonerate ( Slater and Birney , 2005 ) to search for homologs . Hits were manually inspected to ensure compatibility with RNA-seq data , predicted gene models , known protein domains ( using CDD ( Marchler-Bauer et al . , 2015 ) and homologs from other species . P450 genes were submitted to David Nelson’s Cytochrome P450 Homepage ( Nelson , 2009 ) for nomenclature and classification . Sequences and statistics of these genes are in Supplementary files 6 , 7 and 8 . To determine the phylogeny of these gene families , we aligned the putative protein sequences from T . ni and B . mori genomes using MUSCLE ( Edgar , 2004 ) , trimmed the multiple sequence alignments using TrimAl ( Capella-Gutiérrez et al . , 2009 ) ( with the option -automated1 ) , and performed phylogenetic analysis ( PhyML 3 . 0 ( Vastenhouw et al . , 2010 ) , with parameters: -q --datatype aa --run_id 0 --no_memory_check -b −2 ) . Phylogenetic trees were visualized using FigTree ( http://tree . bio . ed . ac . uk/software/figtree/ ) . To curate opsin genes , we used opsin mRNA and peptide sequences from other species ( Zimyanin et al . , 2008; Futahashi et al . , 2015 ) to search for homologs in T . ni . To discriminate opsin genes from other G-protein-coupled receptors , we required that the top hit in the NCBI non-redundant database and UniProt were opsins . To determine transposon age , we calculated the average percent divergence for each transposon family: the percent divergence ( RepeatMasker ) of each transposon copy was multiplied by its length , and the sum of all copies were divided by the sum of lengths of all copies in the family ( Pace and Feschotte , 2007 ) . We used TEMP ( Zhuang et al . , 2014 ) to identify transposon insertions in the Hi5 genome . mirDeep2 ( Friedländer et al . , 2008 , 2012 ) with default parameters predicted miRNA genes . Predicted miRNA hairpins were required to have homology ( exact seed matches and BLASTn e-value <1 × 10−5 ) to known miRNAs and/or miRDeep2 scores ≥ 10 . miRNAs were named according to exact seed matches and high sequence identities ( BLASTn e-value <1 × 10−5 ) with known miRNA hairpins . To determine the conservation status of T . ni miRNAs , putative T . ni miRNAs were compared with annotated miRNAs from A . aegypti , A . mellifera , B . mori , D . melanogaster , H . sapiens , M . musculus , M . sexta , P . xylostella , and T . castaneum: conserved miRNAs were required to have homologous miRNAs beyond Lepidoptera . To compare siRNA abundance in oxidized and unoxidized small RNA-seq libraries , we normalized siRNA read counts to piRNA cluster-mapping reads ( piRNA cluster read counts had >0 . 98 Pearson correlation coefficients between oxidized and unoxidized libraries in all cases ) . Because piRNA degradation products can be 20–22 nt long , we excluded potential siRNA species that were prefixes of piRNAs ( 23–35 nt ) . To search for viral transcripts in T . ni , we downloaded viral protein sequences from NCBI ( http://www . ncbi . nlm . nih . gov/genome/viruses/ ) and used using tBLASTn to map them to the T . ni genome and to the transcriptomes of Hi5 cells and five T . ni tissues . We filtered hits ( percent identity ≥0 . 80 , e-val ≤1 × 10−20 , and alignment length ≥100 ) and mapped small RNA-seq reads to the identified viral transcripts . Candidate genomic hairpins were defined according to Okamura et al . ( 2008b ) ) . Candidate cis-NATs were defined according to ( Ghildiyal et al . , 2008 ) . To determine the genomic coordinates of piRNA-producing loci , we mapped small RNAs to the genome as described ( Han et al . , 2015a ) . We then calculated the abundance of piRNAs in 5 kb genomic windows . For each window , we counted the number of uniquely mapped reads and the number of reads mapped to multiple loci ( multimappers ) by assigning reads using an expectation-maximization algorithm . Briefly , each window had the same initial weight . The weight was used to linearly apportion multimappers . During the expectation ( E ) step , uniquely mapped reads were unambiguously assigned to genomic windows; multimappers were apportioned to the genomic windows they mapped to , according to the weights of these windows . At the maximization ( M ) step , window weights were updated to reflect the number of reads each window contained from the E step . The E and M steps were run iteratively until the Manhattan distance between two consecutive iterations was smaller than 0 . 1% of the total number of reads . To identify differentially expressed piRNA loci , we used the ppm and rpkm values , normalized to the total number of uniquely mapped reads , to measure piRNA abundance . For analyses including all mapped reads ( uniquely mapped reads and multimappers ) , reads were apportioned by the number of times that they were mapped to the genome . To make piRNA loci comparable across tissues , we merged piRNA loci from ovary , testis , female and male thorax , and Hi5 cells . For the comparison between female and male thoraces , the cluster on tig00001980 was removed as this cluster likely corresponds to a mis-assembly . We used Spearman correlations to calculate the pairwise correlations of piRNA abundances . As for defining sex-linked contigs , we calculated M:F ratios and used the same thresholds to determine whether a piRNA cluster was sex-linked . A piRNA locus was considered to be differentially expressed if the ratio between the two tissues was >2 or<0 . 5 and FDR < 0 . 1 ( after t-test ) . Splice sites were deemed to be supported by RNA-seq data when supported by at least one data set . We used AUGUSTUS ( Stanke et al . , 2006 ) , with the model trained for T . ni genome-wide gene prediction , to predict gene models and their splice sites in T . ni piRNA clusters . Total RNAs were extracted from Hi5 cells using mirVana kit as described previously . We then incubated 100 µg total RNA with 25 mM NaIO4 in borate buffer ( 148 mM Borax , 148 mM Boric acid , pH 8 . 6 ) for 30 min at room temperature , beta-elimination was performed in 50 mM NaOH at 45°C for 90 min ( Horwich et al . , 2007 ) . The resultant RNA was collected by ethanol precipitation . sgRNAs for the target loci ( 5′-end of TnPiwi and 5′-end of vasa ) were designed using crispr . mit . edu ( Hsu et al . , 2013 ) to retrieve all possible guide sequences , and guide sequences adjacent to deletion or insertion targets were chosen . Supplementary file 9 lists guide sequences . Donor template sequence was produced as a gBlock ( Integrated DNA Technologies , San Diego , CA ) . A biotinylated forward primer and a standard reverse primer were used in PCR to generate a double-stranded , biotinylated DNA donor . The biotinylated DNA was captured on M-280 streptavidin Dynabeads ( ThermoFisher ) , and the biotinylated strand was separated from the non-biotinylated strand essentially as described in the manufacturer’s protocol . Supplemental file 10 provides a detailed protocol . sgRNAs were transcribed using T7 RNA polymerase , gel purified , then incubated with Cas9 in serum-free Hi5 culture medium supplemented with 18 mM l-glutamine . The resulting sgRNA/Cas9 RNPs were incubated with Trans-IT insect reagent ( Mirus Bio , Madison , WI ) for 15 min at room temperature , then evenly distributed onto 90% confluent Hi5 cells . Culture medium was replaced with fresh medium 12 hr later . Genomic DNA was isolated and analyzed by PCR 48 hr later . Forty eight hours after transfection , Hi5 cells from one 90% confluent well of a six-well plate ( Corning , Corning , NY ) were collected , washed once with PBS ( ThermoFisher ) and lysed in 2 × PK buffer containing 200 μg/ml proteinase K , extracted with phenol:chloroform:isoamyl alcohol ( 25:24:1 ) , and then genomic DNA collected by ethanol precipitation . Deletions in TnPiwi were detected by PCR using primers flanking the deleted region ( Supplementary file 9 ) . To confirm deletions by sequencing PCR products were resolved by agarose gel electrophoresis , purified ( QIAquick Gel Extraction Kit , QIAGEN , Germantown , MD , USA ) , and cloned into pCR-Blunt II-Topo vector ( ThermoFisher ) . The recombinant plasmid was transformed into Top10 competent E . coli ( ThermoFisher ) following supplier’s protocol . PCR products amplified using M13 ( −20 ) forward and M13 reverse primers from a sample of a single bacterial colony were sequenced by GENEWIZ ( South Plainfield , NJ ) . Wild-type Hi5 cells were seeded into a 96-well Transwell permeable support receiver plate ( Corning , Corning , NY ) at 30% confluence and incubated overnight in serum free medium with 100 U/ml penicillin and 100 μg/ml streptomycin . A Transwell permeable support insert plate with media in each well was inserted into the receiver plate , and a single EGFP-positive cell was sorted into each insert well by FACS . After 14 days incubation at 27°C , wells were examined for EGFP-positive cell clones using a DMi8 fluorescent microscope ( Leica ) . EGFP-HA-Vasa-expressing Hi5 cells were seeded on 22 × 22 mm cover slips ( Fisher Scientific , Pittsburgh , PA ) in a well of a six-well plate ( Corning ) . After cells had attached to the coverslip , the medium was removed and cells were washed three times with PBS ( Gibco ) . Cells were fixed in 4% ( w/v ) methanol-free formaldehyde ( ThermoFisher ) in PBS at room temperature for 15 min , washed three times with PBS , permeabilized with 0 . 1% ( w/v ) Triton X-100 in PBS for 15 min at room temperature , and then washed three times with PBS . For antibody labeling , cells were incubated in 0 . 4% ( v/v ) Photo-Flo in 1 × PBS for 10 min at room temperature , then 10 min in 0 . 1% ( w/v ) Triton X-100 in PBS and 10 min in 1 × ADB PBS ( 3 mg/ml bovine serum albumen , 1% ( v/v ) donkey serum , 0 . 005% ( w/v ) Triton X-100 in 1 × PBS ) . Next , cells were incubated with primary antibodies ( mouse anti-GFP antibody ( GFP-1D2 , Developmental Studies Hybridoma Bank , Iowa City , IA ) and rabbit anti-HA Tag antibody ( C29F4 , Cell Signaling , Danvers , MA ) , diluted 1:200 in ADB ( 30 mg/ml BSA , 10% ( v/v ) donkey serum , 0 . 05% ( w/v ) Triton X-100 in 1 × PBS ) at 4°C overnight . After three washes in PBS , cells were incubated sequentially in 0 . 4% ( v/v ) Photo-Flo in 1 × PBS , 0 . 1% ( w/v ) Triton X-100 in PBS , and 1 × ADB PBS , each for 10 min at room temperature . Cells were then incubated with secondary Alexa Fluor 488-labeled donkey anti-mouse ( ThermoFisher ) and Alexa Fluor 680-labeled donkey anti-rabbit ( ThermoFisher ) antibodies , diluted 1:500 in ADB at room temperature for one hour . After washing three times with 0 . 4% ( v/v ) Photo-Flo in 1 × PBS and once with 0 . 4% ( v/v ) Photo-Flo in water , coverslips were air dried in the dark at room temperature . Slides were mounted in ProLong Gold Antifade Mountant with DAPI and examined by confocal microscopy ( TCS SP5 II Laser Scanning Confocal , Leica ) . The T . ni Whole Genome Shotgun project has been deposited at DDBJ/ENA/GenBank under the accession NKQN00000000 . The version described here is version NKQN01000000 . All sequencing data are available through the NCBI Sequence Read Archive under the accession number PRJNA336361 . Further details are available at the Cabbage Looper Database ( http://cabbagelooper . org/ ) .	A common moth called the cabbage looper is becoming increasingly relevant to the scientific community . Its caterpillars are a serious threat to cabbage , broccoli and cauliflower crops , and they have started to resist the pesticides normally used to control them . Moreover , the insect’s germline cells – the ones that will produce sperm and eggs – are used in laboratories as ‘factories’ to artificially produce proteins of interest . The germline cells also host a group of genetic mechanisms called RNA silencing . One of these processes is known as piRNA , and it protects the genome against ‘jumping genes’ . These genetic elements can cause mutations by moving from place to place in the DNA: in germline cells , piRNA suppresses them before the genetic information is transmitted to the next generation . Not all germline cells grow equally well under experimental conditions , or are easy to use to examine piRNA mechanisms in a laboratory . The germline cells from the cabbage looper , on the other hand , have certain characteristics that would make them ideal to study piRNA in insects . However , the genome of the moth had not yet been fully resolved . This hinders research on new ways of controlling the pest , on how to use the germline cells to produce more useful proteins , or on piRNA . Decoding a genome requires several steps . First , the entire genetic information is broken in short sections that can then be deciphered . Next , these segments need to be ‘assembled’ – put together , and in the right order , to reconstitute the entire genome . Certain portions of the genome , which are formed of repeats of the same sections , can be difficult to assemble . Finally , the genome must be annotated: the different regions – such as the genes – need to be identified and labeled . Here , Fu et al . assembled and annotated the genome of the cabbage looper , and in the process developed strategies that could be used for other species with a lot of repeated sequences in their genomes . Having access to the looper’s full genetic information makes it possible to use their germline cells to produce new types of proteins , for example for pharmaceutical purposes . Fu et al . went on to make working with these cells even easier by refining protocols so that modern research techniques , such as the gene-editing technology CRISPR-Cas9 , can be used on the looper germline cells . The mapping of the genome also revealed that the genes involved in removing toxins from the insects’ bodies are rapidly evolving , which may explain why the moths readily become resistant to insecticides . This knowledge could help finding new ways of controlling the pest . Finally , the genes involved in RNA silencing were labeled: results show that an entire chromosome is the source of piRNAs . Combined with the new protocols developed by Fu et al . , this could make cabbage looper germline cells the default option for any research into the piRNA mechanism . How piRNA works in the moth could inform work on human piRNA , as these processes are highly similar across the animal kingdom .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "genetics", "and", "genomics" ]	2018	The genome of the Hi5 germ cell line from Trichoplusia ni, an agricultural pest and novel model for small RNA biology
Understanding allostery in enzymes and tools to identify it offer promising alternative strategies to inhibitor development . Through a combination of equilibrium and nonequilibrium molecular dynamics simulations , we identify allosteric effects and communication pathways in two prototypical class A β-lactamases , TEM-1 and KPC-2 , which are important determinants of antibiotic resistance . The nonequilibrium simulations reveal pathways of communication operating over distances of 30 Å or more . Propagation of the signal occurs through cooperative coupling of loop dynamics . Notably , 50% or more of clinically relevant amino acid substitutions map onto the identified signal transduction pathways . This suggests that clinically important variation may affect , or be driven by , differences in allosteric behavior , providing a mechanism by which amino acid substitutions may affect the relationship between spectrum of activity , catalytic turnover , and potential allosteric behavior in this clinically important enzyme family . Simulations of the type presented here will help in identifying and analyzing such differences . The rise in antimicrobial resistance ( AMR ) is a growing global public health crisis ( Centers for Disease Control and Prevention ( U . S . ) , 2019 ) . As AMR has continued to spread and many antimicrobial agents have become ineffective against previously susceptible organisms , the World Health Organization recently projected that AMR could result in up to 10 million deaths annually by 2050 ( Interagency Coordination Group on Antimicrobial Resistance , 2019 ) . The problem of AMR is particularly urgent given the alarming proliferation of antibiotic resistance in bacteria; pathogens associated with both community-acquired and healthcare-associated infections are increasingly resistant to first-line and even reserve agents ( Lythell et al . , 2020 ) . This not only poses a serious challenge obstacle in fighting common and severe bacterial infections , but also reduces the viability and increases the risks of interventions such as orthopedic surgery and also threatens new antibiotics coming to the market ( Bush and Page , 2017 ) . AMR risks negating a century of progress in medicine made possible by the ability to effectively treat bacterial infections . In spite of the advances in the field of antimicrobial chemotherapy , the efficacy , safety , chemical malleability , and versatility of β-lactams make them the most prescribed class of antibiotics ( Tooke et al . , 2019 ) . Their cumulative use exceeds 65% of all injectable antibiotics in the United States ( Bush and Bradford , 2016 ) . β-Lactam antibiotics work by inhibiting penicillin binding proteins ( PBPs ) , a group of enzymes that catalyze transpeptidation and transglycosylation reactions that occur during the bacterial cell wall biosynthesis ( Tooke et al . , 2019 ) . A damaged cell wall results in loss of cell shape , osmotic destabilization , and is detrimental for bacterial survival in a hypertonic and hostile environment ( Bonomo , 2017 ) . Of the four primary mechanisms by which bacteria resist β-lactam antibiotics , the most common and important mechanism of resistance in Gram-negative bacteria , including common pathogens such as Escherichia coli and Klebsiella pneumoniae , is the expression of β-lactamase enzymes ( Tooke et al . , 2019 ) . These enzymes hydrolyze the amide bond in the β-lactam ring , resulting in a product that is incapable of inhibiting PBPs ( Palzkill , 2018 ) . The Ambler system of classifying β-lactamase enzymes categorizes them , based on amino acid sequence homology , into classes A , B , C , and D ( Ambler , 1980; Bush and Jacoby , 2010 ) . While β-lactamases of classes A , C , and D are serine hydrolases , class B enzymes are metalloenzymes that have one or more zinc ions at the active site ( Palzkill , 2013 ) . Class A enzymes are the most widely distributed and intensively studied of all β-lactamases ( Tooke et al . , 2019 ) . The hydrolytic mechanism in class A ( Figure 1—figure supplement 1 ) , revealed by experiments and QM/MM modeling , is initiated by reversible binding of the antibiotic in the active site of the enzyme ( formation of the Michaelis complex ) . This is followed by nucleophilic attack of the catalytic serine ( Ser70 ) on the carbonyl carbon of the β-lactam ring , resulting in a high-energy acylated intermediate that quickly resolves , following protonation of the β-lactam nitrogen and cleavage of the C-N bond , to a lower energy covalent acyl-enzyme complex ( Chudyk et al . , 2014; Hermann et al . , 2003; Hermann et al . , 2005 ) . Next , an activated water molecule attacks the covalent complex , leading to the subsequent hydrolysis of the bond between the β-lactam carbonyl and the serine oxygen , resulting in the regeneration of the active enzyme and release of the inactive β-lactam antibiotic ( Tooke et al . , 2019; Bonomo , 2017; Palzkill , 2018; Chudyk et al . , 2014; Fisher and Mobashery , 2009; Hermann et al . , 2006; Hirvonen et al . , 2019; Pan et al . , 2017 ) . TEM-1 is one of the most common plasmid-encoded β-lactamases in Gram-negative bacteria and is a model class A enzyme ( Brown et al . , 2009 ) . It has a narrow spectrum of hydrolytic activity that is limited to penicillins and early generation cephalosporins; in contrast , its activity toward large , inflexible , broad-spectrum oxyiminocephalosporins such as the widely used antibiotic ceftazidime is poor ( Palzkill , 2018 ) . However , mutations in the bla_TEM-1 gene have led to amino acid modifications , which allow subsequent TEM-1 variants to hydrolyze broad-spectrum cephalosporins ( so-called ‘extended-spectrum’ activity ) or to avoid the action of mechanism-based inhibitors such as clavulanate that are used in combination with β-lactams to treat β-lactamase producing organisms ( Brown et al . , 2009 ) . Another class A enzyme , KPC-2 ( K . pneumoniae carbapenemase-2 ) , encoded by the bla_KPC-2 gene is an extremely versatile β-lactamase ( Queenan et al . , 2004 ) with a broad spectrum of substrates that includes penicillins , cephamycins , and , importantly , carbapenems ( Queenan et al . , 2004; Yigit et al . , 2003 ) . Currently , predominant strains of K . pneumoniae and other Enterobacterales continue to be identified as responsible for outbreaks internationally . Continued dissemination of KPC makes this one of the β-lactamases of most immediate clinical importance and a key target for inhibitor development . The structure and activity of class A β-lactamases have been well studied ( Palzkill , 2018; Papp-Wallace et al . , 2012; Salverda et al . , 2010 ) . In spite of sequence differences , class A β-lactamases share the same structural architecture ( Philippon et al . , 2016 ) , as evident from the present 47 structures of TEM-1 and 38 structures of KPC-2 , or their engineered variants , deposited in the Protein Data Bank ( PDB ) at the time of this writing . However , despite the wide variety of substrates that TEM-1 and KPC-2 can hydrolyze , their structures are quite rigid . The average mean order parameter , S2 , as calculated from NMR experiments for TEM-1 , is between 0 . 81 and 0 . 94 , and almost all class A β-lactamases are conformationally identical ( Gobeil et al . , 2019; Morin and Gagné , 2009; Savard and Gagné , 2006 ) . Loops ( e . g . active site loops ) play a crucial role in the activity of many enzymes ( Liao et al . , 2018 ) , including β-lactamases . There is increasing evidence that active site conformations may be influenced by distal loops , connected , for example , through active closure and desolvation , and potentially via networks of coupled motions ( Liao et al . , 2018; Agarwal , 2019; Bunzel et al . , 2020; Bunzel et al . , 2021 ) . The active sites of TEM-1 and KPC-2 are surrounded by three loops: ( a ) the Ω-loop ( residues 172–179 ) , ( b ) the loop between α3 and α4 helices , in which a highly conserved aromatic amino acid is present at position 105 , and ( c ) the hinge region , which lies opposite to the Ω-loop and contains the α11 helix turn ( Figure 1 , Figure 1—figure supplement 2 ) . Two highly conserved residues , Glu166 and Asn170 , which are essential for catalysis , influence the conformation of the Ω-loop ( Banerjee et al . , 1998 ) . The conformational dynamics of these loops play an important role in enzyme activity and are probably modulated by evolution ( Pan et al . , 2017; Banerjee et al . , 1998; Escobar et al . , 1994; Guillaume et al . , 1997; Leung et al . , 1994; Zawadzke et al . , 1996 ) . For example , we have recently found that differences in the spectrum of activity between KPC-2 and KPC-4 are due to changes in loop behavior ( Tooke et al . , 2021 ) . There has been extensive discussion about the possible contribution of protein dynamics to enzyme catalysis ( Glowacki et al . , 2012; Kamerlin and Warshel , 2010; Luk et al . , 2013; Singh et al . , 2015 ) . In some enzymes , conformational changes have been identified as necessary in preparing the system for reaction ( Liao et al . , 2018; Agarwal , 2019 ) . Several simulation studies , including long timescale and enhanced sampling molecular dynamics ( MD ) simulations and QM/MM simulations of reactions , have been reported for TEM-1 and KPC-2 β-lactamases ( Chudyk et al . , 2014; Hirvonen et al . , 2019; Bowman et al . , 2015; Galdadas et al . , 2018; Hart et al . , 2016; Tooke et al . , 2021 ) . MD simulations have explored cryptic pocket formation ( Hart et al . , 2016 ) , studied protein-ligand interactions ( Fisette et al . , 2012 ) , predicted antibiotic resistance ( Chudyk et al . , 2014; Hirvonen et al . , 2019; Galdadas et al . , 2018 ) , explained the effects of mutations on enzyme specificities ( Zaccolo and Gherardi , 1999 ) , and investigated conserved hydrophobic networks ( Galdadas et al . , 2018 ) . It remains a challenge to directly link conformational heterogeneity and function . Understanding conformational behavior is relevant to β-lactamase inhibition as well as catalytic mechanism . For organisms producing class A β-lactamases , co-administration of susceptible β-lactams with mechanism-based covalent inhibitors ( e . g . clavulanate ) represents a proven therapeutic strategy and has successfully extended the useful lifetime of penicillins in particular ( Fritz et al . , 2018; Drawz and Bonomo , 2010 ) . However , while the mechanism of direct inhibition by covalently bound inhibitors is well established ( Fritz et al . , 2018 ) , the possibility of exploiting sites remote from the active center in allosteric inhibition strategies is less well explored , and where this has been achieved ( Horn and Shoichet , 2004; Pemberton et al . , 2019; Hart et al . , 2016 ) the structural changes occurring as a result of ligand binding or unbinding to allosteric sites and the relay of structural communication that leads to inhibition are not well understood . The conformational rearrangements that take place upon ligand ( un ) binding in allosteric sites and their potential connection to the β-lactamase active site are the focus of this study . Here , we employ a combination of equilibrium and nonequilibrium MD simulations to identify and study the response of two class A β-lactamases , TEM-1 and KPC-2 , to the ( un ) binding of ligands at sites distant from the active site . Nonequilibrium simulations applying the Kubo-Onsager approach ( Ciccotti and Ferrario , 2016; Ciccotti et al . , 1979 ) are emerging as an effective way to characterize conformational changes and communication networks in proteins ( Abreu et al . , 2020; Damas et al . , 2011; Oliveira et al . , 2019a; Oliveira et al . , 2019b ) . To the best of our knowledge , this is the first application of this nonequilibrium MD approach to study enzymes . We study β-lactamases , whose ultrafast turnover rates can approach the diffusion limits for natural substrates ( ~107–108 M−1s−1 ) ( Fisher and Mobashery , 2009 ) . We perform 10 µs of equilibrium MD simulations of TEM-1 and KPC-2 , with and without ligands present in their allosteric binding sites . These simulations identify conformational changes in the highly dynamic loops that shape the active site and structurally characterize the dynamics of the formation and dissolution of the allosteric pocket . We also carry out an extensive complementary set of 1600 short nonequilibrium MD simulations ( a total of 8 µs of accumulated time ) , which reveal the response of the enzyme to perturbation and identify pathways in the enzymes that connect the allosteric site to other parts of the protein . These simulations demonstrate direct communication between the allosteric sites and the active site . The results show that this combination of equilibrium and nonequilibrium MD simulations offers a powerful tool and a promising approach to identify allosteric communication networks in enzymes . To explore the conformational space of TEM-1 and KPC-2 in the ApoEQ ( no ligand ) and IBEQ ( inhibitor-bound ) states , we started by running a set of equilibrium simulations ( 20 replicas of 250 ns each ) that resulted in 5 µs of accumulated simulation time per system . Conformational changes during the simulations were assessed using their Cα root mean-square deviation ( RMSD ) profiles ( Figure 2—figure supplement 2 ) . The simulated systems were considered equilibrated beyond 50 ns as shown by RMSD convergence . In each case , the proteins remained close to their initial conformation during the course of 250 ns ( Figure 2—figure supplement 2a ) . The average RMSD for ApoEQ and IBEQ states were between 0 . 10 and 0 . 12 nm for all systems ( Figure 2—figure supplement 6 ) . The low RMSD values are consistent with previously published results , which have also shown class A β-lactamase enzymes to be largely rigid and conformationally stable when studied on long timescales and rarely divergent from the initial structure ( Gobeil et al . , 2019; Galdadas et al . , 2018 ) . Conventional RMSD fitting procedure using all Cα atoms failed to separate regions of high versus low mobility . To resolve such regions , we used a fraction ( % ) of the Cα atoms for the alignment . Beyond this fraction , there is a sharp increase in the RMSD value for the remainder of the Cα atoms ( Figure 2—figure supplement 2b ) . At 80% , the core of TEM-1 could be superimposed to less than 0 . 064 and 0 . 074 nm for ApoEQ and IBEQ states , respectively ( Figure 2—figure supplement 2bi ) . In the KPC-2 ApoEQ state , the RMSD of 80% of the Cα atoms was below 0 . 060 nm , while the same subset of atoms had an RMSD below 0 . 066 nm in the IBEQ state ( Figure 2—figure supplement 2bii ) . This 80% fraction of Cα atoms constitutes the core of the enzyme and did not show any divergence from the initial reference structure ( Figure 2—figure supplement 2c ) . RMSD values for the remaining 20% of Cα atoms varied between 0 . 16 and 0 . 23 nm . This apparent rigidity is consistent with the experimental finding , based upon , for example , thermal melting experiments Mehta et al . , 2015 , that KPC-2 is more stable than many other class A β-lactamases such as TEM-1 . Some large conformational changes were observed in all replicates; these involved changes in conformations of the loops that connect secondary structural elements ( Figure 2—figure supplement 3 ) . To further validate the stability of the two systems , we analyzed structural properties including the radius of gyration ( Rg; Figure S5 ) , solvent accessible surface area ( SASA; Figure 2—figure supplement 5 ) , and the secondary structure of each enzyme over the simulated time ( Figure 2—figure supplement 7 ) . The values for these properties are listed in Figure 2—figure supplement 6 . A ligand that binds to an allosteric site can control protein function by affecting the active site ( Laskowski et al . , 2009 ) . This generally occurs by altering the conformational ensemble that the protein adopts ( Laskowski et al . , 2009; Motlagh et al . , 2014 ) . To probe how ligand binding to an allosteric site affects the dynamics of β-lactamases , we calculated the Cα root-mean-square fluctuation ( RMSF ) for both ApoEQ and IBEQ states . Higher RMSF values correspond to greater flexibility during the simulation . Although the Cα RMSF profiles for ApoEQ and IBEQ states are similar , indicating similar dynamics , there are some discernible differences ( Figure 2 ) . In equilibrium simulations of TEM-1 and KPC-2 , the hydrophobic core of the enzyme is stable and shows limited fluctuations . Most of the RMSF variance is observed in loops that connect secondary structural elements ( Figure 2 ) . In TEM-1 IBEQ , higher fluctuations are observed predominantly in three distinct regions when compared with the ApoEQ enzyme; in the loops between helices α7 and α8 ( residues 155–165 ) , α9 and α10 ( residues 196–200 ) , and the hinge region including helix α11 ( residues 213–224 ) ( Figure 2A ) . The α11 and the α12 helices are part of a highly hydrophobic region that also constricts the allosteric pocket in all TEM-1 apo crystal structures . Binding of the ligand disrupts the hydrophobic interactions within this region , resulting in the opening of the allosteric pocket between helices α11 and α12 ( Horn and Shoichet , 2004 ) . It should be noted that the starting ApoEQ structure of TEM-1 was generated from the IB crystal structure , by the removal of the ligand from the allosteric binding site . During the ApoEQ simulations , α12 helix behaves like a lid and closes over the empty , hydrophobic , allosteric binding site , and thus displays high RMSF at the C-terminal end of the enzyme . This conformational change recovers the structure of the apo crystal form , as observed , for example , in PDB id 1ZG4 ( Stec et al . , 2005 ) , as reflected to the RMSD of ~0 . 07 nm after superposition of the structures . The rest of the loops displayed comparable fluctuations in both ApoEQ and IBEQ states . The differences between the ApoEQ and IBEQ states were of similar magnitude in KPC-2 . In KPC-2 IBEQ , more extensive fluctuations than in ApoEQ were also observed in the loops between α7 and α8 ( residues 156–166 ) , the hinge region , around α11 ( residues 214–225 ) , and in the loop between β7 and β8 ( residues 238–243 ) ( Figure 2b ) . Conversely , fluctuations are slightly higher in the ApoEQ than IBEQ state in the loop leading into the Ω-loop from α7 helix ( residues 156–166 ) . Overall , however , RMSFs are similar in analogous regions of the IBEQ and ApoEQ states in both TEM-1 and KPC-2 , highlighting the conservation of structural dynamics in class A β-lactamases . However , there were some fluctuations that were unique and limited to each enzyme ( Figure 2 ) . In both TEM-1 and KPC-2 IBEQ states , interactions of the ligands in their respective allosteric binding sites contribute to enhanced fluctuations ( i . e . larger than in the Apo forms ) of the local structural elements ( Figure 3—figure supplement 1 ) . The sites in which the ligands bind are very different . In TEM-1 , the binding site is deep and forms a hydrophobic cleft . The ligand penetrates to the core of the enzyme and is sandwiched between helices α11 and α12 ( Horn and Shoichet , 2004 ) . The FTA ligand remains tightly bound in the allosteric pocket throughout the simulations ( Figure 3—figure supplement 2 ) . In KPC-2 , the allosteric binding site is shallow and solvent-exposed even in the absence of the ligand . Although the distal end of the pocket is hydrophobic , there are some polar amino acids on the proximal surface ( e . g . Arg83 and Gln86 ) , which are exposed to the solvent . This shallow site forms a part of a larger pocket that is occluded by the side chain of Arg83 ( α7 helix ) . In some of our IBEQ simulations , the Arg83 side chain rotates , leading to the opening of a larger hidden pocket . This enlarged space is now accessible to the ligand for exploring various interactions . The tumbling of GTV increases the fluctuations in the complex ( Figure 3—figure supplement 1c , d ) , however , the ligand does not leave the binding site ( Figure 3—figure supplement 2 ) . To further highlight the structural changes occurring as a result of ligand binding , positional C deviations were calculated between IBEQ and ApoEQ systems for the equilibrated part of the simulations ( Figure 3a , b ) . The Cα deviation values plotted are an average between simulation taken by combining all trajectories from ApoEQ and IBEQ simulations for that particular system . This is one of the simplest approaches , which can determine residues undergoing largest structural rearrangements . The averaged Cα positional deviations are mapped onto the averaged ApoEQ structure to visualize the largest relative displacements in three dimensions ( Figure 3c , d ) . The hydrophobic cores of both TEM-1 and KPC-2 β-lactamase enzymes show little or no conformational change . The major differences between the ApoEQ and IBEQ states are in the loops connecting different secondary structure elements . In TEM-1 , Cα deviations are observed in the loops between α4 and β5 ( residues 112–116 ) , α7 and α8 ( residues 155–166 ) , Ω-loop ( residues 172–179 ) , α9 and α10 ( residues 196–200 ) , hinge and α11 ( residues 213–224 ) , β7 and β8 ( residues 238–243 ) , and β9 and α12 ( residues 267–272 ) . There are some relatively minor deviations observed in loops β1-β2 ( residues 51–55 ) , β2-β3 ( residues 61–65 ) , α2-β4 ( residues 86–93 ) , α6-α7 ( residues 143–144 ) , β8-β9 ( residues 252–258 ) , and at the pivot of the α3 helix ( residues 98–101 ) . The hinge region and residues in helices α11 and α12 display the largest deviations . This is also in agreement with other experimental data that indicate the connection between the active site and the allosteric pocket studied in TEM-1 in the presence of BLIP inhibitor , seems to be mostly due to hinge region motions ( Meneksedag et al . , 2013 ) . The structural dynamics observed in KPC-2 were slightly different from TEM-1 . In KPC-2 , prominent Cα deviations were observed in the loops between β1 and β2 ( residues 51–55 ) , α2 and β4 ( residues 88–93 ) , α4 and β5 ( residues 114–116 ) , α7 and α8 ( residues 156–166 ) , Ω-loop ( residues 172–179 ) , β7 and β8 ( residues 238–243 ) , β8 and β9 ( residues 252–258 ) , in the loop between β4 and α3 leading up to the proximal end of α3 ( residues 94–102 ) and in the hinge/α11 helix ( residues 214–225 ) . There are some minor deviations observed in α1-β1 ( residues 39–42 ) and β9-α12 ( residues 266–270 ) . The most important ligand-induced Cα deviation is observed in the loop connecting the α4 helix to the β5 strand ( residues 114–116 ) . The deviation of the α4-β5 loop together with the deviation observed in the loop between β4 and α3 leading into α3 helix ( residues 96–102 ) has the potential to deform the α3 helix-turn-α4 helix . The β4-α3 and α4-β5 loops form the basal pivot joint of the α3 and α4 helices and maintain the correct positioning of this helix-turn-helix at the periphery of the enzyme active site . The correct positioning of this loop is important as Trp105 lies on this loop . Mutagenesis studies have shown that a highly conserved aromatic amino acid at position 105 in class A β-lactamases ( Tyr105 in TEM-1 , Trp105 in KPC-2 ) is located at the perimeter of the active site and plays a crucial role in ligand recognition via favorable stacking interactions with the β-lactam ring ( Papp-Wallace et al . , 2010b; Doucet et al . , 2004 ) . The aromatic side chain at position 105 coordinates the binding of substrates not only via stacking and edge-to-face interactions but by also adopting ‘flipped-in’ or ‘flipped-out’ conformations ( Galdadas et al . , 2018; Papp-Wallace et al . , 2010a; Papp-Wallace et al . , 2010b ) . This has been proposed based on the conformations observed in the available crystal structures and confirmed by enhanced sampling MD simulations ( Galdadas et al . , 2018; Ke et al . , 2012 ) . Any perturbation that alters the conformation of α3-turn-α4 helix or deforms the α3-α4 pivot region would prevent α3 and α4 helices from correctly shaping the active site of the enzyme . This would result in the aromatic residue at 105 partially detaching from the edge of the active site and being unable to stabilize the incoming substrate as required for efficient catalysis . This explains the loss of β-lactam resistance in strains expressing KPC variants at position 102 or 108 , as established in the MIC experiments reported previously ( Galdadas et al . , 2018 ) . To study signal propagation from the two allosteric sites , we ran 800 short nonequilibrium simulations , with a total sampling time of 4 µs for each system . The nonequilibrium simulations were initiated from regular intervals of the equilibrated part of the long IBEQ simulation , starting at the 50 ns time point ( Figure 2—figure supplement 1 ) . In each simulation , the ligand was removed from its binding site and the resulting system was further simulated for 5 ns . The response of the system to the perturbation was determined using the Kubo-Onsager approach developed by Ciccotti and Ferrario , 2013; Ciccotti et al . , 1979; Ciccotti and Ferrario , 2016 . In this approach , the time evolution of the conformational changes induced by ligand removal can be determined by comparing the ApoNE and IBEQ simulations at equivalent points in time . The subtraction method , applied to multiple pairs of trajectories , effectively removes noise arising from fluctuations of the systems and allows residues that are involved in signal propagation to be identified . The disappearance of the ligand from its binding site generates a temporary localized vacuum , against which there is an immediate structural and solvent response . As the simulation progresses , the cascading conformational changes in response to the perturbation ( removal of ligand ) show the route by which structural response is transmitted through the protein . Video supplements: Signal propagation in TEM-1 and KPC-2 as a result of the perturbation ( ligand removal ) in the allosteric binding site . The disappearance of the ligand from its binding site generates a localized vacuum , against which there is an immediate structural response by the enzyme . As the simulation progresses , the cascading conformational changes in response to the perturbation ( ligand removal ) show the route by which structural response is transmitted through the protein . This approach has identified a general mechanism of signal propagation in nicotinic acetylcholine receptors , by analyzing their response to deletion of nicotine ( Oliveira et al . , 2019a ) . The difference in the position of Cα atoms is calculated between the short ApoNE and IBEQ simulations at specific time points . These differences are then averaged over all pairs of simulations to reveal the structural conformations associated with this response ( Figure 4 ) and their statistical significance . The Cα coordinates of each residue in the ApoNE were subtracted from the corresponding Cα atom coordinates of the IBEQ simulation at specific points in time , namely 0 . 05 , 0 . 5 , 1 , 3 , and 5 ns . This resulted in a difference trajectory for each pair of simulations . The difference trajectories are averaged over the set of 800 simulations for each system . The low standard error ( SE ) calculated for the average between the ApoNE and IBEQ demonstrates the statistical significance of the results . Due to the short timescale ( 5 ns ) of the nonequilibrium simulations , only small amplitude conformational changes will be observed . In TEM-1 , the allosteric site is sandwiched between the α11 and α12 helices . Adjacent to this binding site is the hinge region ( residues 213–218 ) , whose dynamics have previously been examined by NMR and shown to have low order parameters indicating high mobility ( Gobeil et al . , 2019; Savard and Gagné , 2006 ) . This is also the site of perturbation in the nonequilibrium simulations and so the point of origin of the allosteric signal . Located on the loop between the distal end of the α11 helix and β7 is a highly conserved Trp229 residue . The indole ring of Trp229 is sandwiched between two other highly conserved residues , Pro226 and Pro251 , present in loops α11-β7 and β8-β9 , respectively . The π/aliphatic stacked arrangement of tryptophan-proline is a very tight interaction and is similar in geometry to that observed in complexes of proline-rich motif binding families , including the EVH1 and GYF binding domains , with their peptide ligands ( Ball et al . , 2005; Freund et al . , 1999; Reinhard et al . , 1996; Zondlo , 2013 ) . The perturbation destabilizes this stacked arrangement resulting in an extension of an inherently highly mobile region . After 50 ps of simulation , the Cα deviations have propagated and can be observed in the loop between β1 and β2 . Interestingly , the loops at the basal pivot of 3 and 4 also responded rapidly to ligand removal . These loops are ~33 Å away from the allosteric binding site and can affect the spatial position of the turn between helix α3 and α4 . The α3-turn-α4 helix forms the boundary of the active site , and it is on this turn where the Tyr105 residue , important for substrate recognition , is positioned . These results clearly demonstrate the coupling between the distal allosteric site and catalytically relevant regions of the enzyme . As the signal propagates within the protein , there is a gradual and cumulative increase in the Cα deviations in the aforementioned loops . In particular , the loop between the α9 and α10 helices , which is positioned just below the β1-β2 loop , displays high deviations and forms a focal point for the signal to bifurcate in two directions . First , major deviations are observed laterally toward loop α7-α8 and onward into the Ω-loop ( Figure 4a , b ) . Second , more minor deviations move into the loop between α2 and β4 and onward into the basal pivot of α3-turn-α4 helix . There is another shorter route at the top of the enzyme that the signal can take to go from the allosteric binding site to the Ω-loop , via the proximal end of α12 helix and across the loop between β9 and α12 helix ( Figure 4a , b ) . In KPC-2 , the allosteric pocket is shallower and lies between helices α2 and α7 . Residues from three loops ( α6-α7 , α7-α8 , and α2-β4 ) are in close proximity to this binding site . An additional loop , α9-α10 , is linked to this binding site via the distal end of the α2 helix . The perturbation in this binding site results in enhanced mobility of the α2-β4 loop , which leads directly into β4 and onward to the basal pivot of the α3 helix . The proximal end of the α3 helix and the distal end of the α4 helix , which forms the pivot point of the α3-turn-α4 structure , display high deviations ( Figure 4c ) . The highly conserved aromatic amino acid , Trp105 , is located on this turn . The distance between the allosteric binding site and the α3 helix is ~27 Å . Other major deviations are also observed in the Ω-loop as the simulation progresses ( Figure 4c ) . The Ω-loop is directly linked to the allosteric binding site via loop α7-α8 . Some minor deviations are also observed in the loop connecting β9 and the α12 helix . In both TEM-1 and KPC-2 , the removal of the ligand at the beginning of the nonequilibrium simulations does not result in large conformational changes . The subsequent Cα deviations trace the route of the propagating signals ( Figure 3—figure supplement 212 ) . In TEM-1 , α11 and the hinge region , loops β1-β2 and β8-β9 , respond rapidly to the perturbation and display comparable RMSD values to the equilibrated simulations . Similarly , in KPC-2 , only loops α2-β4 and α7-α8 respond rapidly to the perturbation . The other structural elements take longer to respond , and their conformational rearrangements are not fully sampled in the ApoNE simulations . It is worth emphasizing that while the short nonequilibrium simulation can be an excellent tool to study an immediate structural response toward a perturbation , the timescale of nonequilibrium MD does not represent a real timescale and thus should not be compared directly with equilibrium simulations . Nevertheless , nonequilibrium MD can identify the sequence of events and pathways involved . The perturbations of the two enzymes here are different but show some striking common features . In both TEM-1 and KPC-2 systems , even though the point of origin of perturbation ( i . e . allosteric site ) is different , the signal leads to common endpoints at the pivot of α3-turn-α4 helix and in the Ω-loop . Thus , simulations of two different class A β-lactamases , starting from two distinct allosteric sites , identify a common mechanism by which catalytic activity may be disrupted by conformational changes close to the active site . The results from the nonequilibrium simulations also correlate well with experimental data , which suggest that the Ω-loop plays a critical role in ligand binding by altering the conformation of Glu166 and Asn170 which are involved in both acylation and deacylation reactions ( Chudyk et al . , 2014; Pan et al . , 2017; Brown et al . , 2009; Fritz et al . , 2018; Banerjee et al . , 1998 ) . Dynamical cross-correlation analysis provides information about the pathways of signal propagation and also some insights into the timescales of allosteric communication in TEM-1 and KPC-2 β-lactamases . Dynamic cross-correlation maps ( DCCMs ) have been previously used to identify networks of coupled residues in several enzymes ( Agarwal et al . , 2004; Hester et al . , 2019; Agarwal et al . , 2012 ) . Using a similar approach , DCCMs were calculated for the ApoEQ and IBEQ simulations and also for the ApoNE nonequilibrium simulations ( Figure 5 ) . In these figures , the green regions represent no to slightly positive correlations , while yellow regions represent moderate negative correlations . Negative correlations imply residues moving toward or away from each other in correlated fashion ( such as shown by fluctuating hydrogen bonds ) ; for large regions this represents global conformational fluctuations ( also referred to as breathing motions ) ( Agarwal et al . , 2004 ) . The results depicted in Figure 5a indicate that in the case of TEM-1 ApoEQ ( Figure 5a , left ) , β11 helix shows high negative correlation with β12 terminal helix . This represents the lid motion of β12 helix , which moves to shut the empty , hydrophobic , allosteric binding site in the TEM-1 ApoEQ structure ( see above ) . This motion is , however , not observed in the ligand bound TEM-1 IBEQ simulations . The TEM-1 IBEQ system shows a substantial increase in correlations , representing changes in the dynamical communications due to the presence of the allosteric ligand ( Figure 5a , middle ) . The binding of the ligand changes the overall global conformational fluctuations of TEM-1 , as represented by the increase in yellow regions in the DCCMs . Furthermore , a number of negative correlations ( encircled red regions in DCCMs ) also increase in other regions of the protein on ligand binding . The DCCM collectively computed from all nonequilibrium trajectories for TEM-1 ( Figure 5a , right , Figure 5—figure supplement 1 ) also shows a further increase in the areas of negative correlations ( encircled ) . Interestingly , DCCM also identifies the pathway of allosteric communication ( Figure 5—figure supplement 1 ) , with notable correlations between the regions β1-β2:α2-β4 , α3-α4:α2-β4 , β4-α3:α7-α8 , β3-α2:Ω , α9-α10:β1-β2 , β3-α2:β8-β9 , α5-α6:α12 , hinge-α11:α1-β1 , β8-β9:α4-β5 , β7:α12 , and α11:α12 . These results indicate that the presence of ligand in TEM-1 increases the dynamic communication between regions that are independent in the ApoEQ simulations . This is particularly evident in the nonequilibrium trajectories that show the largest changes from the case of ApoEQ TEM-1 , identifying changes in correlation as the system adjusts to the absence of the ligand . KPC-2 shows even more interesting behavior ( Figure 5b ) . Simulations of ApoEQ KPC-2 show overall more correlated regions than TEM-1 ApoEQ system ( as indicated by the more extensive yellow regions in the DCCM ) , with further increases in the presence of the inhibitor ( indicated by a number of orange regions ) . However , the DCCM collectively computed from all nonequilibrium trajectories for KPC-2 shows a reduction in regions of cross-correlations , a contrast from the case of TEM-1 . To obtain a better understanding , the DCCMs from individual 5 ns nonequilibrium trajectories were also computed and analyzed . These reveal interesting trends as depicted in Figure 5—figure supplement 2 . For most nonequilibrium trajectories , the maps are similar with a decrease in dynamic correlations; however , for several trajectories ( shown in Figure 5—figure supplement 2 ) , the maps indicate a significant increase in the correlations . The DCCMs computed from individual trajectories show behavior similar to averaged nonequilibrium trajectories in TEM-1 with a number of regions showing high negative correlations ( as highlighted by widespread presence of small red regions in the DCCMs ) . Overall , these results indicate that the perturbation in KPC-2 generates a dynamical response that is much faster than that observed in TEM-1 . A plausible explanation for the faster response in KPC-2 is that the more solvent-exposed ligand binding site is surrounded by dynamic surface loops that respond to the perturbation more quickly than the allosteric binding site in TEM-1 , which is buried in the hydrophobic core of the protein . This is consistent with the experimental observations that motions can occur on different timescales and can vary greatly between different β-lactamases ( Gobeil et al . , 2019 ) . A number of clinical variants that extend hydrolytic activity to encompass additional β-lactams such as oxyiminocephalosporins , and/or enhance enzyme stability , have been identified for both the TEM-1 and KPC-2 β-lactamase enzymes ( Palzkill , 2018; Clark et al . , 2016; Naas et al . , 2017 ) . Some of these have been crystallized and their protein structure deposited in the PDB . While many of these amino acid substitutions ( e . g . TEM-1 mutations at residues Glu104 in the α3-turn-α4 , Arg164 on the Ω-loop , and Ala237 , Gly238 , and Glu240 on the β7 strand ) directly affect important structural features such as the active site or the Ω-loop , some are of uncertain structural significance . Even when enzyme structures are known , the connections between the positions of clinical variants , protein structure , and their functional implications are often not clear . There is particular uncertainty and interest in the effects of mutations more distant from the active site . To assess how many of these clinically relevant substitutions lie on the allosteric communication pathway , their spatial positions were identified and mapped onto the 3D structures of TEM-1 and KPC-2 . The site of the mutation was plotted as a sphere on its unique Cα position on the structure ( Figure 6 ) , which was rendered to represent the allosteric communication pathways shown in Figure 4 . For TEM-1 , 45 of the 90 , and for KPC-2 15 out of the 25 , amino acid positions known to vary in clinical isolates could be mapped onto the allosteric communication pathway . Notably , in TEM-1 , residues such as Gly92 preceding α4 , His153 at the end of α7 , and Ala224 preceding α11 have all been associated with ESBL and/or inhibitor-resistant phenotypes identified in the clinic ( Palzkill , 2018 ) . Residues such as M182 and A184 , which precede α9 and are not on the communication pathway per se , are however surrounded on all sides by loops that are involved in the communication network ( Figure 6—figure supplement 1 ) . For KPC enzymes , for which less information is available , characterized variants that have emerged in the clinic differ mostly in activity toward ceftazidime 58 and feature substitutions at positions ( 104 , 240 , 274 ) closer to the active site . As more sequences emerge and their phenotypic consequences are described ( Tooke et al . , 2021 ) , however , it will then be of interest to establish the properties of KPC variants featuring substitutions at positions ( e . g . 92 , 93 ) , which lie along the communication pathways described here . We propose that some of these variants differ in allosteric properties , and further , that these differences relate to variances in their clinically relevant spectrum of activity . If our hypothesis is correct , 50% or more of known clinically important variants in these two enzymes may differ in their allosteric behavior , indicating that this is a fundamentally important property in determining their spectrum of catalytic activity . The relationship between sequence ( especially substitutions remote from the active site ) , protein dynamics , spectrum of activity , catalytic turnover , and allosteric behavior will be an important future direction in understanding AMR due to β-lactamase enzymes ( Tooke et al . , 2021 ) . Here , we have identified structural communication between two allosteric binding sites and structural elements , close to the active site , that control enzyme specificity and activity in two distinct , clinically important , class A β-lactamases . The extensive equilibrium MD simulations , with and without ligands , reveal ligand-induced conformational changes , while nonequilibrium MD simulations show that changes at allosteric sites are transmitted to the active site and identify the structural pathways involved . These nonequilibrium simulations identify the initial stages of the dynamic rearrangement of secondary structural elements and highlights the signal propagation routes ( with demonstration of its statistical significance ) . These two complementary approaches together facilitate understanding of how information flows from one part of the protein structure to another . The equilibrium simulations ( of ligand-bound and Apo enzymes ) show that the structural effects of ligand binding to allosteric sites are not restricted to the local binding pocket . Class A β-lactamases are rigid enzymes ( Gobeil et al . , 2019 ) that do not undergo large-scale conformational changes; the observed structural rearrangements ( caused by ligand removal ) are dominated by localized changes in the conformation of loops . Such ligand-induced structural changes are observed in the loops surrounding the active sites including the hinge region , the Ω-loop , and the α3-turn-α4 helix , positioned as far as ~33 Å from the allosteric ligand binding site . In both enzymes , the observed flexible motions lead to an enlargement of the active site , with the potential consequences for the orientation of either mechanistically important regions of the protein or of bound ligand , and , consequently , enzyme activity . The nonequilibrium simulations , using an emerging technique , identify the structural rearrangements arising as a result of a perturbation ( ligand removal ) and demonstrate communication between the allosteric site and the active site . The ordering of these conformational changes shows the initial steps of communication between secondary structure elements . This structural relay constitutes a pathway that enables effective signal propagation within the enzymes . In TEM-1 , the conformational changes initiated at the allosteric site ( which is situated between helices α11 and α12 ) proceed via the β1-β2 loop to the α9-α10 loop . From this point , the signal bifurcates toward the Ω-loop via the α7-α8 loop or toward the α3-α4 pivot via the α2-β4 loop . In KPC-2 , the perturbation caused by ligand unbinding between the α2 and α7 helices results in conformational changes in loop α2-β4 , leading to β4 and onward to the pivot of the α3-turn-α4 helix . These conformational changes are relayed to the Ω loop via the α7-α8 loops . In addition , the signal can also take another route from the α7-α8 loop toward the β9-α12 loop , which lies adjacent to the hinge region . It is worth emphasizing that the TEM-1 and KPC-2 systems display a striking resemblance in that the flow of information is toward a common endpoint , despite the two different points of origin . Thus , even though the propagation pathway taken is different , in each case , the signals accumulate to have a structural impact on the conformation of the Ω loop and the α3-turn-α4 helix . These results demonstrate communication between allosteric ligand binding sites and the active sites of the enzymes , which could be exploited in alternative strategies for inhibitor development . All class A β-lactamase enzymes share conserved structural architecture ( Philippon et al . , 2016; Galdadas et al . , 2018 ) . Mutational studies and the location of sites of substitutions in clinical variants suggest the importance to activity of the hinge region , Ω-loop , and α3-turn-α4 helix , including the spatial position of the conserved aromatic residue at 105 ( or the analogous position in other class A β-lactamases ) ( Palzkill , 2018; Philippon et al . , 2016; Banerjee et al . , 1998; Papp-Wallace et al . , 2010b ) . Perturbations around these sites , as identified in the simulations here , may constitute a general mechanism by which a conformational signal transmitted from an allosteric site is relayed via cooperative coupling of loop dynamics to affect catalytic activity . Exploitation of such signaling networks may constitute a novel strategy for the development of new types of inhibitors for these key determinants of bacterial antibiotic resistance . To study allosteric modulation of class A β-lactamases , we started by identifying crystal structures of TEM-1 and KPC-2 β-lactamases with allosteric ligands bound . From the ~80 structures present in the PDB , there are only two crystal structures of class A β-lactamases that have a ligand bound in an allosteric pocket . For TEM-1 , the 1 . 45 Å crystal structure in complex with FTA [3- ( 4-phenylamino-phenylamino ) −2- ( 1h-tetrazol5-yl ) -acrylonitrile] was chosen as the starting structure ( PDB id: 1PZP ) for this work Horn and Shoichet , 2004 . In this structure , the inhibitor binds between helices α11 and α12 ( Figure 1a ) , in a site ~16 Å away from the active site Ser70 . Two unstructured residues from the C-terminal end ( His289 , Trp290 ) were removed from the crystal structure . For KPC-2 , the 1 . 35 Å crystal structure in complex with a coumarin phosphonate analogue , GTV [ ( 5 , 7-dimethyl-2-oxo-2h-1-benzopyran-4-yl ) methylphosphonic acid] , was chosen as the starting structure ( PDB id: 6D18 ) ( Pemberton et al . , 2019 ) . GTV binds in three sites on KPC-2 ( Figure 1b ) : the first is in the active site ( orthosteric ligand ) ; the second site is adjacent to helix α6 ( allosteric ligand 1 ) ; and the third ( allosteric ligand 2 ) is on the distal end of the enzyme , ~16 Å from the active site Ser70 in between helices α2 and α7 . The orthosteric and allosteric ligand 1 ( Figure 1b ) were discarded because of their direct proximity to the active site and replaced by water . Three unstructured residues from the N-terminal end ( His23 , Met24 , Leu25 ) and seven from the C-terminal end ( Leu288-Gly294 ) were removed from the starting structure to avoid any simulation artifacts arising as a result of terminal fraying during simulations . The protonation states of the amino acid side chains were determined at pH 7 . 0 , using the ProteinPrepare functionality as implemented in the high-throughput molecular dynamics ( HTMD ) framework ( Martínez , 2015; Doerr et al . , 2016 ) . Charges were assigned on the basis of their local environment , via optimization of the hydrogen-bonding network of the protonated structure ( Martínez , 2015 ) . Parameters for the ligands were generated using the Antechamber tool ( Case et al . , 2005 ) . The geometry was optimized at the B3LYP/6-31G ( d ) level and RESP charges were fitted using electrostatic potential obtained at the HF/6-31G ( d ) level . The necessary nonbonded parameters for the dynamics of the ligands were adopted from GAFF2 ( Wang et al . , 2004 ) . All complexes were set up using tleap , as implemented in the Amber MD package . The Amber ff14SB forcefield ( Maier et al . , 2015 ) was used for the protein . In total , four complexes were set up , including an allosteric IB ( inhibitor-bound ) and an Apo ( no ligand ) system for both TEM-1 and KPC-2 β-lactamases . The Apo system was generated by removing the inhibitor from the allosteric binding site . In all simulated complexes , there is no ligand bound to the orthosteric site . Each complex was solvated using TIP3P water in a cubic box , whose edge was set to at least 10 Å from the closest solute atom ( Mark and Nilsson , 2001 ) . The systems were neutralized using K+ and Cl- counter ions . The simulation protocol was identical for each system . The systems were minimized and relaxed under NPT conditions for 5 ns at 1 atm . The temperature was increased to 300 K using a time step of 4 fs , rigid bonds and a cutoff of 9 Å , and particle mesh Ewald summations switched on for long-range electrostatics ( Essmann et al . , 1995 ) . During the equilibration step , the protein’s backbone and the ligand atoms were restrained by a spring constant set at 1 kcal mol−1 Å−2 , while the ions and solvent were free to move . The production simulations were run in the NVT ensemble using a Langevin thermostat with a damping constant of 0 . 1 ps and hydrogen mass repartitioning scheme to achieve a time step of 4 fs ( Feenstra et al . , 1999 ) . The final production step was run without any restraints . All simulations were run using the ACEMD MD engine as implemented in the HTMD framework ( Doerr et al . , 2016 ) . Visualization of the simulations was done using the VMD package ( Humphrey et al . , 1996 ) . The analysis was carried out using GROMACS tools ( Abraham et al . , 2015 ) , MDLovofit ( Martínez , 2015 ) , and in-house scripts ( Oliveira et al . , 2019a ) . All systems were considered equilibrated after 50 ns . The dynamic cross-correlations for Cα-Cα were calculated using cpptraj analysis program ( Roe and Cheatham , 2013 ) . The results were plotted using in-house scripts and visualized using MATLAB ( http://www . mathworks . com ) . An independent-samples Student’s t-test was used to compare the ApoEQ and IBEQ RMSFs and to assess the significance of the differences observed ( Oliveira et al . , 2019a; Roy and Laughton , 2010 ) . The sample size used for the t-test was the 20 RMSF profiles of the ApoEQ and IBEQ independent simulations . The assumption used for the t-test was that the samples from the two states were independent , the dependent variable was normally distributed , and the variances of the dependent variable were equal . The figures were made using PyMol ( http://www . schrodinger . com ) , VMD ( Humphrey et al . , 1996 ) , ChimeraX ( Goddard et al . , 2018 ) , Protein Imager ( 3dproteinimaging . com ) ( Tomasello et al . , 2020 ) , and Molsoft ICM-Pro package ( http://www . molsoft . com ) .	Antibiotics are crucial drugs for treating and preventing bacterial infections , but some bacteria are evolving ways to resist their effects . This ‘antibiotic resistance’ threatens lives and livelihoods worldwide . β-lactam antibiotics , like penicillin , are some of the most commonly used , but some bacteria can now make enzymes called β-lactamases , which destroy these antibiotics . Dozens of different types of β-lactamases now exist , each with different properties . Two of the most medically important are TEM-1 and KPC-2 . One way to counteract β-lactamases is with drugs called inhibitors that stop the activity of these enzymes . The approved β-lactamase inhibitors work by blocking the part of the enzyme that binds and destroys antibiotics , known as the 'active site' . The β-lactamases have evolved , some of which have the ability to resist the effects of known inhibitors . It is possible that targeting parts of β-lactamases far from the active site , known as 'allosteric sites' , might get around these new bacterial defences . A molecule that binds to an allosteric site might alter the enzyme's shape , or restrict its movement , making it unable to do its job . Galdadas , Qu et al . used simulations to understand how molecules binding at allosteric sites affect enzyme movement . The experiments examined the structures of both TEM-1 and KPC-2 , looking at how their shapes changed as molecules were removed from the allosteric site . This revealed how the allosteric sites and the active site are linked together . When molecules were taken out of the allosteric sites , they triggered ripples of shape change that travelled via loop-like structures across the surface of the enzyme . These loops contain over half of the known differences between the different types of β-lactamases , suggesting mutations here may be responsible for changing which antibiotics each enzyme can destroy . In other words , changes in the 'ripples' may be related to the ability of the enzymes to resist particular antibiotics . Understanding how changes in one part of a β-lactamase enzyme reach the active site could help in the design of new inhibitors . It might also help to explain how β-lactamases evolve new properties . Further work could show why different enzymes are more or less active against different antibiotics .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "medicine", "structural", "biology", "and", "molecular", "biophysics" ]	2021	Allosteric communication in class A β-lactamases occurs via cooperative coupling of loop dynamics
Given a sample of genome sequences from an asexual population , can one predict its evolutionary future ? Here we demonstrate that the branching patterns of reconstructed genealogical trees contains information about the relative fitness of the sampled sequences and that this information can be used to predict successful strains . Our approach is based on the assumption that evolution proceeds by accumulation of small effect mutations , does not require species specific input and can be applied to any asexual population under persistent selection pressure . We demonstrate its performance using historical data on seasonal influenza A/H3N2 virus . We predict the progenitor lineage of the upcoming influenza season with near optimal performance in 30% of cases and make informative predictions in 16 out of 19 years . Beyond providing a tool for prediction , our ability to make informative predictions implies persistent fitness variation among circulating influenza A/H3N2 viruses . A general method to predict the evolutionary trajectories of asexual populations would be extremely valuable for understanding the population dynamics of pathogens or of malignant cells . For example , the vaccine against seasonal influenza needs to be updated frequently since virus populations evolve to evade increasing immunity among humans ( Hampson , 2002; Nelson and Holmes , 2007 ) . Reliable prediction of the strains most likely to circulate in the upcoming season , and particularly the ability to predict antigenic change , would be transformative to the vaccine strain selection process . Predictability from genetic sequence data requires heritable fitness variation among the sampled sequences . Neutral evolution - population dynamics in the absence of selective pressure - is by definition unpredictable: all sequences are equally fit . Yet even when selection determines the success of individual lineages , predictability depends on the effect size of fitness-altering mutations . Two competing scenarios of adaptive evolution are illustrated in Figure 1 . If evolution proceeds via rare mutations with large phenotypic effects , the population is homogeneous in fitness most of the time ( Figure 1A ) . In this case large effect mutations can convert any genome into the fittest in a single generation . Prediction from sequence alone is only possible if the time of sampling happens to be during a brief sweep of a large effect mutation . In contrast , continuous accumulation of small effect mutations ( Figure 1B ) results in a gradual change in fitness of lineages and persistent variation in fitness ( Tsimring et al . , 1996 ) . A genealogical tree then potentially contains predictable patterns: the fitness of most lineages decreases over time ( movement to the left in Figure 1 ) , due to a changing environment or the accumulation of weakly deleterious mutations . Only a few adapt rapidly enough to stay among the most fit in the population ( Rouzine et al . , 2003; Brunet et al . , 2007; Desai and Fisher , 2007; Hallatschek , 2011; Goyal et al . , 2012; Desai et al . , 2013; Neher and Hallatschek , 2013 ) and thus have a chance to continue into the future . 10 . 7554/eLife . 03568 . 003Figure 1 . Genealogies in adapting populations . ( A and B ) illustrate the genealogy of two successive samples embedded into the ( Malthusian ) fitness distribution of the population indicated in grey . In absence of adaptive mutations , fitness declines due to a changing environment or accumulation of deleterious mutations . Only one lineage ( thick line ) persists from first sample to second sample . ( A ) Evolution proceeds via rare large effect mutations ( dashed arrows ) that occur in a population with little fitness variance . All individuals are roughly equally likely to pick up the large effect mutation , rendering evolution unpredictable from sequence data alone . ( B ) Conversely , if adaptation is due to many small effect mutations , the successful lineage ( thick ) is always among the most fit individuals . Being able to predict relative fitness therefore enables to pick a progenitor of the future population . DOI: http://dx . doi . org/10 . 7554/eLife . 03568 . 003 In the specific context of human seasonal influenza A/H3N2 viruses , the study of their antigenic evolution has identified specific amino-acid substitutions with large phenotypic effects ( Koel et al . , 2013 ) , that have been responsible for the observed stepwise replacement of antigenic variants over time ( Smith et al . , 2004 ) . Yet , the evolution of seasonal influenza viruses is also marked by the continuous accumulation of mutations that have small or no antigenic effects but nevertheless potentially affect fitness ( Bhatt et al . , 2011; Strelkowa and Lässig , 2012 ) , for example compensatory or permissive mutations ( Gong et al . , 2013 ) . Previous attempts at predicting the evolution of seasonal influenza viruses have tried to identify molecular signatures that are predictive of future success ( Bush et al . , 1999 ) or used clustering approaches based on amino acid sequences ( Plotkin et al . , 2002 ) . Recently , Łuksza and Lässig ( 2014 ) constructed an explicit fitness model based on sequence data from the hemagglutinin ( HA1 ) surface protein . The utility of these explicit models depend on the availability of extensive historical data or a detailed understanding of the influenza virus sequence-to-fitness map . Rather than constructing an explicit fitness model , which is currently impossible for most organisms , we developed a general algorithm to infer fitness from the shape of reconstructed genealogical trees without using any molecular information . Our approach is based on a simple idea: since high ( Malthusian ) fitness implies many offspring , which in turn implies branching , the shape of the tree can be exploited to infer fitness ( Dayarian and Shraiman , 2014 ) . Here , we developed a quantitative model of fitness dynamics on genealogical trees , which is based on recent progress in understanding the statistical structure of genealogies in adapting populations ( Neher and Hallatschek , 2013 ) . Following Neher and Hallatschek ( 2013 ) , our model assumes: 1 ) that the population is under persistent directional selection and 2 ) fitness changes along lineages in small steps through the continuous accumulation of small effect mutations ( Figure 1B ) . This fitness model resembles the well-known infinitesimal model of quantitative genetics ( Falconer and Mackay , 1996 ) in the sense that many small effect mutations give rise to a bell-shaped fitness distribution on which selection acts ( Neher , 2013 ) . However , the infinitesimal model itself provides no insight into the relationship between the structure of genealogical trees and fitness: this insight stems from the more recent work on the dynamics of adaptation in large asexual populations ( Tsimring et al . , 1996; Rouzine et al . , 2003; Desai and Fisher , 2007; Desai et al . , 2013 ; Neher and Hallatschek , 2013 ) and in populations with occasional reassortment ( Neher and Shraiman , 2011 ) . After testing the algorithm on simulated data we apply our algorithm to historical data on human seasonal influenza A/H3N2 virus hemagglutinin sequences . Despite multiple confounding factors – discussed below – we find that our algorithm makes informative predictions about influenza virus evolution . Intuitively , we expect that an exceptionally fit internal node in a genealogical tree will be at the root of a rapidly branching , and hence expanding , clade ( e . g . node 2 in Figure 2A ) . Similarly , extant individuals with high fitness are likely to be recent descendants of internal nodes with high fitness ( e . g . node 3 in Figure 2A ) . By tracing fitness along lineages and integrating across the tree , the algorithm described below makes this intuition precise and quantitative . 10 . 7554/eLife . 03568 . 004Figure 2 . Inferring fitness from genealogical trees . ( A ) The inference algorithm is based on branch propagators associated with each branch of the reconstructed tree ( middle ) . Branch propagators characterize the fitness distribution of child nodes given the fitness of the ancestral node ( left ) . The internal node 2 would have higher marginal fitness estimate ( right ) than node 1 , as node 2 has more children . The inferred distribution of the fitness of the external node 3 has broadened along the branch from node 2 . ( B–D ) Analysis of simulated data . Panel B shows for a typical example that inferred fitness is well correlated with the true fitness with a rank correlation coefficient ρ=0 . 56 . This correlation increases with increasing mutation rate as shown in panel C for 100 simulated data sets each ( boxes cover the interquartile range , red lines indicate the median ) . Panel D shows that the sequence with the highest inferred fitness tends to be similar to the population 200 generations in the future . Both axis show the average Hamming distance to the future population between the predicted and the post-hoc optimal sequence on the y and x-axis , respectively , for 100 simulated data sets . Both distances are relative to the average distance between the present and future population . Parameters: N=20000 , nA=0 . 08 , Γ=0 . 2 , u=0 . 064 ( B , D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03568 . 00410 . 7554/eLife . 03568 . 005Figure 2—figure supplement 1 . Predictability increases with genetic diversity . The prediction performance quantified by the rank correlation coefficient between the inferred and true fitness increases with pairwise diversity . Large Γis superior at small pairwise distances , which corresponds to a regime of few large effect mutations . Smaller Γdoes better in at large pairwise distance where fitness variation is spread among many loci . DOI: http://dx . doi . org/10 . 7554/eLife . 03568 . 00510 . 7554/eLife . 03568 . 006Figure 2—figure supplement 2 . Prediction from continuously sampled sequences . Same as Figure 2B–D , but with continuous sampling of 200 simulated sequences over 100 generations , as opposed to one sample from exactly one time point . Panels B&C shows that the rank correlation does not suffer when sampled continuously , at least at moderate or large mutation rates . Genetic distance of the predicted strain to future population behaves similarly . Parameters: N=20000 , ω=0 . 01 , Γ=0 . 2and u=0 . 064 . DOI: http://dx . doi . org/10 . 7554/eLife . 03568 . 006 As input , our algorithm requires a genealogical tree , e . g . a tree reconstructed from a sample of genomic sequences . For a given tree T , we derived the joint probability distribution P ( x\|T ) for the fitnesses x=x0 , x1 , … of all internal nodes ( corresponding to reconstructed ancestral sequences ) and external nodes ( corresponding to the sampled genomes ) . Fitness xi of each node i is measured relative to the population mean fitness at the time when the corresponding individual was sampled . P ( x\|T ) is given by a product of propagators g ( ·\|· ) for each branch ( 1 ) P ( x\|T ) =p0 ( x0 ) Z ( T ) ∏i=0nintg ( xi1 , ti1\|xi , ti ) g ( xi2 , ti2\|xi , ti ) , where p0 ( x ) is the fitness distribution in the population ( see ‘Materials and methods’ for details ) and the index i runs from 0 ( the root ) through all nint internal nodes . The indices i1 and i2 denote the two children of node i , while Z ( T ) ensures normalization of the distribution . Eq . ( 1 ) has a structure similar to the expression for the likelihood of sampled sequences , given a tree T , defined in phylogenetic analysis ( Felsenstein , 2003 ) . The main difference is that instead of defining the probability of mutation from one character state to another , the branch propagator g ( xj , tj\|xi , ti ) describes the likelihood of the lineage to connect an ancestor with fitness xi at time ti to a child with fitness xj at a later time tj ( child in sense of a subclade in the tree , rather than direct offspring ) . Note that a branch connecting nodes i and j implies that all sampled descendants of i are also descendants of j , i . e . , the ‘branch does not branch’ . This non-branching condition is part of the branch propagator which therefore depends on the fraction ω of the total population that is represented in the sample ( see ‘Materials and methods’ for details ) . Figure 2A illustrates the propagator as function of child fitness xj , which describes the fitness distribution of children , conditioned on ancestral fitness xi . At small Δt=tj−ti , the distribution is peaked around the ancestor . At long times , memory of ancestral fitness is lost and the propagator approaches the population distribution . Backwards in time , g ( xj , tj\|xi , ti ) describes ( using the Bayesian inversion formula [Felsenstein , 2003] ) the fitness distribution of the ancestor i given a sampled child with fitness xj at time tj . Far in the past , the ancestor fitness distribution converges to a narrow peak in the high fitness tail ( Rouzine and Coffin , 2007; Neher and Hallatschek , 2013 ) . See ‘Materials and methods’ for a more detailed discussion . The fitness dynamics along a lineage resemble a random walk on which each step corresponds to a mutation with a certain effect on fitness . This walk is biased towards high fitness by selection , which makes fitter lineages more likely to survive and eventually be sampled . If many mutations contribute , the dynamics of fitness along branches can be approximated by selection-biased diffusion ( SBD ) as described in ‘Materials and methods’ , Equation ( 9 ) – Equation ( 11 ) . The fitness diffusion constant of a branch is given by D=u〈s2〉/2 , where u is the genome wide mutation rate , and 〈·〉 denotes the average over the effect sizes of mutations ( Tsimring et al . , 1996 ) . Fitness diffusion and stochasticity due to finite populations determine the fitness variance σ2 in the population ( Cohen et al . , 2005 ) . Based on the SBD approximation derived in ‘Materials and methods’ , we implemented a program that numerically solves for the branch propagator and , by going up and down the tree using a ‘Message Passing’ ( similar to dynamic programming ) technique ( Mézard and Montanari , 2009 ) , calculates the marginal fitness distribution for each node as illustrated in Figure 2A , for details see ‘Materials and methods’ . To explore the extent to which the idealized SBD model assuming infinitesimal mutations is able to infer fitness when evolution happens via discrete mutations , we simulated a simple model of evolution with fixed fitness variance ( σ=0 . 03 ) ( Zanini and Neher , 2012 ) . In order to mimic adaptive evolution in a changing environment we introduced sites in the simulated genome that allow for beneficial mutations at rate nA=0 . 02 , … , 0 . 16 per generation in a genome otherwise dominated by deleterious mutations . Every 200 generations , we took a random sample of sequences from the simulated population . We recorded the fitness of each sampled sequence , which we will compare with our inferences below . In order to apply the fitness inference method to a reconstructed tree , we needed to parameterize the model and convert branch length measured as similarity between sequences into time . When measuring time in units of σ−1 , the SBD model has only one free dimensionless parameter Γ=Dσ−3 that describes the relative importance of selection and stochastic processes . Γ is inversely proportional to the square root of the logarithm of the population size and hence does not vary greatly ( Tsimring et al . , 1996; Cohen et al . , 2005 ) . We used Γ=0 . 2 and 0 . 5 corresponding to moderate and more rapid diffusion relative to selection , respectively . Coalescent theory of adapting population connects pairwise sequence similarity to Γ . The choice of Γ fixes the conversion from branch length to time via Equation ( 20 ) ( Neher and Hallatschek , 2013 ) . In addition to Γ we need to fix ω . Since we used a sample of 200 sequences out of a total of N=20000 sequences , ω=0 . 01 ( ultimately , ω/σ enters the algorithm , see ‘Materials and methods’ ) . Using these parameters , we applied our method to a reconstructed tree and report the mean posterior fitness as ‘inferred fitness’ for each internal and external node . Figure 2B shows the inferred vs true fitness for a typical simulation . The rank order of fitness is well predicted ( Spearman's correlation coefficients around 0 . 5 ) . Figure 2C shows that fitness rankings improve with increasing mutation rates . This is expected , since increased mutation rates correspond to a larger number of mutations that contribute to fitness and make the SBD model a better approximation . This behavior is consistent across different rates of adaptive mutations and depends weakly on our choice of Γ ( Figure 2—figure supplement 1 ) . Large Γ performs better at low mutation rates when fitness diversity is dominated by only a few mutations , corresponding to more rapid fitness diffusion relative to selection and coalescence . Next , we asked whether sequences that we predict to have high fitness are close in sequence to the progenitor lineage of future populations . Figure 2D shows the Hamming distance Δ ( prediction ) of the sequence of the individual with the highest fitness estimate to the population 200 generations in the future vs the Δ ( minimal ) for the post-hoc optimal pick . The measure Δ ( sequence ) is normalized to the average Hamming distance between the present and future population . In 40 out of 100 simulations , the top-ranked sequence is an almost optimal pick ( points close to the diagonal in Figure 2D ) . In 8 out of 100 cases , the prediction is better than a random pick ( points below the dashed line Figure 2D ) . The fitness inferences shown in Figure 2B–C used 200 sequences sampled from the same generation . However , the influenza data to which we apply our algorithm below is continuously sampled throughout the year . In Figure 2—figure supplement 2 we reproduce panels B–C using 200 sequences sampled from the simulation over a time interval of 100 generation . This gives highly similar results . In general , faithful inference of the posterior fitness distribution requires numerical solution for the branch propagators and knowledge of the parameters Γ and ω/σ . We observed , however , that the ranking of nodes by fitness and the prediction of progenitor lineages depends little on these parameters . This insensitivity suggests that the fitness ranking depends primarily on a more universal quantity on which the inference algorithm builds . In ‘Materials and methods’ , we show that the fitness estimates of internal nodes increase with the total branch length downstream of these nodes–at least for short time periods . The downstream tree length acts as a “polarizer” that pushes the fitness distribution of the node away from the population mean towards high fitness . For given number of descendants , the length of a subtree is maximal if it is star-like . This is intutive , as star-like subtrees indicate rapid branching ( or multiple mergers backwards in time ) which is expected for high fitness nodes . Conversely , prolonged absence of branching of a lineage indicates relatively low fitness . If fitness changes gradually along lineages , high fitness of a node will coincide with both upstream and downstream branching–at least within a certain neighborhood of the tree . The relevant size of the neigborhood will depend on how rapidly fitness decorrelates along lineages . Based on this intuition , we developed a model-independent heuristic ranking algorithm: for each internal and terminal node i , we calculate a local branching index ( LBI ) λi ( τ ) defined as total surrounding tree length exponentially discounted with increasing distance from the focal node . The scale τ of the exponential discounting corresponds to the size of the relevant tree neighborhood or the time over which fitness is ‘remembered’ across the tree . Within the SBD model , τ corresponds to the equilibration time scale of lineage fitness in the high fitness tail , which is of the order Tc/logN , where Tc is the coalescence time scale ( Neher and Hallatschek , 2013 ) . The LBI can be efficiently calculated with the same message passing techniques we used to calculate the posterior fitness distribution . Remarkably , rankings obtained by this simple heuristic are almost as accurate as fitness inference using the more complex SBD model . Figure 3 shows Spearman’s correlation coefficient of λi ( τ ) with true fitness as a function of pairwise difference for different memory time scales τ and compares it to the ranking via mean inferred fitness . The heuristic λi ( τ ) not only correlates well with true fitness in simulations but sequences with the highest λi ( τ ) also tend to be close to the progenitor of future populations ( Figure 3—figure supplement 1 ) . Comparing the performance of the LBI to the full fitness inference in Figure 3 , we concluded that a neighborhood size should be τ≈0 . 0625 of the average pairwise distance in the sample . 10 . 7554/eLife . 03568 . 007Figure 3 . Local tree length as a fitness ranking . Rank correlation between the true fitness and the LBI λi ( τ ) is shown as a function of pairwise diversity in the sample . Different curves correspond to different neighborhood sizes τ , which is measured in units of the average pairwise distance . DOI: http://dx . doi . org/10 . 7554/eLife . 03568 . 00710 . 7554/eLife . 03568 . 008Figure 3—figure supplement 1 . The LBI predicts progenitor sequences . Sequences with the highest LBI in the sample tend to be close to the progenitor of future populations . The measure Δshows the distance of the predicted sequence to the population 200 generations in the future ( relative to the average distance between the two populations ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03568 . 008 Having validated our algorithm on simulated data and presented a model independent method to rank sequences , we attempted to predict progenitor sequences of seasonal influenza A/H3N2 viruses . We used samples of influenza A/H3N2 virus hemagglutinin ( HA1 ) sequences from one year ( May–February , Asia and North America , at most 100 sequence from each region ) to predict the closest relative of the population circulating in the following ( northern hemisphere ) winter ( October–March , Asia and North America ) for the years 1995–2013 . All HA1 domain sequences used for our analysis came from the public domain and are available from Influenza Research Database ( www . fludb . org ( Squires et al . , 2012 ) ) . Next , we built maximum likelihood trees using fasttree ( Price et al . , 2009 ) , collapsed zero-length branches into polytomies , and ranked external and internal nodes using the LBI . We set the memory time scale to τ=0 . 0625 in units of average pairwise distance as suggested by the simulation data . Details of the data sets used for making predictions and discussion of potential biases are given in ‘Materials and methods’ . Figure 4A&B show example trees of the prediction and test sets for 2007 . 10 . 7554/eLife . 03568 . 009Figure 4 . Predicting the evolution of seasonal influenza A/H3N2 viruses . ( A ) A genealogical tree of a sample of HA1 sequences from May 2006 to end of February 2007 . Nodes are colored according to our fitness ranking λi ( τ ) . The highest ranked node is marked by a black arrow . ( B ) A tree of the same sequences from ( A ) ( colored ) and sequences from October 2007 to end of March 2008 ( in grey ) . Our algorithm successfully predicts a sequence genetically close and directly ancestral to viruses circulating the following winter . ( C ) For each year from 1995 to 2013 we predicted a progenitor sequence and calculated its nucleotide distance to the A/H3N2 population of the following winter . Predictions based on terminal or internal sequences are very similar . The figure shows the average Δ ( prediction ) of 50 runs using subsamples of the data . A random pick from the prediction set corresponds to the solid line at 1 . The dashed lines indicate the optimal extant sequence at time of prediction . The distance of the dashed line from the line at 1 indicates the closeness of the optimal extant sequence to future populations . DOI: http://dx . doi . org/10 . 7554/eLife . 03568 . 00910 . 7554/eLife . 03568 . 010Figure 4—figure supplement 1 . Variation of predictions upon variation of the memory time scale of the LBI λi ( τ ) . Each year shows two lines–one for internal and external nodes–that show the variation of the prediction as τ varies from 2−6to 4 in multiples of 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 03568 . 01010 . 7554/eLife . 03568 . 011Figure 4—figure supplement 2 . Comparison to predictions by Łuksza and Lässig ( 2014 ) . In many years , choosing the sequence with the highest LBI results in a very similar sequence to that predicted by Łuksza and Lässig ( 2014 ) . In some years the LBI resulted in a pick closer to the future , in other years the sequences predicted by Łuksza and Lässig ( 2014 ) was a better choice . Łuksza and Lässig aimed at minimizing amino-acid distance at epitope position , rather than nucleotide distance as we do here . The two measures are strongly correlated , but nucleotide distance has better resolution and is hence used here . DOI: http://dx . doi . org/10 . 7554/eLife . 03568 . 01110 . 7554/eLife . 03568 . 012Figure 4—figure supplement 3 . High LBI predicts clade expansion . Each dot corresponds one clade with less than 75% frequency in a sample of sequences from May to February of year t . The excess of points in the upper right corner shows that high LBI is predictive of clade expansion . The x-axis shows its rank according to the LBI in this year , normalized to the iterval [0 , 1] . The y-axis shows the rank according to clade growth measured as the ratio of frequency of this clade in year t+1and year t . Again , rankking is done on a yearly basis and normalized to the interval [0 , 1] . This plot contains data from years 2003–2013 for which there are sufficiently many sequences to calculate meaningful clade frequencies . The pointsin the lower half of the plot correspond to all clades that do not continue into the next year . DOI: http://dx . doi . org/10 . 7554/eLife . 03568 . 012 Figure 4C shows the nucleotide distance of our prediction to the A/H3N2 virus population of the next season , both for the top-ranked internal and external node of each year . Using the highest ranked external node ( Figure 3C , black squares ) is similar to using the highest ranked internal node ( Figure 3C , red diamonds ) in all years but 1997 . The highest ranked internal node predict years 1997–1999 , 2003 , 2006–2009 , and 2013 , reasonably well . Notably , they fail in 1995 , 1996 , and 2002 , while being of intermediate accuracy in the remaining years . The dependence of the prediction accuracy on the neighborhood size τ is shown in Figure 4—figure supplement 1 . We also predicted successful progenitor strains using the fitness inference based on the SBD model which yields results very similar to the ranking by LBI–sometimes slightly better , sometimes worse depending on parameter choice . We compared our predictions to vaccine strain predictions obtained by Łuksza and Lässig ( 2014 ) who predict progenitors of future epidemics as we do here , albeit using an influenza specific model with four parameters , two of which are trained for each individual prediction on data from several preceding years . On average , using the same time cutoffs for prediction ( February to predict October ) as we used above , Łuksa and Lässig achieve an accuracy comparable to our parameter-free ranking based ( see Figure 4—figure supplement 2 ) . Interestingly , these two rather different approaches yield very similar predictions on a year to year basis . One potential explanation for this concordance is an ad hoc aspect of Łuksa and Lässig's model meant to capture epistatic interactions: the total number of synonymous mutations downstream of each clade is used as an additional predictor . The number of synonymous mutations is strongly correlated with tree length and hence with λi ( τ ) . To quantify prediction quality across years , we define the distance measure d= ( Δ ( prediction ) −Δ ( minimal ) ) / ( 1−Δ ( minimal ) ) such that an optimal prediction has d=0 and a random pick has d=1 . The average of d over all years is denoted by d¯ . Figure 5 shows bootstrap distributions of d¯ for our methods and compares it to Łuksza and Lässig ( 2014 ) as well as two naive prediction methods: ( i ) a growth rate estimate of individual clades obtained by fitting an exponential curve to the fraction of the total sequences that are part of this clade in three time intervals between May and February , and ( ii ) the sequence of the most advanced node in a ladderized tree . Predictions with the method described here and by Łuksza and Lässig ( 2014 ) are comparable within errorbars , while the two naive estimators do substantially worse on average . The dependence of the average predictive power of the LBI on the neighborhood size τ is shown in Figure 5—figure supplement 1 . 10 . 7554/eLife . 03568 . 013Figure 5 . Comparison of predictors . Transformed genetic distance d¯ averaged over 1000 bootstrap samples ( bootstrapping years ) to the next influenza season . We compared our method using the sequence of the top ranked internal node , external node , the predictions by Łuksza and Lässig ( 2014 ) , the ancestral sequence of clades with the largest estimated growth rate , and the sequence of the most ‘advanced’ node in a ladderized tree . DOI: http://dx . doi . org/10 . 7554/eLife . 03568 . 01310 . 7554/eLife . 03568 . 014Figure 5—figure supplement 1 . Dependence of prediction accuracy on τ . Predictions for influenza virus A/H3N2 based on the LBI improve with increasing the memory time scale τ . Prediction accuracy is assessed as nucleotide distance to the future sample scaled such that the optimal pick as d=0and a random pick has d=1 , averaged over 50 repeated predictions per year on different subsamples of the data ( at most 100 sequences from Asia and North-America , 70% of the available data in cases fewer than 100 sequences are available ) . The figure shows the average of d over years 1995–2013; the accuracy of predictions by Łuksza and Lässig ( 2014 ) is shown as black line; the value of τ used in the remainder of the manuscript is indicated by the dashed vertical line . DOI: http://dx . doi . org/10 . 7554/eLife . 03568 . 014 Changes in fitness along branches can be associated with the types of mutations on those branches . We found that branches corresponding to the top quartile of differentials of λi ( τ ) are enriched for non-synonymous substitutions over synonymous mutations . Restricting non-synonymous mutations to the epitopes A–D ( used in ( Łuksza and Lässig , 2014 ) and defined in ( Shih et al . , 2007 ) ) increases this enrichment to approximately 2-fold , see Table 1 . Further restriction to the 7 loci identified Koel et al . increases the enrichment slightly , but their number is small and the power to detect additional enrichment is low . These findings are consistent with the notion that influenza evolution is driven by antigenic novelty ( Wiley et al . , 1981; Hampson , 2002; Smith et al . , 2004 ) and provide independent confirmation of the power of the sequences ranking and fitness inference algorithm . 10 . 7554/eLife . 03568 . 015Table 1 . Non-synonymous mutations at epitopes correlate with increasing fitnessDOI: http://dx . doi . org/10 . 7554/eLife . 03568 . 015Quartile# non-syn# syn# epi# Koel2513015543750159178571075184205742110020922211522total68276028960Comparisonenrichmentp-valuenon-syn vs syn1 . 12n . s . epi vs syn1 . 90 . 002Koel vs syn2 . 20 . 08epi vs non-syn1 . 70 . 015Koel vs non-syn2 . 0n . s . For each tree constructed for the years 1995–2013 , we calculated the increment in λi ( τ ) with τ = 0 . 0625 along each branch and determined the likely mutations on each branch . Branches were then sorted into quartiles according to changes in λi ( τ ) . The left table shows the counts of non-synonymous ( non-syn ) , synonymous ( syn ) , non-synonymous mutations at epitope site ( epi ) and non-synonymous mutations at Koel positions ( Koel ) for branches in different quartiles . The right table quantifies the enrichment of certain types of mutations on branches in the top quartile relative the bottom quartile . Non-synonymous mutations at epitopes and Koel positions are approximately twofold enriched relative to synonymous mutations . Enrichment ( odds ratio ) and p-values were obtained using the Fisher exact test as implemented in scipy . stats ( Oliphant , 2007 ) . Starting with a model of adaptive evolution , we developed a probabilistic description of the fitness dynamics on genealogical trees and presented an algorithm to infer fitness of individual nodes in the tree . We validated this algorithm using trees reconstructed from simulated sequences and showed that the sequence with the highest inferred fitness tends to be a close match to the progenitor of future populations . Analysis of the model revealed that a simple quantity–the local branching index ( LBI ) –determines the fitness estimates and can be used to rank sequences by fitness with similar accuracy as the full fitness inference algorithm . The only parameter of the LBI is the size of the neighborhood on the tree and a suitable value can be chosen from simulated data . Our fitness inference framework is based on the selection-biased diffusion model that assumes evolution proceeds via accumulation of many small effect mutations . As expected , its predictive power increases with increasing level of non-neutral genetic diversity ( Figure 2C ) . However , predictive power is retained down to rather low pairwise distances , see Figure 2—figure supplement 1 , where the model is a poor approximation . This suggests that the relationship between fitness and the structure of genealogical trees is more universal than the specific details of the mutation effect distribution that drive evolutionary dynamics ( Neher and Hallatschek , 2013 ) . The essence of this relationship between fitness and tree shape is picked up by the LBI . When applied to influenza A/H3N2 viruses sequences , a ranking by LBI predicts progenitor lineages with high accuracy . One of the dominant paradigms for influenza A/H3N2 virus evolution has been the exploration of ‘neutral’ networks , punctuated by bursts of rapid adaptation through large effect mutations ( Koelle et al . , 2006; Nimwegen et al . , 1999 ) . In contrast , our ability to make meaningful predictions from the shape of genealogical trees of influenza virus sequences suggests that fitness variation persists in A/H3N2 populations . Fitness in the context of seasonal influenza viruses includes antigenic evolution as well as compensatory and deleterious mutations–within HA and other segments–that may contribute to fitness variation , shape the genealogies , and be determinants of future success . This conclusion is consistent with other existing evidence for ubiquitous selection in A/H3N2 populations ( Bhatt et al . , 2011; Strelkowa and Lässig , 2012 ) . The applicability of our fitness inference scheme and the LBI ranking is further supported by the substantial enrichment in the number of non-synonymous substitutions at epitope loci in the lineages with predicted high relative fitness . These epitopes historically have high dn/ds suggesting positive selection . Our model is agnostic to sequence and protein structure but nevertheless associates branches containing these mutations with increasing fitness . It is also clear that large effect mutations , such as the ones associated with antigenic cluster transitions ( Koel et al . , 2013 ) can play an important role in the evolution of human seasonal influenza viruses . Many of the years in which our predictions are suboptimal ( e . g . , 1995 , 2002 , and 2004 ) correspond to antigenic cluster transitions in which antigenic properties changed drastically via specific large effect mutations . We tried to improve predictions by assigning additional positive fitness increments to substitutions at those loci identified by Koel et al . While this did improve results in some years , it also resulted in false positives which erased the overall improvement in predictive power . In some years in which these mutations are important , they tend to occur on many genetic backgrounds . This could explain why these mutations be themselves are not very predictive in our framework . The fact that the branching patterns of reconstructed influenza A/H3N2 trees are predictive is surprising . In addition to occasional large effect effect mutations , e . g . those that cause substantial antigenic change , confounders such as the heterogeneity of sampling , complicated migration patterns , and demographic substructure should hamper prediction . The insensitivity to local oversampling is expected from the structure of our algorithm which senses the total length of subtrees ( rather then the number of leaves ) . Local oversampling will add many very short branches that perturb the total tree length only slightly . Subpopulations of different size , seasonality , and migration patterns , however , will perturb the coalescence patterns in parts of the reconstructed tree and should decrease predictability . Successful prediction therefore reinforces the conclusion that circulating influenza A/H3N2 populations harbor fitness variation . On the other hand , predictions might be improved by combining the shape of genealogical trees with antigenic information ( Bedford et al . , 2014 ) , biophysical and structural knowledge ( Koel et al . , 2013 ) , patterns of past evolution ( Łuksza and Lässig , 2014 ) , and plausible geographic sources ( Russell et al . , 2008; Lemey et al . , 2014 ) . However , each of these refinements introduces additional parameters into the model that need to be trained if not known a priori . A defining feature of our method to predict evolution is that it can operate on a static set of sequences from a single time point and does not require historical data . We use historical data for influenza A/H3N2 only to validate the predictions . In Figure 5 , we compare our results to a method that explicitly uses historical data ( available for the influenza A/H3N2 ) to identify low frequency but expanding clades . By extrapolating their expansion into the future , one can anticipate the dominant strains of next year . Interestingly we found that prediction based on the reconstructed genealogy not only captures similar information , but also performs comparably if not better , even without access to historical data . In summary , we have shown that the shape of reconstructed genealogies holds information about the relative fitness of the sampled individuals that can be exploited to predict the genetic composition of future populations , at least when fitness differences depend on multiple mutations . Since our algorithm requires nothing but a reconstructed genealogy as input , it should be applicable in many scenarios ranging from RNA viruses to cancer cell populations . Our algorithm is based on a branching process approximation to replicating clones within a finite population . Here , we first show how we use this approximation to calculate the probability that offspring of an individual with a certain fitness are sampled . From there , we derive an equation for the branch propagators , that we solve numerically , and combine the propagators into the expression for the posterior fitness distribution given in Equation ( 1 ) . The quantitative probabilistic description of clonal propagation is provided by the distribution P ( n\|x , t ) of the number of offspring n after time t given the ancestor had fitness x . Using a ‘1st-step’ equation , that is , writing an equation for infinitesimal changes at the initial point ( y , t ) , we find for the backwards master equation for P ( n\|x , t ) ( 2 ) P ( n\|x+Δtv , t+Δt ) =[1−Δt ( 2+x+u ) ] P ( n\|x , t ) +Δt〈uP ( n\|x+s , t ) 〉+Δt ( 1+x ) ∑n′=0nP ( n−n′\|x , t ) P ( n′\|x , t ) where the death rate is set to one and the birth rate is given by 1+x ( see also ( Neher and Hallatschek , 2013 ) ) . The first term corresponds to the probability of nothing happening in the time interval Δt and the second term in 〈·〉 corresponds to mutations averaged over the distribution μ ( s ) of possible fitness effects s with the total mutation rate given by u=∫ ds μ ( s ) . The last term corresponds to replication of the individual . At the earlier time point t+Δt , fitness x was larger by Δtv due to the deterioration of the environment with velocity v . So far , this equation holds for arbitrary distribution of fitness effects . To make analytical progress , we assume that the distribution of mutational effects is short-tailed ( exponential or steeper ) and that the total mutation rate u is large compared to the typical effect . In this case , Equation ( 2 ) can be rearranged into a differential equation where mutations are captured by the mean mutational effect and the mutational variance ( Tsimring et al . , 1996; Cohen et al . , 2005; Neher and Hallatschek , 2013 ) . ( 3 ) v∂P ( n\|x , t ) ∂x+∂P ( n\|x , t ) ∂t=− ( 2+x ) P ( n\|x , t ) +u〈s〉∂P ( n\|x , t ) ∂x+u〈s2〉2∂2P ( n\|x , t ) ∂x2+ ( 1+x ) ∑n′=0nP ( n−n′\|x , t ) P ( n′\|x , t ) The second term on the right hand side corresponds to the directional effect of mutations on fitness , while the third term to the diffusive dynamics of fitness due to mutations . To further analyze the behavior of P ( n\|x , t ) , it is useful to consider the generating function ψω ( x , t ) =∑n ( 1−ω ) nP ( n\|x , t ) , which obeys ( 4 ) ∂ψω ( x , t ) ∂t=− ( 2+x ) ψω ( x , t ) + ( u〈s〉−v ) ∂ψω ( x , t ) ∂x+u〈s2〉2∂2ψω ( x , t ) ∂x2+ ( 1+x ) ψω2 ( x , t ) Defining ϕω ( x , t ) =1−ψω ( x , t ) , the fitness diffusion constant D=u〈s2〉/2 , and the variance in fitness σ2=v−u〈s〉 , we have ( 5 ) ∂ϕω ( x , t ) ∂t=xϕω ( x , t ) −σ2∂ϕω ( x , t ) ∂x+D∂2ϕω ( x , t ) ∂x2− ( 1+x ) ϕω2 ( x , t ) with initial condition ϕω ( x , 0 ) =ω . This equation for the generating function can be solved numerically or analytically in limiting cases . To approximate the fitness distribution on a given tree , we will solve this equation numerically . It is also useful to explicitly define the ‘reproductive value’ R ( x , t ) defined as the expected number of offspring of a genotype with fitness x after t generations , R ( x , t ) =∑nnP ( n\|x , t ) . From the definition of the generating function it follows that R ( x , t ) =∂ωϕω ( x , t ) \|ω=0 . Differentiating Equation ( 5 ) w . r . t . ω and noting that ϕω ( x , t ) \|ω=0=0 yields a linear equation for R ( x , t ) ( essentially Equation ( 5 ) without the term ϕ2 ) which can be readily integrated . The expected number of offspring of one individual after time t given it initially had fitness x is ( 6 ) R ( x , t ) =ext−σ2t22+Dt33 This approximation is only valid for times short compared to the coalescence time Tc , but it offers important insight into the dynamics of lineages: Initially , the lineage grows into a clone with rate x . The second term in the exponent describes how this growth slows since the remainder of the population is adapting with rate σ2 . The last term accounts for the fact that the offspring we consider can themselves change in fitness through mutations , the action of which is captured by the fitness diffusion constant D . The generating function ϕω ( x , t ) derived above has the interpretation of the probability that a lineage is represented in a sample of size M from a population of size N with ω=M/N . From its definition , we have ( 7 ) ϕω ( x , t ) =1−∑n=0∞ P ( n\|x , t ) ( 1−ω ) n . Each term ( 1−ω ) n is the probability that none of the n offspring are in the sample . By summing over the distribution of n and subtracting the sum from 1 , one obtains the probability of at least one offspring being sampled . The generating function can be accurately approximated in regimes where ϕω is small and the non-linear term in Equation ( 5 ) can be neglected , as well as the regime of large enough x where ϕ ‘saturates’: ϕω ( x , t ) ≈x , see ( Neher and Hallatschek , 2013 ) . These two asymptotic solutions can be combined to yield the approximation ( 8 ) ϕω ( x , t ) ≈ωxR ( x , t ) x+ω[R ( x , t ) −1] Note that this approximation satisfies the initial condition ϕω ( x , 0 ) =ω , correctly tends to x for x>0 at long times , and recovers the neutral behavior ϕω ( 0 , t ) =ω/ ( 1+ωt ) in the x=σ2=D=0 limit . Having calculated the lineage sampling probability , we are now in a position to derive equations governing the behavior of the branch propagator , that is , the probability of there being an individual with fitness x at time t′ ( the child ) , given it descends from an ancestor with fitness y at time t and all sampled descendants of the ancestor are also descendants of the child . The latter condition amounts to the requirement that in a tree the link between the ancestor and the child does not branch . Using a ‘1st-step’ equation similar to Equation ( 2 ) , we have ( 9 ) g ( x , t′\|y+σ2Δt , t+Δt ) =g ( x , t′\|y , t ) −Δt ( 2+y ) g ( x , t′\|y , t ) +ΔtD∂2g ( x , t′\|y , t ) ∂y2+Δt2 ( 1+y ) [1−ϕω ( y , t ) ]g ( x , t′\|y , t ) . The last term describes a ‘birth’ event in the ancestral lineage with one of the branches surviving up to t′ ( at which time its fitness is in the [x , x+dx] interval ) while the other one is not sampled , which occurs with probability 1−ϕω ( y , t ) at a sampling density ω . The y→y+σ2Δt shift in the argument of the term on the left-hand-side parametrizes the translation of the mean fitness in time Δt . Equation ( 9 ) reduces to the differential equation ( 10 ) ∂tg ( x , t′\|y , t ) =[y−2ϕω ( y , t ) ]g ( x , t′\|y , t ) −σ2∂yg ( x , t′\|y , t ) +D∂y2g ( x , t′\|y , t ) which is complemented with the initial condition g ( x , t\|y , t ) =δ ( x−y ) . In deriving this condition , we have assumed that y≪1 , which is a good assumption when σ ( the standard deviation in fitness ) is small . The fitness differences in a single generation are small in most populations , such that this assumption is not restrictive . Furthermore , violation of this assumption does not change the qualitative behavior of the g ( ·\|· ) . When inferring fitness on trees , we will generally solve this equation numerically . Some limits , however , can be addressed analytically as we will see below . Numerical solutions of g ( x , t′\|y , t ) are shown in Figure 6 . For a fixed ancestor at ( y , t ) , g ( x , t′\|y , t ) is the density of offspring with fitness x at time t′ subject to the following condition: Only one individual from this group of offspring contributes to the sample at present ( this is the condition that the lineage connecting ( x , t′ ) and ( y , t ) is unbranched ) . The propagator g ( x , t′\|y , t ) broadens in x as t−t′ increases as shown in Figure 6A for a case of high ( red , y>2 ) and low ( blue , y=0 ) initial fitness . Figure 6B shows how the integral ∫dx g ( x , t′\|y , t ) increases with t for y>0 but decreases for y<0 . The integral of ∫dx g ( x , t′\|y , t ) differs from the reproductive value R ( y , t−t′ ) , shown as dashed lines in Figure 6B , only in the additional sampling condition . 10 . 7554/eLife . 03568 . 016Figure 6 . Numerical solution for the lineage propagator . Panel A shows g ( x , t′\|y , t ) as a function of x for different t′ at t=0 given the ancestor had Malthusian fitness y=0 ( blue ) or approximately y=2σ ( red ) . In both cases , the offspring tend to get less fit and the distribution broadens due to additional mutations . Saturated colors correspond to small t−t′ , light colors large t−t′ . Panel B shows ∫dx g ( x , t′\|y , t ) as a function of t−t′ for the high ( red ) and low ( blue ) fitness ancestor . The dashed lines show the approximation given in Equation ( 6 ) . In the high fitness case , Equation ( 6 ) overestimates ∫dx g ( x , t′\|y , t ) since it does not account for the non-sampling contribution . Panel C shows g ( x , t′\|y , t ) as a function of y , given the offspring is unfit ( blue ) or fit ( red ) . Ancestors tend to be fit regardless of offspring fitness and both ancestral distributions converge to a common curve far back in time . DOI: http://dx . doi . org/10 . 7554/eLife . 03568 . 016 At fixed ( x , t′ ) , g ( x , t′\|y , t ) is peaked around x for small t−t′ and this peak move to higher fitness as as t−t′ increases and converges against a steady distribution far in the past . This is seen in Figure 6C , where the g ( x , t′\|y , t ) is plotted as a function of y . Far in the past g ( x , t′\|y , t ) has a well defined maximum at y≈3σ . This steady distribution is shaped by two opposing trends: Fit ancestors ( large y ) leave more offspring and are hence more likely sampled . Too fit ancestors , on the other hand , should leave many individuals at time t′ that ultimately contribute to the sample . The width of the steady state distribution is determined the diffusion constant D . As a special case , we will sometimes be interested in a terminal branch propagator , which takes the lineage all the way to the present generation , t′=0 . Marginalizing and multiplying by the sampling probability ω=M/N≪1 defines the probability of the ( y , t ) ancestor to be a direct progenitor of a sampled genome: G ( y , t ) =ω∫dx g ( x , 0\|y , t ) . Interestingly , for positive y , one expects this probability to initially increase with increasing t because the reproductive value - i . e . expected number of surviving offspring - for relatively fit individuals increases with time , so that their offspring constitute a larger fraction of the population and are therefore more likely to appear in the sample . At longer times however G ( y , t ) is expected to start decreasing , because it is increasingly unlikely that the lineage emanating from a highly fit ancestor far in the past , remains unbranched ( i . e . , has only a single descendant in the sample ) . For small times and moderate parental fitness y , the term enforcing non-branching in Equation ( 10 ) can be neglected . In this case , the terminal branch propagator simplifies to ( 11 ) G ( y , t ) ≈eyt−σ2t22+Dt33and is hence identical to the reproductive value Equation ( 6 ) . Armed with branch propagators we can now write down a joint probability of ancestral fitness on any given tree . Let xi denote the fitness of node i starting with i=0 at the root of the tree , i=1 , … , nint for internal nodes , and i=nint+1 , … , nint+next for external nodes . Furthermore , denote the children of node i by ij , where j runs over the number of children . The joint probability distribution of all nodes in the tree is then given by ( 12 ) P ( x\|T ) =p0 ( x0 ) Z ( T ) ∏i=0nint∏jg ( xij , tij\|xi , ti ) where Z ( T ) is a normalization factor , p0 ( x ) is the fitness distribution in the population , and the second product runs over all j children of node i . In contrast to Equation ( 1 ) , Equation ( 12 ) allows for polytomies in the tree . In writing down Equation ( 12 ) , we have made the approximation that the total population size is unconstrained and that different branches of the tree do not interact . In populations dominated by selection , this is a good approximation since coalescent properties depend only weakly on the population size . This joint probability lives in a too high dimensional space to be practically useful , however , the tree structure makes it easy to marginalize the distribution . We commence ‘integrating out’ the independent fitness variables of the leaves , followed by integrating over the fitness values of the parents of these leaves until we arrive at the root of the tree . This defines an iterative ‘message passing’ process ( Mézard and Montanari , 2009 ) in which the ‘message’ node i sends to its parent pi is calculated via ( 13 ) m↑i ( xpi ) =∫dxi g ( xi , ti\|xpi , tpi ) ∏jm↑ij ( xi ) where the product is over all children j of node i ( note that the times ti and tpi are fixed properties of the tree ) . For terminal nodes i without children , m↑i ( xpi ) is simply the terminal branch propagator . Similarly , we calculate “messages” passed downstream to child j of node i: ( 14 ) m↓ij ( xij ) =∫dxi g ( xij , tij\|xi , ti ) m↓i ( xi ) ∏k≠jm↑ik ( xi ) The integrand is the product of the downstream message from the parental node and the upstream messages from all children of node i other than child j . This product is further multiplied by the branch propagator to child j and integrated over the fitness of node i . Having calculated the up and down messages for each branch , we can simply calculate the marginal distributions of fitness xi by multiplying all messages going into a node i . ( 15 ) p ( xi ) =1Zim↓i ( xi ) ∏jm↑ij ( xi ) where Zi assures normalization . Our inference uses the mean marginal fitness to rank internal and external nodes . For a pre-terminal node , the ‘up-message’ ( Equation ( 13 ) ) involves multiplying the terminal branch propagators of all its children . If the node is recent , we can use approximation Equation ( 11 ) and obtain ( 16 ) m↑i ( xpi ) ∼∫dxi g ( xi , ti\|xpi , tpi ) eTtotxi , where Ttot is total tree length downstream of node i , which polarizes the fitness of node i towards the high fitness edge . For a given number of descendants , this total tree length is maximized by a star topology . This corresponds to recent findings that multiple mergers in genealogies are associated with rapid expansion of clones founded by exceptionally fit individuals ( Brunet et al . , 2007; Desai et al . , 2013; Neher and Hallatschek , 2013 ) . The LBI defined as the integrated exponentially discounted tree length surrounding a node can be calculated in a very similar way to the message passing framework used above to evaluate the fitness distributions . The corresponding ‘up’-messages to the parent of node i is simply ( 17 ) m↑i=τ ( 1−e−bi/τ ) +e−bi/τ∑jm↑ijwhere bi is the branch length of node i and the sum runs over the children ij of node i . Similarly , the down message from a parent i to child ij ( 18 ) m↓ij=τ ( 1−e−bij/τ ) +e−bij/τ[m↓i+∑k≠jm↑ik] After having calculated all up and down messages , the exponentially discounted tree length is given by ( 19 ) λi ( τ ) =m↓i+∑j m↑ij The fitness inference algorithm is implemented in Python using the libraries SciPy and NumPy ( Oliphant , 2007 ) . Roughly , we have implemented one class , survival_gen_func , that integrates the fitness propagator on a discrete fitness grid . This class is used by the class fitness_inference to calculate the marginal distribution of fitness at each external and internal node of a given tree . The calculation of the marginals is done using a message passing approach ( Mézard and Montanari , 2009 ) . This fitness inference class is then subclassed to accommodate influenza specific features . All code associated with this manuscript is available at https://github . org/rneher/FitnessInference . To predict the sequence closest to the future population in a multiple sequence alignment , we build a maximum likelihood tree using fasttree ( Price et al . , 2009 ) ( the fasttree code was modified slightly to resolve short branches better ) . The reconstructed tree was passed to the fitness inference class . Following fitness inference , internal or external nodes were ranked by their expected fitness and we report the top ranked node as our prediction . The branch propagator depends on fitness diffusion constant D , the standard deviation in fitness σ , and the sampling fraction ω . For the numerical implementation , we measure time in unites of σ−1 and selection strength in units of σ and the dimensional fitness diffusion constant is Γ=Dσ−3 . The initial condition for the generating function is ϕω ( x , 0 ) =ω/σ in these units . In order to apply our algorithm to a tree reconstructed from sequences , we need to convert branch length into time in units of σ−1 . Given an alignment , we can calculate the average pairwise nucleotide distance π≈2μ〈T2〉 , where 〈T2〉 is the average pair coalescent time and μ is the per site mutation rate . For an adapting population in the SBD model , we have 〈T2〉σ≈Γ−1 ( Neher and Hallatschek , 2013 ) . Given a choice for Γ , the conversion factor β from nucleotide distance to σ−1 units is determined by ( 20 ) π2β=1Γ ⇒ β=Γπ2 . In addition to estimating fitness from the tree , we also measure the frequency changes of clades over time . For influenza A/H3N2 virus data , we partition sequences into three intervals of equal length between May and February and calculate the fraction of sequences that are below every internal nodes in each of these intervals ( using a pseudocount of 5 ) . From these three frequency values , we estimate the expansion rate by fitting a line to the logarithm of the frequencies . We use the population genetics library FFPopSim ( Zanini and Neher , 2012 ) to implement an individual based simulation with fixed fitness variance σ=0 . 03 . Mutations are introduced at random sites in random individuals with rate μ . We varied the total genomic mutation rate u=Lμ between 0 . 016 and 0 . 256 , where the total number of simulated sites is L=2000 . Mutations at all sites are by default deleterious , with effects drawn from an exponential distribution . To emulate a changing environment , we redraw the fitness effect of random positions within the first 500 sites at random with a total rate of nA=0 . 02 , … , 0 . 16 per generation . Beneficial effects are drawn from a gamma distribution with shape parameter 2 and the same scale as the deleterious mutations . Every 200 generations , a random sample of 200 sequences is written to file and later used to predict the sequence closest to the next sample . The simulation code is provided as flusim . cpp in the above mentioned repository . All sequences of influenza A/H3N2 viruses from human hosts from 1968 to 2014 that cover the entire HA1 domain were downloaded from IRD and aligned using the alignment feature provided by IRD with default settings ( Squires et al . , 2012 ) . The alignment was inspected by eye and trimmed to the HA1 domain . A few obvious outliers , lab strains , and sequences with indels or more than 4 ambiguous nucleotides were removed manually . For each strain the location information was converted to longitude and latitude at the country level and the strain was classified into rough geographic regions based on longitude and latitude . Only sequences with geographic information at the country level and date information with at least month accuracy were used . To avoid sampling bias , we subsampled the data to at most 100 sequences from either North America and Asia and used repeated subsamples to assess the robustness of the predictions . In years where less than 100 sequences are available from one of the geographic regions , we repeatedly used 70% of the available data . Increasing the sample size has negligible effect on prediction accuracy beyond a sample size of 100 .	When viruses multiply , they copy their genetic material to make clones of themselves . However , the genetic material in the clone is often slightly different from the genetic material in the original virus . These mutations can be caused by mistakes made during copying or by radiation or chemicals . Further mutations arise when the clones multiply , which means that , after many generations , there will be quite large differences in the genetic material carried by many members of the population . Most mutations have little or no effect on the ‘fitness’ of an individual - that is , on its ability to survive and multiply - but some mutations do have an influence . Some viruses , like seasonal influenza ( flu ) viruses , can mutate so rapidly that the most common strains change from year to year . This is why new flu vaccines are needed every year . To date most attempts to predict the evolution of seasonal flu viruses have focused on identifying specific features within the genetic sequences that might indicate fitness . However , such approaches require lots of information about the viruses , and this information is often not available . To address this problem , Neher , Russell and Shraiman have developed a more general method to predict fitness from virus genetic sequences . First , a ‘family tree’ for a virus population - which shows how each strain of the virus is related to other strains - was constructed by comparing the genetic sequences . The next step was based on the observation that as long as differences in fitness arise from the accumulation of multiple mutations , the branching structure of this family tree will bear a visible imprint of the natural selection process as it unfolds . Using this insight and methods borrowed from statistical physics , Neher et al . then analyzed the shape and branching pattern of the tree to work out the fitness of the different strains relative to each other . Neher et al . tested the method using historical influenza A virus data . In 16 of the 19 years studied , the family tree approach made meaningful predictions about which viruses were most likely to give rise to future epidemics . The ability to predict influenza virus evolution from tree shape alone suggests that influenza virus evolution may be more predictable than previously expected .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology", "physics", "of", "living", "systems" ]	2014	Predicting evolution from the shape of genealogical trees
Glutamate receptors are divided in two unrelated families: ionotropic ( iGluR ) , driving synaptic transmission , and metabotropic ( mGluR ) , which modulate synaptic strength . The present classification of GluRs is based on vertebrate proteins and has remained unchanged for over two decades . Here we report an exhaustive phylogenetic study of GluRs in metazoans . Importantly , we demonstrate that GluRs have followed different evolutionary histories in separated animal lineages . Our analysis reveals that the present organization of iGluRs into six classes does not capture the full complexity of their evolution . Instead , we propose an organization into four subfamilies and ten classes , four of which have never been previously described . Furthermore , we report a sister class to mGluR classes I-III , class IV . We show that many unreported proteins are expressed in the nervous system , and that new Epsilon receptors form functional ligand-gated ion channels . We propose an updated classification of glutamate receptors that includes our findings . Glutamate is the principal excitatory neurotransmitter in the central nervous system of animals ( Fonnum , 1984; Danbolt , 2001; Pascual-Anaya and D'Aniello , 2006 ) . It acts on two families of structurally unrelated receptors: ionotropic glutamate receptors ( iGluRs ) , which are ligand-gated ion channels and G-protein coupled receptors ( GPCRs ) , known as metabotropic glutamate receptors ( mGluRs ) ( Sobolevsky et al . , 2009; Conn and Pin , 1997 ) . While fast excitatory neurotransmission is mediated by iGluRs , metabotropic receptors modulate synaptic transmission strength . iGluRs are formed by four subunits , which can be traced back to bacteria ( Tikhonov and Magazanik , 2009 ) . The current classification of iGluR subunits includes six classes: α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid ( AMPA ) receptors , Kainate receptors , N-methyl-D-aspartate ( NMDA ) receptors ( actually comprising three classes: NMDA1-3 ) and Delta receptors ( Traynelis et al . , 2010 ) . iGluR subunits of the same class assemble into homo- or heterotetramers ( Karakas and Furukawa , 2014; Kumar et al . , 2011 ) and their ligand selectivity is dictated by a small number of residues located in the ligand-binding domain ( Traynelis et al . , 2010 ) . Accordingly , NMDA subunits GluN1 and GluN3 as well as the Delta subunit GluD2 bind glycine and D-serine , while all subunits from the AMPA and Kainate classes bind glutamate ( Traynelis et al . , 2010; Kristensen et al . , 2016 ) . Metabotropic glutamate receptors are class C GPCRs and as such are formed by a single polypeptide . mGluRs also appeared before the emergence of metazoans , being present in unicellular organisms such as the amoeba Dictyostellium discoideum ( Taniura et al . , 2006 ) . mGluRs are presently organized into three classes ( I , II and III ) and all their members respond to glutamate ( Conn and Pin , 1997; Pin et al . , 2003 ) . While the phylogeny of the two families of GluRs is well characterized in vertebrates , that of the entire animal kingdom is only poorly understood . The few studies on iGluR evolution outside vertebrates concentrate on a few phyla , leaving many proteins unclassified ( Greer et al . , 2017; Brockie et al . , 2001; Janovjak et al . , 2011; Kenny and Dearden , 2013 ) . Similarly , the vast majority of mGluRs described so far fall into the three classes described in vertebrates ( Krishnan et al . , 2013; Kucharski et al . , 2007; Dillon et al . , 2006 ) . Although , the existence of three insect mGluRs that cluster apart from classes I-III led to propose the existence of a fourth class ( Mitri et al . , 2004 ) . Here we present what to our knowledge is the most comprehensive phylogenetic study of ionotropic and metabotropic GluRs along the animal kingdom . We have favored the use of more slow-evolving species for the construction of phylogenetic trees . These species are particularly amenable to phylogenetics ( Simakov et al . , 2013; Simakov et al . , 2015; Putnam et al . , 2007 ) as they arguably present lower rates of molecular evolution than other organisms . Our work shows that metazoan evolution of GluRs is much more complex than previously thought . iGluRs present an overall organization into four subfamilies that were already present in the last ancestor of all metazoans . Vertebrate species only retain members of two of these subfamilies . Furthermore , we identify many lineage-specific gains , losses or expansions of GluR phylogenetic groups . Finally , we present experimental evidence showing that unreported GluRs found in the basally divergent chordate Branchiostoma lanceolatum ( amphioxus ) are highly expressed in the nervous system and that members of the unreported Epsilon subfamily , the most phylogenetically spread among unreported groups , can form functional ligand-gated ion channels . We have performed a systematic phylogenetic study of iGluR evolution across the animal kingdom . To increase the confidence on iGluRs evolutionary history phylogenetic trees have been generated using two independent methods ( Bayesian inference and Maximum-likelihood ( ML ) , Figure 1 and Figure 1—figure supplement 1 ) . Our analysis indicates that the family of iGluRs experienced key duplication events that define its present organization into four previously unreported subfamilies , of which two contain the extensively studied vertebrate classes . Assuming ctenophores as the sister group to all other animals ( Moroz et al . , 2014; Ryan et al . , 2013 ) , our data suggest that the three major duplication events leading to this four subfamilies occurred before the divergence of current animal phyla ( see Figure 2 for a summary scheme of iGluRs evolution ) . The first of these duplications produced the separation of the Lambda subfamily , the second lead to divergence of the NMDA subfamily and the third to the split between Epsilon and AKDF subfamilies . The Lambda subfamily is the most phylogenetically restricted , as we could only identify it in porifers . Thus , Lambda would have been lost in two occasions , in the lineage of ctenophores and in a common ancestor of placozoans , cnidarians and bilaterals . On the other hand , the Epsilon subfamily is the best represented among non-bilaterians , being present in all non-bilaterian phyla investigated . Including in porifers , although we could only identify one Epsilon in sponges , GluE_Ifa from the demosponge Ircinia fasciculata . Our data also indicate that this subfamily has been lost in multiple occasions along metazoan evolution , as we could not find it in the protostome , echinoderm or vertebrate species investigated . Interestingly , all ctenophore iGluRs identified , which have been previously reported ( Alberstein et al . , 2015 ) , belong to the Epsilon subfamily . Thus , this phylum would have lost NMDA , Lambda and AKDF proteins . Contrarily , ctenophores would have experienced an important expansion of Epsilon iGluRs , as we report 17 and 10 of these proteins in the two species with genomic information available , M . leidyi and P . bachei , respectively . Although we have not identified NMDA receptors in ctenophores , porifers and placozoans our analysis indicates that this subfamily was already present in the last common ancestor of metazoans . This is because the topology of the tree shows that NMDAs appear in the phylogeny at the same level as the Epsilon subfamily , which has representatives in all non-bilateral phyla . According to our data , NMDA1s on the one hand and NMDA2s and NMDA3s on the other contain members of the cnidarian phylum . Although we have only been able to identify one member more closely related to NMDA2 and NMDA3 than NMDA1 ( GluN2/3_Nve ) , its position in the phylogeny is very well supported by both analyses performed . This indicates that a specific duplication occurred in the ancestor of bilaterians originating NMDA2s and NMDA3s . Moreover , we have also identified a cnidarian-specific NMDA class , that we have termed NMDA-Cnidaria , this class presents representative proteins in 3 of the four species investigated . Among bilaterals we have observed conservation of all NMDA classes with the exception of NMDA2s in echinoderms , which are absent from the two species examined . Interestingly , studied cnidarian species substantially expanded their NMDA subfamily repertoire , with at least six members in Nematostella vectensis . In bilaterians the AKDF subfamily diversified into the known AMPA , Kainate and Delta classes , but also into a fourth new class that we have termed Phi . The phylogenetic spread of these classes is quite variable , as AMPA and Kainate are in all bilateral phyla investigated but Delta and Phi are more restricted . Deltas are almost completely absent from ecdysozoan species , as we could only find a single member of this class in priapulids ( P . caudatus ) and none in arthropods or nematodes . Similarly , Deltas are poorly represented in mollusks and , with the available data , absent in annelids . Finally , we could only identify Phi proteins in cephalochordates , hemichordates and echinoderms , indicating that this class might be lost in the lineages of protostomes and olfactores ( i . e . vertebrates and urochordates ) . The AKDF subfamily also includes proteins from the non-bilateral phyla of porifera , placozoa and cnidarian . The exact organization of these proteins into classes is not as straightforward as for bilateral proteins . The Bayesian and ML analysis only agree in the position of 12 iGluRs from the sponge O . carmela , these would constitute the only clear class in non-bilaterals , which we have termed AKDF-Oca . Another example of a multiple lineage-specific event that occurred during animal evolution of iGluRs can be observed in the evolution of AMPA and Kainate proteins among protostomes . The general iGluRs phylogeny ( Figure 1 ) suggests that ecdysozoan species have expanded their repertoire of Kainate subunits when compared with lophotrochozoans ( e . g . mollusks , annelids ) , since C . teleta and L . gigantea only presents one and two genes coding for Kainate receptors , respectively . Contrarily , we found more AMPA subunits in lophotrochozoans than in ecdysozoan species . To investigate whether the two protostome lineages have alternatively expanded genes coding for AMPA or Kainate subunits we conducted a phylogenetic analysis of these two classes using eight species of ecdysozoans and seven of lophotrochozoans with well-characterized genomes ( Figure 3 and Figure 3—figure supplement 1 ) . Nematodes were left out of the analysis as they lack Kainate receptors ( Brockie et al . , 2001 ) . This analysis retrieved 40 lophotrochozoan genes coding for AMPA subunits but only 15 coding for Kainates . The opposite scenario was observed in the genomes of ecdysozoan species , with 10 AMAP and 40 Kainate proteins , . Yet , among ecdysozoans the priapulid P . caudatus has two AMPA and two Kainate subunits , indicating that the expansion of Kainate receptors might be exclusive to arthropods . Overall the AMPA:Kainate ratio resulted to be around 1:4 in ecdysozoans and 4:1 in lophotrochozoans . All proteins from unreported groups ( i . e . subfamilies and classes ) present well-conserved sequences in iGluR domains , including transmembrane domains or residues involved in receptor tetramerization ( Figure 1—figure supplement 2 and Figure 1—source data 1 ) . Three-dimensional ( 3D ) models of two Epsilon subunits from amphioxus ( GluE1 and GluE7 ) indicate that their general fold is well preserved ( Figure 1—figure supplement 3a ) . The only noticeable distinction in proteins from these groups is an insertion in the intracellular loop between the first and second transmembrane domains in Epsilon proteins . This insertion is particularly distinct in ctenophore iGluRs , having been termed as the cysteine-rich loop ( Alberstein et al . , 2015 ) ( Figure 1—figure supplement 4 ) . We have also identified a sequence difference among Epsilon proteins . Ctenophore iGluRs have two cysteines that form a disulfide bond at loop 1 of the ligand binding domain ( Alberstein et al . , 2015 ) , which are also present in NMDA proteins . Nevertheless , this element is absent from the remaining members of the Epsilon subfamily . The ‘SYTANLAAF’ motif , essential for channel gating ( Traynelis et al . , 2010 ) , is also well conserved in most sequences , in particular the second , fourth and fifth residues ( Figure 1—figure supplement 2 ) . Nevertheless , all members of the Lambda subfamily and some proteins of the Phi class present lower levels of conservation in this sequence . Whether these changes have a functional impact is something that will require further investigation . The Q/R site ( Q586 , residue numbering according to mature rat GluA2 ) and the acidic residue located four positions downstream D/E590 ( Figure 1—figure supplement 4 ) are involved in calcium permeability and polyamine block of AMPA and Kainate receptors ( Bowie and Mayer , 1995; Koh et al . , 1995; Kamboj et al . , 1995 ) . Of these two positions the latter is much better conserved , especially outside ctenophores and the Lambda subfamily . We have identified an acidic residue at position 590 in 84 out of 122 iGluRs from unreported groups , including cnidarian NMDAs . Yet , only 1/3 of these proteins present a glutamine ( Q ) at position 586 . This includes most AKDFs and Epsilon proteins from non-ctenophores , contrarily , none of the Phi subunits presents a Q586 . The key ligand binding residues involved in fixing the amino acid backbone ( α−amino and α−carboxyl ) are Arg485 and an acidic residue at position 705 ( Naur et al . , 2007; Armstrong and Gouaux , 2000; Mayer , 2005; Furukawa et al . , 2005; Yao et al . , 2008 ) . These two positions are well conserved in 94 of the 122 proteins from unreported groups , suggesting that their endogenous ligand is an amino acid ( see Figure 1—figure supplement 3b for a 3D representation of ligand binding by GluE1 and Figure 1—figure supplement 5 for an alignment of iGluR residues involved in ligand binding ) . The residue changes found in the remaining 28 proteins would render them unable to bind an amino acid ( Figure 1—figure supplement 5 ) . This is are particularly common among class Phi proteins from amphioxus and in NMDA-Cnidaria . Residues involved in ligand selectivity show higher variability . These are located at positions 653 and 655 , and are occupied by glycine and threonine in glutamate-binding proteins and by serine and a non-polar residue in glycine-binding iGluRs . However , a recent study of ctenophore receptors has found that position 653 can be occupied by serine or threonine in glutamate-binding iGluRs , and by an arginine in glycine-binding subunits ( Alberstein et al . , 2015 ) . Based on this previous knowledge we have predicted the ligand specificities of most previously unreported receptors . The preferred ligand could be confidently predicted for 72 out of the 94 proteins with well-conserved residues involved in fixing the amino acid backbone . Interestingly , all unreported groups comprise glycine- and glutamate-specific iGluRs . Gly-specific receptors slightly outnumber those predicted to respond to glutamate ( overall ratio about 3:2 ) . The Lambda subfamily would include three proteins specific for glutamate and one for glycine , while seven remain with an unknown selectivity . Of note , the protein predicted to bind glycine ( GluL5_Oca ) displays an arginine at position 653 , a feature which had only been reported in ctenophores ( Alberstein et al . , 2015 ) . This residue would form a salt bridge with Glu423 , which is key for glycine selectivity in ctenophores ( Alberstein et al . , 2015 ) . Most Epsilon and AKDF proteins would preferably bind glycine , although ctenophores present a similar number of Epsilon receptors predicted to respond to glycine or glutamate ( Figure 1 ) ( Alberstein et al . , 2015 ) . In the Phi class we also found a similar number of receptors binding glycine and glutamate . Finally , we could only predict binding specificity for two of the 9 NMDA-Cnidaria proteins , as they present many changes in the residues involved in either amino acid backbone binding or side chain recognition . Interestingly , the 22 proteins for which we could not confidently predict their ligand selectivity ( Figure 1—figure supplement 5 ) , present a limited number of residues occupying position 653 and 655 , suggesting constrained evolution . Of these: ( i ) nine present residues with negative polarity at both positions , being candidates to bind glutamate , ( ii ) six present a Gly653 and a non-polar residue at position 655 , and thus are candidates to bind glycine , ( iii ) five proteins , all from the Branchiostoma genus , present a tyrosine at position 653 . A structural model of one of these receptors , GluE7 ( Figure 1—figure supplement 3c ) , shows that a Tyr653 aromatic side chain would occupy the ligand-binding pocket , strongly suggesting that amino acid binding would be blocked . Finally , ( iv ) two proteins present a phenylalanine in either of the two positions and remain unclassified . We used quantitative PCR ( qPCR ) to investigate gene expression levels of all iGluR subunits identified in B . lanceolatum , including those from the Epsilon and Phi groups . All 24 B . lanceolatum iGluR subunits identified in silico were found expressed in amphioxus , with the exception of Grie5 ( Figure 4a ) . Furthermore , they all showed a significantly higher expression in the nerve cord as compared to the whole body , suggesting tissue-enriched expression . While we observed low expression levels for Epsilon genes coding for subunits with a tyrosine at position 653 ( Grie5-8 ) , which according to the 3D model would block the ligand-binding pocket , the expression of Grif1-2 , also presenting the same tyrosine , reach much higher levels , comparable to those of subunits from the Kainate , Delta or NMDA classes . Thus , the presence of a tyrosine at position 653 does not appear to be directly correlated with low expression levels . Amphioxus genes coding for GluE1 and GluE7 were synthesized in vitro and transiently expressed in HEK293T cells for functional studies . Wild-type GluE1 and GluE7 , which are not predicted to have a canonical signal peptide by SignalP 4 . 1 ( Nielsen , 2017 ) , expressed well but were not trafficked to the plasma membrane ( Figure 4—figure supplement 1a–d ) , even though residues involved in tetramerization ( Salussolia et al . , 2013 ) are well conserved ( Figure 4b ) . We thus synthesized new variants of these genes with the signal peptide from rat GluA2 ( Figure 1—figure supplement 1cd ) . These constructs also expressed well ( Figure 4c ) and now were efficiently trafficked to the plasma membrane , as indicated by the staining observed in non-permeabilized cells ( Figure 4d ) . Furthermore , analysis of receptor oligomerization , performed using non-denaturing gel electrophoresis and immunoblot , clearly indicates that both proteins form homotetramers in vitro ( Figure 4e ) . We next investigated the gating properties of two Epsilon proteins from amphioxus , GluE1 and GluE7 . The presence of a serine and a tryptophan at positions 653 and 704 , respectively , suggested that GluE1 would bind glycine . Indeed , neither glutamate nor aspartate elicited a response in our experimental settings . Instead , glycine application was able to elicit an inward whole-cell current at a membrane potential of −60 mV ( Figure 5a ) . Interestingly , the chemically related amino acids alanine and D-serine only generated very low responses , indicating a high selectivity of the GluE1 homotetramer for glycine . The Epsilon receptor displayed a strong inward rectification , even in the absence of added polyamines in the intracellular solution ( Figure 5b , c ) . This behavior is characteristic of unedited AMPA and Kainate receptors displaying a glutamine ( Q ) and an acidic residue at positions 586 and 590 , respectively ( Bowie and Mayer , 1995; Koh et al . , 1995; Kamboj et al . , 1995 ) and GluE1 presents a glutamine and an aspartic acid at these positions ( Figure 1—figure supplement 4 ) . Glycine-mediated currents showed a slow rate of recovery from desensitization when compared with AMPA or Kainate mammalian receptors , requiring 20–25 seconds until a complete recovery was achieved and a full response of the same magnitude could be recorded ( Figure 5d , e ) . Similar observations have been made with ctenophore receptors activated by glycine in which the recovery from desensitization has an unusually long time constant of 81 seconds ( Alberstein et al . , 2015 ) . Finally , functional studies on receptors formed by GluE7 did not retrieve any positive results . None of the following amino acids: glutamate , aspartate , asparagine , glycine , alanine or D-serine elicited a response in our experimental system . We hypothesize that , as predicted by the 3D model , the presence of a tyrosine at position 653 renders a homomeric form of this receptor unable to function as an amino acid-gated ion channel . We next performed a phylogenetic study of metabotropic glutamate receptors ( Figure 6 and Figure 6—figure supplement 1 ) . This analysis has revealed that the three historical mGluR classes ( I to III ) have a sister group . Following the current nomenclature we have named this as class IV . The existence of this class had already been proposed on the bases of three insect proteins ( Mitri et al . , 2004 ) . Yet , here we show that this class is actually present in all bilateral phyla , excluding vertebrates . Furthermore , we also show that class IV appeared together with classes I-III before radiation of bilateral lineages . We have identified clear orthologues to class I-IV in porifers , placozoans and cnidarians but not in ctenophores . These are organized into four classes , two from cnidarians , and one from placozoans and porifers ( Figure 6 ) . We have also identified non-bilaterian mGluRs that fall outside the above-mentioned classes . Unfortunately , the Bayesian and ML phylogenies do not agree on the exact organization of these early divergent mGluRs , except for the fact that they diverge prior to bilaterian classes . For this reason we have left these sequences unclassified . Whether these sequences belong to one , or even multiple classes that would have been lost in bilateral organisms is something that will require further investigation . Although all class IV proteins show well conserved sequences overall ( Figure 6—figure supplement 2a , Figure 6—figure supplement 3 and Figure 6—source data 1 ) , two residues critical for glutamate binding , Arg78 and Lys409 , are non-conservatively replaced by non-polar or acidic residues in all class IV proteins identified ( Figure 6—figure supplement 2a , residue numbering corresponds to human mGluR1 ) . These changes are predicted to hamper glutamate binding and , indeed , functional studies of a class IV receptor from fruit fly indicated that it does not respond to this amino acid ( Mitri et al . , 2004 ) . All class IV proteins would share this feature . On the other hand , residues involved in contacts with the amino acid backbone are well conserved ( Figure 6—figure supplement 2a ) , suggesting that these proteins might bind an amino acid other than glutamate . Similarly , mGluR residues from most non-bilaterian sequences involved in binding the amino acid backbone are highly conserved . Among non-bilaterian proteins the residues involved in glutamate binding are only conserved in approximately half of the proteins from classes orthologous to I-II-III-IV . Finally , we investigated mGluRs expression in amphioxus following the same procedure described for iGluRs . All five amphioxus mGluRs showed an enriched expression in the nerve cord , including the two class IV genes . Noticeably , these two genes showed significantly higher expression levels than orthologues of vertebrate classes ( Figure 6—figure supplement 2b ) . We have performed what to our knowledge is the most comprehensive phylogenetic study of metazoan glutamate receptors . This has revealed that their evolutionary history is much more complex than what is currently acknowledged , especially for the family of iGluRs . Our study has also revealed the existence of unreported phylogenetic groups in both ionotropic and metabotropic glutamate receptors . Importantly , our data indicate that the evolution of glutamate receptors has not occurred in an unequivocal incremental manner only in those clades with more elaborated neural systems , but it has rather followed an scattered lineage-specific evolutionary history . This means that certain lineages have experienced the gain , loss , expansion or reduction of specific phylogenetic groups . Our phylogenetic analysis indicates that the family of iGluRs is actually divided into four unreported subfamilies that we have termed Lambda , Epsilon , NMDA and AKDF . Interestingly , this general organization was already present in the last common ancestor of all metazoans and later duplications within NMDA and AKDF subfamilies resulted in the formation of well-known iGluR classes . The other two subfamilies are absent from the majority of model species used in neuroscience research . The NMDA subfamily diversified into classes NMDA1-3 but also into the NMDA2/3 and NMDA-Cnidaria . Similarly , the AKDF subfamily diversified into the AMPA , Kainate and Delta classes , but also into the previously unreported Phi class . We have also identified and AKDF class exclusive to porifers , represented by sequences form O . carmela . Most well-studied iGluR classes are the result of duplications in ancestors of current bilateral species , >650 million years ago ( mya ) ( Kumar et al . , 2017 ) , only class NMDA1 originated earlier , as cnidarians present members within this class . The Epsilon subfamily , which includes all iGluRs from ctenophores , is the only subfamily present in all non-bilateral phyla investigated , including sponges . It is thus the subfamily presenting a larger phylogenetic spread , as it is also present in hemichordates and in non-vertebrate chordates . On the other hand , the unreported Phi class shows a more restricted phylogenetic spread , as it is present only in three deuterostome phyla . Moreover , Lambda proteins seem restricted to Porifers , which constitutes an interesting evolutionary case due to maintenance of a glutamate receptor family in a phylum without nervous system . The phylogenetic analysis of metabotropic glutamate receptors has allowed us to unambiguously establish the existence of a sister group to the well-known classes I , II and III . Following the present nomenclature we have named this as class IV . This class had been previously proposed based on the identification of three insect mGluRs that did not cluster with members of known classes ( Mitri et al . , 2004 ) . Here we show that class IV is not restricted to insects , but is actually present in all bilaterian phyla investigated , with the exception of vertebrates where this class has been lost . Interestingly , as it occurs for most well-known iGluR classes , mGluR classes I-IV appeared simultaneously in the ancestor of bilaterals . Our phylogenetic analysis also indicates that the non-bilateral phyla of cnidarians , placozoans and porifers present clear orthologues to classes I-IV , which are organized into four classes , while we failed to find any in the early-branching ctenophores . Finally , we were unable to confidently classify many non-bilateral mGluRs , which might constitute one or more classes . We have identified many examples of lineage-specific evolutionary events . These would antagonize with a model in which species with less elaborated nervous systems would present GluR families with lower complexity . The most noticeable examples are: ( i ) the absence of all subfamilies but Epsilon in analyzed ctenophores , ( ii ) the loss of Delta receptors from arthropods , nematodes and annelid species investigated , ( iii ) the loss of the Epsilon subfamily in vertebrates , echinoderms and protostomes , ( iv ) the loss of the Phi class in vertebrates and studied protostomes , ( v ) the specific expansion of Kainate receptors in arthropods , which contrasts with the expansion of AMPA receptors in its sister lineages of mollusks and annelids , ( vi ) the large expansion of the Epsilon subfamily in ctenophores , placozoans and cephalochordates and , finally ( vii ) the loss of mGluR class IV in vertebrates . Along the same line , it is interesting to note that amphioxus ( B . belcheri and B . lanceolatum ) , with a simple nervous system , have over 20 genes encoding iGluRs , while mammals have 18 . Other non-vertebrate species also present large numbers of iGluRs , including the 19 iGluRs identified in the sponge O . carmela or the 17 present in the ctenophore M . leidyi , to mention a few . Similarly , the cnidarian A . digitifera and the ctenophore M . leidyi have seven mGluRs each , while the placozoan T . adhaerens presents eleven , three more than the eight mGluRs found in the human genome . The large number of GluRs found in many non-vertebrate animals suggests that there has been an evolutionary trend to increase their number in many metazoan lineages . Our experimental results suggest that unreported receptors would play a role in the nervous system , as Epsilon , Phi and mGluR class IV genes are highly expressed in the nerve cord of amphioxus . Nevertheless , whether all these proteins are expressed at the synapse and act as neurotransmitter receptors is an issue that will require further investigation . Their presence in other tissues , such as sensory organs , cannot be ruled out . Those receptors showing more divergent sequences , particularly in residues involved in ligand binding , might respond to other molecules . For instance , they could behave as chemoreceptors , as it is the case of antennal receptors found in insects ( Croset et al . , 2010; Benton et al . , 2009 ) . Proteins from all unreported groups generally present a good conservation of residues involved in binding the amino acid backbone , indicating that their ligand would be an amino acid or a closely related molecule . Interestingly , we could identify proteins predicted to bind either glycine or glutamate in all unreported iGluR subfamilies and classes . If our functional predictions are correct , the ability to recognize one or the other amino acid would have emerged repeatedly in all unreported iGluR phylogenetic groups . Unexpectedly , the nature of the residues conferring amino acid specificity indicates that only a minority of proteins from unreported GluR groups would respond to glutamate . Sequence analysis and structural considerations strongly suggest that class IV mGluRs will not bind glutamate and that among non-bilateral mGluRs only a minority , belonging to classes orthologous to I-II-III-IV , are predicted to bind to this neurotransmitter . Similarly , among unreported iGluR groups , the number of proteins binding glycine outnumbers those binding glutamate . Interestingly , we report a glycine-binding poriferan protein ( GluL5_Oca ) with a structural feature that had only been reported in ctenophores ( Alberstein et al . , 2015 ) . This is an Arg653 that through establishing a salt bridge with Glu423 confers glycine specificity ( Alberstein et al . , 2015 ) . We thus report that this structural element is not exclusive to ctenophores . We have also identified iGluR subunits with important changes in critical ligand binding residues , indicating that they might have evolved new biological functions , for example , response to other , as yet unidentified small molecules . The activation of Epsilon receptors by glycine has been experimentally corroborated by electrophysiological analysis of homotetrameric receptors composed by GluE7 from M . leidy ( Alberstein et al . , 2015 ) and GluE1 from amphioxus ( this study ) . In our hands the amphioxus receptor showed a very high selectivity for glycine , since ion currents could not be elicited by chemically related amino acids such as serine or alanine . Glycine-binding Epsilon subunits from phyla other than ctenophores present structural features similar to those from glycine-binding iGluRs in vertebrates . The greater number of glycine receptors found in non-vertebrate species could be related to the higher abundance of this amino acid in their nerve cord as compared with the mammalian brain ( Pascual-Anaya and D'Aniello , 2006 ) . Altogether , our phylogenetic analysis and experimental findings have uncovered the complex evolution of glutamate receptors within the metazoan kingdom . Our data indicate that the classification of iGluRs is not restricted to the six classes currently recognized . Instead , iGluRs are organized into four subfamilies: Lambda , Epsilon , NMDA and AKDF and ten classes with varying phylogenetic spread . With the data available , the NMDA subfamily is organized into classes NMDA1 , NMDA 2 , NMDA3 , NMDA-Cnidaria and NMDA2/3 , while subfamily AKDF contains classes AMPA , Kainate , Delta , Phi and AKDF-Oca . Both NMDA2/3 and AKDF-Oca are represented by sequences from only one species , further sequencing of non-bilateral species will be required to fully demonstrate their existence . Furthermore , the evolution of mGluRs has generated a sister group to classes I , II and III , class IV . We have also identified classes of non-bilaterian mGluRs orthologous to I-II-III-IV . We propose that the classification of these two families of GluRs , key to the physiology of the nervous system , has to be updated to include our findings . Phylogenetic analysis were performed with sequences from at least two species from each of the following metazoan phyla: Porifera , Ctenophora , Placozoa , Cnidaria , Lophotrochozoa , Ecdysozoa , Hemichordata , Chordata and Vertebrata , with the exception of placozoans for which only one species is available . When possible , we chose slowly evolving species . The complete lists of species used for iGluR phylogenies are given in Figure 1—source data 2 . Species used in the phylogeny of metabotropic glutamate receptors are listed in Figure 6—source data 2 . Sponge sequences were taken from ( Riesgo et al . , 2014 ) , B . lanceolatum sequences were retrieved from unpublished genomic and transcriptomic databases ( access was kindly provided by the Mediterranean Amphioxus Genome Consortium ) , A . digitifera and P . flava sequences were obtained from the Marine Genomics Unit ( Simakov et al . , 2015; Shinzato et al . , 2011 ) and P . bachei sequences from NeuroBase ( Moroz et al . , 2014 ) . GluR sequences were identified using homology-based searches in a two-tier approach . Mouse glutamate receptors were used as search queries ( iGluRs: Gria1-4; Grik1-5; Grid1-2 , Grin1 , Grin2A-D and Grin-3A-B; mGluRs: mGluR1-8 ) . In a first search GluR homologs were identified using the BLASTP tool ( Altschul et al . , 1990 ) with default parameters . Subject sequences with an E-value below 0 . 05 were selected as candidate homologs . These were re-blasted against the NCBI database of ‘non-redundant protein sequences’ using the same BLAST tool . If the first hit obtained in the reciprocal BLAST was a glutamate receptor the sequence was included in the phylogenetic analysis . In a second stage the same mouse sequences were used to perform TBLASTN searches against genomic and , when available , transcriptomic databases . Subject sequences not identified in the first tear and having an E-value below 0 . 05 were selected as candidate homologs . These were re-blasted using BLASTX against the NCBI ‘non-redundant protein sequences’ database . Finally , if the first hit of this search was a glutamate receptor the sequence was also included in the phylogenetic analysis . Identified iGluR sequences in which less than four residues of the SYTANLAAF motif ( Traynelis et al . , 2010 ) were conserved were not considered for the final phylogenetic analysis . mGluR sequences lacking two or more of the seven transmembrane regions were also discarded . The complete reference lists of all iGluRs used in the final phylogeny are given in files Figure 1—source data 2 . The reference list of metabotropic glutamate receptors is presented in Figure 6—source data 2 . The alignments used for the phylogenetic analysis of iGluRs , mGluRs and AMPAs and Kainates from protostomes are provided in Figure 1—source data 3 , Figure 3—source data 1 and Figure 6—source data 3 . The iGluR tree was constructed with 224 sequences identified in 26 non-vertebrate species ( Figure 1—source data 2 ) . The tree also included 18 iGluR sequences from vertebrates and two iGluR proteins from A . thaliana , used as an outgroup ( Chiu et al . , 2002 ) . The phylogenetic analysis of AMPA and Kainate classes in protostomes was inferred using 110 sequences from 15 protostome species ( Figure 3—source data 2 ) and 37 sequences from deuterostomes , of which 4 GluN1 proteins were used as an outgroup . The mGluR tree was constructed with 149 proteins from 29 non-vertebrate species , 38 mGluRs from vertebrate species and 10 sequences from vertebrate metabotropic GABA receptors , used as an outgroup ( Figure 6—source data 2 ) . Protein sequences were aligned with the MUSCLE algorithm ( Edgar , 2004 ) , included in the software package MEGA6 ( Tamura et al . , 2013 ) with default parameters . ProtTest v3 . 4 . 2 was used to establish the best evolutionary model ( Darriba et al . , 2011 ) . Trees were constructed using MrBayes v3 . 2 . 6 ( Ronquist et al . , 2012 ) for Bayesian inference and IQ-TREE ( Nguyen et al . , 2015 ) for Maximum-likelihood analysis . For Bayesian inference phylogenies were stopped when standard deviation was below 0 . 01 and its value was fluctuating but not decreasing . Markov chain Monte Carlo ( MCMC ) was used to approximate the posterior probability of the Bayesian trees . Bayesian analyses included two independent MCMC runs , each using four parallel chains composed of three heated and one cold chain . Twenty-five % of initial trees were discarded as burn-in . Convergence was assessed when potential scale reduction factor ( PSRF ) value was between 1 . 002 and 1 . 000 . In Maximum-likelihood analysis the starting tree was estimated using a neighbor-joining method and branch support was obtained after 1000 iterations of ultrafast bootstrapping ( Hoang et al . , 2018 ) . Gene/protein names were given based on their position in the tree . Phylogenetic trees were rendered using FigTree ( http://tree . bio . ed . ac . uk/software/figtree/ ) . Phylogenetic calculations were performed at the IBB - UAB heterogeneous computer cluster ‘Celler’ and at the CIPRES science gateway ( RRID: SCR_008439 ) ( Miller et al . , 2010 ) . Branchiostoma lanceolatum adults were collected in the bay of Argelès-sur-Mer , France ( latitude 42° 32’ 53’ N and longitude 3° 03’ 27’ E ) with a specific permission delivered by the Prefect of Region Provence Alpes Côte d’Azur . B . lanceolatum is not a protected species . Animals were kept in tanks with seawater at 17°C under natural photoperiod . Adult amphioxus ( B . lanceolatum ) were anesthetized in 0 . 1% diethyl pyrocarbonate ( DEPC; Sigma , D5758 ) PBS buffer . Animals were sacrificed by cutting the most anterior part of the body . The nerve chord was surgically extracted from the animal while submerged in DEPC-PBS using a magnifying glass . Individual nerve chords were snap frozen in liquid nitrogen and stored at −80°C until use . RNA was extracted from whole animals or from dissected nerve chords . Ten nerve chords were used for each RNA extraction , so that biological variability between individuals could be normalized . The tissue was homogenized in 1 mL of TRI Reagent ( Sigma , T9424 ) using a Polytron homogenizer . Homogenates were transferred into an Eppendorf tube and incubated 5 min at room temperature ( RT ) before adding 100 µL of 1-bromo-3-cloropropane . Tubes were vigorously mixed by vortexing for 10–15 s , incubated 15 min at RT and centrifuged at 13000 rpm for 15 min at 4°C . RNA was precipitated from the aqueous phase with 500 µL of isopropanol and 20 µg of glycogen . Tubes were frozen for 1 hr at −80°C and then thawed , incubated at RT for 10 min and centrifuged at 13000 rpm for 10 min at 4°C . The RNA pellet was washed twice with 500 µL of 75% ethanol and air-dried . cDNA was synthesized from 0 . 5 µg of total RNA . One µL of Oligo ( dT ) 15 ( Promega ) , 1 µL of 10 mM dNTP mix ( Biotools ) , RNA and DEPC distilled water were mixed in a PCR tube to a final volume of 14 µL . This mix was incubated at 65°C for 5 min in a T100 Thermal Cycler ( BioRad ) . After cooling tubes on ice for 1 min , we added 4 µL of First Strand 5x buffer , 1 µL of 0 . 1 M DTT and 1 µL of SuperScript III ( Invitrogen ) . Tubes were placed in a T100 Thermal Cycler ( BioRad ) with the following program: 60 min at 50°C , 15 min at 70°C . RNA expression levels were determined using qPCR and the GAPDH gene used as a reference . Primers used for qPCR analysis of iGluRs are in Figure 4—figure supplement 2 and those used for mGluR qPCR in Figure 6—figure supplement 4 . qPCR data for iGluRs and mGluRs are given in Figure 4—source data 1 and Figure 6—source data 4 , respectively . cDNA from nerve chord and whole body samples was diluted 1:10 for the glutamate receptor gene reactions , and 1:100 for the reference gene reaction . For each gene 2 . 5 µL of diluted cDNA were added to 5 μL of iTaq Universal SYBR Green Supermix ( Bio-Rad ) , along with 0 . 5 µL of each primer and 1 . 5 µL of RNase free water . qPCR was run in a C1000 Touch thermocycler combined with the optic module CFX96 . Three technical replicates were performed for all genes analyzed . Primer pairs were designed to detect the expression levels of each glutamate receptor ( Figure 4—figure supplement 2 and Figure 6—figure supplement 4 ) . B . belcheri glutamate receptor sequences were aligned with the genomic sequence of B . lanceolatum , and high identity fragments were used to design primers . All primers were 20–25 base pair long , had GC content over 40–45% and a Tm between 60–65°C . Primers were designed to obtain amplicons between 140–270 base pairs . Values of normalized expression were statistically analyzed using GraphPad Prism5 . No outliers were identified and no data points were excluded . Comparisons between whole body and nerve chord expression levels were done with Student’s T-Test for unpaired samples or the Welch variant of the Student’s T-Test for samples with different variance . For multiple comparisons between the expression levels of genes belonging to the same class one-way ANOVA analysis was performed using Tukey’s Post-Hoc test . Grie1 and Grie7 genes were selected for transient expression in the mammalian cell line HEK293T . We prepared two constructs for each gene . We first introduced an immuno-tag in the N-terminus before the first element of secondary structure . For Grie1 we used the c-Myc tag , which was placed after residue 39 , and for Grie7 we used the hemagglutinin ( HA ) tag introduced after residue 10 of the wild-type sequence . The second set of constructs prepared substituted the wild type N-terminal sequence for the signal peptide from rat GluA2 while maintaining the immuno-tags ( Figure 4—figure supplement 1 ) . Codon-optimized genes for expression in human cells were synthesized and cloned into pICherryNeo ( Addgene , 52119 ) and pIRES2_EGFP ( Addgene 6029–1 ) by the Invitrogen GeneArt Gene Synthesis service . All expression experiments were done with a mycoplasma-free HEK293T cell line kindly provided by Prof . F . Ciruela ( Universitat de Barcelona ) and purchased from the American Type Culture Collection ( ATCC , CRL-3216 , RRID: CVCL_0063 ) . The ATCC has confirmed the identity of HEK293T by STR profiling ( STR Profile; CSF1PO: 11 , 12; D13S317: 12 , 14; D16S539: 9 , 13; D5S818: 8 , 9; D7S820: 11; TH01: 7 , 9 . 3; TPOX: 11; vWA: 16 , 19; Amelogenin: X ) . After the purchase of the cell line , mycoplasma tests are performed in the laboratory on every new defrosted aliquot . The kit used for mycoplasma detection is PlasmoTest ( Invivogen , code: rep-pt1 ) . HEK293T cells were maintained in Dulbecco’s Modified Eagle Medium ( DMEM ) supplemented with 10% FBS and 1% Antibotic-Antimycotic ( Gibco ) in a humidified incubator at 5% CO2 air and 37°C . The day before transfection , cells were plated onto poly-D-lysine coated coverslips in 6-well plates , to reach 60–80% confluence . HEK293T cells were transiently transfected with the following plasmids: empty pIRES2-EGFP , pIRES2-EGFP containing the Grie7_Bbe gene , empty pICherryNeo and pICherryNeo containing Grie1_Bla . Cells were transfected using 3 μg of polyethylenimine and 1 μg of plasmid DNA for each ml of non-supplemented DMEM . Cells were incubated 4–5 hr with transfection medium without supplementation , which was then removed and replaced by supplemented medium . Twenty-four hours after transfection the medium was removed and cells were washed 3 times with PBS . For surface receptor staining , cells were blocked in 2% BSA in PBS for 10 min at 37°C , and incubated for 25 min at 37°C with primary antibodies against HA ( Covance , MMS-101P , RRID: AB_291259 ) , c-Myc ( Cell Signalling , 2272S , RRID: AB_10692100 ) or GluA2 ( Millipore , MAB397 , RRID: AB_2113875 ) . HA and GluA2 antibodies were diluted 1:200 and c-Myc 1:100 in DMEM without supplementation . Cells were washed 3 times with PBS , fixed in 4% paraformaldehyde ( PFA ) for 15 min at RT , rinsed in PBS and incubated 1 hr at 37°C with secondary antibodies Alexa Fluor 555 donkey anti-mouse IgG ( H + L ) ( A-31570 , Invitrogen , RRID: AB_2536180 ) and Alexa Fluor 647 goat anti-rabbit IgG ( H + L ) highly cross-adsorbed ( Life Technologies , A-21245 , RRID: AB_2535813 ) , diluted 1:1000 and 1:500 in PBS , respectively . Finally , coverslips were washed and mounted onto slides with Fluoroshield with DAPI ( Sigma-Aldrich , F6057 ) . For intracellular labeling cells were first fixed in 4% PFA for 15 min at RT , permeabilized with 0 . 2% Triton X-100 in PBS for 10 min , and finally blocked with PBS containing 2% BSA and 0 . 2% Triton X-100 for 20 min . Primary antibodies against HA ( Covance , MMS-101P , RRID: AB_291259 ) and GluA2 ( Millipore , MAB397 , RRID: AB_2113875 ) were diluted 1:1000 and c-Myc ( Cell Signalling , 2272S , RRID: AB_10692100 ) antibody was prepared at 1:100 in PBS . Incubation lasted 25 min at 37°C . Secondary antibody incubations and coverslip mounting were done in the same way as for non-permeabilized cells . Cells were examined using a confocal laser-scanning microscope ( Zeiss LSM 700 ) with a 63x oil objective . HEK293T cells were grown in 6-well plates as described previously and transfected with plasmids expressing amphioxus GluE1 , GluE7 or GluA2 . Twenty-four hours after transfection cells were rinsed with PBS and the content of 4 wells was resuspended in solubilization buffer ( PBS containing 2% N-dodecyl-α-maltopyranoside ( DDM; D310HA , Anatrace ) and the protease inhibitors mix cOmplete EDTA-free Protease Inhibitor Cocktail , Roche ) . Cell lysates were homogenized in a Dounce homogenizer in ice with 20 strokes and kept under orbital agitation for 1 hr at 4°C . Lysates were centrifuged at 89000xg in a Beckman TLA120 . 2 rotor for 40 min at 4°C . The supernatant containing solubilized membrane proteins was recovered in a new tube and stored at −20°C until used . For native gel electrophoresis proteins were resolved in a Mini-PROTEAN TGX Gel 4–20% ( Bio-Rad ) . Samples were mixed with Native Sample Buffer ( Bio-Rad ) and run along with HiMark Pre-Stained Protein Standard ( Life Technologies ) . Electrophoresis was performed in ice at a constant voltage of 100 V for 180 min . Gels were transferred at constant current ( 35 mA ) to polyvinylidene fluoride ( PVDF ) membranes overnight ( 16–18 hr ) at 4°C . After transfer , membranes were blocked for 1 hr with Odyssey Blocking Buffer ( Li-cor ) in TBS , and incubated overnight at 4°C with primary antibodies anti-HA ( Covance , MMS-101P , RRID: AB_291259 ) , anti-c-Myc ( Cell Signaling , 2272S , RRID: AB_10692100 ) or anti-GluA2 ( Millipore , MAB397 , RRID: AB_2113875 ) diluted 1:1000 in TTBS ( TBS containing 0 . 05% Tween-20 ) . After three 15 min washes in TTBS , membranes were incubated with donkey anti-mouse ( Li-cor , 926–32212 , RRID: AB_621847 ) and donkey anti-rabbit ( Li-cor , 926–68073 , RRID: AB_10954442 ) diluted 1:7500 in TTBS for 1 hr . Blots were analyzed in an Odyssey scanner ( Li-cor ) . For denaturing gel electrophoresis ( SDS-PAGE ) protein lysates were denatured by adding loading sample buffer 10x ( 500 mM Tris-HCl pH 7 . 4 , 20% SDS , 10% β-mercaptoethanol , 10% glycerol and 0 . 04% bromophenol blue ) , and incubated for 5 min at 95°C . Protein lysates were loaded in a 10% SDS- polyacrylamide gel and separated at a constant current ( 25 mA ) . Gels were transferred at a constant voltage of 100 V for 90 min in ice . Membranes were blocked for 1 hr with Odyssey Blocking Buffer in TBS , and incubated overnight at 4°C with the same primary antibodies at the same dilution as for native gels in TBS containing 0 . 1% Tween 20 . After three 15 min washes in TTBS , membranes were incubated with secondary antibodies as above . Blots were analyzed in an Odyssey scanner . Models for full-length GluE1 and GluE7 were generated with RaptorX ( Källberg et al . , 2012 ) based on deposited three-dimensional crystal structures of the full-length AMPA-subtype ionotropic glutamate receptor from Rattus norvegicus , GluA2 , bound to competitive antagonists ( PDB codes 4U4G ( Yelshanskaya et al . , 2014 ) and 3KG2 ( Sobolevsky et al . , 2009 ) , respectively ) . Models of their respective ligand binding domains were generated with SWISS-MODEL ( Biasini et al . , 2014 ) using the atomic-resolution crystal structure of the rat GluA2 LBD bound to glutamate as template ( PDB code 4YU0 ) . Model quality was assessed with MolProbity ( http://molprobity . biochem . duke . edu/ , RRID: SCR_014226 ) . MolProbity scores for all models are given in Figure 1—source data 4 . Models were inspected with MIFit ( Smith , 2010 ) and figures were prepared with PyMOL ( www . pymol . org ) . Cells were visualized with an inverted epifluorescence microscope ( AxioVert A . 1 , Zeiss ) and were constantly perfused at 22–25°C with an extracellular solution containing ( in mM ) : 145 NaCl , 2 . 5 KCl , 2 CaCl2 , 1 MgCl2 , 10 HEPES and 10 glucose ( pH = 7 . 42 with NaOH; 305 mOsm/Kg ) . Microelectrodes were filled with an intracellular solution containing ( in mM ) : 145 CsCl , 2 . 5 NaCl , 1 Cs-EGTA , 4 MgATP , 10 HEPES ( pH = 7 . 2 with CsOH; 295 mOsm/Kg ) . Electrodes were fabricated from borosilicate glass ( 1 . 5 mm o . d . , 1 . 16 i . d . , Harvard Apparatus ) pulled with a P-97 horizontal puller ( Sutter Instruments ) and polished with a forge ( MF-830 , Narishige ) to a final resistance of 2–4 MΩ . Currents were recorded with an Axopatch 200B amplifier filtered at 1 KHz and digitized at 5 KHz using Digidata 1440A interface with pClamp 10 software ( Molecular Devices Corporation ) . Whole-cell macroscopic currents were recorded from isolated or coupled pairs of mCherry or EGFP positive HEK293T cells . Rapid application ( <1 ms exchange ) of agonists ( 500 ms pulses ) at a membrane potential of −60 mV was achieved by means of a theta-barrel tool ( 1 . 5 mm o . d . ; Sutter Instruments ) coupled to a piezoelectric translator ( P-601 . 30; Physik Instrumente ) . One barrel contained extracellular solution diluted to 96% with H2O and the other barrel contained 10 mM of the amino acid solution . For measuring current-voltage relationships , 500 ms agonist jumps were applied at different membrane voltages ( −80 mV to +80 mV in 20 mV steps ) and peak currents were fitted to a 5th order polynomial function . To study recovery from desensitization , a two-pulse protocol ( 500 ms each ) was used in which a first pulse was applied followed by a second pulse at different time intervals ( from 2 . 5 s to 25 s ) . The paired pulses were separated 30–60 s to allow full recovery from desensitization . To estimate the percentage of recovery , the magnitude of peak current at the second pulse ( P2 ) was compared with the first one ( P1 ) . Electrophysiological recordings were analyzed using IGOR Pro ( Wavemetrics Inc . ) with NeuroMatic ( Jason Rothman , UCL , RRID: SCR_004186 ) .	Nerve cells or neurons communicate with each other by releasing specific molecules in the gap between them , the synapses . The sending neuron passes on messages through packets of chemicals called neurotransmitters , which are picked up by the receiving cell with the help of receptors on its surface . Neurons use different neurotransmitters to send different messages , but one of the most common ones is glutamate . There are two families of glutamate receptors: ionotropic receptors , which can open or close ion channels in response to neurotransmitters and control the transmission of a signal , and metabotropic receptors , which are linked to a specific protein and control the strength of signal . Our understanding of these two receptor families comes from animals with backbones , known as vertebrates . But the receptors themselves are ancient . We can trace the first family back as far as bacteria and the second back to single-celled organisms like amoebas . Vertebrates have six classes of ionotropic and three classes of metabotropic glutamate receptor . But other multi-celled animals also have these receptors , so this picture may not be complete . Here , Ramos-Vicente et al . mapped all major lineages of animals to reveal the evolutionary history of these receptors to find out if the receptor families became more complicated as brain power increased . The results showed that the glutamate receptors found in vertebrates are only a fraction of all the types that exist . In fact , before present-day animal groups emerged , the part of the genome that holds the ionotropic receptor genes duplicated three times . This formed four receptor subfamilies , and our ancestors had all of them . Across the animal kingdom , there are ten , not six , classes of ionotropic receptors and there is an extra class of metabotropic receptors . But only two subfamilies of ionotropic and three out of four metabotropic receptor classes are still present in vertebrates today . The current classification of glutamate receptors centers around vertebrates , ignoring other animals . But this new data could change that . A better knowledge of these new receptors could aid neuroscientists in better understanding the nervous system . And , using this technique to study other families of proteins could reveal more missing links in evolution .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology", "neuroscience" ]	2018	Metazoan evolution of glutamate receptors reveals unreported phylogenetic groups and divergent lineage-specific events
When complexed with antigenic peptides , human leukocyte antigen ( HLA ) class I ( HLA-I ) molecules initiate CD8+ T cell responses via interaction with the T cell receptor ( TCR ) and co-receptor CD8 . Peptides are generally critical for the stable cell surface expression of HLA-I molecules . However , for HLA-I alleles such as HLA-B35:01 , peptide-deficient ( empty ) heterodimers are thermostable and detectable on the cell surface . Additionally , peptide-deficient HLA-B35:01 tetramers preferentially bind CD8 and to a majority of blood-derived CD8+ T cells via a CD8-dependent binding mode . Further functional studies reveal that peptide-deficient conformers of HLA-B35:01 do not directly activate CD8+ T cells , but accumulate at the immunological synapse in antigen-induced responses , and enhance cognate peptide-induced cell adhesion and CD8+ T cell activation . Together , these findings indicate that HLA-I peptide occupancy influences CD8 binding affinity , and reveal a new set of regulators of CD8+ T cell activation , mediated by the binding of empty HLA-I to CD8 . The major histocompatibility complex class I ( MHC-I ) molecules play a crucial role in adaptive immune responses by presenting antigenic peptides to CD8+ T cells , which enables the immune system to detect transformed or infected cells that display peptides from foreign or mutated self-proteins . Peptides are an integral component of MHC-I molecules . In the MHC-I antigen presentation process , peptides are mainly produced in the cytosol by proteasome and then translocated to the endoplasmic reticulum ( ER ) by the transporter associated with antigen processing ( TAP ) . Peptides are loaded to MHC-I peptide binding groove with the assistance of several ER chaperones , ERp57 , calreticulin and tapasin . There are several quality control components to ensure that most cell surface MHC-I molecules are filled with optimal peptide . However , under certain pathophysiological conditions , MHC-I peptide-deficient or open conformers are also detected on the cell surface . Prior evidence suggests that peptide-deficient conformers of MHC-I molecules appear on the cell surface of activated lymphoid cells ( Madrigal et al . , 1991; Schnabl et al . , 1990 ) , TAP-deficient cells ( Ljunggren et al . , 1990; Ortiz-Navarrete and Hämmerling , 1991 ) or EBV transformed B cells ( Madrigal et al . , 1991 ) . Although the presence of peptide-deficient conformers of MHC-I molecules on the cell surface under certain conditions is established , their functions are poorly understood . In the past few years , peptide-deficient conformers of MHC-I molecules of some allotypes have been shown to be ligands for cell surface receptors such as KIR3DS1 ( Burian et al . , 2016; Garcia-Beltran et al . , 2016 ) , KIR3DL2 ( Goodridge et al . , 2013 ) , KIR2DS4 ( Goodridge et al . , 2013 ) and LILRB2 ( Jones et al . , 2011 ) . However , most of these studies involved a non-classical HLA-I , HLA-F ( Garcia-Beltran et al . , 2016 ) , which has a higher propensity to be expressed in a peptide-deficient version compared to classical HLA-I molecules ( Goodridge et al . , 2010 ) . The paucity of functional studies of peptide-deficient conformers of classic HLA-I could partly be attributed to their general low stability on the cell surface . In a previous study ( Rizvi et al . , 2014 ) , we tested the refolding efficiencies of several HLA-B allotypes in the absence of peptide and found that peptide-deficient conformers of some allotypes such as B35:01 are relatively more stable . Higher stability of peptide-deficient B35:01 is also measurable in this study using a thermal unfolding assay with peptide-deficient HLA-B molecules that were engineered for enhanced stability via leucine zippered sequences ( Figure 1 ) . Peptide-receptive B35:01 molecules are also detectable on the surface of activated T cells ( this study ) and TAP-deficient cells ( Geng et al , submitted manuscript ) . Therefore , B35:01 is a good representative HLA-B to investigate the function of peptide-deficient conformers of HLA-I molecules . In exploring potential binding partners for peptide-deficient conformers of HLA-I molecules , we found that tetramers of peptide-deficient conformers of HLA-B35:01 , in stark contrast to their peptide-filled conformer , stain a majority of blood-derived CD8+ T cells . We hypothesized that the staining is largely CD8-mediated and also that peptide-deficient B35:01 molecules can modulate CD8+ T cell activation . Indeed , we show that CD8 prefers to bind peptide-deficient B35:01 molecules and that peptide-deficient HLA-B35:01 molecules on the cell surface enhance cell adhesion to CD8+ T cells . Although they do not directly activate CD8+ T cells , peptide-deficient HLA-B35:01 molecules on the surface of antigen presenting cells enhance antigen-specific CD8+ T cell responses . Together , these studies indicate key immune regulatory functions for peptide-deficient conformers of HLA-I molecules . We previously quantified refolding efficiencies of HLA-B heterodimers based on in vitro refolding reactions conducted in the absence of peptides . Significant differences in folding efficiencies were noted ( Rizvi et al . , 2014 ) . In the present study , we assessed whether peptide-deficient conformers of HLA-B allotypes also differ in their thermal unfolding characteristics ( Figure 1 ) . The NIH tetramer core facility has developed HLA-I molecules with epitope-linked β2m ( ELBM ) , wherein an HLA-I binding peptide is covalently linked to human β2m via a linker peptide that contains a protease cleavage site . HLA-I heavy chain and β2m are further tethered via leucine zippers ( LZ ) at their C-termini ( Figure 1A ) . Treatment of the HLA-I molecules with protease is expected to release the linked peptides , which are all C-terminally elongated , and thus sub-optimal for binding . When the cleavage is done in the presence of another HLA-I binding peptide , exchange should occur . Cleavage in the absence of peptide can produce peptide-deficient conformers of HLA-I molecules . Peptide-deficient conformers of four HLA-B molecules , B18:01 , B35:01 , B44:02 and B51:01 , were prepared and verified first by SDS-PAGE . The mobility of β2m was increased for all allotypes , consistent with expected reduction in molecular weight after cleavage ( Figure 1B left panels ) . Formation and peptide loading of peptide-deficient conformers of HLA-B molecules were further validated by native-PAGE ( Figure 1B right panels ) . In general , the mobilities for the cleaved HLA-B molecules are clearly different from uncleaved proteins . The reduced intensities of complex bands and increased intensities of the free β2m bands , likely because of heterodimer dissociation during electrophoresis , indicate that these heterodimers are less stable than uncleaved proteins . The HLA-B heterodimers were further loaded with allotype-specific peptides . The observed downward mobility shifts of the complex-specific bands for HLA-B35:01 , HLA-B18:01 and HLA-B44:02 in the native gels are indicative of peptide binding , and provide further evidence that the cleaved forms of those HLA-B molecules are in fact peptide-deficient . A downshift was not observed of the complex-specific bands for HLA-B51:01 , suggestive of low peptide-loading efficiency , also previously noted ( Rizvi et al . , 2014 ) . Thermostabilities of the peptide-deficient HLA-B molecules were assessed by comparing heat-induced unfolding with a thermal shift assay ( Figure 1C ) . A fluorescent dye ( Sypro Orange ) was used that displays enhanced binding to proteins following thermal unfolding . Clear cut transitions are observable for peptide-deficient B18:01 and B35:01 , but not for B44:02 or B51:01 . These findings indicate important thermostability hierarchies among peptide-deficient conformers of HLA-B; allotypes such as B35:01 and B18:01 are more stable in their peptide-deficient conformers compared to allotypes such as B44:02 and B51:01 , consistent with previously described refolding assay ( Rizvi et al . , 2014 ) . As a representative peptide-deficient HLA-B with high thermostability , HLA-B35:01 was used for further functional assessments . To explore potential receptors that are responsive to peptide-deficient HLA-B35:01 , tetramers were generated with peptide-deficient HLA-B35:01 or their peptide-filled versions . Peripheral blood mononuclear cell ( PBMC ) staining of tetramers of peptide-deficient conformers was compared with the peptide-filled HLA-B35:01 . PBMCs obtained from healthy donors were stained with a panel of lymphocyte markers before tetramer staining . As expected , antigen-specific CD8+ T cell populations were rare or absent in PBMCs from healthy B35:01+ donors , as assessed by staining with uncleaved B35:01 ( carrying an epitope LPYPQPQPF from Triticum aestivum ) tetramers ( for example , Figure 2A ) . In contrast , peptide-deficient HLA-B35:01 tetramers bound to most ( over 70% ) of total CD8+ T cells present in the donor ( Figure 2B ) . These findings suggested that observed tetramer binding was unlikely to be linked to specific TCR . Rather , staining was significantly blocked by anti-CD8 ( Clone SK1 , BioLegend ) ( Figure 2C ) , suggesting that peptide-deficient B35:01 tetramers are capable of binding to cell surface CD8 with higher potency compared to the peptide-filled version . In parallel analyses , CD4+ T cells were poorly stained by the peptide-deficient B35:01 tetramers ( Figure 2D ) and staining was not blocked by anti-CD8 ( Figure 2E ) , consistent with the finding of CD8-dependent binding to CD8+ T cells . Similar results were obtained with cells from a B35:01-negative donor ( Figure 2F–J ) , of enhanced CD8-dependent binding of peptide-deficient HLA-B35:01 tetramers to CD8+ T cells ( Figure 2G–H ) , and comparatively poor CD8-independent binding to CD4+ T cells ( Figure 2I–J ) . Binding and inhibition data compiled from multiple B35:01-positive and B35:01-negative donors are shown in Figure 2K and L . CD8 is also found on the surface of other lymphocytes , such as NK cells , although as a CD8αα homodimeric form instead of CD8αβ heterodimeric form ( Moebius et al . , 1991 ) . We examined the ability of peptide-deficient B35:01 tetramers to bind to CD8 on the NK cell surface . B35:01 belongs to Bw6 serotype of HLA-B alleles that lack a binding sequence for engagement of the HLA-B recognizing killer immunoglobulin-like receptor , KIR3DL1 ( Gumperz et al . , 1997 ) . In fact , we found that peptide-deficient B35:01 tetramers exclusively bind to NK cells expressing CD8 and further that binding to the cells is fully blocked by anti-CD8 ( Figure 3A ) . Furthermore , NK cells from different donors have different percentages of CD8 expressing NK cells , and the level of staining with peptide-deficient B35:01 tetramers is directly proportional to the CD8 expressing fraction of NK cells ( Figure 3B ) . Further experiments were undertaken to compare the binding of purified FITC-labeled CD8αα to peptide-deficient or peptide-filled HLA-B35:01 conjugated to streptavidin resin ( Figure 3C ) . Bead-bound CD8 was quantified by fluorimaging analyses of SDS-PAGE-separated samples . Nonlinear curve fitting analyses of the FITC fluorescence signals yielded a KD value of ~20 μM for peptide-deficient B35:01 , significantly stronger binding than that for peptide filled B35:01 , for which a KD value could not be accurately estimated ( Figure 3D ) . CD8 can act as adhesion molecule , co-receptor and immuno-modulator ( Cole and Gao , 2004 ) . Interaction between MHC-I and CD8 is proposed to enhance cell adhesion ( Norment et al . , 1988 ) . We assessed whether the stronger interaction between peptide-deficient HLA-B35:01 and CD8 could enhance cell-cell adhesion . We expressed HLA-B35:01 and a HLA-B35:01 mutated at the CD8 binding residues ( D227K/T228A; B35:01-CD8 null ) ( Purbhoo et al . , 2001 ) , in a TAP1-deficient cell line SK19 ( Yang et al . , 2003 ) . Both proteins are readily detectable on the cell surface ( Figure 4A–B ) . Incubation with a B35:01-specific peptide HPVGEADYFEY ( HPV ) , but not a related truncated and mutated control peptide HGVGEADYFE ( HGV ) , induces binding by the peptide-MHC-I complex-specific W6/32 antibody ( Parham et al . , 1979 ) and reduces binding by the heavy chain-specific HC10 antibody ( Stam et al . , 1990; Gillet et al . , 1990 ) for both B35:01 molecules ( Figure 4C–D ) , indicating that at least a subset are able to be expressed as peptide-deficient conformers , under conditions where TAP , the major source of cellular MHC-I peptides , is absent . To test cell adhesion mediated by peptide-deficient B35:01 or its CD8-null version , two CTL lines were used as CD8 expressing cells . Cell conjugation between the SK19 cells and CTLs was investigated by two approaches , confocal microscopy and flow cytometry . In the microscopy assay , SK19 cells were pre-attached to glass-bottomed petri dish . After co-incubation , CTL line A2-AL9 specific for HLA-A0201 complexed to the HIV-derived AL9 peptide ( AIIRILQQL ) ( Altfeld et al . , 2001 ) showed significantly increased adhesion to SK19 cells expressing HLA-B35:01 compared with SK19 cells lacking HLA-B35:01 ( Figure 4E–F ) . There was also a very marked blocking effect of pre-incubation of SK19 HLA-B35:01 cells with the B35:01 specific peptide HPV upon conjugate formation ( Figure 4G ) , suggesting that peptide-deficient B35:01 is important for mediating cell adhesion . On the other hand , there was no significant cell adhesion enhancement with SK19-HLA-B35:01-CD8 null compared with SK19 cells lacking HLA-B35:01 ( Figure 4H ) , reflecting the significance of CD8-B35:01 binding upon enhancement of cell adhesion . In the flow cytometry assays , SK19 cells were labeled with CFSE and then incubated with CTL . CFSE and CD8 double positive cell populations were identified as SK19-CTL conjugates , and used to quantify the effect of B35:01-CD8 interactions on cell adhesion . Compared with SK19-HLA-B35:01-CD8 null , SK19-HLA-B35:01 showed stronger adhesion to CTL line , A2-AL9 ( Figure 4I upper panel ) . Pulsing of SK19-HLA-B35:01 with peptide HPV strongly reduced cell adhesion , consistent with the microscopy assays . We also generated a CTL line ( B8-RL8 ) specific for HLA-B08:01 complexed with the EBV-derived epitope RAKFKQLL ( RL8 ) , for use as second set of effector cells , and obtained similar results as with the A2-AL9 CTL line ( Figure 4I lower panel ) . Stronger binding of peptide-deficient HLA-B to CD8 ( Figure 4 ) raised the possibility of CD8+ T cell activation regulation by peptide-deficient conformers ( Wooldridge et al . , 2010 ) . We therefore tested whether tetramers of peptide-deficient conformers directly activated CD8+ T cells . Intracellular staining assays were carried out to examine whether CD8 ligation with peptide-deficient B35:01 tetramers induces cytokine expression . We found that cross-linking of CD8 with peptide-deficient B35:01 tetramers at a concentration as high as 40 μg/ml failed to activate primary CD8+ T cells as well as the CTL lines A2-AL9 and B8-RL8 , as assessed by the general absence of changes in the expression of cytokine interferon gamma ( IFN-γ ) ( Figure 4—figure supplement 1 ) . These findings suggested that the ligation of CD8 by peptide-deficient B35:01 did not directly induce significant activation signaling . Although ligation of CD8 by peptide-deficient B35:01 did not induce direct activation of CD8+ T cells , the enhanced cell-adhesion mediated by the interaction ( Figure 2-Figure 4 ) could modulate antigen-specific CD8+ T cell activation . This could not be tested in TAP-deficient SK19 cells , due to very low cell surface expression of both HLA-A02:01 and HLA-B08:01 , the HLA-I molecules recognized by the two CTL lines used in Figure 4 . Lymphocyte activation is previously shown to induce forms of HLA-I molecules that are recognized by HC10 ( Matko et al . , 1994 ) , the antibody specific for peptide-deficient conformations ( Stam et al . , 1990 ) . Consistent with these prior findings , purified CD4+ T cells from different donors were shown to consistently induce HC10-reactive HLA-I molecules on the cell surface following their activation with PHA ( Figure 5A ) . As discussed below in Figure 6 , the presence of peptide-receptive B35:01 was directly measurable on activated CD4+ T cells from HLA-B35:01+ donors . Activated CD4+ T cells were thus used as antigen-presenting cells to present peptide-HLA–B08:01 or peptide-HLA-A02:01 complexes , and provide a parallel source of peptide-deficient B35:01 for further functional assessments of the effects of peptide-deficient B35:01 on antigen-specific T cell responses . Molecular clustering within the immunological synapse is emerging as key mechanism for the control of T cell activation . Therefore , we first used a cell-cell contact assay to determine whether peptide-deficient conformers are clustered within the immunological synapse induced by recognition of RL8-HLA–B08:01 by the B8-RL8 CTL line . CD4+ T cells from a donor expressing both HLA-B08:01 and HLA-B35:01 were pre-activated to induce peptide-deficient conformers on the cell surface . B8-RL8 CTLs were co-incubated with the activated CD4+ T cells pulsed with the antigenic peptide RL8 . RL8 induces stronger clustering of HLA-I peptide-deficient conformers ( measured with the peptide-deficient conformer-specific antibody HC10 , Figure 5D–E , H ) than peptide-filled HLA-I ( measured with the peptide-MHC-I complex-specific W6/32 antibody , Figure 5B–C and H ) in the interface between antigen presenting cells ( APC ) and CTLs . On the other hand , in the absence of RL8 peptide , cell conjugates were strongly reduced and little enrichment was observed in the junctions between CTL and APC ( Figure 5F–G , H ) . We did not observe strong CD8 clustering in the interface , probably due to different kinetics of MHC-I and CD8 clustering which has been reported previously ( Purbhoo et al . , 2004 ) . Similar findings were obtained when activated PBMC rather than activated CD4+ T cells were used as the antigen-presenting cells ( Figure 5—figure supplement 1 ) . We further tested whether the peptide-deficient HLA-B35:01-CD8 interaction has a regulatory effect on cognate antigen-induced target cell lysis . The CTL line specific to the HLA-A2-AL9 complex was first chosen as the effector cell . Primary CD4+ T cells expressing both A02:01 and B35:01 were used as target cells . After incubation at effector-to-target ratios of 1:1 , 5:1 and 20:1 , CTLs exhibited a strong increase in the ability to kill target cells pulsed with the cognate peptide AL9 ( Figure 6—figure supplement 1 , right panels ) compared to target cells pulsed with control peptide SLYNTVATL ( SL9 ) ( Figure 6—figure supplement 1 , left panels ) . Next , primary CD4+ T cells expressing both A02:01 and B35:01 from Donor 24 were activated and cell surface expression of peptide-deficient conformers were detected with peptide-deficient conformer-specific antibody HC10 . Peptide-deficient conformers could be partially blocked by B35:01-specific peptides YPLHEQHGM ( YPL ) , HPNIEEVAL ( HPN ) , HPV and FPTKDVAL ( FPT ) , but not control peptide ( TSTLQEQIGW , TW10 ) ( Figure 6F–I ) , indicating that a subset of the HLA-B35:01 molecules are peptide-deficient . We found that the cognate peptide-induced CD4+ T cell lysis can partly be blocked by B35:01-specific peptides ( Figure 6A ) , suggesting that HLA-B35:01 peptide-deficient conformers do enhance cell lysis induced by cognate peptides . The effect on modulating antigen-specific cell lysis is observed with different HLA-B35:01 peptides and across different donors ( Figure 6B–C ) . Similar effects were also observed in cell lysis assays with the B8-RL8 CTL line ( Figure 6D–E ) . The major functions of MHC-I proteins include presenting antigenic peptides to CD8+ T cells and delivering activation or inhibitory signals to NK cells . It was widely known that the interactions between MHC-I molecules with their receptors are both allotype and peptide dependent . Our studies indicate that peptide-deficient MHC-I molecules are also functional in the immune response . A given peptide-HLA-I complex is typically able to engage only a small percentage of blood-derived CD8+ T cells , those that bear an appropriate TCR . In contrast , peptide-deficient conformers of HLA-B35:01 engaged a majority of CD8+ T cells from multiple donors ( Figure 2 ) . While the pMHC-I/CD8 interaction is generally characterized by very low affinities ( Wang et al . , 2009; Wyer et al . , 1999 ) , we find that peptide-free HLA-B35:01 binds CD8 with significantly higher affinity than their peptide-filled versions ( Figure 3 ) . Thus , unlike peptide-occupied HLA-B that engage CD8+ T cells via a TCR–dependent binding mode , peptide-deficient conformers of HLA-B35:01 engage CD8+ T cells via a binding mode that is largely CD8-dependent . The MHC-I-binding site for CD8 is spatially separated from the peptide-binding domains that are recognized by the TCR , and this spatial segregation allows both TCR and CD8 to bind a single MHC-I molecule simultaneously . In contrast to peptide-loaded MHC-I molecules , peptide-free MHC I molecules are suggested to possess properties similar to molten globules ( Bouvier and Wiley , 1998 ) , and show more protein plasticity based on MD simulations ( van Hateren et al . , 2013 ) . The stronger binding to CD8 of peptide-deficient HLA-B compared to peptide-filled HLA-B ( Figure 3 ) is likely caused by conformational differences between peptide-occupied and peptide-deficient conformers of HLA-B molecules that determine the accessibility or orientation of the CD8 binding site on HLA-I . Peptide-deficient HLA-I molecules are also preferred by ER chaperones tapasin and TAPBPR , which functions to facilitate peptide loading of MHC-I molecules . Crystal structures of tapasin-MHC-I and TAPBPR-MHC-I complexes highlight some common MHC-I binding sites by tapasin/TAPBPR and CD8 ( Blees et al . , 2017; Jiang et al . , 2017; Thomas and Tampé , 2017 ) . Residues at the C-terminal immunoglobulin-like domain of tapasin are positioned close to the CD8 recognition loop ( especially residues 225 and 226 ) of the α3-domain of the MHC-I heavy chains ( Gao et al . , 1997; Wang et al . , 2009 ) , suggesting that the sites co-evolved ( Blees et al . , 2017 ) . A β hairpin of TAPBPR at the N-terminal domain , which reaches under the floor of the peptide-binding groove , is important for sensing the conformation changes of the peptide-binding groove ( Thomas and Tampé , 2017 ) . CD8 also interacts with MHC-I at a similar region ( including residues 115 , 122 and 128 ) ( Gao et al . , 1997 ) . Peptide loading reduces binding affinity between MHC-I molecules from tapasin and TAPBPR resulting in release of MHC-I molecules from tapasin and TAPBPR ( Rizvi and Raghavan , 2006; Wearsch and Cresswell , 2007 ) . CD8 might share a similar mechanism as tapasin and TAPBPR to distinguish MHC-I molecules with different conformations . CD8 functions as an adhesion molecule and co-receptor to enhance the formation of TCR/pMHC complexes and the activation of CD8+ T cells . Although ligation of CD8 with non-cognate peptide-MHC-I complex was proposed to augment CD8+ T cell activation levels ( Anikeeva et al . , 2006; Yachi et al . , 2005 ) , generally , cognate peptide loading of MHC-I molecules is indispensable for CD8 T cell activation . Although a previous study showed that MHC-I molecules with super-enhanced CD8 binding properties bypass the requirement for cognate TCR recognition and nonspecifically activate CTLs ( Wooldridge et al . , 2010 ) , we did not see any direct activation of CTL by HLA-B35:01 peptide-deficient conformers . Nonetheless , our data suggested that preferential engagement of CD8 by peptide-deficient conformers of HLA-B35:01 enhances cognate peptide-induced cell lysis ( Figure 6 ) . The enhancement of T cell activation appears to be caused by enhanced cell adhesion induced by the peptide-deficient conformer-CD8 interaction , or enhanced signaling induced by peptide-deficient conformer enriched within the immunological synapse ( Figure 7 ) . To escape immune surveillance by CD8+ T cells , several pathogens and tumors block HLA-I antigen presentation pathways to prevent antigenic peptide presentation by HLA-I molecules . Interestingly , many pathogen evasion or tumor progression strategies involve the targeting of the TAP transporter , inducing the cell surface expression of partially peptide-deficient HLA-I , as illustrated for HLA-B35:01 ( Figure 4 ) . Peptide-deficient conformers of HLA-I are expected to enhance CD8+ T cells responses against TAP-independent epitopes , and thus counter the pathogen evasion strategies that target the HLA-I antigen presentation pathway . Further studies are needed to quantitatively understand the extent of allele-dependent variations in CD8 binding by both peptide-deficient and peptide-filled conformers of HLA-B , as well as the induction of HLA-I peptide-deficient conformers under different physiological and pathological conditions . Peptide-deficient forms of different HLA-B allotypes were shown to have distinct thermostabilities and are therefore expected to be expressed at different levels on the cell surface . HLA alleles are known to differently associate with disease progression outcomes in major infectious diseases such as acquired immune deficiency syndrome ( AIDS ) ( Carrington and Walker , 2012 ) and with autoimmune diseases such as ankylosing spondylitis ( AS ) ( Brown et al . , 2016 ) , but the general underlying mechanisms are incompletely characterized . AS has been linked to the expression of HLA-B27:05-free heavy chains ( Khare et al . , 1996 ) , which can readily be detected on the surface of TAP-deficient cells ( Allen et al . , 1999 ) . It would be of interest to test whether the interactions between peptide-deficient conformers of HLA-B27:05 molecules and CD8 are involved in the onset and outcome of these diseases . In conclusion , our findings indicate that , without interaction with TCR , the peptide-deficient conformers of HLA-B35:01 are able to interact efficiently with CD8 . The preferential interaction between HLA-I peptide-deficient conformers and CD8 described here identifies a previously unknown mechanism by which CTL can be regulated . Finally , HLA-B peptide-deficient conformer-CD8 interactions may also have physiological regulatory functions in NK cell biology , which requires further study . Blood was collected from consented healthy donors for HLA genotyping and functional studies in accordance with a University of Michigan IRB approved protocol ( HUM00071750 ) . Human melanoma cell line SK-mel-19 ( SK19 ) ( RRID: CVCL_6025 ) ( Yang et al . , 2003 ) and ecotropic virus packaging cell line BOSC ( RRID: CVCL_4401 ) were grown in DMEM ( Life Technologies ) supplemented with 10% ( v/v ) FBS ( Life Technologies ) and 1 × Anti/Anti ( Life Technologies ) ( D10 ) . SK19 cells were gifted by Dr . Pan Zheng and verified for the absence of TAP1 expression . BOSC cells were obtained from the lab of Dr . Kathleen Collins . CTL line A2-AL9 was kindly gifted by Dr . Bruce Walker . CTL line B8-RL8 was generated in the lab by sorting after tetramer staining as previously described ( Dong et al . , 2010 ) . The following monoclonal antibodies were used in this study: Ascites of W6/32 and HC10 from the University of Michigan Hybridoma Core , purified anti-human CD8a ( clone SK1; BioLegend ) , AF700-conjugated anti-human CD8a ( clone HIT8a; BioLegend ) , APC-Cy7-conjugated anti-human CD4 ( clone RPA-T4; BioLegend ) , Pacific Blue-conjugated anti-human CD3 ( clone UCHT1; BioLegend ) , PE-Cy7-conjugated anti-human CD56 ( clone CMSSB; eBioscience ) , purified anti-human CD28 ( clone 28 . 2; BD Biosciences ) and FITC-conjugated anti-human IFN-γ ( clone 4S . B3; BD Biosciences ) . Dead cells were excluded from flow cytometric analyses with 7-amino-actinomycin D ( 7-AAD; BD Biosciences ) or the amine-reactive dye aqua ( 405 nm , Life Technologies ) . Fresh blood was subjected to centrifugation over a Ficoll-Paque Plus ( GE Healthcare Life Sciences ) density gradient , washed twice in PBS and resuspended in RPMI1640 ( Life Technologies ) supplemented with 10% ( v/v ) FBS ( Life Technologies ) and 1 × Anti/Anti ( Life Technologies ) ( R10 ) . Assays were performed either on freshly isolated PBMC used within 2 to 4 hr of cell preparation , or on PBMC cryopreserved in Recovery Cell Culture Freezing Medium ( Life Technologies ) . DNA was extracted from PBMCs using DNeasy Blood and Tissue Kit ( Qiagen ) . The HLA typing was performed by Sirona Genomics ( Mountain View , CA ) , an Immucor Company . The assay , based on a previous publication ( Wang et al . , 2012 ) was performed using the MIA FORA NGS HLA typing assay for the class I loci . The full-length amplicons for the class I loci were amplified and pooled . These samples were then fragmented , and tagged with unique index adaptors . The samples were pooled and sequenced on the Illumina MiSeq , and the HLA type was determined using the MIA FORA NGS HLA typing software . The Sirona Genomic HLA typing method has been validated by the Histocompatibility , Immunogenetics and Disease Profiling Laboratory of the Stanford University School of Medicine using 50 reference cell lines . LZ-ELBM HLA-B molecules were provided by the NIH tetramer core facility . Peptide-deficient conformers of molecules were prepared by incubation with PreScission protease or thrombin overnight at 25°C . Cleaved fragments were removed by centrifuging the sample in a 0 . 5 ml Amicon Ultra filter device for 30 min at 13 , 000 rpm , 4°C . Native-PAGE and SDS-PAGE gels were both run to verify that the cleavage was efficient and HLA-B molecules became peptide-deficient . Peptide exchanges were performed by incubating HLA-B molecules with high affinity peptides together with PreScission protease or thrombin overnight at 25°C . Thermal shift assays were undertaken as previously described ( Del Cid et al . , 2010; Huynh and Partch , 2015 ) . HLA-B molecules ( 8 µM ) were incubated in buffer ( PBS , pH7 . 4 ) and 1 × Sypro Orange Stain ( Invitrogen ) in a total reaction volume of 20 µl . Thermal scans were performed using an ABI PRISM 7900HT Sequence Detection System with temperature increments of 1°C . Fluorescence emission was measured at ROX channel . Fluorescence was normalized within wells as percent maximum fluorescence and plotted against the sample temperature . Peptide-deficient conformers of HLA-B35:01 were prepared by treatment of LZ-ELBM HLA-B35:01 molecules with PreScission protease ( GE Healthcare Life Sciences ) for 2 hr at room temperature to release the tethered peptide , while peptide exchange was performed by adding HLA-B35:01 binding epitopes simultaneously with PreScission protease . HLA-B35:01 molecules were further dialyzed thoroughly with Amicon Ultra Centrifugal Filter Devices ( Millipore ) with a 10 kDa cutoff to remove unbound peptides . The peptide-deficient and peptide-exchanged monomers were verified by SDS or native-PAGE gels . The peptide epitope HPV was used in this study to reconstitute B35:01 . Tetrameric HLA-I reagents were constructed by the addition of streptavidin conjugated to PE ( Prozyme , PJRS25 ) or APC ( Prozyme , PJ27S ) at 4:1 molar ratios following the tetramerization protocol from NIH core facility . PBS + 0 . 5% dialyzed BSA was used as staining buffer . For CD8 blocking , anti-CD8 ( clone SK1; BioLegend ) was incubated with PBMCs at 10 μg/ml for 15 min at room temperature . After washing once , freshly prepared tetramers were added typically at 20 μg/ml and anti-CD8-AF700 ( clone HIT8a; BioLegend ) , anti-CD3-pacific blue ( clone UCHT1; BioLegend ) and anti-CD4-APC-Cy7 ( clone RPA-T4; BioLegend ) were added at concentrations indicated by the manufactures and incubated for another 30 min at room temperature . Cells were washed 3 times and 7AAD was added as a live/dead marker and samples were then analyzed by flow cytometry on a BD FACSCanto II flow cytometer . The FACS data were analyzed with FlowJo software version 10 . 0 . 8 ( Tree Star , San Carlos , CA ) . Soluble human CD8αα ( residues 1–120 ) with his-tag at the N-terminus was expressed in Escherichia coli , refolded , and purified by gel filtration as a ~30 kDa homodimer ( Gao et al . , 1997 ) . The CD8αα concentration was calculated from the extinction coefficient , which was determined by amino acid analysis to be 37150 M−1cm−1 at 280 nm . CD8αα was labeled with FITC according to manufacturer’s protocol ( Thermo Scientific , Rockford , IL , USA ) . Relevant biotinylated HLA-B monomers from the tetramer core were immobilized onto streptavidin-coated agarose resin . FITC-labeled soluble CD8αα was added at different concentrations ( 2 . 5 , 5 , 10 and 20 μM ) to immobilized HLA-B in binding buffer ( PBS + 0 . 5% BSA ) . CD8αα was pulled down after co-incubation with the resin , and the beads were washed with binding buffer . SDS loading buffer was added and samples were denatured by heating for 10 min . Samples were loaded and resolved by SDS-PAGE and visualized by fluorimaging on a Typhoon scanner ( at 520 nm ) . The binding at each concentration was obtained by subtraction of the control response ( resin alone ) from the B35:01 response . HLA-B35:01 in retroviral vector LIC pMSCVneo were prepared as described previously ( Rizvi et al . , 2014 ) . The HLA-B35:01 mutant that cannot bind CD8 ( HLA-B35:01-CD8 null ) was made by introducing D227K/T228A mutations into HLA-B35:01 using the QuikChange II site-directed mutagenesis kit . The primers were 5’- ggtctccacaagctcagccttctgagtttggtcctcgc-3’ ( forward ) and 5’-gcgaggaccaaactcagaaggctgagcttgtggagacc-3’ ( reverse ) . All primers were purchased from Invitrogen . Retroviruses were generated using BOSC cells and used to infect SK19 cells . Cells were infected with HLA-B-encoding viruses or control viruses lacking HLA-B . Infected cells were selected by treatment with 1 mg/ml G418 ( Life Technologies ) , and maintained in 0 . 5 mg/ml G418 . A total of 1 × 105–1 × 106 cells were washed with FACS buffer ( phosphate-buffered saline ( PBS ) , pH 7 . 4 , 1% FBS ) and then incubated with W6/32 or HC10 antibodies at 1:250 dilutions for 30–60 min on ice . Following this incubation , the cells were washed three times with FACS buffer and incubated with GαM-PE at 1:250 dilutions for 30–60 min on ice . Following incubations , the cells were washed three times with FACS buffer and analyzed using a BD FACSCanto II cytometer . For peptide occupancy assay , cells were preincubated with peptides for 2 hr at 37°C before staining . SK19 cells expressing HLA-B35:01 , HLA-B35:01-CD8 null or lacking HLA-B35:01 ( those infected with a control virus lacking HLA-B ) were first labeled with CFSE according to manufacturer’s protocol . For microscope-based assays , SK19 cells were plated to glass-bottomed petri dish the day before the adhesion assay . After washing the dish with medium , CTLs were added at 1:1 and incubate at 37°C for 2 hr . The cells were then washed with 1 × PBS and fixed with 2% PFA . After staining with anti-CD8-APC , imaged using a Leica SP8 confocal microscope . For flow-cytometry-based assays , CFSE-labeled SK19 cells were incubated in suspension with CTLs at 37°C for 2 hr . The cells were then washed with 1 × PBS , fixed with 2% PFA and stained with anti-CD8-APC . CFSE and CD8 double positive cells were quantified by flow cytometry as conjugated cells . PBMCs from donor carrying HLA-B08:01 and B35:01 were preactivated with PHA to express peptide-deficient conformers of HLA-I . Cognate peptide RL8 ( 100 uM ) or DMSO was loaded at 37°C for 2 hr . The PBMCs and HLA-B08:01-RL8 specific CTL line ( CTL B8-RL8 ) were mixed , centrifuged briefly and incubated for 10 min at 37°C to allow immunological synapse formation . Cells were fixed and stained with anti-CD8-APC and W6/32-FITC or HC10-FITC antibodies . Cells were imaged using a Leica SP8 confocal microscope . FITC and APC emission were collected in different channels . Data were processed using Leica Imaging software and ImageJ software . The intensity of HLA-I molecules at the interface was compared with the membrane at a noncontact area and plotted as the fold increase above background . Primary CD8+ T cells were purified from PBMCs by negative selection by magnetic-activated cell sorting ( MACS , Miltenyi Biotec ) , according to the instructions . Tetramers were incubated with primary CD8+ T cells or CTL lines at indicated concentrations in serum free medium for 30 min at room temperature and 5 μg/ml anti-CD28 was then added . FBS was supplemented to a final concentration of 10% . 2 × 105 CD8+ T cells were stimulated at 37°C for six hours in the presence of brefeldin A ( Golgiplug , 1:1000; BD Biosciences ) . Intracellular staining assays were performed to test cytokine expression . Briefly , cells were washed and fixed using 100 μl of 4% formaldehyde at RT for 10 min . After being washed , cells were incubated with 100 μl of 0 . 2% saponin and then stained with fluorochrome-conjugated antibodies specific for intracellular markers at RT for 30 min . After a final wash , flow cytometry measurements were acquired on a BD FACSCanto II flow cytometer . Flow cytometry gates to identify positive cytokine signal were based on unstimulated control tubes . Primary CD4+ T cells were purified from PBMCs by negative selection by magnetic-activated cell sorting ( MACS , Miltenyi Biotec ) , according to the instructions and activated then with PHA . Cells form donors carrying HLA-A02:01 or HLA-B08:01 were pulsed for 2 hr with AL9 or RL8 peptides at 100 μM , respectively , together with or without B35:01 blocking peptides and then incubated with corresponding CTL lines ( A2-AL9 or B8-RL8 ) at indicated ratios for 5 hr at 37°C . Cells were then stained with anti-CD4 , anti-CD8 and live/dead marker 7AAD or Aqua to test the viability of CD4+ T cells by flow cytometry . Statistical analyses ( ordinary one-way ANOVA analysis with Fisher’s LSD or Dunnett test ) were performed using GraphPad Prism version 7 .	The immune system keeps tabs on everything that happens in our body , looking for potential signs of threat . To alert it to any problems , almost every cell produces specific proteins on its surface called human leukocyte antigens class I , or HLA-I for short . These HLA-I molecules are bound to small protein fragments called peptides that have been exported from within the cell and are presented to the cells of the immune system for scanning . When cells are healthy , the peptides all stem from normal proteins . But , if the cell has become infected or cancerous , it contains foreign or abnormal peptides . Some of the HLA-I molecules , however , are empty . These antigens are unstable , and their role is unclear . Now , Geng et al . investigated this further by studying blood samples from healthy donors . The experiments revealed that empty HLA-I molecules help specialized cells of the immune system , the killer T cells , to bind to the antigens , improving their killing ability . It is known that these T cells recognize and bind to the antigens through two receptor proteins , one of which is called CD8 . It was known that when HLA-I molecules carry a peptide , only a small fraction of T cells with a matching receptor can bind . However , Geng et al . found that when HLA-Is were empty , a much larger proportion of the T cells was able to bind to antigens . This indicates that CD8 ‘prefers’ to attach to empty HLA-Is , maybe because binding sites are more accessible . CD8 also enhances the binding between the T cells and the antigen . Empty HLA-Is did not directly activate the T cells but did enhance their immune response . When both full and empty HLA-I were present , the T cells were even more effective at killing their targets . Understanding how killer T cells work is essential for the development of immunotherapies – treatments that help to boost the immune system to fight infections and cancer . Increasing the number of empty HLA-I molecules on cancer or infected cells could enhance T cell killing .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "immunology", "and", "inflammation" ]	2018	Empty conformers of HLA-B preferentially bind CD8 and regulate CD8+ T cell function
Determining the adaptive potential of foundation species , such as reef-building corals , is urgent as the oceans warm and coral populations decline . Theory predicts that corals may adapt to climate change via selection on standing genetic variation . Yet , corals face not only rising temperatures but also novel diseases . We studied the interaction between two major stressors affecting colonies of the threatened coral , Acropora cervicornis: white-band disease and high water temperature . We determined that 27% of A . cervicornis were disease resistant prior to a thermal anomaly . However , disease resistance was largely lost during a bleaching event because of more compromised coral hosts or increased pathogenic dose/virulence . There was no tradeoff between disease resistance and temperature tolerance; disease susceptibility was independent of Symbiodinium strain . The present study shows that susceptibility to temperature stress creates an increased risk in disease-associated mortality , and only rare genets may maintain or gain infectious disease resistance under high temperature . We conclude that A . cervicornis populations in the lower Florida Keys harbor few existing genotypes that are resistant to both warming and disease . Genetic diversity within a population leads to varying levels of stress tolerance among individuals ( Sorensen et al . , 2001 ) , and is critical for species survival and persistence in a changing climate ( Hoffmann and Sgrò , 2011 ) . It is well known that certain corals are more resilient to stress than others , and the genotype of the coral plays a significant role in determining thermal resistance ( Edmunds , 1994; Fitt et al . , 2009; Baird et al . , 2009; Baums et al . , 2013; Kenkel et al . , 2013 ) , with a heritable component ( Kenkel et al . , 2013; Dixon et al . , 2015; Polato et al . , 2013 ) . Tolerance to stress may also be a result of different symbiotic algal species ( Symbiodinium spp . ) or even of different genotypes ( i . e . strains ) of certain Symbiodinium species that reside within the coral host ( Grégoire et al . , 2017; Parkinson and Baums , 2014; Fabricius et al . , 2004 ) . Furthermore , additional threats to corals , such as infectious disease outbreaks and ocean acidification , are affecting populations in combination with temperature anomalies ( Hoegh-Guldberg et al . , 2007 ) . Evidence suggests that many species possess the ability to produce broad-spectrum defense mechanisms ( de Nadal et al . , 2011 ) . Similarly , coral populations showing resilience to high water temperatures constitutively frontload the expression of genes related to heat tolerance in concert with several genes influencing the host innate immune response ( Barshis et al . , 2013 ) , suggesting these corals may also have evolved the general ability to tolerate a multitude of threats . A critical question is whether certain coral genotypes , existing within the same environment , are generally more stress resistant compared with other conspecifics . And additionally , does stress resistance in one trait predict stress resistance in another ? The Caribbean coral species , Acropora cervicornis , was one of the most common corals within the shallow reefs of the Western Atlantic and Caribbean several decades ago ( Pandolfi , 2002 ) . However , over the last 40 years multiple stressors including infectious disease , high sea surface temperatures , overfishing and habitat degradation have caused a 95% population reduction ( Acropora Biological Review Team , 2005 ) . A . cervicornis is now listed as threatened under the U . S . Endangered Species Act . Significant loss of Caribbean Acropora species was attributed to white-band disease outbreaks that spread throughout the region in the late 1970’ s and early 1980’ s ( Aronson and Precht , 2001 ) . While the disease-causing agent of white band has not been identified , the pathogen is likely bacterial in nature ( Peters , 1984; Kline and Vollmer , 2011; Sweet et al . , 2014 ) . This disease continues to cause mortality across populations ( Miller et al . , 2014 ) and especially Florida ( Precht et al . , 2016 ) . Recently , variability of Acropora spp . susceptibility to disease has been explored and documented . For example , 6% of the A . cervicornis tested in Panama were resistant to white-band disease ( Vollmer and Kline , 2008 ) . Additionally , long term monitoring of A . palmata in the US Virgin Islands indicated that 6% of 48 known genets showed no disease signs over eight years; perhaps indicating a small disease resistant population ( Rogers and Muller , 2012 ) . Hence , there are disease resistant variants within some locations , although they may be low in abundance . Regardless , anomalously high water temperatures are increasing the probability of disease occurrence throughout the Caribbean ( Muller et al . , 2008; Miller et al . , 2009; Randall and van Woesik , 2015 ) and field monitoring suggests that bleached corals are more susceptible to disease ( Muller et al . , 2008 ) . Alternatively , recent field-based observations suggest that there is a negative association between heat tolerance and disease susceptibility in A . cervicornis ( Merselis et al . , 2018 ) . Here , we determine whether high water temperature changes the susceptibility of disease resistant variants and explore the potential relationship between disease resistance and susceptibility to high temperatures . Tropical reef-building corals gain a majority of their carbon from their algal symbionts ( Muscatine et al . , 1984 ) and thus the stress response of the coral animal has to be viewed in the context of its symbiotic partner or partners . Prolonged temperature stress causes the disassociation between the coral host and the single-celled algae ( Symbiodinium spp ) residing within its tissues , a phenomenon called bleaching . Studies of the immune protein concentrations within corals indicate a suppressed immune system when corals bleached ( Mydlarz et al . , 2009 ) . Furthermore , immune-related host gene activity is suppressed for at least a year after bleaching occurs , at least for some species ( Pinzón et al . , 2015 ) . While some coral species harbor multiple Symbiodinium species in the same colony or over environmental gradients , others associate with only one Symbiodinium species ( LaJeunesse , 2002 ) . Symbiodinium species differ in their heat tolerance ( Berkelmans and van Oppen , 2006 ) , however , evidence of an association between Symbiodinium species identity and coral host disease susceptibility , is not well studied and equivocal ( Correa et al . , 2009; Rouzé et al . , 2016 ) . Acropora cervicornis can harbor several species of Symbiodinium ( Baums et al . , 2010 ) , but it is often dominated by Symbiodinium ‘fitti’ ( nominem nudum ) . No study has addressed whether different strains of the same Symbiodinium species influence infectious disease susceptibility in the coral host . Yet , different strains of a single Symbiodinium species can affect coral physiology ( Howells et al . , 2011 ) and thus we aimed to determine the influence of Symbiodinium strain diversity on coral disease susceptibility and bleaching in A . cervicornis . Coral nurseries provide a unique opportunity to test the effect of multiple stressors on coral survival and adaptation in a common garden environment . Nurseries propagate colonies via asexual fragmentation providing experimental replicates for each host genotype/S . ‘fitti’ combination . Host genotypes display differences in growth , linear extension , and thermal tolerance , which are maintained within the common garden environment ( Lohr and Patterson , 2017 ) . Here , we use 15 common garden-reared host genotypes infected with known S . ’ fitti’ strains . Genets were exposed to white-band disease homogenates under control conditions and then again after a period of elevated water temperatures . We measured the rate of infection and the performance of the symbiosis under control and treatment conditions to evaluate the hypothesis that infection resistance predicts bleaching resistance in the holobionts . The objectives of the present study were to ( 1 ) determine the relative abundance of genotypes of Acropora cervicornis from the lower Florida Keys that were resistant to disease , ( 2 ) characterize the Symbiodinium strains within each host and explore the potential relationship between the algal symbionts and disease susceptibility , and ( 3 ) quantify the relative change in disease risk when corals were bleached . The photochemical yield ( Fv/Fm ) of all fragments , prior to visual bleaching in August 2015 , averaged 0 . 457 ( ± . 015 SE ) . However , by September 2015 , colonies in the nursery had visually bleached after experiencing temperatures ~2°C above historical averages ( Figure 1 ) , represented by 8 degree heating weeks under NOAA’s Coral Reef Watch products . By this time , all of the corals had visibly turned white and the photochemical yield of the corals dropped to 0 . 148 ( ± 0 . 008 SE ) . While there was a gradual reduction in photochemical yield from August to September , fragments in the first three pre-bleaching trials were significantly higher than the post-bleaching trial ( Figure 2A , Supplementary file 1 , Figure 2—figure supplement 1 ) , as expected . In the August pre-bleaching trials , there were significant differences among the photochemical efficiency of S . fitti associated with different host genets ( X2 = 51 . 173 , df = 14 , p<0 . 001 , Figure 2A ) . Photochemical efficiency also differed among S . fitti associated with different host genets after bleaching ( X2 = 24 . 42 , df = 14 , p=0 . 04 ) , although the relative pattern among S . fitti associated with different host genets changed ( Figure 2A ) . A total of 25 out of the 75 fragments exposed to the disease homogenate showed signs of white-band-disease-associated mortality within the first seven days after exposure during the August , pre-bleaching trials . Only one fragment ( genet 46 ) , out of the 75 total fragments showed signs of disease in the control treatment , within the experimental period . There was high variation in disease susceptibility among genets , with susceptibility values ranging from 0% to 80% ( Figure 2B ) . Four genets showed complete resistance to the disease homogenate , with no replicate fragments showing any signs of tissue loss after disease exposure . The median susceptibility value among the different genets was 20% . Results differed when the same genets were exposed to the disease homogenate after they were bleached . A total of 55 out of 75 fragments lost tissue after exposure to the disease homogenate . Additionally , 13 out of the 75 control fragments died when bleached . Values ranged from 100% susceptibility to disease-induced mortality within five genets , to one genet that maintained disease tolerance , even when bleached ( genet 3; Figure 2B ) . Median susceptibility was 80% among the genets when the corals were bleached . The generalized linear model , which tested whether maximum quantum yield affected disease presence or absence for each replicate genotype indicated there was no significant effect of the photochemical yield on disease susceptibility , even within corals exposed to the disease homogenate , during either the pre-bleaching ( z = 0 . 132 , p=0 . 895 ) or post-bleaching experiments ( z = −1 . 579 , p=0 . 114 ) . Also , there was no effect of the average change in photochemical yield within each genotype on disease susceptibility ( pre-bleaching: z = 0 . 555 , p=0 . 579; post-bleaching: z = −0 . 023 , p=0 . 982 ) . The mixed-effect generalized linear model showed that the treatment effect was significant within both the pre-bleaching ( z = 2 . 263 , p=0 . 0234 ) and post bleaching trials ( z = 3 . 515 , p<0 . 001 ) , with higher disease presence within corals exposed to the disease homogenate . However , there were no significant differences detected among genotypes within trials ( pre-bleaching: z = 0 . 416 , p=0 . 677; post-bleaching: z = −0 . 243 , p=0 . 808 ) , nor was there a significant interaction between treatment and genotype within each trial ( pre-bleaching: z = 0 . 090 , p=0 . 928; post-bleaching: z = 0 . 697 , p=0 . 486 ) . There was also no difference in disease presence or absence among the three trials that created the pre-bleaching experiment ( z = −1 . 308 , p=0 . 191 ) , suggesting that pooling data from these three trials was appropriate . A total of six different S . fitti strains were found within the 15 coral host genets ( Supplementary file 2 ) . Note that no other Symbiodinium clades have been detected in the Mote in situ nursery A . cervicornis fragments above background levels ( Parkinson et al . , 2018a ) or in other offshore A . cervicornis colonies in the Keys ( Baums et al . , 2010 ) . A majority of the host genets tested ( 11/15 ) harbored a single S fitti strain consistently through time ( strain F421; Figure 2A ) . The other four Symbiodinium strains were associated with a single coral genet each ( See Supplementary file 2 ) . There was no significant difference in photochemical efficiency between corals harboring the common F421 Symbiodinium strain and those that harbored the other unique strains , pre- and post-bleaching ( pre-bleaching yield: t = −0 . 13 , df = 13 , p=0 . 99; post-bleaching yield: t = −0 . 244 , df = 13 , p=0 . 811 ) . Similarly , there was no influence of Symbiodinium strain on the amount of change in photochemical efficiency between the pre- and post-bleaching experiments ( change in yield: t = 0 . 172 , df = 13 , p=0 . 866 ) . There was no significant difference in disease susceptibility when corals contained the single Symbiodinium strain F421 compared with the other corals that hosted different Symbiodinium strains ( pre-bleaching disease: X2 = 0 . 039 , df = 1 , p=0 . 842 , post-bleaching disease: X2 = 0 . 079 , df = 1 , p=0 . 779; Figure 2—figure supplement 2 ) . This trend held through time ( Figure 2—figure supplement 3 ) . Among the eleven coral genets that harbored S . ‘fitti’ F421 , disease susceptibility ranged from 0% ( genet 3 ) to 70% ( genet 46 ) during pre-bleaching trials ( Figure 2B ) . S . fitti strain identity also did not influence disease susceptibility for the post bleaching exposures ( Figure 2B ) . Under pre-bleaching conditions , the tested A . cervicornis genets were almost three times as likely to experience disease-induced mortality when exposed to the disease homogenate over healthy homogenates ( median Bayesian relative risk = 2 . 77 , Figure 3A ) . The relative risk analysis also showed evident differences among genets , even though the frequenstist statistics did not indicate significant differences among genets within the generalized linear models . For example , genets 1 , 3 , 41 and 44 showed no increase in disease-induced mortality after disease exposure , with median relative risk values of these four genets near 1 , that is they were resistant . The 11 other genets , however , showed an increase in disease risk after exposure to the disease homogenate ( i . e . they were susceptible ) , with statistically significant increased relative risk values for genets 9 and 10 ( Figure 3A , Supplementary file 3 ) . Post-bleaching , the overall likelihood of disease-induced mortality increased by about three-fold ( median Bayesian relative risk = 3 . 33 , Figure 3B ) when the corals were bleached and exposed to the disease homogenate , compared with corals that were bleached and exposed to the healthy homogenate . Again , substantial variation among genets was detected , although there were now six genets that showed a significant increase in disease risk ( genets 5 , 41 , 44 , 46 , 47 , and 50; Figure 3B , Supplementary file 4 ) . Only one genet , genet 7 , was affected by disease pre-bleaching , but not post-bleaching , whereas genet 3 showed apparent complete immunity whether exposed to a disease homogenate when bleached or not . There were no differences detected in the bacterial community of the homogenates among trials ( F ( 1 , 7 ) =1 . 42 , p = 0 . 219 ) , nor did a single operational taxonomic unit significantly differ in relative abundance among trials ( 1 , 253 OTUs tested using nonparametric Kruskal Wallis tests ) . The PERMANOVA analysis also showed no statistical difference between the bacterial OTU community of the healthy homogenate and the disease homogenate ( F ( 1 , 6 ) = 1 . 962 , p = 0 . 134 ) , which was primarily because one healthy homogenate sample was similar to the disease homogenate samples ( Figure 4 ) . Comparisons of the relative abundances of the major bacterial classes showed no significant differences between the healthy and the disease homogenates using nonparametric tests . However , when the one outlier sample was removed there was a significantly higher abundance of Actinobacteria ( X2 = 4 . 5 , df = 1 , p = 0 . 034 ) and significantly lower abundance of Alphaproteobacteria ( X2 = 4 . 5 , df = 1 , p = 0 . 034 ) within the healthy samples compared with the diseased samples ( Figure 4—figure supplement 1 ) . Our results suggest that disease resistance and temperature tolerance evolve independently within A . cervicornis of the lower Florida Keys . Some genets showed significantly lower levels of bleaching compared with others , but had varying levels of disease susceptibility . These temperature tolerance and disease resistance traits were driven by the host genotype rather than strain variation in Symbiodinium ‘fitti’ , the dominant symbiont in the coral colonies ( Baums et al . , 2010; Parkinson et al . , 2018b ) . In the US Virgin Islands , bleached state , rather than temperature , influenced A . palmata susceptibility to disease; and bleaching resistance conferred disease resistance ( Muller et al . , 2008 ) . Within the present study , however , none of the Florida coral genets appeared to be resistant to bleaching , which may have contributed to high rates of disease risk after bleaching occurred . Variability in disease susceptibility was also reduced as several bleached genets with moderate to low levels of infectious disease susceptibility pre-bleaching showed high mortality levels when exposed to disease . The almost complete loss of white-band disease resistance after temperature-induced bleaching in A . cervicornis suggests that current adaptations to disease infection provide only limited protection against future colony mortality in light of rising ocean temperatures . Without knowing the primary pathogen of white-band disease , it is difficult to definitively differentiate the influence of the coral host state and the pathogenic dose between the pre and post-bleaching experiments . Although the physiological state of the host genotype may have contributed to the higher risk of disease when bleached , the potential pathogenic dose or virulence could have changed between the August ( pre-bleaching ) and September ( post-bleaching ) trials as well . Increased water temperatures can lead to higher growth rates of bacterial pathogens ( Remily and Richardson , 2006 ) and also lead to increased virulence ( Kushmaro et al . , 1998; Toren et al . , 1998; Harvell et al . , 2002; Ben-Haim et al . , 2003 ) . We did not detect a difference in the bacterial community of the homogenates among trials , only among treatments . However , without an identified primary pathogen , it was impossible to know whether pathogenic virulence could have influenced the results . Regardless of the mechanism , higher disease risk is evident when corals experience thermal anomalies and increased disease prevalence is likely as the world’s climate continues to warm . Disease resistance itself , was evident within both the pre and post-bleaching experiments . Prior to bleaching , four out of the 15 genets , or 27% , of the tested population showed complete resistance to disease exposure . In comparison , Panama and the USVI harbored only approximately 6% , of disease resistant genets ( Vollmer and Kline , 2008; Rogers and Muller , 2012 ) . Interestingly , two genets showed resistance after bleaching occurred , one of which was also disease resistant prior to bleaching ( genotype 3 ) . Disease resistance could be provided by a certain gene or set of genes within the host genome ( Libro and Vollmer , 2016 ) , a unique microbiome within the tissue or mucus of the disease resistant corals ( Gignoux-Wolfsohn et al . , 2017 ) , or could be influence by the energy reserves within these particular coral genotypes . For example , each genotype of A . cervicornis interacts with their environment eliciting differential growth rates , bleaching susceptibility and recovery from bleaching ( Drury et al . , 2017 ) . Subsequent studies will focus on identifying the mechanism driving disease resistance within the A . cervicornis corals used in the present experiments , with the recognition that a combination of these potential pathways is also possible . Resistance to disease in Acropora cervicornis populations from Panama may be driven by constitutive gene expression ( Libro and Vollmer , 2016 ) . Particularly , genes involved in RNA interference-mediated gene silencing are up-regulated in disease resistant corals , whereas heat shock proteins ( HSPs ) were down-regulated . Libro and Vollmer ( Libro and Vollmer , 2016 ) postulated that reduced HSPs in disease resistant corals from Panama may indicate high temperature resistance . In the Florida population studied here , however , three coral genotypes that were resistant to disease prior to bleaching showed a similar level of bleaching susceptibility to those that were susceptible to disease . This indicates that there was no obvious tradeoff or shared protection between disease resistance and temperature resistance for A . cervicornis within the lower Florida Keys . Gene flow among A . cervicornis populations is limited ( Baums et al . , 2010; Vollmer and Palumbi , 2007; Hemond and Vollmer , 2010 ) thus spatial variation in genetic trait architecture is possible and would complicate predictions of how corals may adapt to climate change ( Bay et al . , 2017 ) . A higher occurrence of disease resistance of Acropora within the Florida Keys ( 27% ) compared with populations tested in Panama and the USVI ( 6 and 8% respectively ) may be a result of more intense selection events within Florida compared with other locations in the Caribbean . However , there may also be methodological differences among studies that make direct comparisons challenging . Since nursery corals originate from fragments of opportunity within the wild population , the corals used in this study should represent a random subset of the wild population . The density and overall abundance of wild colonies of Acropora spp . within the lower Florida Keys has continued to decline ( Patterson et al . , 2002 ) . Significant direct anthropogenic impacts in the Florida Keys ( Lapointe et al . , 2004; Sutherland et al . , 2011 ) , disease outbreaks ( Aronson and Precht , 2001; Patterson et al . , 2002 ) , as well as several recent bleaching events ( Manzello , 2015; Lewis et al . , 2017 ) , may have fostered the persistence of only extremely hardy coral genets . The documentation of over a quarter of the tested population showing signs of disease resistance under non-bleaching conditions provides a glimmer of hope that natural evolutionary processes may allow for the persistence of a population in peril , such as Acropora cervicornis . Future work should concentrate on determining the degree of spatial variability among temperature and infectious disease resistance traits and their interactions . Disease susceptibility of A . cervicornis was more strongly linked to the coral host genet , rather than the algal symbiont strain . The high variability in disease susceptibility when host genets associated with strain F421 suggests that although Symbiodinium strain can influence phenotypic physiology of the host ( Grégoire et al . , 2017; Parkinson and Baums , 2014; Parkinson et al . , 2015 ) , white-band disease resistance is likely related to A . cervicornis host genotype . Furthermore , corals with the common F421 strain showed similar levels of disease susceptibility , both before and after bleaching , compared with corals that hosted all other strains . Future research should aim to evaluate the influence of additional Symbiodinium fitti strains on host disease resistance , to further evaluate the hypothesis that diversity of Symbiodinium strains within the population has little direct influence on holobiont disease susceptibility . Previous research showed Symbiodinium clade had no influence on disease susceptibility of several diseases infecting various coral species in the Atlantic and Caribbean region ( Correa et al . , 2009 ) . The present results suggest that Correa et al's conclusions may extend from the algal clade ( genus level ) to the Symbiodinium strain ( within species level ) . The dire state of coral populations , such as A . cervicornis , has forced a more interventionist approach to coral conservation because natural population recovery may not be possible on some reefs that are lacking sources of new recruits . Selective breeding of stress resistant host genotypes , experimental evolution of stress-resistant symbiont cultures , and gene therapy are now all being considered ( Mascarelli , 2014; van Oppen et al . , 2015; van Oppen et al . , 2017 ) . Design of effective breeding strategies for hosts and symbionts , however , requires knowledge of how genotypes respond to interacting stressors , not just temperature increases alone . Of particular interest , was the discovery of a genet that became disease tolerant when bleached ( genet 7 ) , although further testing of this genotype should occur to validate the results of the present study , which had limited replication . These results have important implications for selective breeding initiatives . For example , genet 7 would not have been a prime candidate for selective breeding based on disease resistance or bleaching susceptibility alone . Yet its apparent gain of disease resistance after bleaching makes it a potentially valuable genotype . If selective breeding initiatives focus on resistance to single stressors , interactive phenotypes , such as increased disease resistance under bleaching conditions , may be lost from the population . The consequences of these choices may be unpredictable and risky . In conclusion , under non-stressful conditions , disease resistance within the lower Florida Keys A . cervicornis population appears relatively prevalent compared to other regions in the Caribbean , perhaps because of many previous natural selection events over the last several decades . Disease outbreaks within Acropora spp . began in the late 1970 s and early 1980 s and continues to occur on contemporary reefs ( Aronson and Precht , 2001; Miller et al . , 2014; Rogers and Muller , 2012; Patterson et al . , 2002 ) . Resistance to high water temperature anomalies , however , appears decoupled from disease resistance as all genets appeared visibly bleached and showed significant loss of photochemical efficiency during the 2015 bleaching event . Historical records from 1870 to 2007 indicate that Florida , and most of the Caribbean , have had substantially longer return periods between potential bleaching events compared with temperature anomalies observed over the last decade , thus selection strength for thermal tolerance may have been small prior to recent years ( Thompson and van Woesik , 2009 ) . The bleaching event increased the risk of mortality from disease , whether it was from higher disease susceptibility or increased pathogenic load and/or virulence , and caused almost all previously resistant corals to become disease susceptible . Importantly , these results suggest that there is no tradeoff or shared protection between disease resistance and temperature tolerance within Acropora cervicornis of the lower Florida Keys . The present study shows that susceptibility to temperature stress creates an increased risk in disease-associated mortality , and only rare genets may maintain or gain infectious disease resistance under high temperature . We conclude that A . cervicornis populations in the lower Florida Keys harbor few existing genotypes that are resistant to both warming temperatures and infectious disease outbreaks and that recurring warming events may cause continued loss of disease resistant genotypes . A total of 10–12 replicate fragments ( ramets ) each from 15 genotypicially distinct host colonies ( genets ) , as determined via microsatellite genotyping ( see below ) , were collected from the Mote Marine Laboratory in situ coral nursery in August , 2015 . Number of replicates and genotypes were determined based on the maximum number available within the nursery for experimentation with considerations of additional spatial constraints within the wetlab area . Genets were originally collected from nearby reefs ( <20 km maximum linear distance ) and had been growing in the nursery for at least 5 years ( Supplementary file 5 ) . The small spatial scale over which genets were originally sourced suggests that these belong to the same population ( Baums et al . , 2010; Drury et al . , 2016 ) . Each ramet was cut from the donor colony using metal pliers and was approximately 5 cm in length . Ramets were transported in ambient seawater to Mote Marine Laboratory . Corals were mounted on PVC pipe plugs or glass slides using cyanoacrylate gel ( Bulk Reef Supply extra thick super glue gel ) . Permit restrictions meant that three disease exposure trials had to be conducted rather than testing all genotypes at once . The total number of genets varied for each trial , but each of the 15 genets was comparatively represented by the conclusion of the experiments . The sum of the three trials resulted in a total of 5–7 ramets per treatment ( disease vs control ) of each of the 15 different genets; a total of 170 corals ( see Supplementary file 6 for details ) . For each trial , ten aquaria were held within a single raceway , which contained a recirculating water bath kept at approximately 25°C . One ramet of one genet was placed within a 19 L glass aquarium that contained 9 . 5 L of seawater , thus each aquarium contained a single ramet of each genet for each trial with a maximum amount of 15 corals per tank . Water flow within the aquaria was maintained using 340 L per hour submersible powerheads . Temperature , salinity and pH were measured daily to ensure consistency among tanks . Corals were allowed to acclimate to tank conditions for 3 days prior to disease exposure . During the acclimation period the photochemical efficiency of the corals was measured using an Imaging Pulse Amplitude Modulation fluorometer ( IPAM Walz , Germany ) . Measurements were taken at least 1 hr after sunset . PAM fluorometry is a useful tool for quantifying the physiological parameters of the symbiotic algae found within scleractinian corals . Peak photochemical efficiency typical yields values between 0 . 5 and 0 . 7 , depending on the species , whereas reduced values indicate photochemical inhibition ( Fitt et al . , 2001 ) . Within the present study , photochemical efficiency ( Fv/Fm ) was used as a proxy for coral bleaching . Although Fv/Fm was not a direct measurement of bleaching , visual qualitative assessment showed that each genotype was regularly colored during the August trials ( Figure 1 ) . After the acclimation period , five randomly selected tanks were treated with a disease tissue homogenate , whereas the remaining five tanks were treated with a healthy tissue homogenate using a modified protocol developed by Vollmer and Kline ( Vollmer and Kline , 2008 ) . To create the disease homogenate , fragments of A . cervicornis showing signs of active white-band disease were collected from an offshore reef at approximately 7 . 6 m depth ( 24 . 54129° N , 81 . 44066° W ) . Live tissue from diseased fragments was removed by airbrushing off the tissue within 5 cm of the advancing band using 0 . 2 micron filter-sterilized seawater . To increase the likelihood that the disease homogenate contained active and viable pathogens , the homogenate from several different diseased corals was pooled into one sample . Surface area of diseased tissue acquired to create the slurry was approximately 10 cm2 per fragment , equating to ~11 cm2 of coral tissue per 100 ml of slurry . Approximately 100 ml of the disease homogenate was poured into each of the five treatment tanks . Acropora cervicornis fragments from the Mote Marine Laboratory in situ coral nursery were collected to create the healthy tissue homogenate; reducing impacts to the wild population of Acropora cervicornis . The healthy tissue of 11 fragments , all approximately 5 cm in length , was airbrushed using filter-sterilized seawater and collected in 50 ml plastic tubes . Surface area of each healthy fragment was approximately 10 cm2 , equating to ~11 cm2 of coral tissue per 100 ml of slurry . Approximately 100 ml of the healthy tissue homogenate was poured into each of the five control tanks . This procedure was repeated for each of the three August trials . All experimental corals were abraded near the base of the fragment prior to treatment using a sterile scalpel to increase the probability of disease infection . In September , 2015 another set of 10 ramets from the same 15 genets of Acropora cervicornis were collected from the Mote Marine Laboratory in situ coral nursery . By this time , the nursery corals had been experiencing anomalously high water temperatures reaching approximately ~2°C above historical averages , represented by 8 degree heating weeks under NOAA’s Coral Reef Watch products ( www . coralreefwatch . noaa . gov ) ( Strong et al . , 2011 ) . Corals were collected and mounted similarly to the August collection and allowed to acclimate for three days in tanks at 27 . 5°C . During that time the Fv/Fm of each coral was determined using the IPAM . Visual qualitative assessment showed that each genotype was completely white at the onset of the September trial ( Figure 1 ) . Fresh samples of diseased corals were collected from the same reef area as the August experiment , although from different colonies because the original diseased colonies had died . The healthy homogenate was again created from nursery corals that showed no apparent signs of tissue loss . Infectious dose between the pre- and post-bleaching trials was standardized in several ways even though the primary pathogen of white-band disease is unknown . First , the disease samples were collected from the same area ( ~100 × 100 m in reef area ) for each disease trial . Second , the area of diseased tissue used to create the slurry was standardized for the healthy and the disease slurries , and was also consistent among trials . Third , the tissue was collected within a standardized distance away from the disease margin to create the disease slurry . Fourth , disease samples collected in the field showed similar rates of tissue loss within the host colony . However , because the primary pathogen of white-band disease is still unknown , the dose of the infectious agent could not be determined within each slurry combination . This limits the ability to compare results between each experiment , but does not inhibit interpretation of the results within each experiment . Samples of each homogenate were processed for 16S rDNA in an effort to characterize the bacterial community of each trial ( see Materials and methods below ) . For both the August and September experiment , corals were monitored twice a day , in the morning and early evening hours , for seven days post treatment . Signs of disease mortality were recorded when observed and photographs were taken with a ruler . A new infection was defined as recently exposed skeleton caused by tissue sloughing off , often occurring from the base of the fragment and progressing towards the branch tip ( Figure 5 ) . Because the corals were already white during the post-bleaching experiment , the visual sign of mortality was determined by the apparent loss of tissue and simultaneous accumulation of algae on the coral skeleton . During bleaching , this often occurred over the entire fragment rather than a visual progression from the base to the tip . Mortality within the controls often showed the same signs of tissue loss for both the pre-bleaching and post-bleaching experiment . The number of ramets per genet that showed signs of disease mortality was used as the risk input within the relative risk analysis ( see below ) . In this regard , the proportion of ramets of a given genet that showed disease signs reflected the level of disease resistance for that genet . All data are fully available through the Biological and Chemical Oceanography Data Management Office , an open access data repository ( http://www . bco-dmo . org/dataset/642860 ) . We used the Two Sample Welch's T test or Kruskal Wallis tests to determine whether the photochemical efficiency of each coral genet changed between the pre and post-bleaching experiments , depending on the data set passing parametric assumptions . A Kruskal Wallis test with a Dunn's post hoc was used to test for differences in Fv/Fm among coral genets pre- and post-bleaching because the data were not normally distributed . A Two Sample Welch’s T test was used to determine whether corals with single or diverse strains of Symbiodinium fitti showed differing levels of photochemical efficiency pre- and post-bleaching or differed in the change of their photochemical efficiency through time . The Mann-Whitney-Wilcoxon Rank test was used to determine whether corals with single or diverse strains of S . fitti showed differing levels of disease susceptibility either pre or post-bleaching . A binomial generalized linear model within the ‘lme4’ package ( Bates et al . , 2015 ) was used to test whether the Fv/Fm or average change in Fv/Fm for each genotype between August and September , and the interaction of these factors with treatment , influenced the presence or absence of disease on each coral fragment . A binomial generalized mixed-effect linear model within the ‘lme4’ package ( Bates et al . , 2015 ) was used to test whether the fixed effects of treatment ( disease vs control homogenate ) , genotype , and the interaction of two variables significantly influenced the presence or absence of disease manifestation within each replicate fragment . The pre-bleaching and post-bleaching trials were analyzed separately . Trial was added to the pre-bleaching analysis to determine whether there were differences in disease susceptibility among the three trials in August . Tank was identified as the random effect within the model for both the pre- and post-bleaching experiments . A relative risk analysis compares the likelihood of an event occurring between two groups , individuals exposed to a risk factor versus individuals not exposed to a risk factor . Within an epidemiological setting , this analysis incorporates disease within non-exposed individuals thus accounting for chance occurrence . Traditionally , this analysis does not test for statistical significance , however , estimating the relative risk ratio within a Bayesian setting allows for statistical inference from interpretation of the posterior distribution and a comparison of results among genotypes . Within the present study , the relative risk of each genet was calculated as the number of ramets within each genet with disease after exposure to the risk ( disease homogenate ) divided by the number of ramets within each genet with disease that had not been exposed to the risk ( healthy homogenate ) : Relative risk ( RR ) = Risk in exposedRisk in non−exposed , where the risk in exposed individuals was calculated as the incidence ( diseased/total population ) of those exposed to the risk and the risk in non-exposed individuals was calculated as the incidence of those not exposed to the risk . When RR=1 then there is no association between the exposure and disease occurrence . However when RR>1 then there is a positive association and when RR<1 there is a negative association . The posterior distribution of the relative risk was calculated using a Bayesian approach ( Gelman et al . , 2004; Lawson , 2009 ) and estimated using a binomial likelihood distribution and a uniform-Beta prior distribution . To obtain an estimate of relative risk , Markov Chain Monte Carlo simulations were used with Gibbs sampling in OpenBUGS ( MRC Biostatistics Unit , Cambridge , UK , Supplementary file 7 ) . Ninety-five percent credible intervals were calculated for each estimate of relative risk . Credible intervals that did not include a value of one were considered significant , with a credible interval above one signifying a higher risk of disease because of exposure to the disease homogenate . A credible interval below one signified a higher risk of disease from the lack of exposure . Two different relative risk analyses were conducted using the present studies’ data set . The initial relative risk analysis compared the prevalence of disease signs for each genet when exposed to the disease homogenate with those that were exposed to the healthy homogenate prior to bleaching ( non-stressful conditions ) from the August 2015 data set . The second analysis compared the prevalence of disease signs for each genet when exposed to the disease homogenate with those that were exposed to the healthy homogenate under bleached conditions from the September 2015 data set . Host genotype was characterized using four host ( diploid ) microsatellite markers following ( Baums et al . , 2005 ) ; ( Supplementary file 8 ) . Previous work showed that the A . cervicornis corals within the nursery were dominated by Symbiodinium ‘fitti’; no other clades were detected above background level ( ca < 1% , 40 ) . Therefore , each host genet was also sampled twice in August 2015 to determine the multi-locus genotype of the dominant dinoflagellate species , S . ‘fitti’ , using 13 algal ( haploid ) microsatellite markers following ( Baums et al . , 2014 ) . Samples that returned identical multilocus algal genotypes at all loci were considered to belong to the same strain . Multilocus genotypes generated here were added to a database containing 1668 A . cervicornis genets and 345 Symbiodinium fitti strains from across the Caribbean . The probability of identity for the host is 10−4 ( calculated by GenAlEx 6 . 503 , 64 ) and for the symbiont is 10−5 ( calculated after 63 ) among all samples of the respective species in the database . The four disease and four healthy homogenate samples were processed for next generation sequencing analysis of the bacterial community . Total DNA was extracted from each homogenate sample using the MoBio Powersoil DNA isolation kit with an extended bead-beating time of one hour ( MoBio Inc . , Carlsbad , CA ) . The bacterial community of each sample was analyzed using 16S rDNA Illumina sequencing on the MiSeq platform ( see supplemental material for detailed protocol ) . Paired-end sequencing was performed at MR DNA ( www . mrdnalab . com , Shallowater , TX , USA ) using a single flow cell on a MiSeq following the manufacturer’s guidelines . Sequence length averaged 450 base pairs . Sequence data were processed using MR DNA analysis pipeline ( MR DNA , Shallowater , TX , USA ) . Sequences were joined and then depleted of barcodes . Sequences < 150 bp and sequences with ambiguous base calls were removed . Sequences were then denoised , operational taxonomic units ( OTUs ) were generated and chimeras were then removed . OTUs were defined by clustering at 3% divergence ( 97% similarity ) . Final OTUs were taxonomically classified using BLASTn against a curated database derived from NCBI ( www . ncbi . nlm . nih . gov ) and defined based on the homology identified in Supplementary file 9 . Illumina sequencing resulted in an average of 82 , 246 ( ± 11 , 393 SE ) sequence reads per sample and a total of 1257 distinct operational taxonomic units ( OTUs ) . The minimum number of reads within a sample was 39 , 810 and maximum reads reached 134 , 396 . To maintain comparability among samples throughout the statistical analyses , a random subset of the minimum value , 39 , 810 reads , was taken from each sample prior to statistical processing . The percent composition of bacterial groups from each sample was analyzed at the OTU level using a factorial permutation multivariate analysis of variance ( PERMANOVA ) with trial and homogenate type ( healthy or disease ) as two independent variables using the ‘vegan’ package of the statistical program R ( R Foundation for Statistical Computing , Vienna A , 2011; Oksanen , 2013 ) . A similarity percentages ( SIMPER ) analysis within the ‘vegan’ package ( Oksanen , 2013 ) provided the percent dissimilarity between the disease and healthy homogenate caused by each bacterial OTU . The relative abundance of each OTU and each bacterial class was tested for differences among homogenate types using a Kruskal Wallis test . Bacterial OTU data were then processed through non-metric multidimensional scaling ( nMDS ) , which applied the rank orders of data to represent the position of communities in multidimensional space using a reduced number of dimensions . The nMDS results were then plotted in two-dimensional ordination space . The average relative abundance of each bacterial class was also plotted for visualization . The sequencing data are available from GenBank within the National Center for Biotechnology Information ( http://www . ncbi . nlm . nih . gov ) under Accession numbers MG488295 – MG489819 for 16S rRNA gene Illumina sequencing .	The staghorn coral was once prevalent throughout the Florida Reef Tract . However , the last few decades have seen a substantial reduction in the coral population because of disease outbreaks and increasing ocean temperatures . The staghorn coral shows no evidence of natural recovery , and so has been the focus of restoration efforts throughout much of the Florida region . Why put the time and effort into growing corals that are unlikely to survive within environmental conditions that continue to deteriorate ? One reason is that the genetic make-up – the genotype – of some corals makes them more resilient to certain threats . However , there could be tradeoffs associated with these resilient traits . For example , a coral may be able to tolerate heat , but may easily succumb to disease . Previous studies have identified some staghorn coral genotypes that are resistant to an infection called white-band disease . The influence of high water temperatures on the ability of the coral to resist this disease was not known . There also remained the possibility that more varieties of coral might show similar disease resistance . To investigate Muller et al . conducted two experiments exposing staghorn coral genotypes to white-band diseased tissue before and during a coral bleaching event . Approximately 25% of the population of staghorn tested was resistant to white-band disease before the bleaching event . When the corals were exposed to white-band disease during bleaching , twice as much of the coral died . Two out of the 15 , or 13% , of the coral genotypes tested were resistant to the disease even while bleached . Additionally , the level of bleaching within the coral genotypes was not related to how easily they developed white-band disease , suggesting that there are no direct tradeoffs between heat tolerance and disease resistance . These results suggest that there are very hardy corals , created by nature , already in existence . Incorporating these traits thoughtfully into coral restoration plans may increase the likelihood of population-based recovery . The Florida Reef Tract is estimated to be worth over six billion dollars to the state economy , providing over 70 , 000 jobs and attracting millions of tourists into Florida each year . However , much of these ecosystem services will be lost if living coral is not restored within the reef tract . The results presented by Muller et al . emphasize the need for maintaining high genetic diversity while increasing resiliency when restoring coral . They also emphasize that disease resistant corals , even when bleached , already exist and may be an integral part of the recovery of Florida’s reef tract .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology" ]	2018	Bleaching causes loss of disease resistance within the threatened coral species Acropora cervicornis
Psychological theories of suicide suggest that certain traits may reduce aversion to physical threat and increase the probability of transitioning from suicidal ideation to action . Here , we investigated whether blunted sensitivity to bodily signals is associated with suicidal action by comparing individuals with a history of attempted suicide to a matched psychiatric reference sample without suicide attempts . We examined interoceptive processing across a panel of tasks: breath-hold challenge , cold-pressor challenge , and heartbeat perception during and outside of functional magnetic resonance imaging . Suicide attempters tolerated the breath-hold and cold-pressor challenges for significantly longer and displayed lower heartbeat perception accuracy than non-attempters . These differences were mirrored by reduced activation of the mid/posterior insula during attention to heartbeat sensations . Our findings suggest that suicide attempters exhibit an ‘interoceptive numbing’ characterized by increased tolerance for aversive sensations and decreased awareness of non-aversive sensations . We conclude that blunted interoception may be implicated in suicidal behavior . Suicide ranks among the leading causes of death worldwide ( World Health Organization , 2014 ) . In the US alone , suicide increased by nearly 30 percent between 2000 and 2016 ( World Health Organization , 2014; Hedegaard et al . , 2018 ) . For every death by suicide , it is estimated that there are 25 additional suicide attempts ( Hedegaard et al . , 2018 ) , each associated with significant social , emotional , and financial consequences . Experts have strived to understand and prevent death by suicide for decades , and yet , our current scientific grasp of the factors that contribute to suicidal behavior is lacking . Moreover , epidemiological data suggest that we are no better at preventing death by suicide than we were 100 years ago ( Hedegaard et al . , 2018; United States Department of Commerce Bureau of the Census , 1920; Kessler et al . , 2005; Nock et al . , 2008 ) , with suicide rates rising despite the application of prevention and intervention efforts ( Linehan , 2008; Nock , 2016; Paris , 2006 ) . Theoretical models of suicide have invoked the concept of ‘suicidal capacity’ to differentiate the small subset of individuals who attempt suicide from the much larger group of individuals who experience suicidal ideation but never resort to suicidal action ( Ribeiro and Joiner , 2009; Smith and Cukrowicz , 2010 ) . A basic tenet of this concept is the notion that most human beings are ‘hard-wired’ for survival and thus driven to avoid physical pain and threats to bodily homeostasis . Psychological theories of suicide suggest that in an individual with heightened suicidal capacity , certain dispositional ( Klonsky and May , 2015 ) and acquired ( Van Orden et al . , 2010; Van Orden et al . , 2008 ) traits result in a lower aversion to physical threat and a higher likelihood of transitioning from suicidal ideation to action . Consistent with this line of thinking , non-suicidal self-injury ( Klonsky and May , 2014; Franklin et al . , 2011 ) and high levels of fearlessness of the pain involved in dying ( May et al . , 2012 ) are behavioral and clinical factors that have been reported to predict suicide attempts . Thus it seems possible that suicidal behavior might be influenced by one’s ability to access and respond adaptively to homeostatic information regarding the internal state of the body , but few studies have directly investigated this topic . Interoception describes the nervous system’s process of sensing , interpreting , and integrating signals originating from inside the body ( Craig , 2002; Khalsa et al . , 2018 ) . Emerging evidence suggests that dysfunctions of interoception may contribute to certain mental illnesses ( Khalsa et al . , 2018; Khalsa and Lapidus , 2016 ) , including mood and anxiety disorders ( Paulus and Stein , 2010; Avery et al . , 2014; Barrett et al . , 2016; Wiebking et al . , 2015; Harshaw , 2015; Domschke et al . , 2010 ) , substance use disorders ( Paulus and Stewart , 2014; Verdejo-Garcia et al . , 2012 ) , eating disorders ( Kerr et al . , 2016; Berner et al . , 2018; Khalsa et al . , 2015 ) , and nonsuicidal self-injury ( Muehlenkamp , 2012 ) , all of which are associated with an elevated risk of suicide ( Nock et al . , 2010; Harris and Barraclough , 1997; Smith et al . , 2018a ) . Interoception is thought to be substantially supported by the insular cortex , with the primary representation of visceral sensations occurring in the mid-to-posterior insula , and the integration of interoceptive information with cognition , emotion , and other higher order processes occurring in more anterior regions ( Barrett and Simmons , 2015; Critchley and Harrison , 2013; Hassanpour et al . , 2018 ) . To test the hypothesis that abnormalities of interoception are associated with suicidal capacity in individuals with psychiatric disorders , we evaluated interoceptive processing in participants with a history of suicide attempts as compared to a matched psychiatric reference sample of participants with no history of suicide attempts . We measured aversive interoceptive processing across the respiratory and nociceptive domains , via an inspiratory breath-hold challenge and a cold-pressor challenge . We assessed cardiac interoception during a heartbeat perception task as well as during a functional magnetic resonance imaging ( fMRI ) task involving focused attention to heartbeat sensations . We predicted that relative to non-attempters , suicide attempters would 1 ) tolerate aversive interoceptive sensations to a greater extent , 2 ) demonstrate lower interoceptive accuracy , and 3 ) exhibit differences in brain activity in the insular cortex when attending to interoceptive sensations . We found that both participant groups were well-matched in terms of demographic and clinical characteristics , showing no significant differences in age , BMI , sex , diagnosis , or levels of self-reported depression , anxiety , substance use , or eating disorder symptoms ( Table 1 ) . We noticed that the groups showed a significant difference in their usage of psychotropic medication , with a greater proportion of the suicide attempters reporting taking such medications . We provide further details regarding our participants , including psychiatric diagnoses , use of psychotropic medications , missing data values , and scores on self-report measures in Appendix 1 . We found that suicide attempters held their breath for significantly longer than non-attempters , approximately 10 s longer on average across both trials ( F ( 1 , 121 . 84 ) = 4 . 48 , p = 0 . 036 , R2 = 0 . 042 ) ( see Figure 1 ) . We also observed a repetition effect , such that all participants held their breath longer during the second trial ( F ( 1 , 97 . 01 ) = 20 . 18 , p < 0 . 001 , R2 = 0 . 173 ) , replicating previous results with this task ( Willem Van der Does , 1997 ) . We did not find a significant interaction between group and trial . We report a summary of the Linear Mixed Effects ( LME ) output for the model examining breath-hold duration , including fixed effects estimates and standardized regression coefficients in Supplementary file 1 . Concordant with the increased breath-hold duration in suicide attempters , we found that suicide attempters had higher concentrations of exhaled carbon dioxide ( CO2 ) than non-attempters after the breath-hold trials ( F ( 1 , 120 . 33 ) = 5 . 52 , p < 0 . 001 , R2 = 0 . 043 ) . However , we did not find an effect of trial or interaction between group and trial . We also found that suicide attempters had lower concentrations of exhaled oxygen ( O2 ) following the breath-hold trials relative to non-attempters ( F ( 1 , 132 . 27 ) = 5 . 00 , p = 0 . 027 , R2 = 0 . 036 ) . We observed a significant main effect of trial ( F ( 1 , 91 . 09 ) = 6 . 16 , p = 0 . 015 , R2 = 0 . 020 ) , such that reductions in O2 were greater after the second breath-hold across both groups . We report summaries of the LME outputs for the O2 and CO2 models , including fixed effects estimates and standardized regression coefficients in Supplementary file 1 . Despite the prolonged breath-hold duration and elevations in CO2 , we found that suicide attempters did not report any differences in perceived breathlessness ( p = 0 . 70 ) , feelings of suffocation ( p = 0 . 95 ) , fear of suffocation ( p = 0 . 97 ) , urge to breathe ( p = 0 . 76 ) , breathing sensation intensity ( p = 0 . 53 ) , unpleasantness ( p = 0 . 63 ) , task difficulty ( p = 0 . 48 ) , or effort expended during the breath-hold ( p = 0 . 27 ) relative to non-attempters ( Figure 1—figure supplement 1 ) . We found that the cold-pressor challenge elicited increased pain ratings over time in both groups ( F ( 3 , 276 . 16 ) = 86 . 78 , p < 0 . 001 , R2 = 0 . 589 ) . However , this effect was qualified by a significant interaction between timepoint and group ( F ( 3 , 277 . 39 ) = 2 . 89 , p = 0 . 036 , R2 = 0 . 030 ) . On closer examination of the LME fixed effects , we observed that suicide attempters kept their hands submerged in the cold water for significantly longer than non-attempters after reaching their peak pain level ( t ( 278 . 56 ) = 2 . 78 , p = 0 . 006 , β = 0 . 13 ) , without any significant differences in the amount of time taken to reach mild , moderate , and peak pain levels ( Figure 2 ) . Overall , suicide attempters kept their hands submerged in the icy water for approximately 18 s longer than the non-attempters . We report fixed effects and model summary values in Supplementary file 2 . Additionally , although suicide attempters provided slightly lower average ratings of unpleasantness , pain , difficulty , and stress than non-attempters , these differences were not statistically significant ( unpleasantness: U = 1107 , p = 0 . 117 , FDR-p = 0 . 144 , r = 0 . 16; pain: U = 1067 , p = 0 . 122 , FDR-p = 0 . 144 , r = 0 . 15; difficulty: U = 1123 , p = 0 . 090 , FDR-p = 0 . 144 , r = 0 . 17; stress: U = 1095 , p = 0 . 144 , FDR-p = 0 . 144 , r = 0 . 15 ) . Our initial LME model examining heartbeat perception accuracy as a function of group , condition ( i . e . guess , no-guess , and perturbation ) , and their interaction , showed a significant effect of condition ( F ( 2 , 195 . 08 ) = 12 . 72 , p < 0 . 001 , R2 = 0 . 200 ) . However , there was no significant effect of group and no group by condition interaction . By examining the fixed effects we noticed that , relative to guessing trials , accuracies on the no-guess ( t ( 195 . 08 ) = −4 . 61 , p < 0 . 001 , β = −0 . 33 ) and breath-hold perturbation trials ( t ( 195 . 07 ) = −4 . 013 , p < 0 . 001 , β = −0 . 29 ) were significantly lower . We made a post-hoc decision to apply a second model that omitted the guessing score from the analysis , based on a recent study indicating that heartbeat perception accuracy scores are potentially confounded by guessing ( Desmedt et al . , 2018 ) . For the second heartbeat perception model , we examined accuracy as a function of group and condition across the no-guess and breath-hold perturbation trials only ( i . e . after omitting the ‘guess’ trial ) , and found a significant difference between groups ( F ( 1 , 97 . 04 ) = 8 . 64 , p = 0 . 004 , R2 = 0 . 048 ) ( Figure 3 ) and a significant interaction between group and trial ( F ( 1 , 144 . 47 ) = 4 . 37 , p = 0 . 04 , R2 = 0 . 043 ) . In particular , we found that suicide attempters exhibited lower heartbeat perception accuracy during the no-guess condition relative to non-attempters , t ( 144 . 46 ) = −2 . 94 , p = 0 . 003 , β = −0 . 29 ) , and that the difference in accuracy between attempters and non-attempters was attenuated during the perturbation trial ( t ( 97 . 73 ) = 2 . 09 , p = 0 . 04 , β = 0 . 14 ) . We did not observe group differences in ratings of task confidence or difficulty across the no-guess and perturbation trials ( Confidence: U = 1153 . 5 , p = 0 . 453 , FDR-p = 0 . 67 , r = 0 . 08; Difficulty: U = 1014 . 5 , p = 0 . 766 , FDR-p = 0 . 767 , r = 0 . 03 ) . Suicide attempters displayed a tendency to rate their heartbeat sensations as less intense ( U = 1345 , p = 0 . 028 ) , although this was non-significant after applying a Benjamini-Hochberg correction across contrasts ( FDR-p = 0 . 084 , r = 0 . 22 ) ( Figure 3 ) . We report fixed effects and model summary values in Supplementary file 3 . We also found that , relative to non-attempters , suicide attempters exhibited reduced BOLD activation in the right dorsal mid-insula and right posterior insula during interoceptive attention to the heartbeat versus the exteroceptive attention condition ( p < 0 . 005 , corrected at α <0 . 05; Figure 4 ) . There was also a cluster of reduced BOLD activation within the left dorsal mid-insula , but this did not survive correction . The whole-brain analysis revealed four additional clusters with significantly reduced BOLD activation during attention to heart sensations among suicide attempters: one cluster within the right precuneus , one within the right superior temporal gyrus , one within the right posterior cingulate cortex , and one within the right dorsomedial prefrontal cortex ( p < 0 . 0005 , ACF corrected at α < 0 . 05; Table 2 ) . We report exploratory correlations across all behavioral and neuroimaging variables in Appendix 1—figure 1 . We conducted additional analyses to examine potential confounding effects of medication status on our primary interoception variables of interest , due to the statistically significant difference observed in the proportion of individuals taking psychotropic medications in each group . Our results remained largely unchanged after accounting for medication status , as detailed in Appendix 2 . We did not initially account for the role of suicidal ideation in the current study , focusing instead on interoceptive processing differences between individuals with a history of suicide attempts within the last 5 years and individuals with no suicide attempt history . To rectify this issue , we conducted additional analyses examining whether suicidal ideation history might explain the observed abnormalities of interoception across subjective , behavioral , and neural levels . Our observation of diminished interoception in suicide attempters was largely unchanged after accounting for lifetime suicidal ideation intensity , as detailed in Appendix 2 and displayed in Appendix 2—figure 1 . Although our study represents the most comprehensive investigation related to interoception and suicide to date , we must acknowledge certain limitations . We evaluated evidence for interoceptive processing focusing on individuals with a history of suicide attempts within the last 5 years , based on the report that self-reported interoceptive deficits may be greater among individuals with more recent suicide attempts ( Forrest et al . , 2015 ) . An alternative approach for future research might be to compare performance on neural and behavioral constructs related to interoception in individuals with more recent suicide attempts . Another limitation is that , while our findings suggest that individuals with suicide attempts exhibit abnormal interoception , we did not fully examine whether a history of suicidal ideation—versus a suicide attempt—has an independent impact on interoception . To begin to address this point , we conducted additional analyses which suggested that the observation of diminished interoception in suicide attempters was largely unchanged after accounting for lifetime suicidal ideation . But prospective studies are needed to conclusively discern whether the relationship between interoception and suicide attempt history can be attributed to group differences in suicidal ideation . Additionally , after matching our suicide attempter and non-attempter samples on measures of psychopathology , we found that the proportion of participants taking psychotropic medications at the time of data collection was significantly greater in the suicide attempters . Accounting for these differences in subsequent analyses did not substantially affect our results . One possibility is that the greater psychotropic medication usage in this group might reflect an effort by clinicians to reduce further suicide attempts . From our cross-sectional study , it is difficult to judge whether the observed differences in interoception represent predispositions ( i . e . innate characteristics ) , whether they reflect an emerging response at some point during the development of suicidal ideation , or occur as a response to suicidal behavior . Addressing these questions via longitudinal task-based assessments of interoception and/or pain processing would provide valuable insight into the impact of blunted interoception on the emergence of suicidal ideation and the conversion to suicidal behavior ( Millner et al . , 2017 ) . Although not the primary intent of our investigation , we observed several interrelationships within and across levels of analysis raising the possibility of a latent interoceptive awareness trait factor . However , these relationships were inconsistent and were not pre-specified in our hypotheses . Identification of such a latent factor would likely require additional investigation using larger samples and inclusion of individuals not meeting criteria for psychiatric disorders . We also used an imprecise , albeit commonly employed , measure of pain perception in the cold-pressor challenge . It would be advantageous to clarify whether suicidal action is differentially linked to impaired processing of other pain signals . Examples include visceral pain , which tends to be poorly localized , often referred to somatic structures and produces strong autonomic and affective responses , as well as other somatic pain signals ( e . g . thermal or mechanical pain ) , which tend to be discretely localized to somatic structures and produce more variable autonomic and affective responses ( Sikandar and Dickenson , 2012 ) . Beyond stimulating visceral and somatic pain processing via different neuroanatomical pathways , it would be helpful to evaluate the degree to which altered pain responding is directly driven by differences in nociception per se as opposed to indirectly modulated by differences in interoceptive processing ( Pollatos et al . , 2012 ) . Additionally , it is increasingly understood that cardiac interoception is rather difficult to assess ( Khalsa and Lapidus , 2016 ) . Heartbeat perception tasks such as the one employed in the current study are widely used , but have been the subject of criticism ( Desmedt et al . , 2018 ) , and there is evidence to suggest that performance on this type of task can be influenced by one’s a priori knowledge about their heartbeat ( Murphy et al . , 2018 ) . We addressed some of these potential confounds in our heartbeat perception task by including a no-guess trial condition , and a trial in which an inspiratory perturbation was used to putatively increase the intensity of heartbeat sensations . We also conducted analyses with and without the inclusion of the guessing trial . Overall , it appeared that suicide attempters had lower heartbeat perception accuracy across all three trials , with the strongest differences occurring during the least confounded condition ( i . e . the no-guess trial ) . We did not investigate cardiac interoception using a more rigorous and ecologically valid form of perturbation , such as double-blinded infusions of isoproterenol ( Khalsa et al . , 2009 ) , but would expect that blunted interoception in suicide attempters in a similar context would constitute robust evidence replicating the present findings . Lastly , we should note that none of the interoceptive tasks applied in this study have demonstrated sufficient reliability to be considered appropriate for implementation in prognostic assessments of suicidality in clinical settings . We find that suicide attempters exhibit evidence of ‘interoceptive numbing’ characterized by increased tolerance for aversive respiratory and nociceptive sensations , reduced awareness of the heartbeat , and blunted activity in the dorsal mid and posterior insular cortex , a region of the brain associated with the primary representation of visceral afferent signals . The presence of these specific interoceptive deficits among individuals with prior suicide attempts reveals a possible role of interoceptive dysfunction in distinguishing individuals at risk of suicide . We performed a retrospective analysis from a pre-existing dataset containing the first 500 participants of the Tulsa-1000 ( T-1000 ) cohort , a naturalistic longitudinal study of 1000 individuals with mood , anxiety , substance use , and/or eating disorders ( Victor et al . , 2018 ) . Participants were considered eligible for T-1000 study entry if they fulfilled any of the following symptom criteria: Patient Health Questionnaire ( PHQ-9; Kroenke et al . , 2001 ) ≥10 and/or Overall Anxiety Severity and Impairment Scale ( OASIS; Campbell-Sills et al . , 2009 ) ≥8 , and/or Drug Abuse Screening Test ( DAST-10; McCabe et al . , 2006 ) score >3 , and/or Eating Disorder Screen ( SCOFF; Morgan et al . , 2000 ) score ≥2 . Please refer to Victor et al . ( 2018 ) for a detailed description of the T-1000 inclusion criteria and study procedures . All participants provided written informed consent and received financial compensation for their involvement , and all procedures were approved by the Western Institutional Review Board . Participants were included in the suicide attempter group ( n = 34 ) if they endorsed making a suicide attempt at any point during the previous five years as documented in the Columbia Suicide Severity Rating Scale ( CSSRS; Posner et al . , 2011 ) and/or life-chart interviews ( Aupperle et al . , 2020 ) , which were conducted during the baseline data collection period , and used to gather information about each participant’s lifetime psychosocial , medical , educational , occupational , and treatment history ( Victor et al . , 2018 ) . We used a propensity score matching algorithm for psychiatric reference sample identification ( MatchIt package in R De et al . , 2011 , 1:2 nearest neighbor method without replacement ) , resulting in a group of non-attempter participants who denied having ever made a suicide attempt , and who exhibited similar screening symptoms on the PHQ-9 , SCOFF , DAST , and OASIS scales ( Table 1 ) . To maximize the amount of data available for analysis , non-attempter participants ( N = 239 ) were only matched to suicide attempters if their data had been manually checked and they had complete observations of the behavioral and psychophysiological variables examined . Further information regarding participant inclusion and exclusion criteria , matching procedures , and suicide attempt method are provided in Appendix 1 . For a detailed description of general study procedures , please see Appendix 1 . We conducted analyses of demographic , clinical , behavioral , and physiological data using the R base statistical software package version 3 . 5 . 1 ( R Development Core Team , 2013 ) . The ‘TableOne’ package ( version 0 . 9 . 3; Yoshida et al . , 2019 ) was used to display summaries of clinical characteristics between groups . LME analyses were conducted using the ‘lmerTest’ package version 3 . 1 . 1 ( Kuznetsova et al . , 2017 ) . A marginal ANOVA was used on each LME model to examine F-tests for interactions and main effects . In the event of significant interactions , the summaries of LME fixed effects were examined to clarify which factors were driving the effect . The Kenward-Roger approximation of degrees of freedom was used for all LME analyses . R-squared estimates for fixed-effects were computed using the ‘r2glmm’ package in R ( Jaeger , 2017 ) as described in Edwards et al . ( 2008 ) . Tables depicting model output were generated using the ‘sjPlot’ package ( version 2 . 6 . 2; Lüdecke , 2018 ) and figures were created using the ‘ggplot2’ package ( version 3 . 0 . 0; Wickham , 2011 ) . VAS ratings for each task were also compared between groups . Since a proportion of the VAS ratings were not normally distributed , Mann-Whitney tests , which are robust to deviations from normality , were used to compare ratings between groups . Where applicable , a Benjamini-Hochberg correction was applied to minimize the false discovery rate ( FDR ) associated with repeated testing . We provide specific details for the analysis of each task below . The source code for our primary analyses and figures has also been provided .	The human brain closely monitors body signals essential for our survival , including our heartbeat , our breathing and even the temperature of our skin . This mostly unconscious process is called interoception . It helps people perceive potential or actual threats and helps them to respond appropriately . For example , a person charged by a wild animal will act instinctively to run , fight or freeze . Unlike most creatures , humans show an ability to counteract these survival instincts , and are capable of intentionally engaging in behaviors that result in physical harm . Recent increases in the rate of suicide have made it more urgent to try to understand what leads to this behavior in humans . Now , DeVille et al . show that people with psychiatric disorders who have survived a suicide attempt have blunted interoception . In four experiments , people with a history of suicide attempts were compared to another group of individuals without a history of suicide attempts . The groups were carefully matched such that there were no significant differences in the demographic and clinical characteristics of the two groups , including in terms of their age , sex , body mass index and psychiatric symptoms . Both groups completed uncomfortable tasks like holding their breath or keeping their hand in icy cold water . The participants also completed two tasks that required them to focus on their own heartbeat , one of which was paired with functional magnetic resonance imaging . Those with a history of suicide attempts held their breath and kept their hand in cold water for longer , and also were less in tune with their heart rate . This “interoceptive numbing” was associated with less activity in part of the brain called the insular cortex . These differences could not be explained by the individuals having a psychiatric disorder or a history of considering suicide , or by them taking psychiatric medications . Instead , the interoceptive numbing was most often seen in individuals who made an attempt on their own life . The experiments identify physical characteristics that may differentiate people who attempt suicide from those who do not . This lays the groundwork for future research aimed at identifying biological indicators of suicide risk . More studies are needed to verify the results . If the results are verified , the next step would be prospective studies to determine whether measuring interoception can help clinicians predict who is at risk of a suicide attempt . If it does , it might give clinicians a new tool to try to prevent suicide by ensuring those at greatest risk receive appropriate care .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2020	Diminished responses to bodily threat and blunted interoception in suicide attempters
We present a reanalysis of the stochastic model of organelle production and show that the equilibrium distributions for the organelle numbers predicted by this model can be readily calculated in three different scenarios . These three distributions can be identified as standard distributions , and the corresponding exact formulae for their mean and variance can therefore be used in further analysis . This removes the need to rely on stochastic simulations or approximate formulae ( derived using the fluctuation dissipation theorem ) . These calculations allow for further analysis of the predictions of the model . On the basis of this we question the extent to which the model can be used to conclude that peroxisome biogenesis is dominated by de novo production when Saccharomyces cerevisiae cells are grown on glucose medium . Recently a model was presented in which the variation of numbers of a particular type of organelle ( Golgi apparatus , vacuoles or peroxisomes ) observed in cells was proposed as a diagnostic indicator of the relative importance of different processes by which organelles can be formed and destroyed ( Mukherji and O’Shea , 2014; see Mukherji and O’Shea , 2015 for a correction ) . Here we re-examine the mathematical analysis of this model and show that further insight can be gained from considering exact calculations of the equilibrium distributions . For conciseness we will refer to the model , and the analysis in the associated paper ( Mukherji and O’Shea , 2014 ) , by the abbreviation SMOP ( stochastic model of organelle production ) . In the SMOP model , four processes are envisaged for the production and destruction of organelles: de novo synthesis , fission , fusion and decay . These four processes are characterised by one rate constant each , defined in the SMOP paper as kde novo , kfission , kfusion , and γ . Following the definitions in the SMOP paper , the probabilities of each of the four processes occurring in the next small time period δt are given in Table 1 . We also include in this table the total rate of each process that would be observed instantaneously in a large population of N cells . 10 . 7554/eLife . 10167 . 003Table 1 . Definition of terms in the SMOP modelDOI: http://dx . doi . org/10 . 7554/eLife . 10167 . 003ProcessProbability of process occurring in next δt in a particular cell containing n organelles1Process changes n byRate of process in sample of N cells2De novokde novoδt1kde novofnNFissionkfissionnδt1kfissionnfnNDecayγnδt-1γnfnNFusionkfusionn ( n-1 ) δt-1kfusionn ( n-1 ) fnN1For example , if kfission = 0 . 02 then the probability of a cell with n = 2 organelles undergoing a fission event in the next 0 . 1 time units = 0 . 02x2x0 . 1 = 0 . 004 . 2fn is the fraction of cells having n organelles . For example , if 23% of the cells have 2 organelles then f2 = 0 . 23 . If , in a population of 1000 cells , f2 = 0 . 23 and kfission = 0 . 02 then the rate of cells changing from having n = 2 to n = 3 organelles at any one moment due to fission would be 0 . 02x2x0 . 23x1000 = 9 . 2 cells per time unit . Models involving processes of this type are generically termed “birth and death” processes and have a very long history of analysis in the context both of the life sciences ( e . g . evolution; Yule , 1924 ) and in the context of physical processes ( e . g . detection of cosmic rays; Furry , 1937 ) . Accessible discussions can be found in several books ( Bailey , 1990 , which is a reissue of the classic text from 1964; Taylor and Karlin , 1998 ) . In such analyses , the three processes of de novo production , production by fission , and loss by first order decay are often termed immigration , birth and death , respectively . Immigration is used for a process that increases the number of individuals but does not require any other individuals already to be present . Birth is the process by which one individual gives rise to a second individual . Death is a process by which a particular individual is lost from a population with a probability that is independent of any other members of the population . Analyses including a fusion term are much less common . As there are a considerable number of possible combinations of the four processes that might be active , we will use a notation here to define a model by listing in curly brackets the active production processes , followed by the active destruction processes , separated by a semi-colon . Any process that is not mentioned has a rate constant of zero . Thus the model with de novo , fission and decay terms would be denoted {de novo , fission; decay} . During single cell simulations based upon the equations in Table 1 the number of organelles will fluctuate ( Figure 1A ) and one can ask what fraction of time , fn , does a cell spend having n = 0 , n = 1 , n = 2 , etc . organelles . This is equivalent to asking what fraction of a large ensemble of cells have n = 0 , n = 1 , n = 2 , etc . organelles at one moment in time . In treatments of stochastic systems , the values of fn would normally be described as the probability distribution for a cell having n organelles . In terms of a population of cells it can be described as a population distribution . 10 . 7554/eLife . 10167 . 004Figure 1 . The concept of the limiting distribution in a stochastic system . ( A ) The traces show simulations run with the parameters {kde novo = 2 . 0 , kfission = 0 . 9; γ = 1 . 0 , kfusion = 0 . 02} , starting from n = 0 ( black trace ) and n = 50 ( green trace ) . Both simulations “settle down” to stochastic fluctuations about a mean value of <n> = 7 . 1 ( B ) Schematic representation of a set of cells that are all initialised to n = 1 at time t = 0 , and are observed at a time t = τ . ( C ) Distributions calculated with the same parameters as in ( A ) for a set of 1000 cells as in ( B ) , calculated for τ = 0 . 2 ( magenta ) , 0 . 4 ( yellow ) , 1 . 0 ( green ) , 2 . 0 ( blue ) , 5 . 0 ( black ) , 10 . 0 ( cyan ) , 15 . 0 ( red ) . The curves for τ = 10 . 0 and τ = 15 . 0 become very similar as they approach the limiting distribution . These two curves match closely the result ( filled black circles ) of applying the recurrence relation ( Equation 2; Appendix 2 ) . ( D ) The red trace is a time course for parameters {kde novo = 2 . 0 , kfission = 1 . 1; γ = 1 . 0 , kfusion = 0} . Since kfission > γ , and kfusion = 0 , then n diverges; in such a case there is no limiting distribution . DOI: http://dx . doi . org/10 . 7554/eLife . 10167 . 004 The simulations in Figure 1A illustrate the important point that a simulation is always started from some arbitrary starting point , and that a period of time must elapse before the simulations can be considered to be independent of this starting point . If the distribution fn is evaluated at different times after the starting point of simulations ( Figure 1C ) then different distributions are obtained; hence the distribution is “time dependent” . As the probability ( or population ) distribution varies in time ( and since the system can in principle be started from any state ) then it is not strictly possible to talk of “the distribution” for a stochastic system . However in many birth and death models the system will settle down to a limiting distribution , independent of the starting states of the cell ( s ) as in the cyan and red curves in Figure 1C . Such a situation corresponds to a state of dynamic equilibrium that is familiar from chemical kinetics . Thus the terms limiting , equilibrium ( or steady state ) distributions can be used in this context interchangeably . The conditions for the models studied here to have limiting distributions is discussed further below; an example of a set of parameters for which a SMOP model does not yield a limiting distribution is shown in Figure 1D . Assuming that a limiting probability distribution does exist there are two basic sampling methods by which it can be measured , irrespective of whether one is making experimental observations or performing simulations . One method is to note n at a set of time points for a single cell ( such as points drawn from the trajectories in Figure 1A ) , and the other is to take a large number of cells at one point in time , and measure n across this ensemble of cells ( Figure 1B , C ) . The former would require a time dependent set of observations that may be difficult to obtain experimentally . The latter approach is equivalent to making experimental observations on a large field of view of cells , or of making repeated simulations . However there is a large caveat that when the measurements are made one must be convinced that the cells have had “long enough to reach equilibrium” since the last significant perturbation to the system . If this is not the case then the distribution measured will be contaminated with contributions from non-equilibrium distributions ( such as from the magenta , yellow , green and blue curves in Figure 1C ) . Perturbations include the choice of an arbitrary starting point in simulations , and effects such as cell division and change of growth conditions in experimental data . It is implicitly assumed in the SMOP paper that the populations are to be considered to be at equilibrium ( or that the probability distributions are in their limiting form ) for all three cases of the analysis via simulations , from experimental data , or via the fluctuation dissipation theorem; a comment has been added to the original articles to clarify this assumption for the simulations ( see the comment dated November 23 , 2015 on Mukherji and O'Shea , 2014 ) . For the experimental data this seems a reasonable assumption , although dynamic population data are really required to fully settle this issue . The fluctuation dissipation theorem method implicitly assumes a steady state ( Paulsson , 2005 ) . By applying an equilibrium condition it is straightforward to derive precise relations for the distributions in the three scenarios considered in the original SMOP paper , and hence avoid the approximations introduced by the use of the fluctuation dissipation theorem . At equilibrium , the rate at which the population gains cells with n + 1 organelles due to cells with n organelles gaining one organelle must equal the rate at which the cells with n + 1 organelles lose one organelle . The reasoning is the same as for standard treatments of dynamic equilibrium between two states ( as in a chemical reaction ) , and the complete justification of this when there are multiple states ( i . e . cells with n = 0 , n = 1 , n = 2 , etc . , organelles ) is given in Appendix 1 . Thus at equilibrium ( 1 ) ( kde novo + kfissionn ) fnN = ( γ + kfusionn ) ( n + 1 ) fn+1N which gives ( 2 ) fn+1 = ( kde novo + kfissionn ) ( γ + kfusionn ) n + 1fn From Equation 2 the exact distribution of organelle numbers at equilibrium can be calculated for a model involving any combination of the four processes , without recourse to random number based simulations and the attendant issues of ensuring adequate sampling precision . An explicit numerical example of the use of Equation 2 to generate a distribution is given in Appendix 2 . Briefly , an arbitrary value for f0 is chosen; f1 is then calculated from f0; f2 is calculated from f1; f3 is calculated from f2; etc . Finally the entire distribution is normalised , which removes any dependence on the initial choice for f0 . Equation 2 is often termed a recurrence relation ( or sometimes recursion relation or difference equation ) as it allows successive terms in a distribution to be calculated from earlier terms . The recurrence relation readily allows the derivation of precise distributions for the case of the model applied to Golgi ( {de novo; decay} , Appendix 3 ) and vacuoles ( {fission; fusion} , Appendix 4 ) . For the Golgi , a Poisson distribution is obtained as the limiting distribution in accord with the SMOP analysis . However for vacuoles a truncated Poisson distribution is obtained , and not the shifted Poisson distribution that is reported in the SMOP analysis . Although the difference between these distributions is quite subtle ( Appendix 4 ) , the variation of Fano factor with <n> is significantly different: the Fano factor for the truncated Poisson approaches 1 much more rapidly ( Figure 2 , green curve ) than for the shifted Poisson ( Figure 2 , black curve ) . 10 . 7554/eLife . 10167 . 005Figure 2 . Comparison of reported Fano factors for vacuole populations , compared to two different theoretical expectations . Three data points quoted in the SMOP paper are plotted: □ Haploid ( glucose ) ; ○ Diploid ( glucose ) ; ∆ Haploid ( oleate ) . The solid black curve is the expectation from the shifted Poisson distribution ( which is the incorrect distribution given the {fission; fusion} model ) and the solid green curve is the expectation from the truncated Poisson distribution ( which is the correct distribution given the model ) . The dashed line is for a Fano factor of 1 . The solid curves were constructed by calculating a family of distributions and evaluating the mean and Fano factor . DOI: http://dx . doi . org/10 . 7554/eLife . 10167 . 005 In Figure 2 , it can be seen that the experimental values quoted in the SMOP analysis are in excellent agreement with the incorrect prediction , whilst the agreement with the corrected theoretical prediction is much less good . This greatly weakens the argument that the SMOP model makes “quantitatively accurate predictions” or that it therefore correctly accounts for the behaviour of the vacuole population . Having established the value of analysing the SMOP model with the recurrence method , we move on to the case of peroxisomes . Our motivation for this in depth analysis of the SMOP method was sparked by the claim that it could differentiate between fission and fusion dominated mechanisms of peroxisome biogenesis . That the distribution of numbers of organelles can give a clue to the mechanisms by which organelles are formed is a very elegant idea , and the SMOP analysis combines this idea with a very simple kinetic model . As we explored the system further we realised that the fluctuation dissipation theorem result was not necessary for analysis of the system , and that enforcing the equilibrium condition , that was implicit in the work already , greatly simplified the analysis . There are a number of factors that cause us to question the utility of the SMOP model . A main piece of evidence for the correctness of the model was the agreement of the experimentally observed Fano factors for the vacuole data with those from the model . We have shown this agreement to be much less perfect than originally demonstrated . We have also shown that there is a strong interplay between different parameters in the model . This means that the agreement of experimental data with the model is not as compelling as originally presented and that the interpretation of experimental observations back to mechanistic conclusions is open to question . We hope that our analysis will stimulate discussion as to whether , for instance , the SMOP model captures the key features of the underlying processes and is just lacking some details; or whether the model fundamentally lacks key aspects of feedback . The assumption that observations of cells grown in batch culture faithfully report equilibrium distributions also requires further verification . A key conclusion of the SMOP analysis is that the contribution of fission to peroxisome biogenesis is negligible ( <10% ) when yeast cells are grown on glucose , but "dominant" when they are grown on oleate . This is an area of some contention ( Hoepfner et al , 2005; Motley and Hettema , 2007 ) , and a recent model relied on fission of peroxisomes during organelle inheritance as the proliferation mechanism ( Knoblach et al , 2013 ) . Our analysis has shown that the term “dominant” is misleading , and that the data reported for haploid cells grown on oleate indicates approximately equal contributions from the two processes . Nevertheless the model does suggest that the proportion of production by fission increases by about a factor of five on switching to oleate growth . Supporting evidence for this was provided by the observation of the reduction in the inferred fission contribution in cells grown on oleate in which the fission factors Vps1 or Dnm1 ( or Fis1 , an accessory factor of Dnm1 ) were deleted . On the other hand , no data were shown for glucose grown cells harbouring the same deletions . On glucose the Fano factor is reported in the SMOP analysis as 1 . 1 , with <n> = 3 , and from Equations 3 , 4 one obtains kde novo = 2 . 7γ and kfission = 0 . 1γ . The model then implies that on deletion of the fission pathway ( i . e . setting kfission = 0 ) then <n> would only drop by 10% . Peroxisome count data has been reported recently ( Fig S4 , Motley et al . , 2015 ) , with values of <n> = 4 . 9 in WT cells , <n> = 1 . 5 in vps1Δ cells and <n> = 1 . 2 in vps1Δdnm1Δ cells . The drops in peroxisome numbers in vps1Δ and vps1Δdnm1Δ cells are much greater than the 10% estimated above from the SMOP model . There is also peroxisome count data in Kuravi et al . ( 2006 ) , which gives <n> = 1 . 6 ( WT ) , <n> = 1 . 2 ( vps1Δ ) , <n> = 1 . 7 ( dnm1Δ ) , <n> = 0 . 9 ( dnm1Δvps1Δ ) . The drop in <n> for the dnm1Δvps1Δ again conflicts with the idea that fission is such a small contributor to the biogenesis process . The discrepancies between these various data possibly arise from difficulties in quantifying peroxisome numbers , especially when cells contain a large number of small ( and therefore low fluorescence ) peroxisomes as may be the case when fission is a strong contributor to biogenesis . The problem of peroxisome counts depending on the brightness of fluorescent markers has been commented on by Jung et al . ( 2010 ) . There is a continuing push for cell biology to become more quantitative , and to be subject to the use of rigorous models as are common in the physical sciences . Such a push raises significant challenges not only in terms of developing tractable models and justifying the underlying assumptions , but also in terms of the application of the model to complex experimental data . In particular this work highlights the need for greater accounting for detection limits and intensity distributions , as well of time dependent issues , in the reporting and analysis of organelle count data , if these are to be used to infer details of organelle biogenesis mechanisms . All calculations were performed in python 2 . 7 , running either via Cygwin under Windows 8 . 1 , or Linux Mint 17 . 0 . The code used for running stochastic simulations and for calculating distributions via Equation 2 is given in Source code 1 , 2 . Time units are arbitrary . The time step for the simulations in Figure 1 was 0 . 0001 units . Equation 2 can be reformulated in terms of three parameters , e . g . kde novo/γ , kfission/γ , kfusion/γ; thus for example the parameters {kde novo = 2 . 0 , kfission = 0 . 9; γ = 1 . 0 , kfusion = 0 . 02} and any uniformly scaled set of parameters ( e . g . {kde novo = 4 . 0 , kfission = 1 . 8; γ = 2 . 0 , kfusion = 0 . 04} ) yield the same limiting distribution . As in the SMOP paper , the Fano factor is defined here as σ2n .	Any cell that has a nucleus also contains a number of subcellular structures called organelles . The number of organelles inside a cell increases when new organelles are made from scratch ( a process known as de novo synthesis ) , or when an existing organelle divides to produce two organelles in a process called fission . And the number of organelles decreases when an existing organelle decays , or when two organelles fuse together to become one organelle . The actual number of organelles of a particular type inside a cell results from a balance between these creative and destructive processes . Last year researchers at Harvard University developed a model that treats the processes of organelle creation and destruction as if they were chemical reactions , and then used their model to make predictions about the budding yeast S . cerevisiae in three scenarios . The Harvard researchers had to use a number of approximations to make these predictions . Now Jeremy Craven has derived exact solutions to the model for these three scenarios . The exact solutions call into question some aspects of the model , notably the prediction that the production of new peroxisomes – organelles that are involved in breaking down fatty acids and other compounds – is dominated by fission when the yeast cells are grown on a substance called oleate , and by de novo synthesis when they are grown on glucose . Craven's analysis also highlights the need for quantitative time-course imaging data to test theoretical models of dynamic processes in cells .	[ "Abstract", "Introduction", "Analysis", "Discussion", "Materials", "and", "methods" ]	[ "short", "report", "cell", "biology", "computational", "and", "systems", "biology" ]	2016	Evaluation of predictions of the stochastic model of organelle production based on exact distributions
The rate of protein synthesis in the adult heart is one of the lowest in mammalian tissues , but it increases substantially in response to stress and hypertrophic stimuli through largely obscure mechanisms . Here , we demonstrate that regulated expression of cytosolic poly ( A ) -binding protein 1 ( PABPC1 ) modulates protein synthetic capacity of the mammalian heart . We uncover a poly ( A ) tail-based regulatory mechanism that dynamically controls PABPC1 protein synthesis in cardiomyocytes and thereby titrates cellular translation in response to developmental and hypertrophic cues . Our findings identify PABPC1 as a direct regulator of cardiac hypertrophy and define a new paradigm of gene regulation in the heart , where controlled changes in poly ( A ) tail length influence mRNA translation . Cellular growth and function depend heavily on protein synthesis , which is often considered a constitutive activity for a cell . However , it is becoming clear that global protein synthesis rates are not always static , that they vary widely among cell types , and that these differences are necessary for normal tissue development and homeostasis ( Buszczak et al . , 2014 ) . Particularly , the rate of protein synthesis in adult heart is one of the lowest amongst different tissues but increases markedly in response to stress and hypertrophic stimuli ( Garlick et al . , 1980; Lewis et al . , 1984 ) . The molecular basis for these historical observations , however , is still poorly understood . Translation initiation is the rate-limiting step in protein synthesis ( Aitken and Lorsch , 2012; Hinnebusch et al . , 2016; Sonenberg and Hinnebusch , 2009 ) . Interactions between the 5’ m7GpppN cap structure , the pre-initiation factors ( including eIF4A , eIF4E , and eIF4G ) , and poly ( A ) -binding protein C1 ( PABPC1 ) form a stable , looped mRNP complex ( Amrani et al . , 2008; Gallie , 1991; Park et al . , 2011; Safaee et al . , 2012; Tarun and Sachs , 1996; Wells et al . , 1998 ) that stimulates translation while safeguarding the mRNA from exonucleases ( Coller et al . , 1998; Gray et al . , 2000; Kahvejian et al . , 2005; Lewis et al . , 2017; Zekri et al . , 2013 ) . Based on these central roles , PABPC1 is thought to be ubiquitously expressed and serve ‘house-keeping’ roles in protein synthesis . Here , we report that PABPC1 protein expression is post-transcriptionally silenced in adult human and mouse hearts through shortening of its mRNA poly ( A ) tail , which results in reduced polysome association and translation of Pabpc1 transcripts . The developmental silencing of PABPC1 is cardiomyocyte-specific and reversible . We show that Pabpc1 poly ( A ) tail length and protein expression are restored during adult-onset cardiac hypertrophy stimulated by endurance exercise or heart disease . Furthermore , we demonstrate that PABPC1 re-expression and its interaction with eIF4G are necessary and sufficient to globally stimulate translation and physiologic growth of cardiomyocytes . These findings reveal a novel , poly ( A ) tail-based regulatory mechanism in the heart that dynamically controls PABPC1 expression and subsequent protein synthesis in response to developmental and hypertrophic signals . The association of eIF4F complex with the 5’ m7G cap structure is stabilized through eIF4G-PABPC1 interactions , which promote ribosomal recruitment and translation initiation ( Amrani et al . , 2008; Gallie , 1991; Safaee et al . , 2012; Tarun and Sachs , 1996; Wells et al . , 1998 ) . We have discovered that PABPC1 protein levels in the adult mouse heart are drastically lower relative to the embryonic day ( E ) 17 stage ( Figure 1A , B ) . Parallel examination of Pabpc1 mRNA abundance unexpectedly showed only a modest decrease after birth ( Figure 1B ) . A similarly striking reduction in PABPC1 protein , but not mRNA levels , was observed in adult versus fetal human hearts indicating PABPC1 silencing is post-transcriptional and evolutionarily conserved ( Figure 1C , D ) . We inspected PABPC1 mRNA and protein abundance in several other mouse and human tissues and determined that the postnatal silencing of PABPC1 is muscle-specific ( Figure 1A and Figure 1—figure supplement 1A–D ) . Coimmunofluorescent staining of PABPC1 with a cardiomyocyte marker ( TNNT2 ) combined with immunoblot analyses of purified cell types revealed that while PABPC1 is abundantly expressed in all cells within the neonatal heart , it is selectively silenced in adult cardiomyocytes ( Figure 1E , J and Figure 1—figure supplements 1E and 2 ) . 10 . 7554/eLife . 24139 . 003Figure 1 . PABPC1 is dynamically regulated during cardiac development and hypertrophy . ( A–D ) Relative quantification of PABPC1 protein ( immunoblots ) and mRNA ( qPCR ) levels normalized to GAPDH during mouse heart and liver development ( A , B ) and in human fetal and adult hearts ( C , D ) . ( E–J ) Immunofluorescent images of mouse postnatal day 0 ( P0 ) and 8-week-old adult hearts stained for PABPC1 ( red ) , cardiac troponinT ( green ) , and DAPI ( blue ) . Insets G and J show cardiomyocytes , while F and I show non-cardiomyocytes . Immunoblots and quantification of PABPC1 protein and mRNA from wild-type mouse hearts 8 weeks after ( K ) thoracic aortic constriction ( TAC ) or ( L ) 10-week exercise training . Data are mean ± s . d ( n = 3 ) ; p<0 . 05 , unpaired two-tailed t-test . NS , not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 24139 . 00310 . 7554/eLife . 24139 . 004Figure 1—figure supplement 1 . Post-transcriptional silencing of PABPC1 is muscle-specific . ( A ) Immunoblots for PABPC1 from postnatal day 0 ( P0 ) and 8-week-old adult mice from the indicated tissue with GAPDH as a loading control . ( B ) Quantification of PABPC1 protein and mRNA expression ( qPCR ) . Data are mean ± s . d ( n = 3 ) ; p<0 . 005 unpaired two-tailed t-test . ( C ) PABPC1 immuno-histochemistry images based on anti-PABPC1 antibody ( Abcam ab21060 ) from the Human Protein Atlas Database ( www . proteinatlas . org ) showing that PABPC1 is abundantly expressed in other adult tissues but not heart and skeletal muscle . ( D ) RPKM values from the Human Protein Atlas Database showing relative Pabpc1 mRNA levels in various human tissues . ( E ) Western blot of purified P0 and adult cardiomyocytes ( CMs ) and cardiac fibroblasts ( CFs ) . Blotting for Desmin ( DES , CM marker ) and Vimentin ( VIM , CF marker ) show clean separation of cell types . DOI: http://dx . doi . org/10 . 7554/eLife . 24139 . 00410 . 7554/eLife . 24139 . 005Figure 1—figure supplement 2 . Post-transcriptional silencing of PABPC1 is muscle-specific . Single-channel immunofluorescent images of mouse postnatal day 0 ( P0 ) and 8-week-old adult hearts stained for PABPC1 ( red ) , cardiac troponinT ( green ) , and DAPI ( blue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24139 . 005 We sought to determine whether PABPC1 protein is re-expressed in the adult heart during hypertrophy , a condition accompanied by increased size of individual cardiomyocytes , enhanced protein synthesis , and induction of many fetal genes ( Hill and Olson , 2008; Maillet et al . , 2013; Towbin and Bowles , 2002 ) . We used thoracic aortic constriction ( TAC ) and endurance exercise training in mice as models of pathologic and physiologic hypertrophy , respectively ( De Lisio and Parise , 2012; Kalsotra et al . , 2014 ) . PABPC1 protein levels were markedly induced under both conditions without any change in mRNA abundance ( Figure 1K , L ) . These results illustrate that post-transcriptional silencing of PABPC1 in the adult heart is reversed during cardiac hypertrophy . To probe whether the disappearance of PABPC1 protein in the adult heart is due to a reduction of synthesis , we conducted polysome profiling of E18 and adult mouse hearts ( Figure 2A ) . Quantitative PCR ( qPCR ) analyses of fractionated lysates showed a significant shift of Pabpc1 mRNAs away from polyribosomes to the mRNP/monosome fractions in the adult hearts , demonstrating reduced accessibility to the translational machinery in contrast to the E18 hearts . Control Gapdh mRNA remained associated with polyribosomes at both developmental stages ( Figure 2B ) . 10 . 7554/eLife . 24139 . 006Figure 2 . Poly ( A ) tail length determines cell-type and developmental stage-specific translation of PABPC1 . ( A ) . Polysome profile of embryonic day 18 ( E18 ) and adult mouse hearts . ( B ) Percentage of Pabpc1 and Gapdh mRNAs measured by qPCR in each fraction collected from the polysome profiling . ( C ) Neonatal and 8-week-old adult wild-type mice were pulsed with puromycin through an intraperitoneal injection . Forty-five minutes following injection , heart and liver tissues were harvested for immunoblotting with anti-puromycin antibody . De novo protein synthesis was quantified as the ratio of puromycin labeled peptides to total protein . ( D ) Fractional distribution of Pabpc1 mRNAs with short and long poly ( A ) tails in whole heart , C2C12 cells , cardiomyocytes ( CMs ) , cardiac fibroblasts ( CFs ) , whole heart after TAC surgery , and whole heart after exercise ( measured by qPCR following poly ( A ) tail fractionation ) . ( E ) Poly ( A ) tail length status of Pabpc1 mRNA within P0 and adult heart RNP , monosome , and polysome fractions from sucrose gradients . ( F ) Pabpc1 single-molecule RNA-FISH in C2C12 myoblasts and myotubes . Data are mean ± s . d ( n = 3 ) ; p<0 . 05 , p<0 . 005 unpaired two-tailed t-test; NS , not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 24139 . 00610 . 7554/eLife . 24139 . 007Figure 2—figure supplement 1 . Regulation of PABPC1 expression in adult heart is independent of miRNAs or alternative splicing . Targeted deletion of Dicer in adult cardiomyocytes was obtained by treating 8-week-old Dicer f/f; MCM mice with tamoxifen ( 20 mg/Kg/day ) for 5 consecutive days21 . Forty-eight hours after the last injection , heart tissues were harvested . ( A ) Relative PABPC1 protein ( immunoblots ) or ( B ) mRNA ( qPCR ) levels were measured in the indicated samples . Fold change in PABPC1 protein was determined by relative quantification of band intensities , normalized to GAPDH ( shown below the blots ) . Data are mean ± s . d ( n = 3 ) ; unpaired two-tailed t-test . NS , not significant . ( C ) UCSC genome browser tracks of Pabpc1 mRNA in cardiomyocytes and fibroblasts from neonatal and adult mice16 . DOI: http://dx . doi . org/10 . 7554/eLife . 24139 . 00710 . 7554/eLife . 24139 . 008Figure 2—figure supplement 2 . Limited influence for 5’ and 3’ untranslated regions ( UTRs ) of Pabpc1 on luciferase protein translation during C2C12 differentiation . ( A ) Representative immunoblot demonstrating a steady decrease in PABPC1 protein levels during C2C12 differentiation; p38 was used as a loading control . ( B ) Quantification of PABPC1 protein ( immunoblots ) and mRNA ( qPCR ) levels relative to p38 during C2C12 differentiation . ( C ) Schematic of the reporters designed to assess the effect of Pabpc1 5’ and 3’ UTRs on luciferase protein expression . ( D ) Relative luciferase activity derived from C2C12 cells transfected with different reporter constructs shows that only Pabpc1 5’ UTR has a moderate effect ( twofold ) on the reporter activity in differentiated myotubes , whereas the 3’ UTR and the combination of UTRs have no effects . Data are mean ± s . d ( n = 3 ) ; p<0 . 05 , unpaired two-tailed t-test . NS , not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 24139 . 00810 . 7554/eLife . 24139 . 009Figure 2—figure supplement 3 . Experimental design of northern blot and RNA isolation based on poly ( A ) tail length through gradient purification . ( A ) Oligo design for the Pabpc1 northern assay . Oligos were designed for the Pabpc1 mRNA to be targeted in an RNAseH digestion leaving 700nt downstream of the cleavage site and the native poly ( A ) tail . ( B ) Northern blot using a radiolabeled probe against Pabpc1 from E18 and 8-week-old adult mouse hearts following RNAseH cleavage . In the final lane , oligo dT was also added to the RNAseH reaction so that the Pabpc1 transcript was cleaved at both ends , leaving just the 700nt region . A similar RNAseH cleavage assay was used for GAPDH as a control . ( C ) Adult mouse heart total RNAs with and without RNAse T1 treatment were mixed with biotinylated oligo ( dT ) and bound to the streptavidin-conjugated beads . Different salt concentrations were used to elute mRNAs . The differing poly ( A ) tail lengths of eluted RNAs were checked by northern blot analysis using an oligo-dT40 probe . This served as optimizations for further purifications of short and long poly ( A ) tail RNAs that were subjected to qPCR analysis using gene-specific primers . ( D ) Poly ( A ) tail status of P0 and Adult Gapdh in polysome gradient fractions as a control experiment . Gapdh is enriched in the long-tailed and polysome fractions as expected . DOI: http://dx . doi . org/10 . 7554/eLife . 24139 . 009 Intriguingly , in comparison to E18 , the adult heart polysome profiles exhibited a decrease in the number of polyribosomes with a corresponding increase of 80S monosomes , which reflects reduced translation efficiency ( Figure 2A ) . To further quantify global protein synthesis rates in vivo , we pulse-labeled wild-type neonates and adult mice with puromycin followed by immunoblotting using an anti-puromycin antibody ( i . e . SUnSET assays ) ( Goodman et al . , 2011; Schmidt et al . , 2009 ) . We detected a robust decrease in puromycin-labeled peptides in adult versus neonatal hearts but not in liver tissues ( Figure 2C ) . Lower protein synthetic capacity of adult striated muscle compared to other tissues was first observed more than three decades ago ( Garlick et al . , 1980; Lewis et al . , 1984 ) . Our data showing muscle-specific silencing of PABPC1 thus offers a plausible molecular basis for these historical observations . Low-level protein synthesis in the absence of PABPC1 could presumably result from alternative mRNA circularization mechanisms ( Bukhari et al . , 2016; Lin et al . , 2016; Wang et al . , 2015 ) or translation of some mRNAs in a closed loop-independent manner ( Archer et al . , 2015; Costello et al . , 2015 ) . Next , we investigated the molecular mechanisms responsible for inefficient translation of PABPC1 . It appears that suppressed PABPC1 translation in the adult heart is microRNA-independent , as we saw no change in PABPC1 protein or mRNA in tamoxifen-inducible , heart-specific adult dicer knockouts ( Figure 2—figure supplement 1A , B ) , which are defective in microRNA processing ( Kalsotra et al . , 2010 ) . We analyzed available RNA-sequencing data from neonatal and adult mouse cardiomyocytes and fibroblasts ( Giudice et al . , 2014 ) and found no evidence for alternative splicing or a developmental change in Pabpc1 5’- or 3’- untranslated regions ( UTRs ) in either cell types ( Figure 2—figure supplement 1C ) . We reasoned that trans-acting factor ( s ) might bind to the sequence elements within Pabpc1 UTRs to suppress its translation . To test this hypothesis , we constructed luciferase reporters fused to 5’- , 3’- or both mouse Pabpc1 UTRs and transfected them into C2C12 myoblasts . C2C12 cells exhibit marked downregulation of endogenous PABPC1 protein ( ~10 fold ) but not mRNA levels when differentiated into myotubes ( Figure 2—figure supplement 2A , B ) . Amongst the different reporters , only Pabpc1 5’-UTR showed a modest ( <2-fold ) decrease in luciferase ( Rluc/Fluc ) activity during myoblast-to-myotube differentiation ( Figure 2—figure supplement 2C , D ) . Although PABPC1 binding to an A-rich element within its 5′-UTR could auto-regulate its translation ( de Melo Neto et al . , 1995; Kini et al . , 2016 ) , the relatively mild effects of 5’-UTR in reporter assays along with decreasing PABPC1 protein levels during muscle development argue against auto-regulatory feedback as the primary mechanism inhibiting Pabpc1 translation . Besides UTRs , poly ( A ) tail length is another major determinant for protein synthesis; mRNAs containing longer tails being more stable and translated more efficiently ( Eichhorn et al . , 2016; Lim et al . , 2016; Weill et al . , 2012 ) . Therefore , we tested whether Pabpc1 mRNAs are deadenylated in the adult heart . Both RNaseH cleavage assay and qPCR of long- and short-tailed mRNAs ( fractionated by affinity chromatography ) demonstrated significant Pabpc1 deadenylation in adult versus E18 hearts and in myotubes versus myoblasts ( Figure 2C and Figure 2—figure supplement 3A–C ) . Pabpc1 poly ( A ) tail length in E18 hearts was estimated to be ~150 nucleotides ( nts ) , whereas it was reduced to ~20 nts in adult hearts ( Figure 2—figure supplement 3B ) . Gapdh poly ( A ) tail length was measured as a control and showed no difference between E18 and adult hearts . Importantly , shortening of Pabpc1 poly ( A ) tail in the adult heart was cardiomyocyte-specific; and partly reversed after TAC or endurance exercise ( Figure 2C ) . In addition , Pabpc1 poly ( A ) tail length was strongly correlated to its association with monosomes and polysomes . At P0 , Pabpc1 mRNAs were almost exclusively long-tailed and in polysomes , whereas adult Pabpc1 fractionated majorly into short-tailed and monosome fractions ( Figure 2E ) . Gapdh poly ( A ) tail length remained unchanged and primarily associated with polysomes at both developmental stages ( Figure 2—figure supplement 3D ) . We further explored if inhibition of Pabpc1 translation could be due to nuclear retention of Pabpc1 transcripts . Single-molecule RNA FISH , however , showed primarily cytoplasmic staining without any noticeable difference in Pabpc1 mRNA localization between myoblasts and myotubes indicating poly ( A ) tail status does not impact Pabpc1 nucleo-cytoplasmic export in these cells ( Figure 2F ) . Together , these results provide compelling evidence that Pabpc1 poly ( A ) tail length is dynamically regulated during cardiac development and hypertrophy , and that poly ( A ) tail shortening limits Pabpc1 mRNA translation in the adult heart . Because PABPC1 is upregulated in hypertrophy , we tested whether it is required for stimulus-induced growth of cardiomyocytes . PABPC1-depleted neonatal mouse cardiomyocytes were viable but resistant to isoproterenol ( Iso ) or triiodothyronine ( T3 ) -induced hypertrophy ( Figure 3A–I ) . Metabolic labeling of cardiomyocytes with an alkyne-modified glycine analog , L-homopropargylglycine ( HPG ) , to measure newly synthesized proteins revealed that PABPC1 knockdown inhibited the normal surge in protein synthesis rate evoked by Iso or T3 stimulation ( Figure 3J , K ) . Furthermore , we found that PABPC1 deficiency blocked protein but not mRNA upregulation of the hypertrophic markers Acta1 , Myh7 and Anp ( Figure 3L , M ) indicating that while new protein synthesis is impaired , the transcriptional response to hypertrophic stimuli is still functional in these cells ( Kim et al . , 2008; Sergeeva and Christoffels , 2013 ) . These results demonstrate that PABPC1 depletion does not affect the transcriptional regulatory circuits or mRNA stability of these hypertrophy markers . 10 . 7554/eLife . 24139 . 010Figure 3 . Knockdown of PABPC1 in neonatal mouse cardiomyocytes prevents stimulus-induced hypertrophy and protein synthesis . ( A–F ) Primary cardiomyocytes isolated from newborn mice were transfected with siRNA against control Luciferase or Pabpc1 . Twelve hours following transfection , cells were treated with isoproterenol ( Iso ) or triiodothyronine ( T3 ) for 36 hr to induce hypertrophy . Cells were stained for Desmin and Actin by immunofluorescence to verify cardiomyocyte identity and measure cell area . ( G , H ) Immunoblots demonstrating efficient PABPC1 knockdown 48 hr after siRNA treatments with either Iso or T3 . ( I ) Quantification of cell area 36 hr post Iso or T3 treatments . ( J , K ) Measurement of new protein synthesis using Click-iT homopropargylglycine assay after 2 hr of Iso or T3 treatments . ( L ) Quantification of mRNA ( qPCR ) from neonatal cardiomyocytes for each condition shows significant upregulation of mRNA for Acta1 , Myh7 , and Anp in response to Iso or T3 treatments . ( M ) Representative immunoblot showing that protein levels of ACTA1 , MYH7 , and ANP are increased after Iso or T3 treatments in the control Luciferase knockdown but synthesis is prevented when Pabpc1 is knocked down . Data are mean ± s . d ( n = 3 ) ; p<0 . 05 , p<0 . 01 , one-way analysis of variance ( ANOVA ) plus Dunnett’s post-hoc test . NS , not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 24139 . 01010 . 7554/eLife . 24139 . 011Figure 3—source data 1 . Source data for cell area of cultured neonatal cardiomyocytes treated with siRNA and either Iso or T3 . DOI: http://dx . doi . org/10 . 7554/eLife . 24139 . 011 PABPC1 contains four RNA recognition motifs ( RRM1-4 ) , a linker region and a C-terminal MLLE domain ( Gray et al . , 2000 ) . While all four RRMs are capable of binding to poly ( A ) RNA , RRM2 preferentially interacts with eIF4G to stimulate cap-dependent translation ( Safaee et al . , 2012 ) . Hence , we investigated if PABPC1 interactions with eIF4G are necessary for triggering cardiomyocyte hypertrophy and new protein synthesis . The M161A mutation was previously shown to disrupt PABPC1 interactions with eIF4G ( Kahvejian et al . , 2005 ) . However , we decided to use the recently available structural information ( Safaee et al . , 2012 ) to engineer additional mutations within the RRM2 domain of PABPC1 to completely abolish its interactions with eIF4G ( Figure 4—figure supplement 1A ) . We confirmed that although PABPC1mRRM2 fails to interact with eIF4G1 in coimmunoprecipitation and in vitro binding assays , it binds to poly ( A ) RNA with similar affinities as wild-type PABPC1 ( Figure 4—figure supplement 1B–F ) . Next , we carried out rescue experiments wherein endogenous PABPC1 in cardiomyocytes was silenced using a 3’-UTR-targeted siRNA followed by adenoviral transduction of GFP , siRNA-resistant wild type , or PABPC1mRRM2 cDNAs ( Figure 4—figure supplement 1G ) . Supplementing PABPC1-deficient cardiomyocytes with wild type PABPC1 re-sensitized the hypertrophic growth response to Iso ( Figure 4A–M ) , reversed the block in new protein synthesis ( Figure 4N–P ) and restored translation of hypertrophic markers in these cells ( Figure 4Q ) . PABPC1mRRM2 , however , failed to rescue any of these phenotypes ( Figure 4A–Q ) underscoring that PABPC1–eIF4G interactions are essential to evoke cardiomyocyte hypertrophy and stimulate translation in response to hypertrophic signals . 10 . 7554/eLife . 24139 . 012Figure 4 . PABPC1–eIF4G1 interactions control stimulus-induced new protein synthesis and hypertrophy . ( A–L ) Representative images of neonatal cardiomyocytes infected with adenovirus expressing GFP , wildtype PABPC1 , or a PABPC1 RRM2 mutant ( that does not interact with eIF4G1 ) , transfected with siRNA against endogenous Pabpc1 or Luciferase , and treated with isoproterenol ( Iso ) or vehicle ( Veh ) . Quantification of ( M ) cell areas , ( N ) total protein content , ( O , P ) rate of new protein synthesis measured by Click-iT homopropargylglycine fluorescence assay after respective treatments . ( Q ) Representative immunoblots of hypertrophy markers . Data are mean ± s . d ( n = 3 ) ; p<0 . 05 , one-way analysis of variance ( ANOVA ) plus Dunnett’s post-hoc test . NS , not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 24139 . 01210 . 7554/eLife . 24139 . 013Figure 4—source data 1 . Source data for cell area of cultured neonatal cardiomyocytes after treatment with siRNA , adenovirus , and Iso . DOI: http://dx . doi . org/10 . 7554/eLife . 24139 . 01310 . 7554/eLife . 24139 . 014Figure 4—figure supplement 1 . PABPC1mRRM2 can bind to poly ( A ) RNA but does not interact with eIF4G1 . ( A ) Schematic of the PABPC1 protein with key domains labeled . Mutations of the RRM2 domain ( PABPC1mRRM2 ) with eight amino acid replacements to interrupt PABPC1-eIF4G binding were made using the available structural information5 . ( B ) C2C12 myoblasts were infected with FLAG-PABPC1 , FLAG-PABPC1mRRM2 , and control GFP adenoviruses followed immunoprecipitation by anti-FLAG antibody and blotting for endogenous eIF4G1 . FLAG-PABPC1 construct efficiently immunoprecipitated eIF4G1 , whereas the negative control FLAG-GFP and the PABPC1mRRM2 did not . ( C ) Constructs used for bacterial expression and purification for mouse PABPC1 , PABPC1mRRM2 , and a PABP-interacting fragment of mouse eIF4G15 . ( D ) Equimolar amounts of purified His-eIF4GI 61-225 and GST-PABPC1 or GST-PABPC1mRRM2 were mixed and incubated with or without RNase A . Protein complexes were isolated using Glutathione-Magnetic beads , separated on 10% SDS-PAGE gel and visualized by Coomassie staining . GST-PABPC1 showed binding to His-eIF4GI 61-225 whereas GST-PABPC1mRRM2 or empty beads did not . ( E ) A gradient of concentrations of GST , GST-PABPC1mRRM2 , and GST-PABPC1 along with 32P labeled poly ( A ) 25 oligo were used to capture Protein-RNA probe complexes and un-bound RNA probes respectively in a filter binding assay . Following incubation , radioactive signal was measured to quantify how much 32P-poly ( A ) 25 remained bound to the protein and how much washed through to the nucleic acid membrane . GST did not bind any 32P-poly ( A ) 25 . Both wild-type and mutant PABPC1mRRM2 were capable of binding the 32P-poly ( A ) 25 . ( F ) Quantification of the Geiger counts on the protein membrane . ( G ) Schematic depicting the experimental set up of PABPC1 rescue experiments in neonatal cardiomyocytes . DOI: http://dx . doi . org/10 . 7554/eLife . 24139 . 014 To further determine if PABPC1 upregulation is sufficient to drive cardiac hypertrophy , we generated a doxycycline-inducible , cardiomyocyte-specific PABPC1 transgenic mouse model ( Figure 5 and Figure 5—figure supplement 1A , B ) . Ectopic expression of FLAG-tagged PABPC1 protein in adult mouse hearts showed the expected cytoplasmic distribution ( Figure 5E , F ) and was estimated approximately 12-fold higher over endogenous levels ( Figure 5—figure supplement 1C , D ) . Remarkably , forced PABPC1 expression in the adult myocardium led to increased atrial and ventricular size ( Figure 5A–D ) , a significant elevation in heart-to-body weight ratios , and larger cardiomyocyte cross-sectional areas compared to uninduced littermate controls ( Figure 5G–J ) . We also observed significant increase in global protein synthesis rates along with upregulation of many physiological but not pathological hypertrophic markers in PABPC1 transgenic hearts ( Figure 5K , L ) . Notably , PABPC1 transgenic mice did not exhibit premature lethality or cardiac dilation even when induced for 6 months . Moreover , long-term PABPC1 induction did not cause a drop in performance in treadmill tests , deterioration in cardiac contractility , or deficits in systolic and diastolic function ( Figure 5—figure supplement 1E , F ) . Also , histologically , we did not observe any myofiber disarray or fibrosis ( Figure 5 and Figure 5—figure supplement 2 ) suggesting hypertrophy in these animals is compensated and does not progress to heart failure , but rather mimics the physiologic form . 10 . 7554/eLife . 24139 . 015Figure 5 . Forced expression of PABPC1 in adult cardiomyocytes induces physiologic hypertrophy . ( A–H ) Representative whole heart , H&E , immunofluorescent , and WGA-stained sections of 2-week doxycycline ( Dox ) -induced MHCrtTA transgenic controls and TRE-PABPC1; MHCrtTA bitransgenic mice . ( I ) Heart-to-body weight ratios ( n = 6 ) . ( J ) Cell area quantified from WGA-stained sections ( n = 3 ) . ( K ) Global rate of translation based on puromycin incorporation in hearts of injected mice ( n = 6 ) . ( L ) Relative mRNA levels of indicated physiological and pathological hypertrophy markers normalized to GAPDH ( qPCR , n = 9 ) . Data are mean ± s . d; p<0 . 05 unpaired two-tailed t-test . NS , not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 24139 . 01510 . 7554/eLife . 24139 . 016Figure 5—source data 1 . Source data for cardiomyocyte areas performed on WGA stained heart tissue sections of 2-week doxycycline-induced MHCrtTA transgenic controls and TRE-PABPC1; MHCrtTA bitransgenic mice . DOI: http://dx . doi . org/10 . 7554/eLife . 24139 . 01610 . 7554/eLife . 24139 . 017Figure 5—source data 2 . Source data for Figure 5—figure supplement 1 . Sheet 1: Source data for exercise performance between 6-month doxycycline induced MHCrtTA transgenic controls and TRE-PABPC1; MHCrtTA bitransgenic mice . Sheet 2: Source data for cardiac function tests between 6-month doxycycline-induced MHCrtTA transgenic controls and TRE-PABPC1; MHCrtTA bitransgenic mice . DOI: http://dx . doi . org/10 . 7554/eLife . 24139 . 01710 . 7554/eLife . 24139 . 018Figure 5—figure supplement 1 . Generation of tetracycline-inducible , heart-specific PABPC1 transgenic mouse model . ( A ) The TRE-PABPC1 construct expresses mouse PABPC1 containing an N-terminal Flag tag driven by a TRE and a CMV minimal promoter . TRE-PABPC1 mice were mated with Myh6-rtTA ( MHCrtTA ) mice to generate TRE-PABPC1; MHCrtTA bitransgenic mice as shown . Eight-week-old adult bitransgenic and MHCrtTA control animals were fed 2 g/kg doxycycline ( Dox ) containing diet for 2 weeks to induce FLAG-PABPC1 expression specifically in cardiomyocytes . ( B ) Immunoblot against FLAG shows that exogenous PABPC1 in the heart is only expressed when Dox is present in the diet . ( C ) Immunoblot against PABPC1 shows the relative amount of induction relative to the wild-type mice . ( D ) Quantification of the blot in C Data are mean ± s . d ( n = 6 ) ; p<0 . 05 , unpaired two-tailed t-test . ( E ) MHCrtTA control and bitransgenic mice were induced with 2 g/kg Dox diet for 6 months and then subjected to acute exercise performance test to measure effects of PABPC1 expression on relative Max VO2 consumption , distance traveled , or work completed . ( F ) Echocardiograms were performed to calculate ejection fraction ( EJ ) , fractional shortening ( FS ) , and left ventricular internal dimension at systole and diastole ( LVID ) . Data are mean ± s . d ( n = 6–12 ) ; p<0 . 05 , unpaired two-tailed t-test . NS , not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 24139 . 01810 . 7554/eLife . 24139 . 019Figure 5—figure supplement 2 . Induced expression of cardiomyocyte specific PABPC1 does not lead to damage . Histological sections of 6-month 2 g/kg doxycycline ( Dox ) -induced MHCrtTA transgenic and TRE-PABPC1; MHCrtTA bitransgenic mice . Hematoxylin and eosin or trichrome staining shows normal cytoarchitecture of the heart , no myofiber disarray or presence of inflammatory cells , and no signs of damage or interstitial fibrosis in either genotype . Immunohistochemistry with Ki67 antibody shows absence of any proliferating cells with PABPC1 overexpression . DOI: http://dx . doi . org/10 . 7554/eLife . 24139 . 019 In conclusion , we have uncovered that poly ( A ) tail length shortening suppresses Pabpc1 mRNA translation in mature cardiomyocytes , thereby reducing overall protein synthesis rates in the adult heart . Intriguingly , despite having a short poly ( A ) tail , Pabpc1 transcripts remain stable and are not completely deadenylated or degraded . This might be due to binding of trans-acting factor ( s ) , the presence of RNA structural element ( s ) , or RNA modifications that act to stabilize Pabpc1 in adult cardiomyocytes ( Lewis et al . , 2017 ) . We further demonstrate that PABPC1 protein expression and poly ( A ) tail length are partially restored during adult-onset cardiac hypertrophy to activate new protein synthesis and physiologic growth of cardiomyocytes . PABPC1 was also shown to be both necessary and sufficient to drive cardiomyocyte hypertrophy . Thus , dynamic control of PABPC1 serves an adaptive role in stimulating a beneficial form of cardiac hypertrophy that may be physiologically advantageous to the failing or dilated myocardium . Two genome-wide studies found a weak correlation between poly ( A ) tail lengths and translational efficiency of mRNAs in cell culture ( Chang et al . , 2014; Subtelny et al . , 2014 ) . While Subtelny et al . observed clear coupling between poly ( A ) tail length and translation at early developmental stages in Xenopus and zebra fish , the association became less apparent in the adults . Thus , it is plausible that at steady states , poly ( A ) tail length is less critical for translation but becomes a determinant in certain tissue contexts , developmental stages , cell cycle regulation , daily rhythmic oscillations of protein synthesis , or cellular stress ( Besse and Ephrussi , 2008; Kojima et al . , 2012; Park et al . , 2016 ) . Furthermore , stronger relationships between poly ( A ) tail length and mRNA translatability may emerge under conditions where PABPC1 protein concentrations are limiting; our data from adult cardiomyocytes supports this hypothesis . Taken together , our findings not only provide insight into the long-standing question of how the heart regulates protein synthesis during development and hypertrophy but also provide a new direction to explore therapeutic interventions . These results set the stage for elucidating the signaling cascades that regulate Pabpc1 poly ( A ) tail length in cardiomyocytes , determining mechanistically how those dynamics are achieved , and probing the general roles for poly ( A ) tail length in translation control within the heart and other somatic tissues . Mouse PABPC1 cDNA containing an N-terminal FLAG-tag was expressed from a transgene with a TRE/minimal CMV promoter , a genomic fragment including α -MHC untranslated exons 2 and 3 with intron 2 ( Kistner et al . , 1996 ) , the Pabpc1 ORF and the bovine growth hormone polyadenylation site and 3′ flanking genomic segment for proper mRNA 3′ end formation . The linearized transgene construct was subjected to pronuclear injection using standard methods to generate PABPC1 transgenic mice that were maintained on an FVB background . MHC-rtTA transgenic mice ( FVB/N-Tg ( Myh6-rtTA ) 1Jam ) expressing a codon-optimized rtTA variant specifically in heart were commercially obtained ( RRID: MMRRC_010478 ) ( Valencik and McDonald , 2001 ) . All mice reported were the F1 progeny of TRE-PABPC1 and MHC-rtTA matings and were , therefore , hemizygous for one or both transgenes . PABPC1 expression in 8- to 12-week-old bitransgenic animals was induced through doxycycline ( Dox ) in the food ( 2g Dox/kg food , Harlan , KY ) . Cardiac-specific Dicer knockouts were generated by crossing Dicerf/f mice , with Tam-inducible MerCreMer transgenic mice as previously reported ( Kalsotra et al . , 2010 ) . DNA was extracted from tail clips using DirectPCR lysis reagent ( Viagen Biotech , Los Angeles , CA ) and genotyped by PCR using transgene-specific primers ( Supplementary file 1 ) . Both male and female mice and littermate controls were used whenever possible . Human fetal ( 22-week old ) and adult ( 51-year-old Caucasian male ) heart RNAs and proteins were purchased from Clonetech Laboratories , Inc . , Mountain View , CA We followed the NIH guidelines for use and care of laboratory animals , and all experimental protocols were approved by IACUC ( Institutional Animal Care and Use Committee at University of Illinois , Urbana-Champaign . Mice performed an exhaustive acute exercise bout on a motorized treadmill ( Columbus Instruments , Columbus , OH ) to determine exercise capacity in Dox-induced TRE-PABPC1 × MHC-rtTA ( n = 15 ) and control MHCrtTA ( n = 13 ) mice using a previously validated equation ( Fernando et al . , 1993 ) . Prior to the maximal exercise test , mice were acclimated to treadmill running for 2 concurrent days with the following protocol: warm-up 5 min at 8 m/min , 5 min exercise at 10 m/min , and 5 min cool-down at 8 m/min at 0% incline . Three days later , mice performed an exhaustive , incremental exercise bout that began at 11 m/min and increased 1 m/min every 2 min until exhaustion as previously described ( De Lisio and Parise , 2012 ) . VO2 was determined for each stage using the following equation: PredictedVO2 ( ml/min ) =0 . 127×weight ( g ) + ( 0 . 040⋅runningspeed ( m/min ) ) −0 . 974 Exhaustion was determined if mice failed to continue running after spending 5 s off of the treadmill bout and not responding to manual stimulation . The researcher evaluating exhaustion was blinded to mouse genotype . PABPC1 expression was determined in the hearts of exercise trained C57BL6/J from a previous study ( De Lisio and Parise , 2012 ) . Six-week-old C57BL6/J mice were randomized into sedentary ( SED , n = 12 ) or endurance trained ( EX , n = 10 ) groups and performed a progressive exercise training program on a motorized treadmill ( Columbus Instruments , Columbus , OH ) . Mice were trained 3 days/ week ( M/W/F ) for 8 weeks . The exercise protocol consisted of: 10-min warm-up at 12 m/min , 45-min training period that began at 14 m/min ( week 1 ) that increased to a speed of 22 m/min ( week 8 ) , and 5-min cool-down at 10 m/min . Mice were trained at the same time each day and sedentary mice were placed onto the treadmill to mimic the stress of handling and treadmill exposure . Transthoracic echocardiography was performed on lightly anesthetized mice using 1 . 5% isoflurane mixed with 95% oxygen as previously reported ( Kalsotra et al . , 2010 ) . Mice were stabilized on a heated platform and taped to ECG electrodes . Evaluation of cardiac function was done using a Visual Sonics Vivo 770 ultra sound using a 30 MHz probe . Two-dimensional guided M-mode tracings were recorded in both parasternal long and short axis views at the level of papillary muscles . Image analysis was done using Visual Sonics software version 2 . 3 . 0 . Heart tissues from doxycycline-induced MHCrtTA ( n = 3 ) transgenic and TRE-PABPC1; MHCrtTA bitransgenic mice ( n = 3 ) were harvested and fixed overnight in 10% neutral-buffered formalin , embedded in paraffin , and sectioned ( 3 μm thickness ) . Hematoxylin and eosin ( H&E ) and trichrome stainings were performed using standard histological methods as previously described ( Bhate et al . , 2015 ) . For immunohistochemistry , unstained slides were deparaffinized in xylene ( two treatments , 5 min each ) , rehydrated sequentially in ethanol ( 2 min each in 100% , 95% , and 80% ) , and washed for 3 min in water . Antigen retrieval was performed by boiling the sections in Tris buffered solution ( 20 mM Tris-Cl pH 9 , 1 mM EDTA , 0 . 02% Tween 20 ) for 20 min at 111°C in a steam cooker then cooled for 20 min under tap water . For ki-67 staining , the endogenous peroxidase activity was quenched with a solution of 3% hydrogen peroxide solution ( Fisher Scientific ) . After washing , sections were blocked ( 2% normal goat serum , 1% bovine serum albumin ( BSA ) , 0 . 1% Triton X-100 , 0 . 05%Tween 20 in 1X ~ PBS ) for 30 min and incubated with primary anti-Ki-67 antibody at 4°C for 12 hr . After several washes , sections were incubated with HRP-conjugated goat anti-mouse IgG light-chain-specific antibody for 2 hr . For visualization of signal , DAB kit ( Vector Labs ) was used according to the manufacturer’s instructions . All intermediate washing steps were done using 1X PBS , 0 . 5% Tween 20 , pH 7 . 2 ( 1X PBST ) , and all antibodies were diluted in 1X PBST with 1% bovine serum albumin . Slides were sealed with a coverslip after lightly counterstaining with hematoxylin and photographed with an EVOS XL microscope . Wheat germ agglutinin stain ( WGA , L4895 SIGMA ) was used to stain heart cross-section according to manufacturer’s instructions . Proteins were labeled using a protocol adapted from the SUnSET method ( Goodman et al . , 2011; Schmidt et al . , 2009 ) . P0 and Adult mice were injected with puromycin made in sterile PBS ( 0 . 04 μmol/ per g of body weight ) . After 45 min hearts and liver were harvested and protein lysates were prepared . Proteins were separated by 10% SDS-PAGE . Puromycin-labeled peptides were identified using the mouse monoclonal antibody 12D10 ( 1: 5000 dilution ) . Protein synthesis levels were determined by densitometry analysis of whole lanes . Normalization of the free puromycin was analyzed as previously described ( Goodman et al . , 2011 ) . Whole hearts ( pooled batches of 2 to 3 hearts ) were extracted and washed in ice-cold PBS containing cyclohexamide . Blood was removed by squeezing the heart with blunt forceps and quickly pulverized under liquid nitrogen using previously cooled , RNAse free mortar and pestle . The powder obtained was transferred to a 10 cm plate , previously cooled on dry ice for 10 min . Afterward , 1 mL of lysis buffer ( 10 mM Tris-HCl at pH 8 . 0 , 150 mM NaCl , 5 mM MgCl2 , 1% Nonidet-P40 , 40 mM dithiothreitol , 500 U/mL RNAsin [Promega] , 40 mM VRC [New England Bio Labs] ) supplemented with 1% deoxycholate [Fluka] was added to the tissue powder . Next , re-suspended powder was scraped from the plate and transferred to a 2-mL tube with pipetting 10x to lyse the cells . The cell nuclei were removed by centrifugation ( 12 , 000 g , 10 s , at 4°C ) , and the supernatant was supplemented with 500 μL of 2X extraction buffer ( 0 . 2 M Tris-HCl at pH 7 . 5 , 0 . 3 M NaCl ) , 150 μg/mL cycloheximide , 650 μg/mL heparin , and 10 mM phenyl-methylsulfonyl fluoride , and centrifuged ( 12 , 000 g , 5 min , at 4°C ) to remove mitochondria and membranous debris . The supernatant was layered onto a 10 mL linear sucrose gradient ( 15%–45% sucrose [w/v] , supplemented with 10 mM Tris-HCl at pH 7 . 5 , 140 mM NaCl , 1 . 5 mM MgCl2 , 10 mM dithiothreitol , 100 μg/mL cycloheximide , 0 . 5 mg/mL heparin ) and centrifuged in a SW41Ti rotor ( Beckman ) for 120 min at 38 , 000 rpm and 4°C , with the brake off . Polysome profiles were recorded using a UA-6 absorbance ( ISCO ) detector at 254 nm . Fractions ( 12 × 1 mL ) were collected and RNAs were recovered by extraction with an equal volume of Trizol . RNAs were reverse transcribed using random hexamer primers and Maxima Reverse Transcriptase kit ( Thermo Scientific ) . The cDNA was diluted to 25 ng/μL with nuclease free water and used for gene-specific qPCR assays . Proteins from hearts or purified cell fractions were isolated as previously described ( Bhate et al . , 2015 ) . In brief , frozen heart tissue or purified cell fractions were homogenized with cold homogenization buffer ( 10 mM HEPES-KOH , pH 7 . 5 , 0 . 32 M Sucrose , 5 μM MG132 , 5 mM EDTA-free Proteinase inhibitor [Pierce 88666 , Thermo Fisher] ) using a bullet blender ( Next Advance ) . Samples were sonicated in the presence of 0 . 1% SDS and clarified by centrifugation ( 20 , 000 rcf at 4°C ) . The protein content was measured using the BCA protein assay kit ( Thermo Scientific ) . Protein lysates ( 100–150 μg of protein loaded per lane ) were resolved by 10% SDS– polyacrylamide gel electrophoresis gels and transferred onto PVDF membranes ( Immobilon , Millipore ) . Membranes were blocked in Tris-buffered saline ( TBS ) containing 5% non-fat dry milk and 0 . 2% Tween 20 ( TBST ) , prior to incubation with primary antibody . The membranes were then washed with TBST followed by incubation with an appropriate horseradish peroxidase-conjugated secondary antibody for 2 hr . Blots were treated with Clarity Western ECL kit , visualized on a ChemiDoc XRS+ ( BioRad ) , and quantified using Image Lab Software ( RRID: SCR_014210 ) according to standard procedures with experimental bands normalized to a control protein . Please refer to supplementary information for product numbers and Research Resource Identifiers of antibodies used in this study ( Supplementary file 1 ) . Cardiomyocytes and cardiac fibroblasts at specified time points were isolated from FVB/NJ wild-type mice as previously reported ( Giudice et al . , 2014 ) . Briefly , neonatal cardiomyocytes were isolated with a neonatal rat/mouse cardiomyocyte isolation kit ( Cellutron Life Tech Highland Park , NJ , USA; nc-6031 ) using manufacturer’s instructions . Cells from 24 to 36 hearts were pooled , pre-plated for 2 hr on an uncoated dish to separate fibroblasts from cardiomyocytes . The supernatant containing the cardiomyocytes was removed , plated on SureCoat ( Cellutron Life Tech; sc-9035 ) coated plates , and incubated at 37°C in a humidified incubator with 5% CO2 . After 12 hr , media was changed to NW ( Cellutron Life Tech; m-8032 ) and cultured until use . The purity of cultures was determined by Western blot as well as immunofluorescence staining with anti-Desmin ( Abcam , ab15200 ) and anti-Vimentin ( Abcam , 11256 ) antibodies . The plate containing fibroblasts was washed with PBS three times before extracting protein/RNA . Adult cardiomyocytes were isolated from 2-month-old FVB mice using a Adumyt Cardiomyocyte Isolation Kit ( Cellutron Life Technology , ac-7034 ) according to the manufacturer's instructions . Mice were treated with anticoagulant ( 500 U heparin i . p . ) 30 min prior to heart extractions . Langendorf perfusion was carried out at 37°C . Cardiomyocytes and fibroblasts were separated using the same pre-plating method as described above . To evaluate the role of PABPC1 proteins in cardiac hypertrophy , isolated neonatal cardiomyocytes were cultured in serum containing NS media ( Cellutron Life Tech ) for 12 hr . After 12 hr , the cells were washed twice with pre-warmed Serum-free media ( NW , Cellutron Life Tech ) before transfecting with the siRNA against endogenous Pabpc1 ORF or 3’- UTR using Lipofectamine RNAiMAX Transfection Reagent ( Thermo Fisher ) . siRNA against Luciferase was used as control . After 12 hr , cells were washed with NW media and treated with 20 nM isoproterenol ( Iso ) or 20 ng of triiodothyronine ( T3 ) and cultured for additional 36 hr . C2C12 cells ( ATCC , CRL-1772; RROD: CVCL_0188 ) were cultured in DMEM ( Dulbecco's Modified Eagle's Medium ) supplemented with 10% FBS , 2 mM glutamine , 100 units/mL penicillin and 100 μg/mL streptomycin and were maintained at 37°C in 5% CO2 as previously descried ( Singh et al . , 2014 ) . The C2C12 cells were seeded in six-well plates . After 12–16 hr , when cell confluence reached approximately 100% the differentiation of C2C12 myoblasts into myotubes was induced by the addition of differentiation medium ( DMEM containing 2 . 5% horse serum ) . Starting at the beginning of differentiation , the C2C12 cells were cultured in six-well plates and harvested for RNA extraction at 0 , 1 , 2 , 3 , and 4 days after differentiation . Total RNA samples were extracted using TRIZOL ( Invitrogen ) according to the manufacturer's instructions . Upon DNAse treatment ( Promega ) , RNAs ( ~2 . 5 μg ) were reverse transcribed using random hexamer primers and Maxima Reverse Transcriptase kit ( Thermo Scientific ) . The cDNA was diluted to 25 ng/μL with nuclease-free water and used for downstream qPCR assays . The 3’ UTR of Pabpc1 was cloned at the Not1 and Xho1 site of the psiCheck2 plasmid ( downstream of RLuc ) . The 5’ UTR of Pabpc1 was cloned at Nhe 1 site ( Upstream of RLuc ) . The double clone included both 5’UTR and 3’UTR and were cloned in their respective sites ( Upstream and downstream of RLuc ) . C2C12 cells were plated on six-well plates at approximately 90–100% confluency and reverse transfections were performed using Mirus-Trans 20 reagent according to slightly modified manufacturer’s protocol . Briefly , we transfected 3–5 μg of the respective plasmid at very high density of cells . The media was replaced after 24 hr of transfection with fresh low-serum differentiation media and cultured for 5 days before determination of luciferase activities using the Dual-Luciferase system . The Firefly and Renilla luciferase activities were measured at 24 hr after transient transfection for myoblast and 5 days after changing to low-serum differentiation media for myotube using the Dual-Glo Luciferase assay system ( Promega ) according to the manufacturer's instructions . Each plasmid was tested in three independent experiments . Luciferase activity was normalized using the Firefly luciferase activity levels and expressed as relative luciferase units ( RLU ) to reflect the influence of 5’ UTR and 3’UTR activity on translation of Renilla luciferase . The PABPC1-binding domain of eIF4G1 ( residues 161–225 ) was cloned into the pET41a vector . The ORF of PABPC1 and PABPC1mRRM2 were cloned into the pGX-2T vector to be expressed as N-terminal 6X His-tagged and C-terminal GST tagged fusion proteins respectively in E . coli BL21 ( strain α-DE3 ) . Cells expressing GST-fusion proteins were lysed in PBS ( 19 mM Na2HPO4 , 0 . 9 mM KH2PO4 , 2 . 5 mM KCl , 140 mM NaCl [pH 7 . 4] ) and the proteins were purified by affinity chromatography on glutathione-magnetic beads ( Thermo Fisher , 88821 ) . Cells expressing His-tagged proteins were lysed in lysis buffer ( 20 mM Tris-Cl , 300 mM NaCl , 1% NP40 . pH-7 . 4 ) , and the proteins were purified by affinity chromatography on Ni2+-NTA ( Qiagen , 30230 ) . The beads were washed using lysis buffer supplemented with 20 mM imidazole then eluted in 300 mM Imidazole buffer ( 20 mM Tris-Cl , 300 mM Imidazole , 10% Glycerol , 300 mM NaCl , pH 7 . 4 ) . Equimolar amounts of His-eIF4GI ( 161-225 a . a ) and GST-PABPC1 or GST-PABPC1mRRM2 were mixed and incubated with or without RNase A in 1XPBS for 3 hr at 4°C . Protein complexes were separated using Glutathione-Magnetic beads by incubating for 2 hr at 4°C . For control , purified GST protein was incubated with His-eIF4GI ( 161-225 a . a ) . Magnetic beads with the protein complex were washed with ice cold 1X PBS and directly boiled in 2X laemmli buffer before separating on 10% SDS-PAGE gel . The protein complexes were visualized using Coomassie stain . Poly ( A ) 25 RNA ( IDT ) was radiolabeled at the 5’ end with 50 mCi γ−32P-ATP using T4 polynucleotide kinase ( NEB ) . For filter binding assays , 1 fmol of 32P-poly ( A ) 25 was incubated with purified GST , GST-PABPC1 , or GST-PABPC1mRRM2 in a final volume of 100 µL ( Tris-HCl buffer ( pH 8 . 0 ) , 70 mM KCl , 10% glycerol , 0 . 05% IGEPAL , 1 mM DTT , 100 ng/mL BSA supplemented with 0 . 4 unit/mL RNasin ) . The reaction was incubated at 40°C for 3 hr on a 96-well dot-blot ( Minifold I; Schleicher & Schuell Cat . 10447900 ) . A Hybond N+ nylon membrane was placed beneath a Amersham Hybond ECL Nitrocellulose Membrane and pre-equilibrated with 1X TBE . These membranes , used to capture Protein-RNA probe complexes and unbound RNA probes respectively , were further sandwiched in dot blot assembly . The incubated mixture was drawn through slowly by vacuum . Following incubation , membranes were separated , air dried , and measured by autoradiography for 6 hr . After 6 hr , the dots from membranes were cut out and corresponding radiation was quantified using a scintillation counter . C2C12 cells were expanded according to previously described methods . In brief , cells were thawed from a low passage number and expanded by splitting before 40% confluency . At 95–100% confluency , cells were infected with AdPABPC1 , AdPABPC1mRRM2 , or AdGFP virus at 5 × 109 o . p . u and incubated for 4 days . After 4 days , the cells were washed with PBS and lysed on ice in lysis buffer ( 50 mM HEPES [pH 7 . 5] , 150 mM NaCl , 0 . 5 mM EDTA , 10% glycerol , 1% Triton X-100 , 5 mM dithiothreitol ) supplemented with complete EDTA free protease inhibitor ( Sigma-Aldrich , P0044 ) and phosphatase inhibitor ( Thermo Scientific , 88666 ) . Of cell lysate , 1–5 mg was incubated with 50% slurry of 200 μL of anti-Flag magnetic beads ( Sigma-Aldrich , M8823 ) in each IP reaction and incubated at 4°C for 6 hr . After washing , the proteins were separated by 8% SDS-PAGE , transferred to PVDF membranes , and detected with anti-eIF4G1 antibody ( #2617 , Cell Signaling Technology ) . Adenovirus was produced using the methodology as previously described ( Bhate et al . , 2015 ) . Briefly , ORF encoding FLAG-tagged PABPC1 and PABPC1mRRM2 were cloned into the p-Adeno-X-ZsGreen1 vector ( Clonetech , 632267 ) using the In-Fusion kit ( Clonetech , 639646 ) as per the manufacturer’s instructions . High-titer adenoviruses were generated by transfecting Ad-293 cells ( ~70% confluent ) in T-25 flasks with linearized recombinant adenoviral plasmid using Mirus TransIT-2020 reagent . Virus was harvested once a cytopathic effect was observed . Next , two viral amplification steps were performed and the viral particles were purified using a CsCl gradient accurding to the Adeno-X Adenoviral System 3 user manual . After purification of viral particles , the titer was determined by ultraviolet spectrophotometry at 260 nm . Neonatal cardiomyocytes cultured in NW media were transfected with siRNA target the 3’UTR of endogenous Pabpc1 . After 6 hr , the cardiomyocytes were infected with 5 × 109 o . p . u . of Ad-PABPC1 , Ad-PABPC1mRRM2 , or Ad-GFP adenovirus . Cells were washed and treated with 20 nM Isoproterenol following a 12 hr incubation with virus . After 36 hr of Isoproterenol treatment , cells were collected to harvest protein and RNA for further analysis . Protein synthesis was measured using the L-homopropargylglycine ( HPG ) Click-iT ( ThermoFisher , C10428 ) metabolic labeling reagents according to the manufacturer’s protocol . Briefly , cultured neonatal cardiomyocytes were washed twice with warmed PBS and incubated in methionine-free DMEM supplemented with Iso or T3 for 1 hr . The medium was replaced with methionine-free DMEM to which 50 μM of the methionine analog HPG was added . Cells were again treated with same concentration of Iso or T3 and incubated for 60 min for incorporation of the AHA into nascent proteins with or without Iso/T3 . After incubation , the dishes were rinsed twice . Newly synthesized proteins labeled with Click-iT HPG were conjugated with the carboxytetramethylrhodamine alkyne ( TAMARA ) using the Click-iT TAMRA Protein Analysis Kit ( Cat . no . C33370 ) . Protein samples were separated on 10% SDS-PAGE and was visualized using 532 nm excitation . The gel was subsequently stained with Coomassie blue for normalization . To determine the cellular location of Pabpc1 mRNA , we performed RNA-FISH on undifferentiated and differentiated C2C12 cells using the Stellaris RNA-FISH kit ( Biosearch Technologies ) . In brief , 48 fluorescently labeled oligonucleotide probes targeting Pabpc1 mRNA were designed using the custom Stellaris Probe Designer ( Supplementary file 1 ) . Undifferentiated and differentiated C2C12 cells were fixed in a 3 . 7% formaldehyde buffer in 1X PBS . The cells were then permeabilized using 70% ethanol for 48 hr at 4°C . All subsequent steps were performed in the dark to minimize loss of fluorescent signal . Hybridization with the probes was performed by washing the cells with Stellaris Wash Buffer A before incubating overnight at 37°C with hybridization buffer containing the probe set or hybridization buffer only as a negative control . Cells were then washed for 1 hr with Wash Buffer A before incubating with NucBlue ( ThermoFisher Scientific ) nuclear stain for 5 min . After incubation with DAPI , cells were washed in Wash Buffer B for 5 min and mounted with CC/Mount ( Sigma-Aldrich ) . Images were obtained using a Zeiss LSM 700 . To assess the Pabpc1 poly ( A ) tail lengths by northern blot , we mixed total RNA with 0 . 5 μM DNA oligonucleotides that hybridize to PABPC1 or GAPDH in 15 μL . Where indicated , oligo-dT40 was also included at 0 . 5 μM . After incubation at 65° for 5 min and chilling on ice , the following components were added to the reaction in 30 μL total volume: 1X RNase H buffer ( Promega ) , 10 mM DTT , 15 ng/μL poly ( A ) ( Sigma ) , 20 U RNasin ( Promega ) , and 1 U RNase H . The reaction proceeded at 37° for 2 hr and was stopped by addition of 270 μL of G-50 buffer ( 0 . 25% SDS , 0 . 3 M NaOAc , 20 mM Tris pH 6 . 8 , and 2 mM EDTA ) . RNA was isolated by standard phenol:chloroform:isoamyl alcohol ( 25:24:1 ) extraction followed by ethanol precipitation . Northern blots were performed as previously described ( Bresson and Conrad , 2013 ) . RNA probes were generated by incorporation of 32P-UTP into in vitro transcribed RNAs using T7 templates generated by PCR . This protocol was based on one previously reported ( Kojima et al . , 2015 ) . Herein , a 1X SSC solution contains 150 mM NaCl and 15 mM sodium citrate at pH 7 . 0 . Briefly , 100 μL of GTC buffer ( 4M guanidine thiocyanate , 25 mM sodium citrate , pH 7 . 1 ) was mixed with ~5–15 μg RNA , 2 μL β-mercaptoethanol , and 3 . 75 μL of 50 μM biotinylated oligo dT ( Promega ) in a final volume of ~115 uL . To this mixture , 209 μL of dilution buffer ( 3X SSC , 5 mM TrisHCl pH 7 . 5 , 0 . 5 mM EDTA , 0 . 125% SDS , 5% β-mercaptoethanol ) was added . The samples were heated to 70°C for 5 min and centrifuged at 12 , 000 x g at room temperature . The supernatant was then mixed with MagneSphere Streptavidin paramagnetic particles ( Promega; 150 μL of manufacturer’s slurry ) that had been washed three times with 0 . 5X SSC and Igepal-CA-630 was added to 0 . 1% . The RNA and dT and bead solution was allowed to bind for 15 min at 25°C while rotating and subsequently washed three times with 0 . 5X SSC containing 0 . 1% Igepal-CA-630 . Elutions were performed by incubation of the beads in 400 μL SSC at the indicated concentration plus 0 . 1% Igepal-CA-630 for 5 min at 25°C . RNA was isolated by standard phenol:chloroform:isoamyl alcohol ( 25:24:1 ) extraction followed by ethanol precipitation . RNase T1 treatment and poly ( A ) tail northern blots were performed as previously described10 . After standardization the elution conditions were then used on 2 . 5 μg of total RNA from heart tissue . In order to elute short poly ( A ) tailed mRNA [Poly ( A ) <30] , we used elution buffer containing 0 . 075X SSC while for eluting long polyA mRNA [poly ( A ) >30] we used nuclease-free water . Eluted sample was purified and used for cDNA synthesis and downstream qPCR analyses . All quantitative experiments ( for example , qPCR , Western blots , cell areas and counts ) have at least three independent biological repeats . Differences between groups were examined for statistical significance using Student’s t-test ( for two groups ) , or one-way ANOVA plus Dunnett’s post-hoc test ( for more than two groups ) using the GraphPad Prism 6 Software ( RRID: SCR_002798 ) . Results were expressed as mean ± s . d . , unless otherwise specified . p<0 . 05 , p<0 . 005 , *p<0 . 001 were considered statistically significant .	Hundreds of thousands of different proteins are needed to build and maintain the cells in the human body . All proteins are produced when copies of genetic information in the form of molecules of messenger RNA ( mRNAs ) are translated by the ribosome . The rate at which proteins are made varies widely between different tissues and at different times . In particular , the adult heart has one of the lowest rates of protein production , though this rate can increase markedly during exercise and heart disease . The mechanisms that drive this kind of dynamic change in protein production remain unclear . A better understanding of this process would tell scientists more about how and why cells regulate the translation of mRNAs in general , and might uncover new ways for treating heart disease . Molecules of mRNA are composed of smaller building blocks called nucleotides . All mature mRNAs in humans have a long stretch at one end that contains the nucleotide adenosine – commonly referred to as A for short – repeated 200 to 300 times . This sequence is called the poly ( A ) tail , and specific proteins can bind to this tail and determine the final fate of the mRNA , such as how efficiently it is translated . One such poly ( A ) binding protein , called PABPC1 , is known to promote mRNA translation . Now , Chorghade , Seimetz et al . examine how PABPC1 regulates protein production in mice and human cells . The experiments reveal that , before birth , ample amounts of PABPC1 are found in the heart , but that this protein is undetectable in the hearts of adults . Further experiments showed that the levels of the mRNA for PABPC1 in the heart are the same before birth and in adulthood . So why is the mRNA for PABPC1 translated inefficiently in adult hearts ? In general , mRNAs with long tails tend to be translated more efficiently compared to those with short tails , and it turns out that the mRNA for PABPC1 has a substantially shorter poly ( A ) tail in the adult heart . This tail shortening limits the translation of this specific mRNA , which leads to reduced protein production . Chorghade , Seimetz et al . also showed that the length of the poly ( A ) tail on the mRNA and the levels of the PABPC1 protein are restored in adult hearts during a condition known as hypertrophy . This state of hypertrophy can be triggered by exercise or heart disease and is marked by an increase in the size of individual heart cells and enhanced protein production . Finally , Chorghade , Seimetz et al . found that experimentally raising the levels of PABPC1 in adult hearts could , by itself , make the heart cells produce more protein and the heart grow more . Further studies will explore if other mRNAs in the heart also undergo similar changes and whether this is a general mechanism for controlling protein production .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "biochemistry", "and", "chemical", "biology" ]	2017	Poly(A) tail length regulates PABPC1 expression to tune translation in the heart
A polyubiquitin comprises multiple covalently linked ubiquitins and recognizes myriad targets . Free or bound to ligands , polyubiquitins are found in different arrangements of ubiquitin subunits . To understand the structural basis for polyubiquitin quaternary plasticity and to explore the target recognition mechanism , we characterize the conformational space of Lys63-linked diubiquitin ( K63-Ub2 ) . Refining against inter-subunit paramagnetic NMR data , we show that free K63-Ub2 exists as a dynamic ensemble comprising multiple closed and open quaternary states . The quaternary dynamics enables K63-Ub2 to be specifically recognized in a variety of signaling pathways . When binding to a target protein , one of the preexisting quaternary states is selected and stabilized . A point mutation that shifts the equilibrium between the different states modulates the binding affinities towards K63-Ub2 ligands . This conformational selection mechanism at the quaternary level may be used by polyubiquitins of different lengths and linkages for target recognition . Ubiquitin is a 76-residue signaling protein found ubiquitously in cells . Multiple ubiquitins are covalently linked to form a polyubiquitin , which can then be attached to a substrate protein . The process is known as ubiquitination , a post-translational modification of the substrate protein . Three classes of enzymes catalyze ubiquitination: ubiquitin-activation enzyme ( E1 ) , ubiquitin-conjugation enzymes ( E2 ) , and ubiquitin-protein ligases ( E3 ) . E2 and E3 dictate ubiquitin linkage and substrate specificities . Additionally , deubiquitinases ( DUBs ) are responsible for specifically erasing ubiquitin signals from a substrate protein ( Clague et al . , 2012 ) . Catalyzed by a linkage-specific E2 , two or more ubiquitins are linked up via an isopeptide bond between the carboxylate at the C-terminus of one ubiquitin ( referred to as the distal unit ) and the ε-amine of a lysine residue ( Lys6 , Lys11 , Lys27 , Lys29 , Lys33 , Lys48 , or Lys63 ) or the α-amine at the N-terminus of another ubiquitin ( referred as the proximal unit ) . Lys48-linked polyubiquitin is the most abundant linkage in cells , and signals substrate proteins for proteasomal degradation ( Kravtsova-Ivantsiv et al . , 2013; Lu et al . , 2015 ) . Other types of linkages mostly perform non-degradative functions ( Xu et al . , 2009; Kulathu and Komander , 2012 ) . Lys63-linked polyubiquitin is another common linkage , and has been found to be involved in DNA damage response ( Hoege et al . , 2002 ) , multivesicular body mediated protein sorting ( MacDonald et al . , 2012 ) , NF-κB signaling ( Chen and Chen , 2013; Zinngrebe et al . , 2014 ) , and oxidative stress response ( Silva et al . , 2015 ) . Lys63-linked polyubiquitin also exists as an unanchored form without being attached to a substrate protein , and the unanchored form may serve as a scaffold for recruiting proteins in the signaling pathways of innate immunity and protein aggregate removal ( Zeng et al . , 2010; Hao et al . , 2013 ) . How does Lys63-linked polyubiquitin perform different functions ? To do so , Lys63-linked polyubiquitin has to specifically recognize multiple target proteins . Structural studies of Lys63-linked diubiquitin ( K63-Ub2 ) have indicated that K63-Ub2 mostly adopts open extended conformations in the absence of a ligand and in complex with many target proteins ( Datta et al . , 2009; Komander et al . , 2009; Sato et al . , 2009a; Weeks et al . , 2009; Sekiyama et al . , 2012 ) . It has been proposed that such an open extended structure differentiates K63-Ub2 from Lys48-linked polyubiquitin that predominantly exists in a closed structure ( Varadan et al . , 2004; Fushman and Walker , 2010; Hirano et al . , 2011 ) . Notwithstanding , K63-Ub2 has been found in closed conformations when complexed with the Npl4 zinc-finger domain ( NZF ) of TAK1-binding proteins ( Kulathu et al . , 2009; Sato et al . , 2009b ) for the activation of NF-κB signaling pathways . K63-Ub2 is also selectively recognized by the fourth zinc-finger ( ZnF4 ) of A20 , a ubiquitin-editing enzyme for the termination of NF-κB signaling ( Wertz , 2014; Wertz and Dixit , 2014 ) . The crystal structure of the complex between ZnF4 and ubiquitin monomer indicated that K63-Ub2 likely exists in a closed conformation when binding to ZnF4 ( Bosanac et al . , 2010 ) . So are the closed-state structures already present for the free K63-Ub2 , but have hitherto eluded structural characterization ? Or is the closed-state structure induced by a specific ligand ? To recognize a target protein and to perform specific functions , a protein has to fluctuate among a variety of conformational states ( Henzler-Wildman and Kern , 2007 ) . For a ubiquitin monomer , studies have shown that the protein exists as an ensemble of tertiary conformations which can accommodate different target proteins ( Lange et al . , 2008 ) . As ubiquitin mainly functions as polyubiquitins , a more pertinent question is how does a polyubiquitin fluctuate at the quaternary level and achieve its target recognition specificity ? Nuclear magnetic resonance ( NMR ) is well suited to characterize protein ensemble structures and to visualize protein dynamics . NMR depiction of dynamic fluctuation for multi-domain proteins such as pre-mRNA splicing factor U2AF and DNA-binding protein CAP , has uncovered conformational selection and equilibrium shift mechanisms for these systems ( Mackereth et al . , 2011; Tzeng and Kalodimos , 2012; Huang et al . , 2014 ) . Among the NMR techniques , paramagnetic relaxation enhancement ( PRE ) is exquisitely sensitive to transient and fleeting interactions between proteins ( Sekhar and Kay , 2013; Xing et al . , 2014 ) . Owing to dipolar interactions between the paramagnetic center and protein nuclei , PRE is proportional to the inverse sixth power of the distance between the paramagnetic center and protein nuclei , and is ensemble-averaged over all the conformational states sampled ( Clore and Iwahara , 2009 ) . To explore the target recognition mechanism for K63-Ub2 , we used PRE NMR in conjunction with other biophysical methods . We characterized the arrangements between the ubiquitin units , and we determined the ensemble structure for K63-Ub2 in the absence of a ligand . Our findings indicated a conformational selection mechanism at the quaternary level , whereby a target protein selects and stabilizes one of the preexisting conformational states of ligand-free K63-Ub2 . We first compared the chemical shift differences between the subunits in K63-Ub2 and ubiquitin monomer . Except for residues near the covalent ubiquitin linkage , the differences in chemical shifts are small ( <0 . 04 ppm; Figure 1—figure supplement 1 ) . Residues with relatively large chemical shift differences ( >0 . 01 ppm ) can be tentatively mapped , and form rather contiguous surfaces on each subunit ( Figure 1—figure supplement 1 , insets ) . However , it is unclear whether the perturbations are due to the covalent linkage , or due to non-covalent interactions between the two subunits . We also performed a half-filtered NMR experiment , which failed to reveal any inter-subunit nuclear Overhauser effects ( NOEs ) between the proximal and distal units of K63-Ub2 . Our data are consistent with the previous NMR studies of K63-Ub2 ( Tenno et al . , 2004; Varadan et al . , 2004 ) ; in the latter work , the authors failed to detect cross-saturations between the two subunits . Together , the diamagnetic NMR studies indicated that the closed-state structure of K63-Ub2 , if existing , is loosely packed and possibly adopts multiple conformations . To visualize the quaternary arrangement between the subunits of K63-Ub2 , we resorted to PRE NMR . We prepared K63-Ub2 protein with the proximal unit 15N-labeled and the distal unit unlabeled . A cysteine point mutation was introduced to Asn25 or Lys48 in the distal unit of K63-Ub2 . An MTS paramagnetic probe was conjugated at N25C or K48C site . Each of the conjugation sites was designed so that the paramagnetic probe was away from the binding partners of K63-Ub2 , which include Rap80 tUIM domain ( Sekiyama et al . , 2012 ) , TAB2 NZF domain ( Kulathu et al . , 2009; Sato et al . , 2009b ) , and A20 ZnF4 domain ( Bosanac et al . , 2010 ) , and therefore the conjugation does not interfere with ligand binding ( Figure 1—figure supplement 2 ) . Indeed , the binding affinities between K63-Ub2 and tUIM or NZF remain unchanged for the paramagnetically tagged K63-Ub2 proteins ( Figure 1—figure supplement 3 ) . In addition , the paramagnetic NMR spectrum can be overlaid onto the diamagnetic spectrum , except for residues that are completely broadened out due to the PRE effect ( Figure 1—figure supplement 4 ) . Together , the modifications do not perturb the conformational space of K63-Ub2 or have an effect on K63-Ub2 function . We previously reported that ubiquitin monomer dimerizes non-covalently with an apparent KD value of 4 . 9 ± 0 . 3 mM ( Liu et al . , 2012 ) . Therefore , the PRE effect could arise both intramolecularly and inter-molecularly . The inter-molecular PREs were measured on an equimolar mixture of K63-Ub2 ( each at 500 µM ) , with paramagnetic tagging and isotope labeling on different subunits in separate proteins . With a paramagnetic probe conjugated at either N25C or K48C site and using the inter-molecular data for reference , our measurements revealed large intra-molecular inter-subunit PREs for many residues in the proximal unit ( Figure 1A , B ) . At relatively low protein concentration ( 50 µM ) , the inter-molecular contribution to the overall PRE is negligible . We found that the relative decreases in peak intensities between the paramagnetic and diamagnetic spectra recorded at 50 µM are highly correlated with the relative decreases between the paramagnetic and inter-molecular spectra recorded at 500 µM ( Figure 1—figure supplement 5 ) . This corroborates the PRE measurement at the higher concentration . In addition , when a different paramagnetic probe , EDTA-Mn2+ , was conjugated at N25C , the PRE profile is similar to that using the MTS probe ( Figure 1—figure supplement 6 ) . Thus , the intra-molecular inter-subunit PREs are independent of the paramagnetic probe used , and reveal intrinsic structural features of ligand-free K63-Ub2 . 10 . 7554/eLife . 05767 . 003Figure 1 . Intra-molecular inter-subunit paramagnetic relaxation enhancements ( PREs ) measured for ligand-free K63-Ub2 . With the paramagnetic probe conjugated at ( A ) N25C or ( B ) K48C in the distal unit , the PRE 1H Γ2 rates were measured for amide protons of the 15N-labeled proximal unit . The red spheres indicate the observed PREs . The error bar indicates 1 SD in the PRE measurement . The blue lines are the back-calculated PRE values for residues 1–71 . Residues that are completely broadened out are denoted with asterisks at the top . DOI: http://dx . doi . org/10 . 7554/eLife . 05767 . 00310 . 7554/eLife . 05767 . 004Figure 1—figure supplement 1 . NMR chemical shift differences between ubiquitin monomer and K63-Ub2 . Overlay of 2D 1H-15N correlation spectra for ( A ) ubiquitin monomer and K63-Ub2 proximal unit , and ( B ) ubiquitin monomer and K63-Ub2 distal unit . The ubiquitin monomers ( also the reactants for preparing K63-Ub2 ) were modified ( either by appending an Asp at the C-terminus or by mutating Lys63 to an Arg ) so that only a single product ( diubiquitin ) was obtained . Residues with relatively large chemical shift differences ( >0 . 01 ppm ) were mapped to the surface of ( C ) the proximal unit and ( D ) the distal unit , and are colored red . The chemical shift differences in ppm is calculated as ( ΔδH2/2 + ΔδN2/5 ) ^0 . 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 05767 . 00410 . 7554/eLife . 05767 . 005Figure 1—figure supplement 2 . Illustration of cysteine point mutation and conjugation of an MTS paramagnetic probe to K63-Ub2 , which is complexed with ( A ) Rap80 tUIM , ( B ) TAB2 NZF , or ( C ) A20 ZnF4 . The corresponding Protein Data Bank ( PDB ) codes are 2RR9 , 2WX0 , and 3OJ3 , respectively . The modifications are away from the bound ligands in the structures and also from the other subunit . DOI: http://dx . doi . org/10 . 7554/eLife . 05767 . 00510 . 7554/eLife . 05767 . 006Figure 1—figure supplement 3 . Isothermal calorimetry ( ITC ) measurements for the binding affinities ( A–C ) between K63-Ub2 and tUIM , and ( D–F ) between K63-Ub2 and NZF . The titrations were performed for ( A , D ) wild type K63-Ub2 , ( B , E ) N25C mutant conjugated with an MTS paramagnetic probe , and ( C , F ) K48C mutant conjugated with an MTS paramagnetic probe . The binding affinities KD values were averaged over four independent titrations with SD reported . The averaged enthalpy and entropy changes ( ±SD ) are also labeled . DOI: http://dx . doi . org/10 . 7554/eLife . 05767 . 00610 . 7554/eLife . 05767 . 007Figure 1—figure supplement 4 . Overlay of 2D NMR spectra for wild type protein and paramagnetically tagged K63-Ub2 proteins at ( A ) N25C site and ( B ) K48C site . Peaks that disappear in the paramagnetic spectra are labeled . DOI: http://dx . doi . org/10 . 7554/eLife . 05767 . 00710 . 7554/eLife . 05767 . 008Figure 1—figure supplement 5 . Correlations of the paramagnetic effects measured at two different concentrations . Peak intensities were compared between the spectra for the wild type diamagnetic protein and for the paramagnetic protein collected at 50 µM concentration ( x-axis ) , or were compared between the spectra for the 1:1 mixture of diamagnetic and paramagnetic proteins and for the paramagnetic protein collected at 500 µM concentration ( y-axis ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05767 . 00810 . 7554/eLife . 05767 . 009Figure 1—figure supplement 6 . Comparison of the intra-molecular inter-subunit paramagnetic relaxation enhancement ( PRE ) data with an EDTA-Mn2+ ( red circles ) or MTS probe ( blue circles ) conjugated at N25C site . Lines simply connect the data points . The PREs obtained using EDTA-Mn2+ are larger than those using the MTS probe owing to the larger paramagnetic dipole of the former . DOI: http://dx . doi . org/10 . 7554/eLife . 05767 . 00910 . 7554/eLife . 05767 . 010Figure 1—figure supplement 7 . Comparison between all known structures of K63-Ub2 in the open state . With the distal unit superimposed , the positions for the proximal unit are compared . By enforcing negative inter-subunit paramagnetic relaxation enhancement ( PRE ) restraints , open-state conformations of K63-Ub2 can also be obtained . The atomic probability map plotted at 10% threshold ( gray meshes ) encompasses the known structures of K63-Ub2 in the open state . DOI: http://dx . doi . org/10 . 7554/eLife . 05767 . 01010 . 7554/eLife . 05767 . 011Figure 1—figure supplement 8 . Intra-molecular inter-subunit paramagnetic relaxation enhancements ( PREs ) arising from K63-Ub2 open state are negligible . With an MTS probe conjugated at N25C site or K48C site , the PRE values were calculated for the known structures of K63-Ub2 in the open state , which include ( A ) 3H7P , ( B ) 3HM3 , ( C ) 2JF5 , ( D ) 2RR9 , ( E ) 3A1Q , and ( F ) 2NZV . DOI: http://dx . doi . org/10 . 7554/eLife . 05767 . 011 K63-Ub2 has been generally considered to only exist in the open state for both ligand-free and many ligand-bound forms . For the known open-state structures of K63-Ub2 ( Sato et al . , 2008; Datta et al . , 2009; Komander et al . , 2009; Sato et al . , 2009a; Weeks et al . , 2009; Yoshikawa et al . , 2009; Sekiyama et al . , 2012 ) , the intra-molecular inter-subunit PREs calculated with an MTS probe attached at either N25C or K48C site are essentially zero for residues in the proximal unit ( Figure 1—figure supplement 7 ) . Alternatively , an open extended conformation of K63-Ub2 can be simply obtained by restraining the inter-subunit PRE target value to zero for residues in the proximal unit—the resulting conformational space encompasses all known K63-Ub2 structures in the open state ( Figure 1—figure supplement 8 ) . As such , the large inter-subunit PREs should only arise from the closed state of K63-Ub2 , and ligand-free K63-Ub2 should exist in both open and closed states . The existence of the closed state for ligand-free K63-Ub2 is corroborated by small angle X-ray scattering ( SAXS ) analysis . The SAXS data collected for K63-Ub2 at higher concentrations display larger particle size than those at lower concentrations , indicative of high-order oligomers for the former ( Figure 2—figure supplement 1 ) . At lower concentrations , the Dmax value is smaller , and the data recorded at 1 mM and 500 µM are similar ( Dmax = 67 . 2 and 65 Å , respectively ) . The Dmax values are smaller than the calculated values for all known open-state structures ( 84 . 0 ± 3 . 3 Å ) . Significantly , the experimental paired-distance distribution function P ( r ) at 1 mM is much narrower than those computed for the known open-state structures , with a large probability of distribution at ∼30 Å ( Figure 2A ) . Further , the theoretical scattering profiles for the open-state structure models ( Figure 1—figure supplement 8 ) all differ from the experiment curve ( Figure 2—figure supplement 2A ) . 10 . 7554/eLife . 05767 . 012Figure 2 . Small angle X-ray scattering ( SAXS ) analysis of ligand-free K63-Ub2 . ( A ) Paired-distance distribution curves transformed from the experimental data ( black line ) or calculated for the known structures of K63-Ub2 . Except for the Protein Data Bank ( PDB ) structures 3H7P , 3HM3 , and 2JF5 , the bound ligand was removed before calculation . ( B ) Comparison between experimental ( gray dots ) and simulated scattering data ( transparent cyan line ) , affording a χ2 value of 1 . 24 . The simulated curve was obtained by linearly combining the theoretical curves calculated for open-state ( red line ) and closed-state ( blue line ) solution structures at 30:70 ratio . DOI: http://dx . doi . org/10 . 7554/eLife . 05767 . 01210 . 7554/eLife . 05767 . 013Figure 2—figure supplement 1 . Concentration dependence of small angle X-ray scattering ( SAXS ) profiles for ligand-free K63-Ub2 . The paired-distance curve displays a narrower distribution at lower protein concentration . The larger particle size at higher protein concentration can be attributed to the non-covalent interactions between two or more K63-Ub2 . DOI: http://dx . doi . org/10 . 7554/eLife . 05767 . 01310 . 7554/eLife . 05767 . 014Figure 2—figure supplement 2 . Comparison between experimental and theoretical scattering curves for ( A ) open-state and ( B ) closed-state structures . Each line represents the theoretical curve calculated for one conformer in one of the ensemble structures obtained . For reference , the calculated scattering curves for the known structures of K63-Ub2 in either open or closed states are also shown ( with the bound ligand removed if present ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05767 . 014 To characterize the closed-state structure of ligand-free K63-Ub2 , we performed rigid-body simulated annealing refinement against the inter-subunit PREs . The linker between the two subunits ( Lys63 side chain in the proximal unit and C-terminal flexible residues 72–76 of the distal unit ) was given full torsional freedom . A grid search was performed by varying the number of conformers representing the closed state ( from a single conformer to a five-conformer ensemble ) , and by varying the overall population of the closed state ( from 10% to 90% ) . A single-conformer representation for the closed state does not satisfy the inter-subunit PREs , as assessed by the PRE Q-factor ( Iwahara et al . , 2004 ) . This means that the closed state of ligand-free K63-Ub2 should exist in multiple conformations . The PRE Q-factor rapidly decreases as the number of conformers representing the closed state increases , and levels off with four or more conformers ( Figure 3A ) . On the other hand , a closed-state population of at least 30% is required to achieve a good fit to the PRE data ( Figure 3A ) . For reasons that will be discussed below , the population of the K63-Ub2 closed state is about 70% . At a 70% population for the closed state with a four-conformer representation , the back-calculated PREs agree well with the experimental ones , affording a PRE Q-factor of 0 . 22 and correlation coefficient of 0 . 94 ( Figures 1 , 3B ) . Importantly , the two paramagnetic conjugation sites , N25C and K48C , provide cross-validating PRE measurements—when refining the ensemble structure of the K63-Ub2 closed state against the N25C data alone , the PRE values predicted for the K48C site largely agree with the experimental values , affording a free Q-factor of 0 . 46 ( Figure 3—figure supplement 1 ) . On the other hand , the SAXS profiles computed for the PRE-based closed-state structures differ from the experiment curve , with the calculated intensities larger at scattering angles between 0 . 5 and 1 nm−1 ( Figure 2—figure supplement 2B ) . 10 . 7554/eLife . 05767 . 015Figure 3 . Ensemble refinement of the closed-state structure of K63-Ub2 against intra-molecular inter-subunit paramagnetic relaxation enhancements ( PREs ) . ( A ) Heat map of PRE Q-factor upon varying the number of conformers and the population of the closed state . ( B ) The correlation between observed and calculated PREs , with a four-conformer representation at 70% closed-state population . The PRE ensemble Q-factor is 0 . 26 and 0 . 18 for the N25C site ( open circles ) and K48C site ( closed circles ) , respectively , and 0 . 22 for both sites . The diagonal indicates a perfect correlation to guide the eyes . DOI: http://dx . doi . org/10 . 7554/eLife . 05767 . 01510 . 7554/eLife . 05767 . 016Figure 3—figure supplement 1 . Cross-validation of paramagnetic relaxation enhancement ( PRE ) data . The closed-state structure of K63-Ub2 was refined against N25C data only with a four-conformer representation . The PREs for the K48C site were back-calculated ( blue line ) , which agree well with the experimental data ( red circles ) with a free Q-factor of 0 . 46 . Residues that are completely broadened out are denoted with asterisks at the top . DOI: http://dx . doi . org/10 . 7554/eLife . 05767 . 016 To better visualize the ensemble structure of K63-Ub2 in the closed state , we projected the position of the proximal unit relative to the distal unit using spherical coordinates ( Figure 4—figure supplement 1 ) . Upon reducing the dimensionality , we found that the closed-state structures exist in two clusters , namely C1 and C2 ( Figure 4A ) . For each four-conformer structure , one of the conformers falls into C2 , while the other three are in C1 . The proximal unit of ligand-free K63-Ub2 utilizes distinct interfaces to interact with the distal unit in C1 and C2 states ( Figure 4B , C ) , affording buried solvent-accessible surface areas of 283 . 9 ± 139 . 7 Å2 and 200 . 5 ± 59 . 6 Å2 , respectively . Significantly , the crystal structures of K63-Ub2 in complex with the ZnF4 domain of A20 ( Bosanac et al . , 2010 ) and with the NZF domain of TAB2 or TAB3 ( Kulathu et al . , 2009; Sato et al . , 2009b ) are found within or near the C1 and C2 clusters , respectively ( Figure 4A ) . The root-mean-square difference ( RMSD ) between the conformers in C1 and the ZnF4-bound structure of K63-Ub2 is as small as 3 . 93 Å ( Figure 4B ) , while the RMSD between C2 conformers and the NZF-bound structure is as small as 1 . 68 Å ( Figure 4C ) . We predicted the inter-subunit PREs for two known complex structures in closed states ( Figure 4—figure supplement 2A , B ) . Linearly combining the two sets of PREs at a 3:1 ratio and 70% total population , the resulting PREs agree well with the experimental values , although some details differ ( Figure 4—figure supplement 2C ) . On the other hand , the paired-distance distribution profiles computed for A20 ZnF4 and TAB2/TAB3 NZF complexed K63-Ub2 ( with bound ligand removed ) display narrower distributions compared to those computed for the open-state structures or to the experimental data ( Figure 2A ) . 10 . 7554/eLife . 05767 . 017Figure 4 . Ensemble structure of K63-Ub2 in closed state . ( A ) Projection of the ensemble structures in two dimensions with spherical coordinates . K63-Ub2 crystal structures in closed states are also projected . ( B , C ) Comparison of the ligand-free K63-Ub2 structure with K63-Ub2 crystal structures in complex with A20 ZnF4 and NZF TAB2 , respectively . With the distal unit superimposed , the other ubiquitin subunit in the crystal structure is shown as a gray cartoon , affording root-mean-square ( RMS ) differences of 3 . 93 Å and 1 . 68 Å for C1 and C2 closed states , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 05767 . 01710 . 7554/eLife . 05767 . 018Figure 4—figure supplement 1 . Definition of the spherical coordinate system . The origin is set at the center-of-mass of the distal unit , with north pole indicated . With the distal unit fixed , the polar angle of the proximal unit defines the relative orientation of the vector connecting the centers-of-mass of the distal and proximal units ( dashed lines ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05767 . 01810 . 7554/eLife . 05767 . 019Figure 4—figure supplement 2 . Inter-subunit paramagnetic relaxation enhancements ( PREs ) predicted from the crystal structures of K63-Ub2 in the closed state . ( A , B ) PREs predicted for the proximal unit in the Protein Data Bank ( PDB ) structures 3OJ3 and 2WX0 , with an MTS probe conjugated at K48C site . Averaged PRE values are shown as black lines , with the SDs shown as gray bars . ( C ) Linear combination of the two sets of predicted PREs at 3:1 ratio with 70% total population ( 52 . 5% for 3OJ3 and 17 . 5% for 2WX0 ) . The experimental PRE data at K48C are shown as red circles . DOI: http://dx . doi . org/10 . 7554/eLife . 05767 . 019 Taken together , the ensemble structure refinement revealed that in addition to the open state , ligand-free K63-Ub2 also exists in at least two distinct closed states at a significant combined population . As C1 is represented with multiple conformers , it is possible to further partition the closed state into more conformational states . As the ligand-bound closed structures are similar to ligand-free structures of K63-Ub2 in either C1 or C2 state , a cognate ligand of K63-Ub2 can be preferentially recognized and accommodated into one of the preexisting conformations . As there are some differences between the ligand-free and ligand-bound K63-Ub2 in the closed state ( Figure 4B , C ) , the binding may require some induced fit , especially towards the end of the binding process . On the other hand , the open-state conformation of K63-Ub2 may specifically recognize its corresponding ligand like Rap80 tUIM ( Sekiyama et al . , 2012 ) via a conformational selection mechanism . How does K63-Ub2 inter-convert among the preexisting conformations ? To address this , we introduced a charge reversal mutation to residue Glu64 in the proximal unit , resulting an E64RP mutant of K63-Ub2 . Glu64 is located at the interface between the two subunits in both C1 and C2 closed states , opposing the positively charged residues Arg72 and Arg74 in the distal unit ( Figure 5—figure supplement 1 ) . We reasoned that this mutation should affect the conformational space of K63-Ub2 . Indeed , the E64RP mutation results in chemical shift perturbations ( CSPs ) in the K63-Ub2 distal unit ( Figure 5—figure supplement 2 ) . Although the perturbations are small , almost the same residues are perturbed upon E64RP mutation as upon the covalent linkage of ubiquitin monomers ( Figure 5A and Figure 1—figure supplement 1D ) . However , the NMR peaks for the perturbed residues in the mutant do not simply move in the direction towards the chemical shift values of the ubiquitin monomer . This can be either due to altered non-covalent interactions around the mutation site , or to a change in the relative population of the conformational states . Therefore , it is difficult to quantitate the CSPs in terms of K63-Ub2 structural change . 10 . 7554/eLife . 05767 . 020Figure 5 . Changes in NMR parameters for K63-Ub2 upon E64RP mutation . ( A ) Chemical shift differences of the distal unit upon mutation . Inset , residues with relatively large chemical shift differences ( >0 . 01 ppm ) are mapped to the surface of the distal unit ( colored red ) . ( B ) Decreases in intra-molecular inter-subunit PREs upon mutation with an MTS paramagnetic probe conjugated at K48C site . Lines simply connect the data points . PRE values characteristic of C1 and C2 states are indicated with cyan and yellow strips , respectively . Error bars indicate the SD in PRE measurements . DOI: http://dx . doi . org/10 . 7554/eLife . 05767 . 02010 . 7554/eLife . 05767 . 021Figure 5—figure supplement 1 . Structural basis for the perturbation of K63-Ub2 conformational space upon E64RP mutation . Representative ( A ) C1 and ( B ) C2 closed-state structures . Glu64 in the proximal unit opposes Arg72 and Arg74 in the distal unit , shown as sticks . Judging from the structures , the charge reversal mutation could have a larger impact on the stability of C1 closed state than on C2 closed state of K63-Ub2 . DOI: http://dx . doi . org/10 . 7554/eLife . 05767 . 02110 . 7554/eLife . 05767 . 022Figure 5—figure supplement 2 . Overlay of 2D NMR spectra for wild type K63-Ub2 and E64RP mutant at 50 µM with distal unit 15N-labeled . Residues with relatively large chemical shift differences ( >0 . 01 ppm ) are labeled . DOI: http://dx . doi . org/10 . 7554/eLife . 05767 . 022 Thus we measured the intra-molecular inter-subunit PREs for the E64RP mutant of K63-Ub2 , using the same paramagnetic conjugation scheme . The PRE profile of the mutant is similar to that of the wild type , indicating similar ensemble structures for the mutant ( Figure 5B ) . However , the magnitude of the PRE decreases by about 50% . As the inter-subunits’ PRE arises only from the closed-state structures of K63-Ub2 , smaller PREs indicate that the E64RP mutation destabilizes the closed state , reducing the closed-state population to half of that of the wild type . At the same time , the E64RP mutation should result in an increase in the population for the open state . In a conformational selection mechanism , a K63-Ub2 ligand ( tUIM , NZF , or ZnF4 ) is preferentially recognized by one of the preexisting conformational states . Since the relative populations of the conformational states are perturbed upon the E64RP mutation , we expect that the binding affinities of K63-Ub2 towards the respective ligands differ . Importantly , the mutation is away from the binding interfaces between K63-Ub2 and its ligands ( Figures 6A , B , 7 ) , and therefore should not directly affect the interactions between K63-Ub2 and its ligands . 10 . 7554/eLife . 05767 . 023Figure 6 . The interactions between K63-Ub2 mutant with tUIM and with NZF . ( A , B ) Illustration of the E64RP mutation in the complex structures ( Protein Data Bank [PDB] codes 2RR9 and 2WX0 ) . The point mutation is distant from the K63-Ub2 interfaces with tUIM and NZF . ( C , D ) Isothermal calorimetry ( ITC ) measurements for the bindings with tUIM and NZF . The raw data ( top panels ) are converted to heat per injection , and the fitted curves using one-site binding model are shown as solid lines ( bottom panels ) . The binding affinities KD , binding enthalpy changes ΔH , and entropy changes ΔS values are averaged over four independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 05767 . 02310 . 7554/eLife . 05767 . 024Figure 7 . The interaction between K63-Ub2 and A20 ZnF4 at 313 K . ( A , C ) NMR titrations of ZnF4 into wild type and mutant K63-Ub2 with 15N-labeling at the distal unit . Residues 50–62 ( labeled ) experience slow timescale exchange and gradually disappear upon ZnF4 titration . ( B , D ) Fittings of chemical shift perturbations to binding isotherms . The chemical shift perturbations are calculated as ( ΔδH2 + ΔδN2 ) ^0 . 5 in Hz units . Inset , ZnF4-binding surface on K63-Ub2 distal unit ( residues 50–62 ) is mapped ( colored orange ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05767 . 02410 . 7554/eLife . 05767 . 025Figure 7—figure supplement 1 . Isothermal calorimetry ( ITC ) measurements for the bindings between A20 ZnF4 and ( A ) wild type and ( B ) E64RP mutant of K63-Ub2 proteins . The heat exhausted could not be fitted to a binding isotherm . DOI: http://dx . doi . org/10 . 7554/eLife . 05767 . 02510 . 7554/eLife . 05767 . 026Figure 7—figure supplement 2 . Overlay of NMR spectra for K63-Ub2 and K63-Ub2 mixed with equimolar A20 ZnF4 at ( A ) 303 K and ( B ) 313 K . The protein concentrations are 50 µM . The timescale of the exchange between ZnF4-free and bound species is slow for residues 50–62 of K63-Ub2 distal unit , whose peaks disappear upon ZnF4 titration . At 313 K , the exchange is slightly faster , which allows the fitting of the KD value . DOI: http://dx . doi . org/10 . 7554/eLife . 05767 . 026 Using isothermal calorimetry ( ITC ) , we evaluated the binding affinities between wild type K63-Ub2 and tUIM , and between wild type K63-Ub2 and NZF . The respective KD values are 9 . 7 ± 0 . 3 µM and 12 . 2 ± 0 . 6 µM ( Figure 1—figure supplement 3 ) , which agree with the literature values ( Kulathu et al . , 2009; Sekiyama et al . , 2012 ) . We attempted to measure the binding affinity between A20 ZnF4 domain and K63-Ub2 . However , the heat was too small to be fitted ( Figure 7—figure supplement 1 ) . Thus we resorted to NMR titration for the KD measurement . Upon titrating A20 ZnF4 , a large set of residues in the distal unit of K63-Ub2 was perturbed ( Figure 7A ) . Among the perturbed residues , residues 50–62 are located at the interface with ZnF4 ( Bosanac et al . , 2010 ) , and their peaks disappear upon A20 ZnF4 titration , indicating a slow exchange between ZnF4-free and bound species ( Figure 7—figure supplement 2 ) . We were only able to fit the CSPs at an elevated temperature ( 313 K instead of 303 K ) , and determined the KD value at 384 . 1 ± 39 . 4 µM . A number of other residues in the K63-Ub2 distal unit also experience CSPs upon A20 ZnF4 titration . However , their peaks shift progressively at increasing ZnF4 concentrations , which indicates a fast exchange and should belong to a separate binding event . Upon the E64RP mutation , the binding between K63-Ub2 and the open-state ligand tUIM becomes tighter , with the KD value decreasing by more than fourfold to 2 . 2 ± 0 . 1 µM ( Figure 6C ) . On the other hand , the binding towards closed-state ligands weakens—the KD value of K63-Ub2 binding towards NZF increases by ∼50% to 17 . 8 ± 1 . 1 µM ( Figure 6D ) . Importantly , the enthalpy change ΔH values are almost identical for the bindings involving the wild type and mutant proteins ( Figure 6 and Figure 1—figure supplement 3 ) . For the interaction between K63-Ub2 mutant and A20 ZnF4 , more peaks at the interface ( residues 50–62 of the distal unit ) can be traced , which can be attributed to a faster exchange than that of the wild type . Significantly , the KD value increases by almost threefold to 1199 . 9 ± 104 . 9 µM ( Figure 7 ) . Together , the binding affinity towards the open-state ligand increases at the expense of the binding affinities towards the closed-state ligands , and the changes in binding affinities are caused entropically due to the perturbation of K63-Ub2 conformational space . The difference in binding affinity can be accounted for by the difference in the conformational energy of K63-Ub2 . Based on the PRE measurement , the closed-state population decreases by ∼50% upon E64RP mutation ( Figure 5 ) . At the same time , the population for the open state increases by the same amount . Thus , the gain in conformational energy will be manifested as the free energy difference for the increase in binding affinity between K63-Ub2 and its open-state partner tUIM ( −0 . 89 ± 0 . 04 kCal/mol ) . On the other hand , the C1 and C2 closed states forfeit 0 . 71 ± 0 . 08 kCal/mol and 0 . 23 ± 0 . 09 kCal/mol in binding free energies , respectively . Our calculation indicated that only when the closed-state population decreases from ∼70% to ∼35% upon the mutation , in which C1 state population decreases from ∼52 . 5% to ∼22 . 5% and C2 state population decreases from ∼17 . 5% to ∼12 . 5% , could the binding free energy change be the same as the conformational energy change . Indeed , upon the point mutation , a larger decrease was observed for the PRE corresponding to the C1 state than the PRE for the C2 state ( Figure 5B ) , and the binding affinity of K63-Ub2 towards a C1 state ligand decreases more than the affinity towards a C2 state ligand ( Figures 6 , 7 and Figure 1—figure supplement 3 ) . Although the SAXS data have insufficient resolution to resolve multiple closed states , and may include contributions from high-order non-covalent oligomers , linearly combining the SAXS data calculated for the closed and open states at 70% and 30% and without further refinement , we were able to recapitulate the experimental SAXS data with a χ2 value of 1 . 24 ( Figure 2B ) . In this study , we have shown that about 70% of K63-Ub2 exists in the closed state , whereas only about 30% of the protein exists in the open state . Our findings disagree with many previous structural characterizations of K63-Ub2 ( Varadan et al . , 2004; Datta et al . , 2009; Dikic et al . , 2009; Komander et al . , 2009; Weeks et al . , 2009; Komander and Rape , 2012 ) , which reported that ligand-free K63-Ub2 exists only in the open state . Nevertheless , the closed-state population is in line with our previous finding that ubiquitin monomer non-covalently dimerizes with an apparent KD value of 4 . 9 ± 0 . 3 mM ( Liu et al . , 2012 ) . In the non-covalent dimer of ubiquitin monomer , a ubiquitin adopts an array of orientations in respect to the other . With a covalent linkage , the non-covalent interaction between ubiquitin becomes intra-molecular and restricted . Therefore , a significant population of diubiquitin should exist in the closed compact conformation regardless of the ubiquitin linkage . Indeed , studies have indicated that K48-Ub2 mainly exists in the closed state ( Cook et al . , 1992 , 1994; Phillips et al . , 2001; Varadan et al . , 2002; Eddins et al . , 2007; Hirano et al . , 2011; Ye et al . , 2012 ) . For ligand-free K63-Ub2 , however , only a recent single-molecule fluorescence resonance energy transfer ( smFRET ) study has provided direct experimental evidence and indicated that the protein exists in a closed state at a population of ∼75% ( Ye et al . , 2012 ) . Here , using paramagnetic NMR , SAXS , and mutational analysis , we found that the population of the closed state of ligand-free K63-Ub2 is ∼70% . But why have only open-state structures been reported for ligand-free K63-Ub2 ? A possible explanation is that the open-state structure is more readily captured owing to non-covalent interactions between neighboring unit cells , and becomes enriched during crystallization processes . Further , we have identified two distinct closed states , namely C1 and C2 , with different populations ( Figure 4A ) . The PRE NMR provides 1/r6 ensemble-averaged distance information ( Clore and Iwahara , 2009 ) . So the inverse problem is to determine the constituting conformational states that give rise to the ensemble-averaged PRE observables . Here by projecting the structures with spherical coordinates , we were able to visualize the distinct conformational states of ligand-free K63-Ub2 . In comparison , the smFRET study on ligand-free K63-Ub2 ( Ye et al . , 2012 ) measured just a single distance between the N-terminus of the distal unit and the C-terminus of the proximal unit , and could not distinguish multiple closed states or provide a structural description for each state . Similarly , SAXS analysis is unable to reveal how two nearly globular proteins are docked to each other in atomic detail ( Figure 2A ) . Our structural analysis based on the PRE revealed that the C1 and C2 closed states utilize different binding interfaces . Importantly , the ligand-free structures of K63-Ub2 are similar to the respective ligand-bound structures ( Kulathu et al . , 2009; Sato et al . , 2009a; Bosanac et al . , 2010 ) ( Figure 4B , C ) . This suggests a conformational selection mechanism for K63-Ub2 target recognition . This mechanism is further supported by conformational energy analysis . The inter-conversion between K63-Ub2 conformational states and ligand binding are coupled equilibria , and the population for each conformational state weights on the binding affinity towards a respective ligand . For the interaction between K63-Ub2 and tUIM , an open-state ligand , the change in binding free energy accompanying an E64RP mutation can be fully accounted for by the increase in the relative population of the open state . On the other hand , the change in conformational energy also accounts for the difference in binding affinities between wild type and mutant K63-Ub2 towards closed-state ligands TAB2/TAB3 NZF and A20 ZnF4 . Together , our ensemble structural refinement and mutational analysis revealed that ligand-free K63-Ub2 adopts at least three conformational states , including one open state and two closed states , each of which can accommodate cognate ligands . Closed compact structures have been reported for ligand-free diubiquitins with Lys48 ( Cook et al . , 1992 ) , Lys11 ( Matsumoto et al . , 2010; Castaneda et al . , 2013 ) , Lys29 , and Lys33 ( Kristariyanto et al . , 2015; Michel et al . , 2015 ) linkages . These structures are different from the C1 or C2 closed-state conformations of K63-Ub2 , and therefore are involved in different functions . As such , a covalent ubiquitin linkage dictates how the two subunits non-covalently interact with each other in a diubiquitin , and resulting quaternary arrangements encode specific cell signals . For K63-Ub2 , the open state recognizes tUIM of Rap80 and is involved in DNA damage repair , the C2 closed state recognizes the NZF domain of TAK1 binding proteins and is involved in the activation of NF-κB signaling , while the C1 state recognizes the ZnF4 domain of A20 and is involved in the termination of NF-κB signaling pathways ( Figure 8 ) . Constructed from repeating units of diubiquitins , a polyubiquitin should exist in a combination of quaternary structures of the diubiquitins and participate in diverse functions . 10 . 7554/eLife . 05767 . 027Figure 8 . Proposed mechanism for K63-Ub2 signaling . In the absence of a ligand , K63-Ub2 alternates between an open and two closed states . A specific ligand can be accommodated and bound to one of the three preexisting conformations , eliciting the downstream signal . DOI: http://dx . doi . org/10 . 7554/eLife . 05767 . 027 Human ubiquitin was cloned into a pET11a vector; single-point mutations including N25C , K48C , E64R , K63R , and 77D were introduced using QuikChange ( Stratagene ) . BL21 star cells were used for protein expression , and were grown in either LB medium ( for preparing unlabeled proteins ) or in M9-minimum medium ( for preparing isotope-enriched proteins ) . All ubiquitin proteins were purified on Sepharose SP and Sephacryl S100 columns ( GE Healthcare , Piscataway , NJ ) in tandem . Ligation between two ubiquitin molecules was based on the established protocol ( Pickart and Raasi , 2005 ) . Briefly , the proximal unit carrying a 77D mutation ( Asp77 appended at the C-terminus ) was mixed equimolarly with the distal unit carrying a K63R mutation , to thus ensure a single ligation product . With the addition of 2 . 5 µM E1 and 10 µM E2 ( Mms2/Ubc13 complex from yeast ) , 2 mM ATP , and 5 mM MgCl2 , ligation between two ubiquitin monomers was allowed to proceed for 5 hr at 30°C . The reaction was quenched with 5 mM DTT and 2 mM EDTA . The product was purified on a Sephacryl S100 column . Tandem ubiquitin-interacting motif ( tUIM ) from human protein Rap80 encompassing residues 79–124 , the fourth ZnF4 domain of A20 ( residues 590–635 ) , and NZF domain from human TAK1-binding protein 2 ( TAB2 , residues 663–693 ) were cloned into the pGEX vector . Since tUIM has no 280 nm UV absorption , a C121Y mutation was introduced . It has been shown that modification to this residue has no effect on the structure of tUIM or the interaction between tUIM and K63-Ub2 ( Sekiyama et al . , 2012 ) . All three proteins were expressed in BL21 star cells in LB medium . The proteins were purified off a GST affinity column . With the GST tag removed by TEV protease , the proteins were further purified through a Sephacryl S100 column . For the purification of ZNF4 and NZF , the buffer also contains 5 mM DTT and 50 µM ZnCl2 . All purified proteins were confirmed by electrospray mass spectrometry ( Bruker Daltonics , Germany ) . A single-point cysteine mutant of K63-Ub2 , either N25C or K48C in the distal unit , was reacted with a fivefold excess of S- ( 2 , 2 , 5 , 5-tetramethyl-2 , 5-dihydro-1H-pyrrol-3-yl ) methyl methanesulfonothioate ( MTS; from Toronto Research Chemicals , Canada ) or a fourfold excess of [N- ( 2-maleimido ) ethyl]ethylenediamine-N , N , N′ , N′-tetraacetate ( pre-incubated with a twofold excess of MnCl2 , EDTA-Mn2+ ) for 3 hr at room temperature . Unreacted probe was removed by desalting . The conjugation product was confirmed by mass spectrometry for a mass difference of 184 Da for MTS and 414 Da for EDTA conjugation . The NMR buffer contains 100 mM NaCl , 10 mM sodium acetate at pH 6 . 0 , and 10% D2O . The paramagnetic NMR data were collected on a 500 µM sample at 303 K on Bruker 850 MHz or 600 MHz instruments , each equipped with a cryogenic probe . Transverse relaxation rates of amide protons for the 15N-labeled subunit of K63-Ub2 protein were measured using the standard pulse scheme with a 4 ms delay between the two time points , Ta and Tb ( Iwahara et al . , 2007 ) . Inter-molecular PREs were measured for a 500 µM equimolar mixture of K63-Ub2 , with paramagnetic conjugation and isotope labeling on two separate proteins . The peak intensity at the second time point Tb is given as follows: ( 1 ) Ib=Iaexp[−R ( Tb−Ta ) ] , in which the relaxation rates R can be diamagnetic relaxation rates R2 , or ( R2 + Γ2 , inter ) for the equimolar mixture , or ( R2 + Γ2 , inter + Γ2 ) for the paramagnetic sample . Ia and Ib are the peak intensities at Ta and Tb . Using the equimolarly mixed sample as the PRE reference , the intra-molecular inter-subunit PRE 1H Γ2 value can be determined as follows: ( 2 ) Γ2=1Tb−TalnIinter ( Tb ) Ipara ( Ta ) Iinter ( Ta ) Ipara ( Tb ) . The same scheme was used for determining intra-molecular inter-subunit PREs for the E64RP mutant of K63-Ub2 . At 50 µM concentration , the percentage of K63-Ub2 dimer can be negligible and the Γ2 , inter term disappears . Thus the peak intensities for a single time point measurement scheme are defined as: ( 3 ) Idia , 50=I0exp[−R2T] , ( 4 ) Ipara , 50=I0exp[− ( R2+Γ2 ) T] , in which I0 is the intensity at the beginning of the pulse sequence , and T is ∼9 . 2 ms ( Iwahara et al . , 2007 ) . Owing to the Γ2 , inter term , peak intensities at 500 µM concentration are calculated as below: ( 5 ) Ipara , 500=I0exp[− ( R2+Γ2 , inter+Γ2 ) T] , ( 6 ) Iinter , 500=I0exp[− ( R2+Γ2 , inter ) T] , in which Ipara , 500 and Iinter , 500 are peak intensities for 500 µM paramagnetic sample and 500 µM equimolar mixture , respectively . Taking the ratios of the four equations above , the following relationship can be obtained: ( 7 ) Idia , 50Ipara , 50=exp ( Γ2T ) =Iinter , 500Ipara , 500 . Refinement against experimental restraints was conducted using Xplor-NIH ( Schwieters et al . , 2006 ) . A three-conformer representation for each paramagnetic probe at each conjugation site ( N25C or K48C ) was employed , to thus recapitulate the conformational flexibility for the paramagnetic probe ( Iwahara et al . , 2004 ) . With the protein backbone fixed , the dihedral angles for the rotatable bonds between the paramagnetic center and the protein backbone were optimized . Excluding structures with large van der Waals violations , intra-molecular and intra-subunit PREs were calculated for each structure , which were employed as the target values to restrain the spatial distribution of the paramagnetic probes in the subsequent calculations . The starting coordinates of each ubiquitin subunit were taken from Protein Data Bank ( PDB ) structure 1UBQ ( Vijay-Kumar et al . , 1987 ) , and two ubiquitin molecules were patched together with an isopeptide bond between the Lys63 side chain of the proximal unit and the C-terminus of the distal unit . While keeping the coordinates of the distal unit fixed ( conjugated with the paramagnetic probes at N25C and K48C sites , each in three-conformer representation ) , the proximal unit of K63-Ub2 was treated as a rigid body that can rotate and translate as a whole . The Lys63 side chain of the proximal unit and residues 72–76 of the distal unit were given full torsional freedom . To initiate the ensemble rigid-body simulated annealing , the coordinates for the proximal unit ( also including residues 72–76 of the distal unit ) were replicated to make additional members of the ensemble . Each ensemble member was subjected to random rotational and translational movement , and was allowed to overlap . The simulated annealing ensemble refinement was performed with a target function that comprises the inter-subunit PRE restraints for residues 1–71 , theoretical intra-subunit PRE restraints to confine the spatial distribution of the paramagnetic probes , the van der Waals repulsive term , and covalent energy terms . Square-well energy potential was used for the PRE term—no energy penalty was given when the back-calculated PRE value was within ± the experimental error of the target value . Residues that are completely broadened out in the paramagnetic spectrum were given a large PRE target value with the lower bound at 120 s−1 . An apparent PRE correlation time ( τc = 7 . 2 ns ) was estimated based on the rotational correlation time of the diamagnetic protein ( ∼7 . 6 ns for the closed state ) and the large electron relaxation time of the nitroxide spin radical ( ∼150 ns ) ( Tang et al . , 2007 ) . The population of the closed state was varied from 10% to 90% in 10% increments , which was implemented as a scaling factor for the back-calculated PRE value . In simulated annealing refinement , the PRE energy force constant was ramped from 0 . 01 to 1 kcal mol−1 s2 , and the temperature was cooled from 3000 to 25 K . For each combination of closed-state population and number of conformers , 160 structures were calculated . The agreement between the observed and calculated PRE rates was assessed with PRE Q-factor for both conjugation sites ( Iwahara et al . , 2004 ) . To better visualize the structures , a spherical coordinate system was constructed , with the origin set at the center-of-mass of the distal unit . Analysis of the buried interfaces and rendering of an atomic probability map ( Schwieters and Clore , 2002 ) were performed using Xplor-NIH ( Schwieters et al . , 2006 ) . Structure figures were illustrated using PyMOL ( The PyMOL Molecular Graphics System , Version 1 . 7; Schrödinger , LLC ) . Solution SAXS was performed at 303 K on the SAXSess mc2 platform ( Anton Paar , Graz , Austria ) equipped with a sealed-tube X-ray source and a CMOS diode array detector . The proteins were extensively dialyzed to the same buffer used for NMR , and the SAXS profile for the matching buffer was recorded for background subtraction . To remove high molecular weight aggregate , the protein samples were centrifuged at 15 , 000 rpm for 30 min prior to each experiment , and the upper portion of the supernatant ( the concentration was measured again at UV 280 nm ) was pipetted and loaded into a quartz cuvette . The sample was placed 306 mm from the detector with a slit width of 10 mm . The SAXS data were collected in 30 min increments for a total of 5 hr ( 10 hr for the 0 . 5 mM sample ) ; no difference was found between the first and last frames of SAXS data . The maximum distance of the particle ( Dmax ) was extrapolated from the paired-distance distribution function P ( r ) after indirect Fourier transformation of the I ( q ) scattering curve . The data collected at 1 mM were used for further analysis . The theoretical P ( r ) curve was calculated for each known structure of K63-Ub2 using CPPTRAJ in the AMBER 14 package ( UCSF ) . The bound ligand was removed if present , and any missing residues from the crystal structure were patched using Xplor-NIH ( Schwieters et al . , 2006 ) . With a water layer of ∼3 . 5 Å thickness padded to protein structure ( 291–349 explicit water molecules added depending on the PDB structure ) , the calculation of paired-distance distribution was performed at 0 . 5 Å resolution . The theoretical P ( r ) curve was smoothed using a 10-point spline function for plotting , and was normalized to a total area of 1 . The theoretical scattering I ( q ) profiles were calculated using CRYSOL ( Svergun et al . , 1995 ) without fitting to the experimental data , and were scaled by the first point of the experimental scattering data . The isothermal calorimetry ( ITC ) binding experiment was performed on a VP-ITC instrument ( GE Healthcare ) at 303 K . All protein samples were prepared in the same buffer as in the NMR experiments . A 20 µM sample of K63-Ub2 protein , either wild type or E64RP mutant , was placed in the reservoir . The titrant , 300 µM tUIM , TAB2 NZF , or A20 ZnF4 proteins , was titrated into K63-Ub2 drop-wise . Dilution heat was subtracted by titrating tUIM or NZF into a matching buffer and was measured for each experiment . After converting to heat per injection , the curves were fitted using a one-site binding model using Origin 8 . 1 software . All ITC titrations were performed at least four times . NMR titration of A20 ZnF4 was performed on the Bruker 850 MHz instrument at 303 K or 313 K , by titrating 50 , 100 , 150 , 250 , 350 , or 450 µM ZnF4 into wild type or E64RP mutant K63-Ub2 protein ( distal unit 15N-labeled ) . The exchange timescale at 313 K is faster than at 303 K , which allowed us to fit the binding isotherm from the chemical shift perturbations of residues at the ZnF4 binding interface . Based on the binding affinities measured by ITC and by NMR , the binding free energies differ by −0 . 89 ± 0 . 04 , 0 . 23 ± 0 . 09 , and 0 . 71 ± 0 . 08 kCal/mol upon E64RP mutation , for the bindings towards tUIM , NZF , and ZnF4 , respectively . The probability of open state Po can be defined , ( 8 ) Po=e−εo/kBT1+e−εo/kBT , in which εo is the free energy of the open state relative to the closed state , and kB is the Boltzmann constant . Thus , the energy for the open state can be calculated: ( 9 ) εo=kBT ln ( 1/Po−1 ) . The difference in conformational energy for the open state between the mutant and wild type K63-Ub2 can be calculated as below: ( 10 ) ΔΔGconformation= ( εo , mt−εo , wt ) NA=RT ln ( 1/Po , mt−1 ) −RT ln ( 1/Po , wt−1 ) , in which εo , mt and εo , wt are the conformational energies for the open state of the mutant and of the wild type , respectively , and NA is the Avogadro constant . As the overall population of the closed state decreases by ∼50% for the mutant , as estimated from the PRE , the population of the open state follows this relationship: ( 11 ) Po , mt=1−0 . 5× ( 1−Po , wt ) . As the differences in the binding affinities towards the respective ligands of K63-Ub2 are caused entropically , the difference in conformational free energy should be equal to the difference in binding free energy . Solving Equations 10 and 11 , one could determine that the open-state population increases from ∼30% for the wild type to ∼65% for the mutant , which corresponds to a conformational energy change of −0 . 88 kCal/mol . At the same time , the closed-state population drops from ∼70% to ∼35% . Further , the ratio between C1 and C2 closed states can be determined at about 3:1 for the wild type K63-Ub2 . Thus , upon the point mutation , the population of C1 state drops from 52 . 5% to 22 . 5 % , and the population of C2 state drops from 17 . 5% to 12 . 5 % , which correspond to conformational energy changes of 0 . 80 and 0 . 24 kCal/mol , respectively . Such changes are in line with the design of the E64RP mutant , and are also consistent with the relative decreases in PRE values .	Proteins can be tagged with other molecules that indicate what the cell should do with that protein . For example , proteins tagged with a small protein called ubiquitin—which is linked to other ubiquitin molecules to form ‘polyubiquitin’—may be destroyed or relocated within a cell . Like all proteins , a ubiquitin is made up of chains of amino acids . Specific amino acids form the linkages between individual ubiquitins to form a polyubiquitin , and the nature of these linkages influences the effect that a polyubiquitin has on the tagged protein . One linkage involves a lysine amino acid at position 63 ( known as Lys63 ) . This linkage is found in the polyubiquitin that is involved in repairing damaged proteins and relocating target proteins to a part of the cell where they are utilized for immune response . To perform these different roles , the polyubiquitin must be able to distinguish between a variety of target proteins . The shape that a protein takes on determines how it works , and most proteins constantly and rapidly switch between different shapes . Previous work suggested that the Lys63-linked polyubiquitin could only take on an elongated ‘open’ structure by itself . It was not clear whether the protein could take on a compact ‘closed’ structure without first binding to a target protein . Liu et al . used a technique known as nuclear magnetic resonance ( NMR ) to explore the high-resolution structures of the Lys63-linked ubiquitin chain when they are not bound to other proteins . The results showed that a large percentage of the protein was in a closed state , and that there were at least one open shape and two kinds of closed shapes . Liu et al . suggest that the shape of the unbound Lys63-linked ubiquitin chain determines what other proteins can be bound , and that the binding stabilizes the shape of the ubiquitin . This mechanism of binding is known as conformational selection . Further work is required to analyze whether other polyubiquitin chains recognize their partners in a similar manner .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics" ]	2015	Lys63-linked ubiquitin chain adopts multiple conformational states for specific target recognition
In mammals , the neocortical layout consists of few modality-specific primary sensory areas and a multitude of higher order ones . Abnormal layout of cortical areas may disrupt sensory function and behavior . Developmental genetic mechanisms specify primary areas , but mechanisms influencing higher order area properties are unknown . By exploiting gain-of and loss-of function mouse models of the transcription factor Emx2 , we have generated bi-directional changes in primary visual cortex size in vivo and have used it as a model to show a novel and prominent function for genetic mechanisms regulating primary visual area size and also proportionally dictating the sizes of surrounding higher order visual areas . This finding redefines the role for intrinsic genetic mechanisms to concomitantly specify and scale primary and related higher order sensory areas in a linear fashion . The mouse neocortex is patterned into functionally distinct fields that include the primary sensory areas ( visual , somatosensory and auditory ) , which receive modality-specific sensory inputs from thalamocortical axons ( TCAs ) originating from nuclei of the dorsal thalamus ( O'Leary et al . , 2013 ) . In the cortex , the connections of TCAs establish precise topographic representations ( or maps ) of the sensory periphery ( Krubitzer and Kaas , 2005; O'Leary et al . , 2013 ) . Primary areas are flanked by higher order sensory areas ( HO ) , which are interconnected with them and also contain topographic maps ( Felleman and Van Essen , 1991 ) . In mammals , this evolutionarily conserved general layout of the intra-areal neural circuits is responsible for the orderly progression of sensory information , sensory perception and the integration of higher cortical functions ( Felleman and Van Essen , 1991; Geschwind and Rakic , 2013; Krubitzer and Kaas , 2005; Laramée and Boire , 2014; O'Leary et al . , 2013 ) . Disrupted layouts of cortical area layouts appear to be associated with neurodevelopmental disorders including autism ( Courchesne et al . , 2011; Voineagu et al . , 2011 ) . Studies of cortical arealization , the mechanisms that pattern the neocortex into areas , have focused almost exclusively on the primary areas and have led to the prevailing model that genetic mechanisms intrinsic to the neocortex control arealization during early cortical development ( Greig et al . , 2013; Krubitzer and Kaas , 2005; O'Leary et al . , 2013 ) . For example , the graded expression of the homeodomain transcription factor Emx2 in neocortical progenitors determines the size and position of the primary visual area ( V1 ) in mice ( Bishop et al . , 2000; Hamasaki et al . , 2004 ) . Although higher order areas outnumber primary areas by roughly 10-fold ( Marshel et al . , 2011; Wang and Burkhalter , 2007 ) , mechanisms that specify them and define their proportions relative to primary areas have yet to be explored . To investigate the impact of altered primary area size on higher order areas , we have used the cortical visual area V1 as a model . Previous studies have shown that genetic manipulation of patterning genes , including Fgf17 and Emx2 , results in altered V1 size ( Cholfin and Rubenstein , 2007; Hamasaki et al . , 2004 ) . Here we have analyzed transgenic mice that overexpress Emx2 ( ne-Emx2 ) and show area patterning defects including a V1 that is ~150% of the normal size , while retaining overall normal neocortex size ( Hamasaki et al . , 2004 Leingärtner et al . , 2007 ) . By revealing the targeting patterns of TCAs projecting from thalamic sensory nuclei into the cortex ( Fujimiya et al . , 1986 ) , the perimeters of primary sensory areas and the border between the neocortex and entorhinal cortex ( ECT ) can be visualized by serotonin ( 5HT ) staining using a single postnatal day ( P ) 7 tangential section of the flattened cortical hemisphere ( Figure 1A ) . The staining shows that , in addition to the previously reported enlarged V1 ( Hamasaki et al . , 2004 , ; Leingärtner et al . , 2007 ) , the cortical tissue that is nested between V1 and the surrounding primary areas ( primary somatosensory cortex: S1 , auditory areas: Aud ) and the ECT laterally appears qualitatively larger in ne-Emx2 brains , when compared to wildtype ( wt ) sections ( Figure 1A ) . We have defined this caudal cortical territory , which lies outside of V1 , S1 , and the auditory areas and shows no or weak 5HT staining as a joint higher order cortical area complex and have termed it HO-5HT . The 5HTstaining revealed that this region contains the higher order visual areas surrounding V1 ( Wang and Burkhalter , 2007 ) , the retrosplenial cortex ( RSC ) medially , and the ventral posterior temporal cortex laterally . The accurate distribution of staining across cortical layers can only be estimated using tangential sections . However , using P7 sagittal section , we confirmed in layer 4 that the caudal 5HT-positive cortical area ( corresponding to V1 ) and the anteriorly adjacent 5HT-negative area between V1 and S1 ( corresponding to HO-5HT ) appears larger in ne-Emx2 brains than in wt ones ( Figure 1B ) . Next , we labeled TCAs projecting to V1 by filling the dLG with crystals of the lipophilic neuronal tracer DiI . On the P7 sagittal sections that were derived from five different medial to lateral levels , anterograde DiI labeling in the cortex revealed that TCAs from the dLG terminate in a smaller region in wt than in ne-Emx2 brains ( Figure 1—figure supplement 1 ) . Across genotypes , the DiI staining revealed a sharp border with adjacent cortical tissues that did not receive TCAs input from the dLG ( Figure 1—figure supplement 1 ) . This finding is consistent with the 5HT staining and indicates a well-defined border between V1 neighboring higher order areas that is anteriorly shifted in ne-Emx2 brains . 10 . 7554/eLife . 11416 . 003Figure 1 . Increased V1 and higher order sensory area sizes in ne-Emx2 cortices ( A ) Serotonin ( 5HT ) staining on postnatal day ( P ) 7 tangential sections of the flattened cortex reveals targeting patterns of TCAs revealing primary sensory area borders and the border of the neocortex to the ECT . 5HT staining is not detectable in the region containing the retosplenial cortex and the higher order sensory areas surrounding V1 ( HO-5HT ) . In Emx2-overexpressing brains ( ne-Emx2 ) , V1 and HO-5HT appear larger ( compare dotted outlines in higher magnification images ) , compared to wt brains . ( B ) Targeting of TCAs in cortical layer 4 ( L4 ) was revealed on P7 sagittal cortex sections by 5HT staining , whereas L4 genetic area borders were revealed by in situ hybridization for Rorb . In ne-Emx2 brains , the V1 border shifts anteriorly . Higher order areas surrounding V1 are characterized by low 5HT/Rorb staining ( between arrowheads , HO-5HT and HO-Rorb ) , which in ne-Emx2 brains appear overall larger ( compare area between arrowheads ) . ( C ) In L5 , an expansion ( see arrows ) of corticotectal projection neurons ( retrogradely labeled by DiI injections into the superior colliculus ) is apparent in ne-Emx2 brains , to the expense ( see dotted lines ) of L5 corticospinal projection neurons ( retrogradely labeled by DiI injections into the pyramidal decussation ) . Main axes: A: anterior; M: medial; F/M: frontal/motor cortex; S1: primary somatosensory cortex; Aud: auditory areas; V1: primary visual cortex , ECT: entorhinal cortex . 5HT , serotonin; L5 , cortical layer 5; TCAs , thalamocortical axons; wt , wildtype . DOI: http://dx . doi . org/10 . 7554/eLife . 11416 . 00310 . 7554/eLife . 11416 . 004Figure 1—figure supplement 1 . Anterior shifted cortical boundary formed by TCAs from the dLG in ne-Emx2 brains . ( A ) In P7 brains , DiI ( red dye ) crystals were inserted into the thalamic dorsal lateral geniculate nucleus ( dLG ) to label the projections of TCAsinto the primary visual cortex ( V1 ) . In sagittal sections at five different medial to lateral levels , TCAs in Emx2-overexpressing brains ( ne-Emx2 ) extend more anteriorly , compared to wt brains ( compare dotted lines ) . Across genotypes , DiI labeling shows a sharp border to DiI-negative cortical tissues . ( B ) Images show representative DiI staining close to the injection sites in the dLG revealing robust and comparable dye filling in the dLG in wt and ne-Emx2 brains . dLG , dorsal lateral geniculate nucleus; TCA , thalamocortical axons; wt , wildtype . DOI: http://dx . doi . org/10 . 7554/eLife . 11416 . 004 Cortical areas can also be distinguished by area-specific gene expression patterns , which overlap with anatomical area borders and shift similarly when area patterning is disrupted ( O'Leary et al . , 2013 ) . For example , Rorb expression is strongly induced by thalamic input to primary areas ( Jabaudon et al . , 2012 ) like S1 and V1 but is low in areas that do not receive their major inputs from the principal thalamic sensory nuclei , such as cortical higher order areas surrounding V1 ( Chou et al . , 2013; Wang and Burkhalter , 2007 ) . In situ hybridization ( ISH ) on sagittal sections adjacent to 5HT-stained ones revealed sharp Rorb gene expression borders between areas in layer 4 . Notably in ne-Emx2 brains , the high-to-low Rorb expression border is located more anteriorly , and the area showing low Rorb expression and resembling HO-5HT ( Chou et al . , 2013 ) is larger than in wt sections ( Figure 1B ) . This reveals that characteristic molecular markers that delineate the borders between V1 and surrounding higher order areas remain expressed at normal levels , but their sharp expression borders shift anteriorly in ne-Emx2 . Projection neurons in layer 5 , which extend axons into subcortical targets , are similarly determined by a molecular code ( Greig et al . , 2013 ) . We therefore predicted that the areal shifts in ne-Emx2 brains would be accompanied by corresponding changes in layer 5 output projections . We labeled two distinct types of layer 5 subcerebral projection neurons by inserting DiI crystals either into the superior colliculus , which labels corticotectal projections from V1 and HO , or else into the pyramidal decussation , which labels corticospinal projections from the frontal cortex and S1 ( Greig et al . , 2013; Zembrzycki et al . , 2015 ) . We found that layer 5 corticotectal projections extended more anteriorly in ne-Emx2 sagittal sections . Vice versa , the layer 5 corticospinal projections extended less posteriorly in ne-Emx2 brains . These staining patterns are consistent with an altered balance of projection neuron identity in layer 5 ( Greig et al . , 2013; Zembrzycki et al . , 2015 ) and an overall expansion of visual areas , demonstrating that areal patterning changes in ne-Emx2 brains are not limited to the cortical layers that receive thalamic input . These findings complement previous reports describing Emx2 patterning functions ( Bishop et al . , 2000; Hamasaki et al . , 2004; Leingärtner et al . , 2003; Leingärtner et al . , 2007 ) and indicate for the first time that V1 and higher order area sizes are altered concomitantly in ne-Emx2 brains at the level of area-specific connectivity and gene expression in multiple cortical layers . Taken together , our results suggest that changes in primary area size are paralleled by similar changes in higher order area size . It is commonly assumed that areal patterning changes also alter area-specific functional neuronal properties and topographic sensory maps , but this has never been demonstrated conclusively . Therefore , to compare functional neuronal properties of an enlarged visual cortex to a normal-sized one , we have used Fourier intrinsic signal optical imaging to construct topographic visual response maps to light bars that were moved across the visual field of the retina ( up and down: elevation maps; left to right: azimuth maps ) ( Kalatsky and Stryker , 2003 ) . Visual responses in V1 of wt and ne-Emx2 mice produced intrinsic signal maps that were indistinguishable in strength , and the axes of azimuth and elevation were organized in the same way in all tested brains ( Figure 2A ) , revealing that functional topographic organization of the visual cortex was intact . However , the representations of elevation and azimuth were expanded in ne-Emx2 animals , and their retinotopic maps were overall larger ( elevation: 138% ± 8 . 7% of wt; azimuth: 143% ± 8 . 2% of wt ) . For example , the green region in the response maps representing ~20 to ~30 degrees of elevation/azimuth is clearly enlarged in ne-Emx2 brains , compared to wt brains ( Figure 2A ) . To investigate the relationship between the location and size of the V1 functional response area and the histochemically delineated V1 , as indicated by 5HT staining , multiple injections of DiI were placed lining the border of the V1 optical response map after the imaging procedure . On 5HT-stained , flattened tangential cortical sections , the DiI injection sites were found in all cases to be located near the border of the 5HT staining in V1 , confirming the overall enlarged V1 perimeters in ne-Emx2 brains compared with the wt brains ( Figure 2B ) . This shows that the 5HT-stained V1 area accurately corresponds to the intrinsic functional V1 map , suggesting that enlarged HO in ne-Emx2 brains have not acquired ectopic V1-like functional properties . 10 . 7554/eLife . 11416 . 005Figure 2 . Enlarged functional V1 topographic maps in ne-Emx2 mice . ( A ) Fourier intrinsic signal optical imaging reveals topographic visual response maps to light bars that were moved across the visual field of the retina ( up and down: elevation maps; left to right: azimuth maps ) . Visual responses in V1 produced intrinsic signal maps that were indistinguishable in strength and the color-coded axes of azimuth and elevation were organized in the same way in all tested brains ( wt: n = 6 , ne-Emx2: n = 6 ) . The elevation/azimuth representations were expanded in ne-Emx2 animals , revealing that overall their V1 retinotopic maps were larger . ( B ) 5HT staining performed on flattened tangential cortex section reveals cortical area borders including V1 . On representative images ( n = 6 per genotype ) , the red dots lining the perimeter of the 5HT-stained V1 indicates DiI injection sites that were made after the recordings adjacent to the border of the derived V1 intrinsic response maps , determined by Fourier intrinsic optical imaging . ( C ) Schematics depict caudal cortical sensory areas and main sensory thalamus divisions . In wt brains ( n = 15 ) , cortical dual tracer injections ( red tracer ( DiI ) injected around V1/HO border; green tracer ( DiD ) into HO; injection site location ( arrowheads ) was identified by 5HT staining ) showed retrogradely labeled red cells in the dLG and the PO , whereas green labeled cells were only present in the PO . Dotted lines show that dual tracer injections in ne-Emx2 brains ( n = 17 ) were administered at more anterior coordinates ( red tracer into V1; green tracer around the V1/HO border ) compared with wt brains . In ne-Emx2 brains , retrogradely labeled red cells were apparent in the dLG , whereas green cells were present in the dLG and the PO , revealing normal thalamocortical connectivity patterns , but an anterior shifted V1/HO border in ne-Emx2 brains . 5HT , serotonin; dLG , dorsal lateral geniculate nucleus; PO , posterior thalamic nucleus; S1 , primary somatosensory cortex; VP , ventroposterior nucleus; V1 , primary visual cortex; wt , wildtype . DOI: http://dx . doi . org/10 . 7554/eLife . 11416 . 005 To further characterize the shifted border between visual areas in ne-Emx2 animals , we used additional neuronal tracing approaches . Stereotypically , V1 is connected with the dLG , whereas the HO areas are wired to the posterior thalamic nucleus ( PO ) ( Leyva-Díaz and López-Bendito , 2013; López-Bendito and Molnár , 2003 ) . To first label axonal connections between the cortex and the thalamus in wt brains , we administered dual tracer injections into locations that approximate to HO ( DiD: green dye ) and another injection around the approximated border area between V1 and HO ( DiI: red dye ) . After diffusion of the tracers , we performed 5HT staining on flattened cortex sections to identify the areas in which the injections were administered and analyzed the patterns of retrograde dye labeling on coronal sections of the thalamus . In representative cases ( Figure 2C ) where 5HT staining confirmed that DiI was injected at the border between V1 and HO and DiD was injected into HO , retrogradely labeled green DiD cells were apparent in the dLG and the PO . Conversely , red DiI cells were only labeled in the dLG . In ne-Emx2 brains , we administered similar dual tracer injections: DiI was targeted to V1 and a DiD injection was administered around the approximate border between V1 and HO . Due to their enlarged V1 , all ne-Emx2 injections were administered at more anterior coordinates than in wt brains ( compare dashed lines in Figure 2C ) . In representative cases ( Figure 2C ) where 5HT staining confirmed that the DiI injection was administered into V1 and DiD was injected around the V1/HO border , red cells were found in the dLG , whereas green-labeled cells were apparent in the PO and the dLG . The dual tracings in wt and ne-Emx2 brains are consistent with the predicted connectivity of cortical neurons around the injection sites ( Leyva-Díaz and López-Bendito , 2013; López-Bendito and Molnár , 2003 ) . Although located more anteriorly in ne-Emx2 brains , the subcortical connectivity patterns around the V1/HO borders were similar , demonstrating that these neurons show connectivity patterns that are consistent with their intrinsic areal identity and not their topographic location on the cortical sheet . Taken together , our results indicate that increased V1 size in ne-Emx2 brains is accompanied by a concomitant enlargement and anterior shift of HO . To define the individual magnitudes of the V1 and HO size increases in ne-Emx2 brains , we next used gene expression domains as molecular markers delineating visual areas and quantified them ( Figure 3 ) . An accurate assessment of area sizes using flattened and/or sectioned cortical tissues could potentially be hampered by imperfect flattening of the tissues or by cutting artifacts . Therefore , we have used RNA in situ hybridization on intact whole brains ( whole mount in situ hybridization: WMISH ) at P7 for quantification purposes , which has the advantage that quantifications can be made using single images without sectioning and artifacts that may arise from such tissue processing . We first used a set of two marker genes , Unc5d and Igfbp5 , whose expression delineates V1 at P7 ( Chou et al . , 2013 ) . The gene expression domains on WMISH-stained brains were outlined and their sizes quantified as a measure of V1 area size . The mean value of wt brains was defined as 100% and the area size percentages of ne-Emx2 brains displayed accordingly as ‘percent of wt’ ( Figure 3A ) . V1 gene expression domains of both markers were larger in ne-Emx2 brains ( Unc5d-V1: 148% ± 6 . 1% , p = 0 . 0003; Igfbp5-V1: 148 ± 4 . 5% , p < 0 . 0001 ) . The magnitude of the increased in V1 size labeled genetically in ne-Emx2 brains is comparable to V1 area measurements derived from 5HT-stained P7 flattened cortical sections ( Figure 3A: 5HT-V1: 142 . 1% ± 3 . 1% , p < 0 . 0001 ) , indicating that molecular markers on whole brains reliably delineate V1 and can therefore be used to quantify and compare area sizes between samples and mouse lines . 10 . 7554/eLife . 11416 . 006Figure 3 . Proportionally increased V1 and HO sizes in ne-Emx2 cortices . ( A ) Schematic shows sensory area outlines in the caudal neocortex ( 12 ) . WMISH with the molecular V1 marker genes Unc5d ( wt: n = 5 , ne-Emx2: n = 6 ) and Igfbp5 ( wt: n = 11 , ne-Emx2: n = 6 ) at P7 highlights increased V1 size in ne-Emx2 brains using whole un-sectioned brains . Quantification of V1 size using 5HT-stained P7 flattened cortical sections similarly reveals larger V1 sizes in ne-Emx2 brains ( n = 11 ) , compared to wt brains ( n = 13 ) . ( B ) WMISH for molecular markers that label both V1 and HO ( dotted outlines: Cdh8 , Lmo4: high expression in HO , lower in V1; for each probe and genotype n = 6 ) reveal that V1 as well as HO sizes in ne-Emx2 are larger compared with wt brains . Cdh8 is not expressed around the anteromedial edge of V1 ( arrowheads ) . Quantifications in Figures 3 and 4 show mean values as percent of wt , error bars indicate standard error of the mean; asterisks highlight statistical significance according to unpaired to t-test . 5HT , serotonin; S1 , primary somatosensory cortex; WMISH , whole mount in situ hybridization; V1 , primary visual cortex; wt , wildtype . DOI: http://dx . doi . org/10 . 7554/eLife . 11416 . 00610 . 7554/eLife . 11416 . 007Figure 3—figure supplement 1 . Lmo4 expression delineates primary and higher order cortical area boundaries in whole mount brains . ( A ) WMISH for Lmo4 using P7 brains reveals gene expression borders that delineate cortical regions . Their approximated outlines are annotated in the lower panels ( A-D: dotted lines ) . Using images showing a lateral view of the brain , auditory areas ( Aud ) can be delineated by a ring of strong Lmo4 expression . Similarly , a line of strong Lmo4 expression delineates the border between the neocortex and theECT . The perimeter of the primary visual cortex ( V1 ) cannot be clearly seen in the lateral view . ( B ) In the dorsolateral view , Lmo4 expression readily reveals the ECT and auditory area outlines . In addition , the V1 dimensions and the caudal border of the primary somatosensory cortex ( S1 ) are clearly identifiable . ( C ) In the dorsal view , primary cortical area borders are identifiable using Lmo4 staining . In addition , lower Lmo4 expression domains in the caudal cortex appear to outline a cortical region that includes higher order visual areas around V1 . In addition , this region ( HO ) contains adjacent cortical areas like the ventral posterior temporal cortex up to the border of the auditory areas and ECT but seem to exclude the retrosplenial cortex ( RSC ) medially , which appears to be delineated by a narrow higher expression stripe of Lmo4 ( arrowheads ) . ( D ) Lypd1 expression is low in F/M cortex , S1 and V1 , but very strong around the medial cortical pole , where its expression appears to overall overlap with the RSC . Comparing Lypd1 and Lmo4 expression in the medial cortex suggests that the outlined HO in ( C ) and in Figures 3–4 do not contain much of the RSC . The Lypd1 gene expression domain around the RSC in ne-Emx2 brains appears enlarged compared with wt brains . Aud , auditory areas; ECT , entorhinal cortex; F/M , frontal/motor; RSC , retrosplenial cortex; S1 , primary somatosensory cortex; V1 , primary visual cortex; wt , wildtype . DOI: http://dx . doi . org/10 . 7554/eLife . 11416 . 00710 . 7554/eLife . 11416 . 008Figure 3—figure supplement 2 . Lmo4 and Cdh8 expression marks primary and caudal extrastriate cortical regions . ( A ) Primary cortical area borders are apparent by Lmo4 staining on WMISH brains at P7 . ( A-B: raw images in upper panels , annotated images in lower panels ) . The low Lmo4 expression domain in the caudal cortex appears to outline a joint HO that contains higher order visual areas around V1 and the ventral posterior temporal cortex up to the borders to the auditory areas and ECT laterally , but seem to exclude the RSC medially . In ne-Emx2 brains the Lmo4-positive HO complex overall appears larger than compared to wt brains . ( B ) Low Cdh8 expression labels V1 , which is surrounded by a domain of higher Cdh8 expression labeling higher order visual areas between V1 and S1 in the medial cortex up to the borders to the auditory areas laterally and the ECT cortex caudally , respectively . In ne-Emx2 brains the Cdh8-positive HO complex overall appears larger compared to wt brains . Aud , auditory areas; ECT , entorhinal cortex; F/M , frontal/motor; RSC , retrosplenial cortex; S1 , primary somatosensory cortex; V1 , primary visual cortex; WMISH , whole mount in situ hybridization; wt , wildtype . DOI: http://dx . doi . org/10 . 7554/eLife . 11416 . 008 We next have used additional markers to quantify higher order area sizes ( Figure 3B ) . Previous studies have parsed higher order visual areas using neuroanatomical tracers ( Wang and Burkhalter , 2007 ) ( see also schematic in Figure 3 ) and have revealed genes that are expressed at different levels in V1 and higher order visual areas ( Chou et al . , 2013 ) . For example , Cdh8 and Lmo4 expression is higher in the area surrounding V1 , where higher order visual areas are located ( Chou et al . , 2013; Marshel et al . , 2011; Wang and Burkhalter , 2007 ) . The domains of high Cdh8 and Lmo4 expression appear to label higher order visual areas uniformly ( Chou et al . , 2013 ) , without revealing subdivisions between them ( compare schematic in Figure 3 showing approximate location and outline of higher order visual areas as identified by Wang and Burkhalter , 2007 to Cdh8 and Lmo4 gene expression domains around V1 ) . On P7 WMISH-stained wt brains , we quantified the V1 and HO sizes in the medial cortex on the basis of low and high gene expression domains ( see dotted lines in Figure 3 , Figure 3—figure supplement 2 ) . Anatomically , these gene expression domains surrounding V1 , which show much stronger staining compared with V1 , extend anteriorly up to the S1 border , laterally to the border of the auditory areas and the ECT and medially up to the border to the RSC , respectively ( Figure 3—figure supplements 1 , 2 ) . Hence , compared with the above-mentioned HO complex that was identified using 5HT staining ( Figure 1 ) , the higher order area complex labeled by Cdh8 and Lmo4 relates to a smaller cortical region that more closely relates to higher order visual areas , but excludes the RSC . The size ( Cdh8-V1: 145 . 3 ± 7 . 4% , p < 0 . 0001; Lmo4-V1: 146 . 6% ± 6 . 7% , p = 0 . 0005 ) and shape of the gene expression domains in V1 were similar in Cdh8- and Lmo4-stained brains . Similarly , the gene expression domains nested around V1 largely overlapped between the two probes . The only apparent difference between them is around the anteromedial edge of the higher order visual areas ( Wang and Burkhalter , 2007 ) , where Cdh8 is expressed at much lower levels compared to more lateral regions around V1 across genotypes ( arrowheads in Figure 3B ) . The wt values of the measurements were again defined as 100% . The overall shapes of the two HO marker gene domains were similar and the sizes larger in ne-Emx2 brains compared with wt brains ( Cdh8- HO: 145 . 7 ± 6 . 4% , p = 0 . 0015; Lmo4- HO: 144 . 9 ± 3 . 8% , p = 0 . 00157 ) . The analysis of different area-specific sets of marker genes , either showing unique expression in V1 , or discernable expression levels between visual areas , revealed an increase in visually-related HO in ne-Emx2 brains that was proportionate to the V1 size increase . The extrastriate areas that we have identified on the basis of 5HT staining ( Figure 1 ) included the RSC , which is not a higher order visual area ( Garrett et al . , 2014; Marshel et al . , 2011; Vann et al . , 2009; Wang and Burkhalter , 2007 ) raising the possibility that only related primary and higher order areas ( e . g . vision ) could scale proportionately . To test this possibility , we have used WMISH of Lypd1 on P7 wt and ne-Emx2 brains as a specific marker labeling the caudomedial cortex , where the RSC is located ( Figure 3—figure supplement 1 ) . We found that the specific Lypd1 gene expression domain in the caudomedial cortex is significantly enlarged in ne-Emx2 brains ( 114 . 3 ± 5 . 2% , p = 0 . 0225 , n = 4 ) , compared with wt brains . This size increase is not proportionate to the size increases of V1 and the higher order visual area complex labeled by Cdh8 and Lmo4 in ne-Emx2 brains ( Figure 3 ) suggesting that increased V1 size is specifically accompanied by a proportionate size increase of related higher order visual areas . To test if related HO size matches V1 size only when it is larger than normal , or if primary area size bi-directionally is accompanied by according scaling of related higher order areas , we next analyzed HO sizes excluding the RSC in brains with a smaller than normal V1 ( Figure 4 ) . Constitutive Emx2 mutant mice have an overall smaller brain and visual cortex , but homozygous mutants die perinatally ( Bishop et al . , 2000 ) , preventing the analysis of cortical areas , which arise at later stages . To overcome this limitation , we generated a novel mouse line with floxed Emx2 alleles ( Figure 4—figure supplement 1 ) , allowing conditional inactivation of Emx2 . We crossed Emx2 floxed mice with Emx1-IRES-Cre expressing mice ( Gorski et al . , 2002 ) to generate conditional , cortex-specific Emx2 mutant mice . These cKO mice are viable , fertile and have an anatomically normal neocortex ( Figure 4—figure supplement 2 ) . Confirming the prediction that reduced Emx2 expression levels in cortical progenitors would lead to smaller visual areas , cKO brains show areal patterning changes ( Figure 4—figure supplement 3 ) that are similar to those previously reported in heterozygous Emx2 mutant brains ( e . g . larger frontal cortex ) ( Hamasaki et al . , 2004 ) , but are opposite to those apparent in ne-Emx2 brains ( e . g . smaller frontal cortex ) ( Hamasaki et al . , 2004; Leingärtner et al . , 2007 ) . As in ne-Emx2 brains , V1 in cKO brains was characterized using 5HT staining and DiI injections into the dLG ( Figure 4—figure supplement 4 ) , revealing that V1 in cKO is greatly reduced in size . 10 . 7554/eLife . 11416 . 009Figure 4 . Proportionally decreased V1 and HO sizes in cKO cortices . WMISH for V1 ( A: Unc5d; wt: n = 5 , ne-Emx2: n = 6 , Igfbp5; wt: n = 11 , ne-Emx2: n = 8 ) or V1 and HO marker genes ( B: Cdh8; wt: n = 6 , ne-Emx2: n = 6; Lmo4; wt: n = 5 , ne-Emx2: n = 5 ) at P7 conversely reveals decreased sizes ( ~70% of wt size ) of V1 and HO in brains that were derived from Emx1-IRES-Cre-mediated cortex-specific conditional Emx2 mutant brains ( cKO ) , compared with wt brains . Quantification of V1 size using 5HT staining on P7 flattened cortical sections ( A: wt: n = 15 , cKO: n = 10 ) reveals similar reductions of V1 size in cKO brains . ( C ) The ratio between quantified V1 and HO sizes derived from WMISH-stained brains with decreased ( cKO ) , normal ( wt ) , and increased ( ne-Emx2 ) V1 sizes demonstrates linear scaling of HO size in response to bi-directional changes of V1 size . S1 , primary somatosensory cortex; V1 , primary visual cortex; WMISH , whole mount in situ hybridization; wt , wildtype . DOI: http://dx . doi . org/10 . 7554/eLife . 11416 . 00910 . 7554/eLife . 11416 . 010Figure 4—figure supplement 1 . Generation of Emx2-floxed mice and confirmation of cortex specific Emx2 deletion . ( A ) Targeting strategy of the Emx2 locus: Blue shapes indicate exons of the Emx2 gene . Red and yellow triangles indicate LoxP and FRT sites . RV , EcoRV; H , HindIII indicate restriction enzyme cutting sites . The targeting construct included LoxP sites flanking Emx2 exons 1 and 2 followed by a FRT-site-flanked PGK-Neo cassette . ( B ) Southern blot hybridization of wt and Emx2 floxed heterozygous ( +/– ) embryonic stem cell clones with probes A , B , and C ( denoted by green arrows in A ) . Blots of HindIII digested genomic DNA that were hybridized with probe A revealed a 14kb Emx2 wt band and a 10 kb Emx2 floxed band . EcoRV digestion and hybridization with probe B revealed a 5 . 5kb Emx2 wt band and a 1 . 7 kb Emx2 floxed band . EcoRV digestion and hybridization with probe C revealed a 8 . 5kb Emx2 wt band and a 7 . 5kb Emx2 floxed band . ( C ) Genomic DNA from wt and Emx2 heterozygous ( +/– ) embryonic stem cell clones was amplified by PCR using combinations of the primers ( F1 , R1 , F2 , R2 , genomic location indicated in A ) revealing a Emx2 wt PCR product ( ~300 bp ) and a Emx2 floxed PCR product ( ~400 bp ) . ( D ) WMISH at E11 using a Emx2 probe showed that Emx2 expression was intact in the entire embryo in homozygous Emx2 floxed embryos ( Emx2fl/fl: considered as wt control ) . Conversely , WMISH using Emx1-IRES-Cre-mediated conditional Emx2 knock out animals ( Emx2fl/fl- Emx1-IRES-Cre+: cKO ) reveals specific deletion of Emx2 activity from cortical progenitors in the dTel , while leaving Emx2 expression in the rest of the embryo intact . dTel , dorsal telencephalon; PCR , polymerase chain reaction; WMISH , whole mount in situ hybridization; wt , wildtype . DOI: http://dx . doi . org/10 . 7554/eLife . 11416 . 01010 . 7554/eLife . 11416 . 011Figure 4—figure supplement 2 . Normal cortical neuroanatomy in cKO brains . On P7 sagittal sections , Nissl staining was performed to reveal neuroanatomy . Major forebrain structures are annotated in the wt section . In addition , ISH for the layer-specific marker genes Cux2 ( layers 2-4 ) and Tbr1 ( layer 6 ) was carried out . All stains reveal comparable cortical neuroanatomy and comparable cortical layering across genotypes . The only apparent neuroanatomical difference in cKO brains is that the hippocampus appears smaller in cKO brains ( arrowhead ) compared with wt brain . CTX , cortex; HC , hippocampus; ISH , in situ hybridization; OB , olfactory bulbs; STR , striatum; wt , wildtype . DOI: http://dx . doi . org/10 . 7554/eLife . 11416 . 01110 . 7554/eLife . 11416 . 012Figure 4—figure supplement 3 . Area patterning changes following cortex-specific deletion of Emx2 . ( A ) Tangential flattened sections of 5HT-immunostained cortices ( wt: n = 15 , cKO: n = 10 ) reveal sensory areas and their borders . Total cortical surface area was reduced by 11 . 70 ± 5 . 25% in cKO brains , without being statistically significant ( p > 0 . 05 ) . Further area measurements were normalized to cortical surface area , generating size ratios to facilitate comparisons of area measurements derived from different experiments and mouse lines . Mean ratios of wt brains were defined as 100% and variance calculated using standard error of the mean . Quantified area parameters are depicted in the schematics , below the individual graphs . ( B ) F/M cortex is larger in cKO brains ( F/M area ratio: 131 . 14 ± 3 . 22% ) . S1 size is not affected ( PMBSF area ratio: 98 . 03 ± 4 . 37% ) , but overall , indicated by the relative location of PMBSF barrel ‘C3’ ( C3 length ratio: 84 . 44 ± 1 . 16% ) , S1 was located more posteriorly in cKO specimens compared with wt ones . ( C ) The size of V1 ( V1 area ratio: 68 . 50 ± 2 . 95%; V1 length ratio: 72 . 07 ± 4 . 39% ) was reduced and its relative position shifted medially ( V1 medial shift ratio: 82 . 44 ± 7 . 29% ) in cKO brains compared with wt brains . 5HT , serotonin; F/M , frontal/motor; PMBSF , posterior medial barrel sub field; wt , wildtypeDOI: http://dx . doi . org/10 . 7554/eLife . 11416 . 01210 . 7554/eLife . 11416 . 013Figure 4—figure supplement 4 . Posteriorly shifted V1 border in cKO cortices . Adjacent P7 sagittal sections show thalamocortical projections from the dLG to V1 , which were either labeled by inserting DiI crystals into the dLG ( counterstained with DAPI ) , or by 5HT immunostaining . Arrowheads denote the anterior border of the V1 area that is innervated by TCAs from the dLG . Both TCA labeling methods reveal that this sharp border is shifted posteriorly in cKO brains compared with wt brains . 5HT , serotonin; DAPI , 4' , 6-diamidino-2-phenylindole; S1 , primary somatosensory cortex; TCAs , thalamocortical axons; V1 , primary visual cortex . DOI: http://dx . doi . org/10 . 7554/eLife . 11416 . 013 To complement the quantification of V1 and its related HO sizes in ne-Emx2 brains , we next have used cKO brains to perform WMISH with both sets of marker genes noted above ( Figure 3 ) and measured their sizes ( Figure 4 ) . Measurements of the molecular V1 marker domains ( Unc5d-V1: 67 . 1 ± 2 . 6% , p = 0 . 00029; Igfbp5-V1: 68 ± 3 . 3% , p = 0 . 00017 ) , as well as the V1 expression domains of Cdh8 ( 67 . 7 ± 4 . 5% , p < 0 . 0001 ) and Lmo4 ( 68 . 6 ± 0 . 8% , p < 0 . 0001 ) revealed that the molecularly defined V1 in cKO was smaller than in wt brains . These reductions matched the reduced V1 in 5HT-stained flattened cortex sections ( Figure 4A: 5HT-V1: 68 . 5 ± 3% , p < 0 . 0001 ) . Subsequent quantification of the gene expression domains of Cdh8 and Lmo4 surrounding V1 ( Figure 4B: Cdh8- HO: 66 . 8 ± 2 . 9% , p < 0 . 0001; Lmo4- HO: 69 . 3 ± 2 . 4% , p < 0 . 0001 ) revealed that the cortical region that contains visually-related HO was also reduced in cKO brains to a degree proportional to the reduction in V1 size . These data demonstrate that when V1 size is reduced , related HO size is reduced to a similar extent . In order to reveal a correlation between primary and related higher order area size between brains with larger and smaller than normal visual areas , we calculated the ratios between the genetically defined V1 and related HO sizes ( Figure 4C ) . Ranging over an ~80% variation of the normal V1 size , Figure 4C reveals that related HO size is bi-directionally altered in a linear fashion ( Cdh8 regression: y = −0 . 0071x + 1 . 7463; Lmo4 regression: y = 0 . 0208 + 1 . 7463 ) . Taken together , our results are consistent with a proportional scaling relationship between the size of primary and related higher order visual areas: The size of V1 is determined by the activity of transcription factors including Emx2 during development , and this mechanism likewise controls the linear matching of the proportions of higher order visual areas in the mouse neocortex . The present findings address the mechanisms that specify and regulate the size of higher order sensory areas , an issue that has been largely neglected . They reveal a novel , prominent role for intrinsic genetic information in this process . Genetically altering the size of V1 over a range of ~80% of its normal size using a Emx2 gain-of function mouse line and a novel conditional Emx2 loss-of function mouse line showed that the specification of both primary and related higher order cortical areas during development was linearly scaled by driving the unique properties that characterize both , V1 and higher order visual areas . Regardless of whether V1 was larger or smaller than in wt mice , related HO exhibited normal cytoarchitecture , genetic profiles , functional properties , and characteristic patterns of connectivity that resulted in an overall uniformly altered ‘visual cortical field’ in the occipital cortex that remained accurately and proportionally subdivided into V1 and higher order visual areas . This demonstrates that Emx2 ( and perhaps additional intrinsic area patterning regulators ) specify a ‘sensory cortical field’ that includes primary and higher order areas and a defined border between them . This model of cortical area patterning is not consistent with the possibility that the core properties of primary and higher order areas are specified sequentially or through parallel genetic mechanisms . Our results further reveal that higher order areas do not have a fixed size . Rather their relative size is flexible . By using mouse models with bi-directional changes of V1 size as a model , our study revealed that higher order areas scale linearly together with their related primary sensory areas , This observation is important for at least two reasons: ( i ) it re-emphasizes a sequential model of primary sensory area formation that likewise influences the properties of related higher order areas . In this model , cortical intrinsic mechanisms specify all generic primary and higher order visual cortex properties during early development . Much later during postnatal development , geniculocortical input is needed to terminally differentiate the genetic distinctions between V1 and HO ( Chou et al . , 2013; Vue et al . , 2013 ) . ( ii ) It contradicts the hypothesis that cortical structure/function evolution mainly is driven by a disproportionate increase in the size of related higher order areas relative to primary areas . To the contrary , our results show that primary and related higher order areas remain proportionate when primary area size is altered through genetic mechanisms , suggesting that an increase in the complexity of connections and micro-circuits among higher order cortical processing centers likely accounts for gains in cortical functions that are characteristic for gyrencephalic mammals with larger cortical surface areas , compared to simpler lissencephalic mammals . In summary , the newly discovered linear scaling relationship between primary and related higher order areas has major implications for the basic understanding of the development and organization of the neocortical bauplan and its evolution and variability in normal and affected conditions . All experiments were approved and conducted following the guidelines of the Institutional Animal Care and Use Committee at the Salk Institute and were in full compliance with the guidelines of the National Institutes of Health for the care and use of laboratory animals . When mice were mated , the morning of the identified vaginal plug was designated as E0 . 5 . The morning on which pups were born was designated P 0 . 5 . Transgenic mice overexpressing Emx2 under the Nestin promoter ( ne-Emx2 ) were previously described ( Hamasaki et al . , 2004 ) . For generating Emx2 floxed mice ( Emx2fl/fl ) , gene targeting was carried out using homologous recombination in embryonic stem cells . A targeting construct was designed in which the 5’ loxP site was upstream of the Emx2 transcriptional start site and the 3’ loxP site downstream of Exon 2 , followed by a FRT-site-flanked PGK-Neo cassette , Figure 4—figure supplement 1 ) . Targeted embryonic stem cell clones were screened by Southern blot with probes A , B , and C and by PCR to identify Emx2floxed-neo/+ clones ( Figure 4—figure supplement 1 ) . Positive clones were injected into C57BL/6J blastocysts at the Salk Transgenic Core Facility and chimeras were mated to C57BL/6J females to obtain germline transmission . Heterozygous mice were mated with mice expressing FLPe ( Rodríguez et al . , 2000 ) to remove the neo cassette and then mated to obtain homozygous Emx2fl/fl mice . Cortex specific deletion of Emx2 ( cKO ) was obtained by crossing Emx2fl/fl mice with Emx1-IRES-Cre mice ( Gorski et al . , 2002 ) . Specificity of Emx1-IRES-Cre-mediated deletion of Emx2 floxed alleles was analyzed by WMISH ( described below ) staining using a full-length Emx2 antisense RNA probe on E11 embryos . Genotyping was performed using primers for Emx2 floxed alleles ( Emx2 forward: GAC-TCC-TTT-CCC-AAA-TAA-CCC-C , Emx2 reverse: GTA-AGC-GGG-CGG-GGA-CTG-GTT-C ) and for the Cre recombinase ( cre forward: GCT-AAA-CAT-GCT-TCA-TCG-TCG-G , cre reverse: GAT-CTC-CGG-TAT-TGA-AAC-TCC-AGC ) , and the ne-Emx2 transgene ( nestin forward: TCA-ACC-CCT-AAA-AGC-TCC , Emx2 reverse: GGA-CGG-AGA-GAA-GGC-GGT ) . Tissues were dissected , washed in phosphate-buffered saline ( PBS ) , fixed overnight in 4% phosphate-buffered paraformaldehyde ( PFA ) , washed in PBS , and cryopreserved in 30% sucrose in PBS . Postnatal brains were perfused with PFA , postfixed overnight in PFA , washed with PBS , and cryopreserved in 30% sucrose in PBS . Tissues were embedded in Tissue-Tek OCT ( Sakura Finetek , Japan ) and sectioned on a cryostat ( Leica , Germany ) . Antisense RNA probes were labeled using a DIG-RNA labeling kit ( Roche , Switzerland ) . ISH on 18-μm cryostat sections and WMISH using P7 brains were carried out as previously described ( Chou et al . , 2013; Hamasaki et al . , 2004; Zembrzycki et al . , 2007; Zembrzycki et al . , 2015 ) . For tangential cortical sections , cortical hemispheres were dissected , flattened , postfixed between slide glasses , and then cryoprotected . Tangential sections were cut into 40-μm slices from flattened cortical hemispheres with a sliding microtome and then they were immunostained for Serotonin ( 5HT , ImmunoStar , Hudson , WI ) . Immunostaining was developed using the diaminobenzidine colorimetric reaction and the Vectastain kit ( Vector , Burlingame , CA ) . For Nissl staining , sections were stained with 0 . 5% cresyl violet and then dehydrated with graded alcohols . Lipophilic tracers DiI and DiD ( all from Molecular Probes , Eugene , OR ) were used to label corticothalamic- , thalamocortical- , corticotectal- , and corticospinal projections . For each experiment 4–6 brains with comparable tracer injection sites were cut and used for further data analysis , representative example images are shown in the figures . Analysis of thalamocortical axons by thalamic DiI injections ( Figure 1B ) : P7 brains were fixed in 4% PFA , hemisected , and a coronal cut between the diencephalon and mesencephalon was made in order to expose thalamic nuclei at the section surface and DiI crystals were implanted to cover the dorsal lateral geniculate nucleus ( dLG ) . After incubation for 1 to 2 months at 30°C to 60°C , preparations were sectioned sagittally on a vibratome ( Leica ) . Sections were counterstained with DAPI ( Vector ) and analyzed under a fluorescence microscope to determine the tangential distribution of labeled thalamocortical axons in the neocortex . Analysis of layer 5 subcerebral projection neurons ( Figure 1C ) : Corticospinal neurons in cortical layer 5 were retrogradely labeled by inserting DiI crystals into the pyramidal decussation in 4% PFA fixed brains . Layer 5 corticotectal neurons were labeled in 4% PFA fixed brains by implantation of small DiI crystals into the upper layers of superior colliculus . Brains were incubated at 37°C for 2–3 months before 100 μm sagittal vibratome sections were cut and analyzed under fluorescent light . Analysis of area-specific thalamocortical and corticothalamic connectivity of caudal cortex ( Figure 2 ) : P7 pups were anesthetized by hypothermia and a small area of skull was removed to expose the cortical surface . DiI crystals and a small piece of DiD were implanted into cortical locations around V1 and the V1/HO border . After 1 day of survival , brains were removed after 4% PFA perfusion and their cortices and thalami dissected . Cortices were then flattened and stained for 5-HT to reveal primary sensory areas relative to the dye injection sites . The thalami were preserved sectioned coronally , stained with DAPI and DiI/DiD labeled cells analyzed under a fluorescence microscope . Data collection and analyses were performed blind to genotype and the conditions of the experiments , data were collected and processed randomly , and no data points were excluded . No statistical methods were used to predetermine sample sizes , but our sample sizes were similar to those reported in previous publications ( for example , ( Chou et al . , 2013; Zembrzycki et al . , 2013 ) . Data met the assumptions of the statistical tests used , and the data distribution was assumed to be normal but was not formally tested . Statistics were calculated with Microsoft Excel . Quantifications show mean values of the tested groups and are displayed as a percentage of the wt group . Quantified sample sizes ( number of brains: n ) are indicated in the figure legends . The examples shown in each figure are representative and were reproducible for each set of experiments . Individual experiments were successfully repeated at least three times using different litters . Area size measurements on 5HT stained sections and statistical analysis was performed as previously described ( Leingärtner et al . , 2007; Zembrzycki et al . , 2015; 2013 ) . To quantify molecular V1 and HO sizes , gene expression domains were quantified on single images of WMISH-stained brains ( examples of measured area outlines are shown as dashed lines in Figures 3 , 4 ) using ImageJ ( Rasband 1997−2013 ) . Derived wt mean values were defined as 100% and values of the other mouse lines calculated accordingly . Statistical significance was determined using unpaired two-tailed t test , p values < 0 . 05 ( indicated as * ) were considered as statistically significant . Variance is indicated in the main text sections reflecting standard error of the mean . Intrinsic signal optical imaging was performed as previously described ( Kalatsky and Stryker , 2003 ) . To determine the spatial relationship between V1 functional maps and V1 , defined histochemically by 5-HT staining , animals were imaged to determine the V1 map , and after completion , small DiI injections were made outside of the functional map perimeters . Animals were then perfused with 4% PFA , cortices dissected , flattened , sectioned tangentially , and stained for 5-HT .	The neocortex is the most recently evolved part of the human brain . It is associated with higher thought processes , including language and the processing of information from our senses . Anatomically , the neocortex is organised into different regions called ‘primary areas’ and ‘higher order areas’ , and perturbations to this organisation are associated with disorders such as autism . There are many more higher order areas than primary areas in a mammalian brain . But , while primary areas are known to be specified by developmental genes in the embryo , little is known about how the development of higher order areas is controlled . Recent findings suggested that primary areas might themselves influence the emergence of higher order areas via a series of developmental events . Now , Zembrzycki , Stocker et al . have investigated the developmental mechanisms that organise the mouse neocortex into its different regions . The experiments involved mouse mutants that produce either too much or too little of a protein called Emx2 . This protein is known to determine the size and position of the primary visual area ( commonly called V1 ) during embryonic development . In the mutant mice with too much Emx2 , the primary visual area was about 150% larger than it should be , even though the neocortex was a normal size . Zembrzycki , Stocker et al . then went on to show that the higher order areas associated with the primary visual area also grew proportionally larger in these mutant mice . The opposite was true for mice that didn’t produce Emx2 in their brains , and these mice had a much smaller primary visual area than normal mice . Together , these findings reveal a previously unknown linear relationship between the size of the primary visual area and higher order visual areas that is controlled by the genes that pattern the neocortex during development . This and other new insights will inform future studies of the development and organization of the neocortex and our understanding of how it evolved .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "neuroscience" ]	2015	Genetic mechanisms control the linear scaling between related cortical primary and higher order sensory areas
Combinatorial cis-regulatory networks encoded in animal genomes represent the foundational gene expression mechanism for directing cell-fate commitment and maintenance of cell identity by transcription factors ( TFs ) . However , the 3D spatial organization of cis-elements and how such sub-nuclear structures influence TF activity remain poorly understood . Here , we combine lattice light-sheet imaging , single-molecule tracking , numerical simulations , and ChIP-exo mapping to localize and functionally probe Sox2 enhancer-organization in living embryonic stem cells . Sox2 enhancers form 3D-clusters that are segregated from heterochromatin but overlap with a subset of Pol II enriched regions . Sox2 searches for specific binding targets via a 3D-diffusion dominant mode when shuttling long-distances between clusters while chromatin-bound states predominate within individual clusters . Thus , enhancer clustering may reduce global search efficiency but enables rapid local fine-tuning of TF search parameters . Our results suggest an integrated model linking cis-element 3D spatial distribution to local-versus-global target search modalities essential for regulating eukaryotic gene transcription . The existence and importance of long-range interactions between distal cis-control elements and cognate core promoter factors in regulating gene expression programs that govern cell-fate in animals have been extensively studied in traditional biochemistry , genetics , and genomics ( Levine and Tjian , 2003; Levine et al . , 2014 ) . However , these earlier classical studies were unable to capture the three dimensional ( 3D ) spatial organization or the temporal dynamics of the functional interactions between sequence-specific transcription factors ( TFs ) and distal enhancers . The more recent development of Chromosome Conformation Capture ( 3C ) and high throughput sequencing based techniques have provided additional insights into long-distance chromatin looping , genome folding , and topological domains in the context of whole animal genomes but without providing direct spatial information ( Dostie et al . , 2006; Lieberman-Aiden et al . , 2009; Dixon et al . , 2012; van de Werken et al . , 2012 ) . Indeed , emerging evidence suggests that proximity ligation frequency based distances measured by 3C assays may be limited in its capacity to accurately capture 3D molecular proximity ( Gavrilov et al . , 2013; O'Sullivan et al . , 2013; Belmont , 2014 ) . The inherent constraints of using fixed cells or population based assays to dissect the nature of 3D enhancer organization and transcription factor search dynamics can , however , be partly overcome by single live-cell imaging . Recent advances in fluorescence super resolution microscopy and protein labeling chemistry make possible the visualization and tracking of individual transcription factors as they diffuse and bind to specific targets in the nucleus of living mammalian cells ( Mazza et al . , 2012; Gavrilov et al . , 2013; Izeddin et al . , 2014; Chen et al . , 2014b ) . If specific and stable TF:DNA binding events can be localized and visually reconstructed at single-molecule resolution within an intact nucleus , we would have an opportunity to map and decipher critical spatial features linked to the 3D organization of the functional genome and simultaneously measure differences in the dynamic nature of the TF target search process in distinct compartments within living cells . In our recent work ( Chen et al . , 2014b ) , we described a single-cell , single-molecule imaging strategy to study the in vivo Sox2 and Oct4 target search process and dissect the kinetics of enhanceosome formation at endogenous single-copy gene loci in live embryonic stem ( ES ) cells . We found that Sox2 and Oct4 search for their cognate targets via a trial-and-error mechanism in which these two TFs undergo multiple rounds of diffusion and non-specific chromatin collisions before stably engaging with a specific target via an ordered assembly mechanism . Single-molecule in vitro measurements indicate that Sox2 can also slide along short stretches of naked DNA to search for its target . Although our findings revealed significant mechanistic insights of the in vivo TF target search process , these initial single molecule tracking ( SMT ) studies were constrained to investigate the average behavior of TF dynamics in single cells . We were not able to address whether TFs behave differently within distinct sub-nuclear territories such as active gene enriched euchromatic regions vs the more tightly compacted regions of heterochromatin nor whether the 3D spatial distribution of enhancer sites might affect target search dynamics . To develop new approaches to probe 3D genome organization and address some of these important unresolved questions regarding the dynamic TF target search process , here we took advantage of further developments in super resolution microscopy ( Chen et al . , 2014a ) and fluorescent dye chemistry ( Grimm et al . , 2015 ) . We applied lattice light-sheet single-molecule imaging to selectively localize , track , and map endogenous Sox2 binding sites in single , living ES cells . Two-color imaging enabled us to quantify the spatial distribution of Sox2 binding sites ( enhancers ) with respect to euchromatic vs heterochromatic regions . We also measured potential differential rates of Sox2 diffusion and binding modes within enhancer clusters compared to heterochromatic regions . SMT and Monte Carlo simulations of the Sox2 target search process revealed two distinct behaviors—a 3D diffusion dominant long-range mode when traveling between clusters and a local binding dominant search mode within individual binding clusters . These studies suggest that enhancer clustering may reduce global target search efficiency but enable rapid local fine-tuning of search parameters that govern spatially controlled gene regulation in the nucleus . We also probed potential links between enhancer clustering and epigenetic regulation . Together , these results reveal principles that integrate 3D enhancer organization with dynamic in vivo TF-DNA interactions that may play a key role in regulating stem cell pluripotency . The combination of methods described here also open new avenues for studying single live-cell genome spatial organization and function . Although numerous studies have been conducted to investigate Sox2:enhancer interactions by biochemical and genomic approaches , no direct sub-nuclear global spatial information of Sox2 enhancer sites has been attained . This aspect of dissecting TF function presents a particular challenge , because the majority of Sox2 molecules ( >74% ) in the nucleus are in a dynamically diffusing state ( Kaur et al . , 2013; Chen et al . , 2014b ) . Our recent single-molecule tracking ( SMT ) experiments found that Sox2 interactions with DNA consist of two distinct populations: non-specific collisions of short duration ( residence time ∼0 . 7 s ) and specific ‘stable’ interactions of much longer duration ( residence time ∼12 s ) ( Chen et al . , 2014b ) . Since only ∼3% of the Sox2 molecules in the nucleus are bound to specific DNA sites at a given window of time , it is impossible to infer the spatial distribution of Sox2 enhancer sites simply from fluorescence fluctuations captured by wide-field imaging or from conventional super resolution images of live or fixed cells . Currently , the only information we have that can distinguish site-specific binding from non-specific binding events or rapidly diffusing molecules is the relatively long specific residence times of Sox2 at putative cognate recognition sites ( Chen et al . , 2014b ) . Therefore , we set out to devise a time-resolved , live-cell imaging strategy to selectively localize , track , and map these longer lived ‘stable’ Sox2 binding events that likely represent site specific Sox2 binding events to generate a super resolution 3D Sox2/enhancer site map for the whole nucleus . To achieve this , we implemented a lattice light-sheet based single-molecule imaging strategy ( Figure 1A , see Figure 1—figure supplement 1A–B for details of optical layout ) . We first used an improved labeling method in which a HaloTag ligand based on a newly developed fluorophore , Janelia Fluor 549 ( JF549 ) ( Grimm et al . , 2015 ) , at ultralow concentrations ( ∼0 . 1 fM ) was gradually diffused into HaloTag-Sox2 expressing ES cells to fluorescently tag individual Sox2 molecules . During the labeling , we performed iterative cycles of z tiling by light-sheet microscopy of ES cell nuclei that allowed us to image Sox2 at single molecule resolution in 3D in a series of time-lapse movies . Background fluorescence contributed by rapidly diffusing free JF549-HaloTag ligand was negligible under these imaging conditions as single molecules were only detectable inside cell nuclei but not in the cytoplasm or other regions lacking Sox2 binding sites ( Video 1 ) . Light-sheet imaging turned out to be critical for the success of this strategy because the selective plane illumination not only preserved the photon budget by preventing out-of-focus molecules from photo-bleaching but also significantly increased the signal-to-noise ratio . With 3D localization at high precision ( xy: 14 nm , z: 34 nm , Figure 1—figure supplement 1C , Video 2 ) coupled to single molecule tracking , we were able to selectively preserve the global positions where single Sox2 molecules dwell ( <50 nm ) for at least 3 s . The average residence time of selected molecules was ∼6 . 92 ± 0 . 51 s ( n = 9 cells ) ( Figure 1B ) , consistent with the notion that most of these events likely reflect the longer residence times representing specific Sox2-enhancer interactions ( Chen et al . , 2014b ) . We next calculated the number of local neighbors for each Sox2 enhancer site to generate a color-coded heat map for visualizing this data ( Figure 1C and Video 3 ) . As can be seen in Figure 1C , many local density hot spots were observed within a single nucleus , suggesting that instead of being uniformly distributed throughout the nucleus , Sox2 bound enhancers form locally enriched distinct higher density clusters ( EnCs ) . 10 . 7554/eLife . 04236 . 003Figure 1 . Localization of Sox2 stable binding sites in 3D by lattice light-sheet , single-molecule imaging . ( A ) Whole-nucleus single molecule imaging was performed by lattice light-sheet microscopy with 300 nm z steps and 50 ms per frame . HaloTag-Sox2 molecules were labeled by membrane permeable JF549 dye . The imaging scheme was cycled every 3 s for ∼500 times . The 3D positions of single molecule localization events were tracked ( for more details , see ‘Materials and methods’ ) . Any Sox2 molecules that dwelled at a position for more than 3 s were counted as stable bound events . See Videos 1 and 2 for the exemplary raw data . ( B ) Upper: out of total localized and tracked events , only ∼11 . 6 ± 3 . 2% had residence times longer than 3 s . ∼88 . 4 ± 6 . 5% Sox2 molecules appeared in single frames ( n = 9 cells ) . Lower: residence time histogram of stable bound Sox2 molecules . The average residence time detected by this imaging set-up is ∼6 . 92 ± 0 . 51 s ( n = 9 cells ) . ( C ) 3D density map of stable Sox2 binding sites in single ES cell nucleus . For fair comparisons between experimental conditions , we only considered 7000 stable binding sites for each cell . The color map reflects the number of local neighbors that was calculated by using a canopy radius of 400 nm . The unit of the x , y , z axes is nm . See Video 3 for the full 3D rotation movie . DOI: http://dx . doi . org/10 . 7554/eLife . 04236 . 00310 . 7554/eLife . 04236 . 004Figure 1—figure supplement 1 . Optics layout , PSF , and localization uncertainty estimation . ( A ) Schematic of the light path used for the optical lattice microscope . A collimated circular laser beam is passed through two pairs of cylindrical lenses to illuminate a thin stripe across the width of a ferroelectric spatial light modulator ( SLM , Forth Dimension Displays , SXGA-3DM ) . The remainder of the excitation optical path serves to create a demagnified image of the SLM ( 81 . 6 nm pixels ) at the focal plane of the excitation objective . First , a lens is used in a 2F configuration to create a diffraction pattern at its front focal plane that is the Fourier transform of the electric field reflected from the SLM . A custom opaque mask with transmissive annuli ( Photo-Sciences Inc . ) is placed at this plane , and a specific annulus is chosen to remove the unwanted diffraction orders and enforce a limit on the minimum field of view . The electric field transmitted through the mask is then imaged in series onto each of a pair of galvanometers ( Cambridge Technology , 6215H ) and the rear pupil plane of the excitation objective . The galvanometers serve to translate the light sheet through the specimen in x and z . Finally , the field is reverse transformed by the excitation objective to create the desired lattice light sheet at its front focal plane . The fluorescence generated within the specimen is collected by a detection objective ( Nikon , CFI Apo LWD 25XW , 1 . 1 NA , 2 mm WD ) whose focal plane is co-incident with the light sheet . Its high NA is essential to maximize the xy resolution and to optimize the light collection for single molecule detection . The excitation objective ( Special Optics , 0 . 65 NA , 3 . 74 mm WD ) was custom designed to fill the remaining available solid angle above the cover slip . A tube lens images the fluorescence from the illuminated slice within the specimen onto a sCMOS camera ( Hamamatsu Orca Flash 4 . 0 v2 ) capable of frame rates down to 1 ms . A 3D image is produced from a stack of such 2D slices , either by moving the light sheet and detection objective together through the specimen ( the former with the z galvo , the latter with a piezoelectric stage [Physik Instrumente , P-621 . 1CD] ) or , far more commonly , by translating the specimen with a second piezo stage through the stationary light sheet along an axis s in the plane of the specimen cover slip . The specimen holder and specimen piezo are mounted on a trio of closed loop micropositioning stages ( Physik Instrumente M-663 for horizontal motion in the cover slip plane , M-122 . 2DD for vertical travel ) . ( B ) The XY and XZ point spread function profile of a fluorescent bead ( Emission: ∼590 nm; Voxel size , 100 nm in each direction , the radius of the bead is 50 nm ) . Scale bar , 500 nm . ( C ) 3D single-molecule localization was performed using 3D Gaussian model ( Equation 1 ) by FISH-QUANT ( Mueller et al . , 2013 ) . x , y , z localization uncertainty for each 3D localization event was calculated by using the published estimator , Equation 2 . The localization histogram was fitted by Extreme Fit by Matlab . The mean and center values were labeled in each plot . DOI: http://dx . doi . org/10 . 7554/eLife . 04236 . 00410 . 7554/eLife . 04236 . 005Video 1 . Single-molecule light-sheet imaging of Sox2 in GFP-HP1 ES cells . HaloTag-Sox2 is gradually labeled with JF549 ligand by diffusion . Light-sheet imaging was performed with a z step of 200 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 04236 . 00510 . 7554/eLife . 04236 . 006Video 2 . Single-molecule , light-sheet imaging of HaloTag-Sox2 in single live ES cells . The z step size is 300 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 04236 . 00610 . 7554/eLife . 04236 . 007Video 3 . Reconstructed Sox2 stable binding sites in the live ES cell nucleus . HaloTag-Sox2 stable binding sites ( 7000 , >3 s ) were localized , tracked , and reconstructed with a color map same as Figure 1C . The unit is nm . 2 cells were shown here . DOI: http://dx . doi . org/10 . 7554/eLife . 04236 . 007 To test whether the clustering behavior of stable Sox2 binding sites was due to potential artifacts introduced by our imaging strategy , we also inspected ES cells that stably expressed a control HaloTag fusion protein , the histone subunit ( HaloTag-H2B ) using the same imaging set-up followed by an identical computational pipeline and presentation scheme ( Figure 2A and Video 4 ) . In contrast to Sox2 , we observed dramatically decreased clustering behavior of HaloTag-H2B ( Figure 2A and Video 4 ) . In order to establish a more quantitative description of the Sox2-enhancer clustering behavior , we adapted a pair-correlation function used by cosmologists to describe the clustering behavior of stars in galaxies ( Peebles , 1973; Peebles and Hauser , 1974 ) ( Figure 2B , see details in Equations 3–7 ) . Briefly , the pair correlation function , g ( r ) , describes the density of spots in a volume element at a separation r from single spots relative to the average density in the whole volume . If enhancer sites were uniformly distributed ( Figure 2—figure supplement 1A and Video 5 ) the pair correlation function would equal 1 ( Figure 2B and Figure 2—figure supplement 1D , gray diamonds ) because the local densities around each position would be invariant and equal to the average density in the entire volume . However , when spots are highly clustered , the g ( r ) will start with values much greater than 1 and gradually decrease as r increases , indicating that the local molecular densities around individual spots would be much higher than the average density in the volume . As expected , the g ( r ) function of Sox2 stable binding sites agreed well with a highly clustered behavior while by contrast , the g ( r ) function of H2B suggests a much more random and uniform distribution in the nucleus ( Figure 2B ) . We next extended the previously established fluctuation model for describing two dimensional heterogeneous protein distribution in membranes ( Sengupta et al . , 2011 ) to fit the g ( r ) function calculated from our 3D dataset ( Equations 10–13 ) . This model extracted two key parameters related to molecular clustering: the fluctuation range ( ε ) and the fluctuation amplitude ( A ) ( Supplementary file 1 ) . Specifically , ε is proportional to the average size of clusters while A is proportional to the relative molecular density within clusters . We observed , on average , a 14 fold higher fluctuation amplitude of Sox2-enhancers compared with those of H2B . However , we did observe a certain degree of H2B density fluctuations at much larger scales ( Supplementary file 1 ) , probably reflecting chromatin density variations in the nucleus as reported previously ( Young et al . , 1986 ) . Because we use the 7000 most stable H2B spots to calculate the pair-correlation functions , according to Nyquist sampling theorem , our results are more sensitive to large-scale H2B density fluctuations in the nucleus and may overlook smaller-scale local H2B clustering . The mathematic tools established here should also serve as the basis for future comparisons when we carry out perturbation experiments that will be instructive for dissecting the function and molecular mechanisms underlying enhancer clustering . To determine whether the blinking of stably bound fluorescently tagged Sox2 molecules might influence or distort the observed ‘stable’ binding of Sox2 in the clusters , we plotted the number of detected events as a function of frame number . These plots show an initial decay that eventually reaches a plateau ( Figure 2—figure supplement 2D ) . Such a temporal decay profile is more consistent with a bleaching dominant mechanism in which an equilibrium has been achieved between photo-bleaching and the ongoing fluorescent labeling of HaloTag-Sox2 molecules . Perhaps the strongest argument that the Sox2 clustering pattern we observe is not likely an artifact of the imaging modality can be derived from the fact that chromatin bound HaloTag-H2B molecules using precisely the same imaging strategy failed to show such a prominent clustering pattern . 10 . 7554/eLife . 04236 . 008Figure 2 . Clustering of Sox2 bound enhancers in the nucleus . ( A ) 3D density map of H2B distribution ( n = 7000 ) in single ES cell nucleus . The imaging condition and analysis parameter set-ups were the same as HaloTag-Sox2 in Figure 1 . The color map reflects the number of local neighbors that was calculated by using a radius of 400 nm . The unit of the x , y , z axes is nm . See Video 4 for the full 3D rotation movie . ( B ) Upper: The pair correlation function g ( r ) measures the relative density of enhancer sites in a volume element at a separation r from single enhancer sites , given that the average density of enhancer sites in the whole volume is ρ¯ . See Equations 3–7 for calculation details . Lower: Pair correlation function of Sox2 stable binding sites ( red dots , n = 6 ) , H2B ( blue squares , n = 6 ) , and simulated uniformly distributed particles ( gray diamond , n = 5 , Video 5 ) fitted with the fluctuation model ( dotted lines ) ( See Equations 10–13 ) . The obtained fluctuation amplitude and range for each curve are in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04236 . 00810 . 7554/eLife . 04236 . 009Figure 2—figure supplement 1 . Quantification of clustering by pair correlation function . ( A ) 3D density map of simulated uniform sites ( n = 7000 ) in single ES cell nucleus . The color map reflects the number of local neighbors that was calculated by using a canopy radius of 400 nm . The unit of the x , y , z axes is nm . See Video 5 for the full 3D rotation movie . ( B ) 3D density map of Sox2 stable binding sites ( n = 7000 ) in single TSA treated ES cell nucleus . The color map reflects the number of local neighbors that was calculated by using a canopy radius of 400 nm . The unit of the x , y , z axes is nm . ( C ) Pair correlation function of Sox2 binding sites that have residence times less than 3 s ( yellow , n = 6 ) fitted with the fluctuation model ( dotted lines ) ( See Equations 10–13 ) . See Video 6 for the full 3D rotation movie . ( D ) The room-in view of pair correlation function and fluctuation model fitting of the indicated conditions . The obtained fluctuation amplitude and range for each curve are in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04236 . 00910 . 7554/eLife . 04236 . 010Figure 2—figure supplement 2 . Temporal profiles of individual clusters and the number of localization detections per frame . ( A–C ) Time counting of the arrival events of Sox2 stable binding sites within individual clusters . Cumulative Density Function is plotted as the function of the frame number . The time interval between two frames is 3 s . ( D ) To test whether photo-bleaching plays a dominant role in our imaging strategy , the number of 3D localization detections is plotted as a function of the Frame Number . The time interval between two frames is the same as in ( A–C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04236 . 01010 . 7554/eLife . 04236 . 011Video 4 . Reconstructed H2B distribution in the live ES cell nucleus . HaloTag-H2B sites ( 7000 ) were localized , tracked , and reconstructed with a color map same as that of Figure 2A . The unit is nm . DOI: http://dx . doi . org/10 . 7554/eLife . 04236 . 01110 . 7554/eLife . 04236 . 012Video 5 . Uniformly distributed , simulated positions in a nucleus . Uniformly distributed positions ( 7000 ) were presented with a color map same as that of Figure 1C . The unit is nm . DOI: http://dx . doi . org/10 . 7554/eLife . 04236 . 012 To test the contribution , if any , of non-specific interactions to the dramatic clustering behavior observed for Sox2 long-lived binding sites within the cell , we also investigated the clustering behavior of shorter-lived ( <3 s ) Sox2 binding sites that were initially filtered out in our mapping experiments ( Figure 1B ) . If the recorded Sox2 stable binding events mainly reflect random non-specific interactions , the clustering behavior of shorter lived binding sites should be similar to that observed for the long lived putative ‘specific’ binding sites . Instead , we found the shorter-lived Sox2 binding sites showed greatly reduced fluctuation amplitudes of the pair correlation function curves ( Figure 2—figure supplement 1C–D ) . We also note that in many cases , we observed little or no clustering of short-lived Sox2 binding sites within the same territories where longer-lived stable Sox2 binding site clusters can clearly be observed ( Videos 3 and 6 ) . These results suggest that the long-residence time filtering strategy that we deployed here likely enriches for specific binding site signals above the background of non-specific interactions consistent with what we observed previously ( Chen et al . , 2014b ) . 10 . 7554/eLife . 04236 . 014Video 6 . Transient Sox2 binding sites in the live ES cell nucleus . HaloTag-Sox2 transient binding sites ( 7000 , <3 s ) were displayed with a color map same as Figure 1C . The unit is nm . DOI: http://dx . doi . org/10 . 7554/eLife . 04236 . 014 To further study the dynamic properties of EnCs , we used a time-counting analysis method ( Cisse et al . , 2013 ) to probe the temporal profiles of arrival times of stable binding events within individual clusters . Interestingly , we did not observe significant bursting behaviors as described for Pol II clusters ( Figure 2—figure supplement 2A–C ) . These results are consistent with a model wherein Sox2 EnCs are relatively stable during the period ( ∼20 min ) of image acquisition . Because Sox2 bound enhancers are chromatin based structures , we note that previous FRAP ( Fluorescence recovery after photo-bleaching ) experiments on core histone components ( Kimura and Cook , 2001 ) found that large-scale chromatin structures in live cells appeared stable with a half-life of >2–4 hr which is much longer than the duration of our imaging experiments . These findings suggest that the enhancer clustering we observed here likely reflects the average 3D genome organization within reasonably short temporal length scales . It has long been proposed that the 3D space inside a cell nucleus is sub-divided into highly active gene enriched regions ( so-called ‘euchromatin’ ) and largely inactive gene regions ( i . e . , ‘heterochromatin’ ) . To probe the spatial relationship between Sox2 EnCs and heterochromatic regions ( HCs ) , we generated dual labeled ES cell lines that stably express HaloTag-Sox2 and GFP-HP1 . HP1 protein is enriched in peri-centromeric and peripheral HCs ( Grewal and Elgin , 2002 ) that form non-diffraction limited structures in the nucleus ( Figure 3B , E , Figure 3—figure supplement 1C , Figure 3—figure supplement 2 , and Videos 7 , 8 , 9 ) . To map the EnCs and HCs in the same cell , we first deployed a wide-field , two-color imaging scheme ( Figure 3A ) in which we used a low-excitation , long-acquisition time imaging condition ( 2 Hz ) to map Sox2 stable binding sites in the nucleus while we recorded the images of GFP-HP1 before and after the SMT experiment ( Video 7 ) . After localization and tracking of stable binding sites , we used a 2D kernel density estimator to generate an intensity map of EnCs in the nucleus ( See ‘Materials and methods’ for details of image acquisition and registration; Figure 3—figure supplement 1B ) and then superimposed the EnC intensity map with the HC map as two different color channels ( Figure 3B and Figure 3—figure supplement 1C ) . We observed that EnCs and HCs are generally not co-localized spatially ( Figure 3B ) . To gain a more quantitative measurement of these two distinct sub-nuclear regions , we tested the pixel-to-pixel correlation between EnC and HC intensity maps from individual cells . Pixels with high levels of EnC intensities generally showed low levels of HC intensities and vice versa ( Figure 3C ) . The Pearson correlation test gave an averaged coefficient ( Rho ) of 0 . 11 ± 0 . 028 ( n = 8 ) ( Figure 4—figure supplement 1B ) , suggesting that the location of EnCs and HCs is indeed very weakly correlated in the nuclear volume of ES cells . We also used pair cross-correlation analysis ( Veatch et al . , 2012 ) to characterize the spatial relationship between EnC and HC regions ( see Equations 8–9 for details of calculation ) . Unlike autocorrelation which measures the degree of self-clustering , pair cross-correlation examines the degree of co-clustering and co-localization between two types of molecules . As expected , the EnC and HC were shown to be clustered as their self-cross ( auto ) correlation curves start with values significantly above 1 and gradually converge to 1 with increased correlation radii ( Figure 3D ) . By contrast , the pair cross-correlation function between EnC and HC intensity maps showed no apparent spatial correlation other than weak exclusion in the range of radii smaller than 600 nm ( Figure 3D ) . To minimize the possibility that some bias may have been introduced by our 2D wide-field imaging and analysis pipeline , we also analyzed the spatial distributions of HCs and EnCs in single cells by lattice light-sheet imaging . Indeed , single molecule tracking confirmed and further strengthened the segregated relationship between HCs and EnCs in the 3D nucleus ( Figure 3E and Video 9 ) . Importantly , these results , taken together , suggest that Sox2 specific binding sites appear less frequently in HCs as most of the stable/specific binding sites were found to be outside of HCs . Consistent with this notion , we observed that levels of Sox2 in HCs are generally significantly lower than Sox2 levels in surrounding sub-nuclear regions , consistent with a reduced association of Sox2 to HCs ( Figure 3—figure supplement 2 ) . 10 . 7554/eLife . 04236 . 015Figure 3 . Sox2 enhancer clusters and heterochromatin regions are not co-localized . ( A ) Two color imaging to probe the spatial relationship between enhancer clusters and heterochromatin regions . Sox2 stable binding sites were mapped by low-excitation 2D single molecule imaging condition ( Video 7 ) . 2D kernel density estimator was used to generate the 2D intensity map of enhancer clusters in the nucleus ( Figure 3—figure supplement 1B ) . The intensity map of heterochromatin regions was obtained by using the GFP-HP1 channel ( Figure 3—figure supplement 1A ) . The composite image was constructed by merging the two intensity maps as two separate color channels . ( B ) Single-cell exemplary images of the HC , EnC intensity maps , and the composite . See Figure 3—figure supplement 1C for more examples . ( C ) The pixel-to-pixel intensity plot from the HC and EnC intensity maps shown in ( B ) . The x , y value of each point is the intensity of HC ( x ) and that of EnC ( y ) from the same pixel . Pixels with low Sox2 EnC and HC intensity values were considered as background signals ( blue points ) . The percentile of points in each quarter ( over the total number of red points ) was indicated in the corner of the region . ( D ) Pair auto- and cross-correlation function of HC ( auto , green ) , EnC ( auto , pink ) , HC ⋆ EnC ( blue ) , and permutated ( gray ) images to investigate the spatial relationship between HC and EnC regions in single cells . ⋆ , denotes the cross-correlation operator . See Equations 8–9 for calculation details . Permutation was performed by randomizing pixels spatially within the nucleus mask for both HC and EnC images prior to calculating the cross-correlation function . ( E ) 3D spatial relationship between heterochromatin regions and Sox2 enhancer clusters determined by two color lattice light-sheet imaging . The HaloTag-Sox2 over-labeled image ( left ) shows fluorescent intensities contributed by all JF549-HaloTag-Sox2 molecules and the single-molecule tracked image ( right ) only shows the stable Sox2 binding site distribution . See Videos 1 , 8 , and 9 for the exemplary raw data and the full rotation movie . DOI: http://dx . doi . org/10 . 7554/eLife . 04236 . 01510 . 7554/eLife . 04236 . 016Figure 3—figure supplement 1 . Heterochromatin and Sox2 EnC spatial relationship . ( A ) Wide-field GFP-HP1 image was first processed by Matlab function to subtract background signals . Then , the normalized intensity map of heterochromatin was calculated ( See details in ‘Materials and methods’ and Video 7 ) . ( B ) 2D localized stable binding events ( residence time > 2 s ) were used for intensity estimation by a customized 2D kernel density estimator ( See details in ‘Materials and methods’ ) . ( C ) The intensity map of heterochromatin and that of the Sox2 enhancer clusters were used to generate the composite image on the right . Data from three cells were shown . Scale bar: 2 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 04236 . 01610 . 7554/eLife . 04236 . 017Figure 3—figure supplement 2 . Probing Sox2 levels in heterochromatin regions . ( A ) Wide-field fluorescent images of GFP-HP1 , over-labeled JF549-HaloTag-Sox2 , and merged from single live cells . Upper: intensity profiles from the two separate channels along the indicated path were plotted in ( B ) . Sox2 intensity drops were observed in heterochromatin regions . Lower: the relative Sox2 intensity levels in the heterochromatin regions were compared with surrounding regions by Equation 16 . The resulting ratios ( n = 19 regions ) were plotted in ( C ) . Scale bar: 2 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 04236 . 01710 . 7554/eLife . 04236 . 018Video 7 . Map stable Sox2 binding sites in GFP-HP1 labeled cells . Low excitation and long acquisition time ( 500 ms ) wide-field imaging was used to map Sox2 stable binding sites in the GFP-HP1 labeled cells . DOI: http://dx . doi . org/10 . 7554/eLife . 04236 . 01810 . 7554/eLife . 04236 . 019Video 8 . Two color light-sheet imaging of Sox2 over-labeled GFP-HP1 ES cells . HaloTag-Sox2 is over labeled with JF549 ligand . Light-sheet imaging was performed with a z step of 200 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 04236 . 01910 . 7554/eLife . 04236 . 020Video 9 . 3D spatial relationship between heterochromatin and Sox2 enhancer clusters . ( A ) 3D reconstruction of over-labeled JF549 HaloTag-Sox2 and GFP-HP1 in single cell nucleus . ( B ) 3D reconstruction of JF549 HaloTag-Sox2 stable binding events ( 7000 ) ( residence time >6 s ) and GFP-HP1 in single cell nucleus . Scale bar , 2 µm . The color map reflects the number of local neighbors that was calculated by using a canopy radius of 400 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 04236 . 020 To investigate the spatial relationship and inferred functional correlation between Sox2 enhancers and RNA Pol II distribution in the nucleus , we generated an ES cell line stably expressing HaloTag-Sox2 and a Dendra2 tagged Rpb1 mutant that is resistant to α-amanitin ( Cisse et al . , 2013 ) . These dual labeled ES cells were able to proliferate in the presence of α-amanitin , indicating that the tagged Rbp1 replaced the endogenous subunit in the RNA Pol II complex without interfering with its normal transcription function . To acquire super-resolution images of Pol II and Sox2 EnCs in the same cell , we first mapped Sox2 EnC clusters by deploying low-excitation and long-acquisition times ( 2 Hz ) for detecting stable DNA bound JF646–HaloTag-Sox2 molecules . Next , we performed live-cell PALM experiments by photo-activating Dendra2 tagged Pol II molecules ( See ‘Materials and methods’ for details of image acquisition and registration ) . The final reconstructed images are shown in Figure 4A . We also performed pair auto- and cross-correlation analysis with Pol II and EnC intensity maps ( Figure 4C ) . Interestingly , results from autocorrelation analysis suggested that Pol II molecules are somewhat more evenly distributed in the nucleus than the highly clustered Sox2-enhancers . Specifically , Sox2 EnC autocorrelation curves generally start with higher values ( higher packing densities ) and more quickly converge to 1 with increased correlation radii ( tighter packing ) ( Figure 4C ) compared with Pol II autocorrelation curves . However , it is worth noting that we did detect significant and distinct local Pol II density fluctuations ( Figure 4A , C ) consistent with previous reports using other imaging modalities ( Cisse et al . , 2013; Zhao et al . , 2014 ) . To better quantify the spatial relationship between the distribution patterns of Pol II and Sox2 EnCs , we next determined the pixel-to-pixel correlation between EnC and Pol II intensity maps generated from individual cells ( Figure 4—figure supplement 1A ) . The Pearson correlation test gave an average coefficient ( Rho ) of 0 . 44 ± 0 . 046 ( Pol II high mask ) ( n = 8 , average p-value from each test <4 . 36E-45 ) , suggesting that unlike the relationship between EnCs and HCs , EnC regions are generally correlated with Pol II occupancy in ES cells . Pair cross-correlation function also suggested a significant degree of co-localization/clustering between Sox2 EnCs and Pol II–enriched regions as the cross-correlation curves start with values significantly greater than 1 and gradually converge to 1 ( Figure 4C ) . However , we note that , due to the tighter clustering of Sox2 enhancers , most Sox2 EnC regions contained significant levels of Pol II whereas only a subset of Pol II enriched regions overlap with Sox2 EnCs ( Figure 4A–B ) . The partial overlap between Sox2 EnCs and Pol II enriched regions is entirely consistent with previous genome-wide analysis showing that Sox2 only targets a subset of transcribed genes involved in maintaining ES cell identity ( Chen et al . , 2008 ) . Many actively transcribed genes ( including house-keeping genes ) are likely subject to regulation by TFs other than Sox2 or Oct4 . These results also suggest that Sox2 enhancer driven gene regulation is largely confined locally within distinct EnCs . Although not detectable in our assays , we assume that sub-nuclear regions outside Sox2 EnCs contain different actively transcribed cis-element clusters that also overlap with other Pol II enriched regions . 10 . 7554/eLife . 04236 . 021Figure 4 . Sox2 targets a subset of Pol II-enriched regions in the nucleus . ( A ) Upper left: a live-cell 2D PALM super-resolution image of Dendra 2 Pol II . Upper Right: Sox2 enhancer clusters mapped by time-resolved , 2D single-molecule imaging/tracking . Stable binding events ( >2 s ) were shown . The color map that reflects number of local neighbors was displayed at the bottom right corner of each image . The canopy radius for calculation is 400 nm . Lower: the superimposed image of Pol II and Sox2 EnCs; Scale bar: 2 µm . ( B ) Selected zoomed-in views from ( A ) ; only a subset of Pol II enriched regions are targeted by Sox2 . ( C ) Upper: single-cell exemplary images of the Pol II and EnC intensity maps calculated by 2D kernel density estimation . Lower: pair auto- and cross-correlation function of Pol II ( auto , green ) , EnC ( auto , pink ) , Pol II ⋆ EnC ( blue ) , and permutated ( gray ) images to investigate the spatial relationship between Pol II enriched and EnC regions in single cells . ⋆ , denotes the cross-correlation operator . See Equations 8–9 for calculation details . Permutation was performed by randomizing pixels spatially within the nucleus mask for both Pol II and EnC images before calculating the cross-correlation function . DOI: http://dx . doi . org/10 . 7554/eLife . 04236 . 02110 . 7554/eLife . 04236 . 022Figure 4—figure supplement 1 . Spatial correlation between Sox2 EnCs and Pol II enriched regions . ( A ) Left: the representative pixel-to-pixel intensity plot calculated from Pol II and EnC intensity maps shown in ( B ) . Right: Pearson correlation coefficients calculated from the pixel-to-pixel correlation of Sox2 EnC & Pol II high regions ( nucleolus and heterochromatin regions excluded , green ) , Sox2 EnC & HC ( Blue ) , and EnC & Pol II permutated ( Black ) intensity plots . Please see the left panel ( Pol II & EnC ) and Figure 3C ( HC & EnC ) for representative pixel-to-pixel intensity plots . DOI: http://dx . doi . org/10 . 7554/eLife . 04236 . 022 In our recent study , we found that in ES cells , Sox2/Oct4 search for their target binding sites via a 3D diffusion dominant mechanism with an average dynamic 3D searching time ( τ3D ) of 3–4 s ( Chen et al . , 2014b ) . However , we were not able to determine whether Sox2 might actually behave differently in distinct sub-nuclear compartments and how enhancer clustering might influence the TF search process . In light of our new finding that the 3D space within the ES cell nucleus can be divided into distinct EnC and HC regions , it became possible to probe the behavior of Sox2 target search dynamics in different chromatin compartments . To address this important and functionally relevant question , we took advantage of recently developed HaloTag dyes ( JF549 and JF646 ) for multiplexing SMT experiments ( Grimm et al . , 2015 ) . We dual labeled HaloTag-Sox2 molecules in the same cells with JF549 and JF646 HaloTag-ligands ( Figure 5A ) . Next , we mapped Sox2 EnC clusters by deploying low-excitation and long-acquisition times ( 2 Hz ) for detecting stable bound JF646–HaloTag-Sox2 molecules . At the same time , we tracked the fast diffusing/binding dynamics of JF549-HaloTag-Sox2 molecules by using high-excitation and short-acquisition times ( 100 Hz ) ( Figure 5—figure supplement 1A and Video 10 ) . We also generated a binary mask for enhancer cluster regions and divided the Sox2 single-molecule tracks into in-mask fragments and out-mask fragments ( See ‘Materials and methods’ for details ) . We note that tracking was performed without knowledge of the mask thus ensuring an unbiased track division . We calculated diffusion coefficients from in-mask track segments ( n = 6 cells ) ( Figure 5A ) and found that most of the molecules inside EnCs are in the bound state ( 64 ± 7 . 8% ) and only ∼36% are rapidly diffusing . These results suggest that Sox2 molecules in EnCs generally spend less time in diffusion before engaging with chromatin and thus have a shorter τ3D . Similarly , we investigated the fast diffusing/binding population of Sox2 within HCs using an analogous strategy ( Figure 5B and Figure 5—figure supplement 1B , Video 11 ) . Consistent with our previous observations ( Figure 3 ) , stable DNA binding events/sites are distinctly low ( 16 ± 4 . 5% ) in HCs , compared with their frequency in EnCs ( 64 ± 7 . 8% ) and in whole nuclei ( 38 ± 4 . 3% ) ( Figure 5A–C ) . However , interestingly , we observed a significant population of Sox2 molecules ( 26 ± 8 . 4% ) within HCs that diffuse with much slower rates ( 0 . 61 ± 0 . 13 μm2s−1 ) than the average Sox2 diffusion rates ( ∼2 . 7 ± 0 . 63 μm2s−1 ) in whole nuclei ( Figure 5B–D ) . This finding suggests that , in certain regions inside HCs , Sox2 diffuses slower . In good agreement with this observation , a previous report demonstrated via Fluorescence Correlation Spectroscopy ( FCS ) measurements that even GFP molecules diffuse much slower in heterochromatic regions possibly due to molecular crowding effects ( Bancaud et al . , 2009 ) . We pooled and analyzed all the SMT tracks obtained from single cells together and found that the majority of Sox2 molecules are in a diffusing mode ( 62 ± 4 . 3% , n = 12 cells ) ( Figure 5C ) , consistent with a 3D diffusion dominant search mechanism . 10 . 7554/eLife . 04236 . 023Figure 5 . Two-color imaging reveals differential Sox2 behavior within enhancer clusters vs heterochromatin . ( A ) Two color single-molecule imaging to probe Sox2 binding and diffusion dynamics in enhancer clusters . EnC regions were first mapped by the low-excitation , long-acquisition time condition . Then , the diffusion coefficient histogram of tracks within the EnC regions was calculated and displayed in the lower panel ( n = 6 cells ) . See Figure 5—figure supplement 1A and Video 10 for more details . The obtained histogram was well fitted with two Gaussian peaks to a fast diffusion ( green , D = 1 . 4 ± 0 . 18 μm2s−1 ) and a bound ( red , D = 0 . 017 ± 0 . 006 μm2s−1 ) population . ( B ) Two color imaging to characterize Sox2 binding and diffusion dynamics in heterochromatin regions . Heterochromatin regions were first mapped by using the HP1-GFP marker . Then , the diffusion coefficient histogram of tracks within the heterochromatin regions was calculated and displayed in the lower panel ( n = 9 cells ) . See Figure 5—figure supplement 1B and Video 11 for more details . The histogram was well fitted with three Gaussian peaks to a fast diffusing ( pink , D = 1 . 58 ± 0 . 25 μm2s−1 ) , a slow diffusion ( green , D = 0 . 61 ± 0 . 13 μm2s−1 ) , and a bound ( red , D = 0 . 023 ± 0 . 011 μm2s−1 ) population . ( C ) Whole-cell Sox2 binding and diffusion dynamics . Single-molecule tracks were shown in the right panel . Data can be fitted by two Gaussian peaks to a fast diffusing ( pink , D = 2 . 7 ± 0 . 63 μm2s−1 ) and a bound ( red , D = 0 . 021 ± 0 . 008 μm2s−1 ) population ( n = 12 cells ) . Scale bar: 2 µm . ( D ) Histograms from ( A–C ) were overlaid . DOI: http://dx . doi . org/10 . 7554/eLife . 04236 . 02310 . 7554/eLife . 04236 . 024Figure 5—figure supplement 1 . Regional specific diffusion and binding dynamics . ( A ) Enhancer cluster regions were first mapped by low excitation , long acquisition ( 2 Hz ) imaging of JF646-HaloTag-Sox2 . Single-molecule stable binding localization events were used to generate the density intensity map by 2D kernel density estimation and the binary EnC mask was obtained by thresholding . Single-molecule tracks were generated by using the information from a fast acquisition ( 100 Hz ) imaging condition . Tracks were divided to in-mask and out-mask fragments . Diffusion coefficients of in-mask tracks were calculated ( See details in ‘Materials and methods’ and Video 10 ) . ( B ) Heterochromatin regions were first mapped by using the HP1-GFP marker . HC mask was obtained by thresholding . Single-molecule tracks were generated by using the information from a fast acquisition ( 100 Hz ) imaging condition . Tracks were divided to in-mask and out-mask fragments . Diffusion coefficients of in-mask tracks were calculated ( See details in ‘Materials and methods’ and Video 11 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04236 . 02410 . 7554/eLife . 04236 . 025Video 10 . Tracking Sox2 binding/diffusion dynamics within enhancer clusters . Two color single molecule imaging was performed with JF646 channel ( Left ) for mapping the enhancer cluster regions and JF549 ( right ) for tracking fast Sox2 diffusion/binding dynamics . DOI: http://dx . doi . org/10 . 7554/eLife . 04236 . 02510 . 7554/eLife . 04236 . 026Video 11 . Tracking Sox2 binding/diffusion dynamics in heterochromatin regions . Two color imaging was performed with the GFP channel ( upper ) for mapping the heterochromatin regions and JF549 ( lower ) for tracking fast Sox2 diffusion/binding dynamics . DOI: http://dx . doi . org/10 . 7554/eLife . 04236 . 026 These results suggest that the Sox2 target search process is likely modulated by the spatial organization of enhancer clusters in the ES cell nucleus . Specifically , inside individual EnCs , Sox2 molecules appear to spend significantly more time binding to either naked DNA or chromatin with relatively short 3D diffusion periods . By contrast , Sox2 molecules that travel from one EnC to the next EnC navigate and tunnel through HCs by a 3D diffusion dominant long-range mode . To further dissect the potential effects of enhancer clustering on TF target search dynamics , we investigated the search process carried out by Sox2 confronted with different degrees of enhancer clustering . The manipulations required to modulate enhancer clustering posed significant experimental challenges . Moreover , because of its probabilistic nature , the target search process cannot be adequately described by ordinary differential equations nor traditional binding kinetic equations , because they typically rely on mass reaction rates and assume that substrate concentrations are invariant across a large field of view . As discussed extensively in the literature ( Robert and Casella , 2005 ) , one of the most effective ways to dissect a random process based behavior is through computer-generated Monte Carlo algorithms that simulate the Brownian motion of TFs in a confined 3D sphere ( cell nucleus ) with multiple target traps ( Equations 14–15 , Figure 6A , Video 12 , See ‘Materials and methods’ for parameter selection criteria . See Figure 6—figure supplement 1A–C for the validation of TF Brownian simulation ) . With such a set-up , we can arbitrarily manipulate the distribution of target sites in the nucleus , precisely control the initial TF injection position and then record the first 3D passage time ( τ3D ) —the duration from the initial injection to the point when the TF hits a target for the first time in the nucleus . 10 . 7554/eLife . 04236 . 027Figure 6 . Enhancer clustering modulates global search efficiency and uncouples target search to a long-range and a local component . ( A ) Monte Carlo simulation of TF target search in the nucleus to test the effects of target site distribution on the first passage 3D time ( τ3D ) . Fold of Delay is defined as the ratio of the average τ3D in the clustered case to the average τ3D in the uniform case . In this experiment , the TF injection site is randomly selected in the nucleus with no overlap with targets . The degree of clustering is tuned by changing the indicated S . D . of the Gaussian distribution . See ‘Materials and methods’ for detailed simulation parameters . TF target search simulation experiments were performed independently 100 times of total 10 repeats for assessing the standard deviation . The Fold of Delay was plotted as a function of S . D . ( Sigma ) in the lower panel . ( B ) Monte Carlo simulation of TF target search in the nucleus to test the effects of releasing Radius ( Kaur et al . ) on the first passage 3D time ( τ3D ) . The injection site is randomly constrained in a shell with the indicated releasing radius relative to the center of the cluster . Fold of Delay is defined same as in ( A ) . TF target search simulation experiments were performed independently 100 times of total 10 repeats for assessing the standard deviation . The Fold of Delay was plotted as a function of Releasing Radius ( Kaur et al . ) in the lower panel . ( C ) The histogram distribution of τ3D for the indicated condition is fitted with both either the single-component ( upper ) or the two-component ( lower ) decay model ( Equations 17–18 ) . The experimental conditions were the same as ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04236 . 02710 . 7554/eLife . 04236 . 028Figure 6—figure supplement 1 . TF 3D Brownian motion simulation . ( A ) An exemplary track of TF 3D Brownian motion simulated by using Equations 14–15 ( See ‘Materials and methods’ for details of parameter set-ups ) . ( B ) Mean square displacement plot fitted with a linear model to extract the diffusion coefficient by MSDanalyzor . 100 simulated tracks were pooled for the calculation . ( C ) Validation of the Brownian motion simulation by calculating the diffusion coefficient from tracks generated by an assigned diffusion coefficient . Independent 100 tracks of total 30 repeats were used for assessing the standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 04236 . 02810 . 7554/eLife . 04236 . 029Figure 6—figure supplement 2 . Effects of number of clusters and distance-between-targets on TF target search . ( A ) To exclude the possibility that increased search times that we observed as target sites become more clustered is due to direct contacts between targets , we performed simulation experiments as described in ( A ) using targets with a smaller radius ( 30 nm ) but maintained the minimal distance between targets as the same ( 80 nm ) . In the right panel , the Fold of Delay was plotted as a function of Gaussian Sigma of the sites spatial distribution . ( B ) Monte Carlo simulation of TF target search in the nucleus to test the effects of number of clusters ( Nc ) on the first passage 3D time ( τ3D ) . Fold of Delay is defined as the ratio of the average τ3D in the clustered case to the average τ3D in the uniform case . In this experiment , the TF injection site is randomly selected in the nucleus with no overlap with targets . The targets do not overlap with each other ( The minimal distance between two targets is two times of the radius of the target [80 nm] ) . The total number of sites remain consistent as 7000 . The centers of clusters were randomly generated again for each simulation experiment . See ‘Materials and methods’ for detailed simulation parameters . TF target search simulation experiments were performed independent 100 times of total 10 repeats for assessing the standard deviation . The Fold of Delay was plotted as a function of Number of Clusters in the right panel . DOI: http://dx . doi . org/10 . 7554/eLife . 04236 . 02910 . 7554/eLife . 04236 . 030Video 12 . TF target search simulation . An example of TF target search simulation in a single nucleus . DOI: http://dx . doi . org/10 . 7554/eLife . 04236 . 030 Having developed this simulation program ( Video 12 ) , we first tested how enhancer clustering would affect the global target search efficiency by injecting the TF randomly into nuclei with different degrees of enhancer clustering ( Figure 6A ) . It is important to note that overlaps between targets were not allowed in our simulation experiments . Specifically , the minimal distance ( 80 nm ) allowed between the centers of two targets is twice that of the target radius ( 40 nm ) . Interestingly , we found that it took increasingly longer τ3D for TFs to reach their target as enhancer sites became more densely packed . This suggests that enhancer clustering may actually decrease global TF target search efficiency in the nucleus . In support of this result , other groups observed similar effects of receptor clustering on ligand binding ( Goldstein and Wiegel , 1983; Care and Soula , 2011 ) . Specifically , the ‘apparent’ macroscopic ligand binding association rates decrease with increased densities of receptors within clusters while the microscopic rates remained the same . To further minimize the possibility that merged targets might create larger binding sites , we reduced the radius of targets to 30 nm while maintaining the minimal distance between the centers of two targets at 80 nm . In this case , there was no possibility of contact or merging between targets . Under these conditions , very similar simulation results were observed ( Figure 6—figure supplement 2A ) . Thus , it seems unlikely that the effects of target site clustering on τ3D would be due to target site fusion . It is also important to note that , in our simulation experiments , the TF binding probability to target is 1 . Since we defined the ‘Fold of Delay’ as τ3D in the ‘clustered’ case normalized by τ3D in the ‘uniform’ case , the binding probability ( 1 or not ) should be identical under both uniform and cluster conditions . Consequently , the trends that we observe for ‘Fold of Delay’ should not alter significantly when the TF binding probabilities vary . As expected , when we increased the number of clusters in the nucleus while holding the total number of sites constant , it took progressively shorter τ3D for TFs to reach their target ( Figure 6—figure supplement 2B ) . Under these conditions , we essentially increased the degree of randomness of enhancer distribution by dispersing the same amount of targets into more randomly localized clusters . We next probed the TF rebinding time in an individual cluster ( Figure 6B ) . Specifically , we injected TFs at different radii of release relative to the center of an enhancer cluster . We found that τ3D becomes reduced as the injection site approaches the center of the EnC ( or when the local concentrations of enhancer sites increase ) . This result suggests that the TF target search dynamics is spatially modulated by enhancer density fluctuations in the nucleus such as we find in EnCs vs HCs ( Figure 5 ) . Specifically , the higher the local concentration of target sites , the shorter the time ( τ3D ) it will take for a TF to reach a target site within an EnC . This relationship can also be verified mathematically by the Smoluchowski equations ( Equations 19–21 ) . We next tried fitting the τ3D histograms derived from different degrees of clustering to single or two-component decay models ( Equations 17–18 ) . Interestingly , a single component model failed to fit the data when the enhancer sites become more and more densely clustered while a two-component model fits the entire range of cluster density data well ( Figure 6C ) . These results suggest that the enhancer clustering behavior itself may be sufficient to bifurcate the target search process into at least two components: a local search mode inside enhancer clusters and a long-range mode for searching outside of clusters . Together , these simulation results help clarify the effect of enhancer clustering on global TF target search efficiency in a non-equilibrium state and also reinforce the notion that the Sox2 target search process can follow two distinct modes: a local search process dominated by a binding dominant mechanism and a long-range mode for TFs to search between EnCs that is dominated by a 3D exploration mechanism as suggested previously by our independent SMT experiments ( Figure 5 ) . As a first step towards deciphering the mechanisms that underlie enhancer clustering , we next asked whether modulation of the epigenome would change the Sox2 enhancer clustering behavior in single live cells . Specifically , we applied our single-molecule , light sheet imaging strategy to map Sox2-enhancer 3D organization in TSA treated ES cells ( Figure 2—figure supplement 1B and Video 13 ) . Interestingly , the pair correlation function of Sox2 EnCs in TSA treated cells showed profiles of significantly decreased clustering , more similar to those of H2B , indicated by the decreased fluctuation amplitudes and increased fluctuation ranges ( Figure 2—figure supplement 1D , Supplementary file 1 and Figure 7A ) . Thus , it seems that TSA treatment , thought to decondense chromatin , makes specific and stable Sox2 binding sites become more randomly distributed in the nucleus . One possibility is that dysregulation of histone deacetylation activities after TSA treatment significantly alters the Sox2 binding profile in the genome to a more random state; the other possibility is that TSA treatment redistributes the 3D localization of existing Sox2 binding sites in the nucleus . To distinguish between these two possible mechanisms , we performed Sox2 ChIP-exo experiments in TSA treated ES cells and compared the resulting Sox2 genome-wide binding profile to Sox2 chromosomal localizations in wild type ( WT ) ES cells . Upon TSA treatment , we observed a much more random distribution of Sox2 ChIP-exo peaks across different chromosomes and with regard to transcription start sites ( Figure 7B , C , Supplementary file 1 ) , favoring the scenario that TSA treatment significantly increased the chances for Sox2 to bind more randomly throughout the genome . These results suggest that a finely balanced epigenetic regulation can influence the maintenance of normal enhancer clustering in the nucleus . 10 . 7554/eLife . 04236 . 013Figure 7 . Epigenetic perturbation of enhancer clustering and genome-wide binding . ( A ) The fluctuation range ( x ) and amplitude ( y ) were obtained by fitting the pair-correlation function of the indicated dataset with the fluctuation model . Figure 2 and Figure 2—figure supplement 1 , Equations 10–13 . Supplementary file 1 . Data from the same condition were grouped in separate ellipses . ( B ) Sox2 ChIP-exo peak density distribution in the wild-type and TSA treated ( red dotted ) cells across chromosome 1 , 2 , 3 . In the upper panels , each chromosome was divided to 500 bins . The color map correlates with the number of peaks in each bin . Top 7000 binding sites were considered in each condition . ( C ) Cumulative density histogram of the distances to transcription start sites ( TSS's ) of Sox2 ChIP-exo peaks in WT , Sox2 ChIP-exo peaks in the TSA treated cells ( red dotted ) , and random genomic positions ( gray ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04236 . 01310 . 7554/eLife . 04236 . 031Video 13 . Reconstructed Sox2 stable binding sites in the TSA treated live cell nucleus . HaloTag-Sox2 stable binding sites in the TSA treated live cell nucleus ( 7000 , >3 s ) were localized , tracked , and reconstructed with a color map same as that of Figure 1C . The unit is nm . DOI: http://dx . doi . org/10 . 7554/eLife . 04236 . 031 Single-molecule tracking experiments coupled with in silico simulations reveal that enhancer clustering favors local spatial fine-tuning of search parameters at the expense of global search efficiency . In particular , we find that inside enhancer clusters , Sox2 displays significantly faster forward association rates ( Figures 5 and 6 ) , thereby increasing local TF concentrations , allowing rapid rebinding to stretches of open chromatin and probably also facilitating the local target acquisition process . The shortened τ3D provides a greater opportunity for re-cycling pre-assembled TF complexes and taking advantage of cooperative interactions between TFs on chromatin . Interestingly , our simulation studies suggest that even subtle changes in the position of target genes within individual clusters can lead to alterations in local target search features . For example , gene targets at the center of EnCs can capitalize on different target search features relative to genes in the periphery of enhancer clusters ( Figure 8B ) . These results suggest that the local TF target search mode may be exquisitely modulated within distinct sub-nuclear environments and serve as an important mechanism for fine-tuning the rates of TF complex assembly at specific cis-regulatory elements . Two-color imaging revealed that enhancer clusters are spatially segregated from heterochromatic regions but overlap with a subset of Pol II enriched clusters ( Figure 4 ) . Single molecule tracking of Sox2 binding and diffusion dynamics in EnCs vs HCs indicates that in contrast to the previously estimated fraction ( ∼74% ) of Sox2 molecules engaged , on average , in 3D diffusion , the majority ( ∼64% ) of Sox2 molecules within EnCs were found to be in a chromatin bound state ( Figure 5 ) . This new finding suggests that local higher concentrations of ‘open’ chromatin in EnCs likely give rise to a dramatically reduced τ3D leading the target search process to switch from a 3D diffusion dominant mode to a binding dominant search mechanism perhaps more similar to the action of LacI in bacteria ( Elf et al . , 2007 ) . By contrast , we found a relatively low Sox2 bound fraction in heterochromatic regions . Intriguingly , in certain HC regions , Sox2 diffuses with slower rates compared to the average rate in the nucleus . This result is consistent with a previous report that even GFP molecules diffuse much slower in HCs possibly due to molecular crowding effects ( Bancaud et al . , 2009 ) . Our findings suggest that the Sox2 target search process can be divided into at least two distinct modes in the ES cell nucleus ( Figure 8A ) , ( 1 ) a chromatin binding dominant , local mode within individual EnCs possibly involving Sox2 sliding along short stretches of naked DNA as demonstrated by our in vitro TIRF single-molecule experiments ( Chen et al . , 2014b ) and ( 2 ) a 3D diffusion dominant long-range mode in which Sox2 molecules must tunnel through HCs and travel between EnCs . Our simulations also suggest that enhancer clustering itself is sufficient to generate these two modes of target search . Together , these findings reveal previously unappreciated principles governing Sox2 target search patterns within distinct sub-nuclear regions of ES cells and provide insights into how enhancer clustering can modulate local target search dynamics that could ultimately influence local transcription rates . From the earliest cloning and characterization of classical sequence specific transcription factors , a striking yet puzzling feature was the discovery of simple repetitive largely unstructured amino acid motifs ( i . e . , Gln-rich , Pro-rich , acidic repeats ) that serve as ‘activation domains’ ( ADs ) coupled to DNA binding domains ( Courey and Tjian , 1988 ) . More recent evidence suggests that such simple repetitive amino acid motifs , now referred to as low-complexity ( LC ) sequences , are found in a variety of regulatory proteins ( such as FUS , TAF15 , and EWS ) and can be induced to form fibrous polymers in vitro to mediate interactions with the CTD of RNA polymerase in a phosphorylation-state dependent manner ( Kwon et al . , 2013 ) . However , in vitro polymer formation required protein concentrations ( 0 . 7–2 mM ) , ∼1000× greater than typical concentrations of TFs ( low micro-molar ) found in vivo ( Chen et al . , 2014b ) . One mechanism proposed to enhance polymer formation involves RNA molecules seeding higher-order assemblies via the intrinsic RNA binding capacity of select regulatory proteins ( Schwartz et al . , 2013 ) . Interestingly , the activation domains of Sox2 are predicted to be unstructured LC domains enriched for G/S/P residues . Importantly , the C-terminal Sox2 AD contains five repeats of degenerate ( G/S/D/H ) Y ( G/S/D/H ) sequences that have been reported to form fibrous polymers in vitro ( Kwon et al . , 2013 ) . We speculate that the in vivo Sox2-enhancer clustering observed in our live cell studies opens the possibility that local higher concentrations of both TFs and specific DNA binding sites within EnCs may promote the formation of Sox2 LC AD containing polymers at least transiently . We envision that these Sox2-enhancer clusters could serve as multivalent docking sites for dynamic TF recruitment via weak protein:protein interactions potentially directed by LC containing proteins . Such ‘clouds’ of weak multivalent protein:protein interactions would be assisted by stronger sequence specific protein:DNA transactions that together build an activated enhanceosome . These transiently formed EnC clusters may , in turn , regulate local TF concentrations and dictate local target search dynamics of key transcriptional pre-initiation components including Pol II , GTFs , and chromatin remodeling complexes . It is tempting to speculate that the Sox2 enhancer cluster and its associated co-factors could thus form the local hub for coordinated and synergistic gene regulation ( Figure 8C ) . Whether the EnCs we observe represent actively transcribed regions remains unclear but the significant co-localization between EnCs and Pol II would be consistent with such an interpretation . Since interactions between classical LC activation domains such as those reported in the original studies of Sp1-Sp1 and Sp1-TAF4 interactions have also been suggested to be important for DNA loop formation and transcription activation in vitro ( Mastrangelo et al . , 1991; Su et al . , 1991; Freiman and Tjian , 2002 ) , it will be instructive in the future to probe whether these prevalent LC mediated transactions also contribute to the maintenance and structural integrity of enhancer clusters . Another intriguing feature of this revised model of gene regulation is that physical proximity but not direct or stable interactions between distal enhancer elements and gene proximal promoters is necessary for delivering transcription activation by cis-elements at a distance . We envision that enhancer clustering with its higher local TF/cofactor concentrations accompanied by altered target search features may be sufficient to serve as an alternative mechanism for achieving distal enhancer directed transcription activation long recognized as a hallmark of mammalian gene control . Our results also suggest that gene and promoter positioning in relationship to EnC and HC territories is critical for optimal fine-tuning of transcriptional activities . Important questions left unresolved by our present study include: how many genes/promoters are present within a cluster; what is the relationship between our 3D clusters and TADs ( topologically associated domains ) ; and are the enhancer binding sites within a cluster all from a given chromosome or is there evidence of transvection occurring as well ? The enhancer clustering behavior that we observed fits generally with the concepts deduced from studies of linearly arrayed enhancers in the genome identified by ChIP-seq analysis ( Whyte et al . , 2013 ) and topological domains identified by Hi-C experiments ( Dixon et al . , 2012 ) . These studies , taken in aggregate , suggest that gene transcription is compartmentalized within topological or spatial territories in the nucleus segregated from other silent regions . However , given the orthogonal nature of these diverse methods of probing genome organization , it is difficult at this stage to firmly establish either direct structural or functional links between the genome-wide ensemble studies and our observation of enhancer clustering by single molecule imaging . We note , for example , that enhancer clustering does not appear simply to result from differential chromatin packaging . Specifically , the fluctuations of Sox2 enhancer densities in the nucleus are smaller in sizes ( ε ) but much larger in amplitudes ( A ) than chromatin ( H2B ) densities ( Figure 7A ) . Two distinct mechanisms could account for this observation , ( 1 ) Sox2 binding sites are already clustered in the linear genome and chromosome folding based on the polymer model ( reviewed in Tark-Dame et al . , 2011 and Fudenberg and Mirny , 2012 ) , automatically leads to 3D cluster formation even without TF directed chromatin looping interactions; ( 2 ) extensive active and TF directed long distance chromatin looping brings distal Sox2-enhancer sites along the linear chromosome to form local 3D clusters . To distinguish between these two potential scenarios , we analyzed the linear arrangement of Sox2 binding sites genome wide using ChIP-exo . Indeed as suspected , many Sox2 binding sites already form clustered arrays along the linear chromosome ( Whyte et al . , 2013 ) . Interestingly , disrupting the epigenome alters the accessibility of linearly arranged Sox2 binding sites as well as the extent of 3D enhancer clustering ( Figure 7 ) . This finding suggests that linearly arrayed Sox2 binding sites likely contribute substantially to the formation of enhancer clusters in 3D but do not exclude the possibility that TF directed chromatin looping also contributes to such an organization of actively transcribed loci . Future modeling and simulation experiments will be required to functionally link the linear genomic localization of different TF binding sites with their 3D spatial distributions in the nucleus to gain further insights into genome organization and chromatin folding . It also remains unclear what forces or exogenous structures are in play to maintain the star-burst arrangement of enhancers in the nuclear volume . Our studies provide a basis for understanding how the 3D organization of enhancers into localized clusters could affect TF target search dynamics and influence local transcription rates . Our imaging analysis of live ES cells suggests that the nucleus is partitioned into multiple levels of spatially segregated functional domains . For example , many Pol II enriched regions do not overlap with Sox2 EnCs although , as might be expected , most Sox2 EnCs do overlap with Pol II clusters ( Figure 4 ) . We also observed extensive residual ‘dark’ spaces that are not significantly occupied by either heterochromatin or Sox2 binding sites ( Figure 3B and Video 9 ) . It seems likely that other uncharacterized enhancer bearing sub-nuclear domains occupy these ‘dark’ territories and influence local gene activity not detected by our current assays . We speculate that galaxies of such 3D clusters of cis-regulatory domains are formed by specific binding of different combinations of TFs we have long suspected but could not discern from classical bulk biochemistry . It will be interesting in the future to complete this 3D mapping of the nucleome to discern which other cadre of factors might reside in these ‘dark’ regions . Another aspect to address will be the degree of spatial overlap between different functional regions created by the stable binding and local concentrations of different classes of TFs . Ultimately we would like to have a more complete understanding of how the 3D organization of these cis-elements specifically influences gene activity and what gene products and mechanisms underlie the formation of these clusters . Addressing these questions will be essential for a deeper understanding of how enhancer-mediated gene regulation works . The ongoing development of simultaneous multi-color super-resolution imaging systems , enhanced dye chemistry , and single gene locus labeling strategies will be essential to address these fundamental questions . Mouse D3 ( ATCC , USA ) ES cells were cultured on 0 . 1% gelatin coated plates in the absence of feeder cells . The ES cell medium was prepared by supplementing knockout DMEM ( Invitrogen , Carlsbad , CA ) with 15% FBS , 1 mM glutamax , 0 . 1 mM nonessential amino acids , 1 mM sodium pyruvate , 0 . 1 mM 2-mercaptoethanol , and 1000 units of LIF ( Millipore , USA ) . 1 day before imaging experiment , cells were plated onto a clean cover glass pre-coated with Matrigel ( BD Biosciences , USA , 356230 ) . In the TSA perturbation experiments , ES cells were treated with 50 nM TSA ( Sigma-Aldrich , USA: T8552 ) for 8 hr prior to imaging and ChIP-exo mapping . Mouse HP1 ( Cbx5 gene: NM_007626 ) cDNA was first amplified by PCR from ES cell cDNA libraries and then inserted into a custom-constructed Piggybac transposon vector that harbors the E1F alpha promoter , the internal ribosome entry site ( IRES ) , and the PuroR gene . eGFP cDNA was further cloned to fuse with HP1 at its N-terminus . Stable cell lines were generated by co-transfection of stable HaloTag-Sox2 ES cells established in our previous work ( Chen et al . , 2014b ) with the HP1 overexpression piggybac vector and a helper plasmid that over-expresses Piggybac transposase ( Supper Piggybac Transposase , System Biosciences , USA ) . 48 hr post-transfection , cells were subjected to puromycim ( Invitrogen Carlsbad , CA ) selection ( 1 µg/ml ) . After 3 days of selection , cells were maintained in their culturing medium with a 0 . 5 µg/ml final concentration of puromycin . Similarly , Dendra2-Rpb1 mutant cDNA was cloned into the piggybac vector and co-transfected into the HaloTag-Sox2 ES cells with the helper plasmid . α-amanitin ( Sigma-Aldrich , USA: A2263 ) selection was conducted by using a final concentration of 3 μg/ml . 10 days after selection , stable cell clones appear on the place . For the long-term maintenance , 1 μg/ml α-amanitin was supplemented into the culturing medium . For electroporation , ES cells were first dissociated by trypsin into single cells . Transfection was conducted by using the Nucleofector Kits for Mouse Embryonic Stem Cells ( Lonza , USA ) . All imaging experiments were performed in the ES cell imaging medium , which was prepared by supplementing FluoroBrite medium ( Invitrogen , Carlsbad , CA ) with 10% FBS , 1 mM glutamax , 0 . 1 mM nonessential amino acids , 1 mM sodium pyruvate , 10 mM Hepes ( pH 7 . 2–7 . 5 ) , 0 . 1 mM 2-mercaptoethanol , and 1000 units of LIF ( Millipore , USA ) . For 3D lattice light-sheet imaging condition , we optimized HaloTag-JF549 concentrations in the medium to a final concentration of ∼0 . 1 fM . The ligand molecules gradually diffuse into the cell and label the HaloTag-Sox2 molecules . Optimal single-molecule labeling density was achieved when the labeling rates equilibrated with the photo-bleaching rates . Due to the light-sheet selective plane illumination , the relative long acquisition time ( 40 ms ) , and ultralow ligand concentration in the medium , negligible fluorescent background signals were observed . For 2D wide-field imaging condition , we first tested the optimal HaloTag-JF549 and HaloTag-JF646 labeling concentrations . Briefly , several concentrations of HaloTag-JF549 and JF646 ( 0 . 5 nM , 1 nM , 2 nM , and 5 nM ) were used to treat cells for 15 min and then cells were washed with imaging medium for three times . The cover glasses were then transferred to live-cell culturing metal holders and mounted onto the microscope one by one . Proper HaloTag-JF549 or HaloTag-JF646 labeling concentrations were determined by the criterion that single-molecules can be easily detected under 2D imaging mode after a minimal 2–5 s pre-bleaching . After fixing the labeling concentration for each cell line , we then proceeded to perform the 2D single-molecule imaging experiments . 3D single-molecule tracking experiments were performed via lattice light sheet plane illumination microscopy using a modified version of the multi-Bessel microscope described previously ( Gao et al . , 2012 ) . The modification consists of a massively parallel array of coherently interfering beams comprising a non-diffracting 2D optical lattice , rather than a set of seven noninterfering Bessel beams . This creates a coherent structured light sheet that can be dithered to create uniform excitation in a 400 nm thick plane across the entire field of view . The experimental hardware is the same as before , except that a binary spatial light modulator ( SXGA-3DM , Forth Dimension Displays , Valencia , CA ) is placed conjugate to the sample plane , and a binarized version of the desired structured pattern at the sample is projected on the display . For imaging , a 500 mW cw488 nm ( Coherent , Santa Clara , CA ) or a 500 mW cw561 laser ( MPB Lasertech , Edmonton , AB ) were used . A custom 0 . 65 NA objective for excitation ( Special Optics , Wharton , NJ ) and a 25× , 1 . 1 NA objective for detection ( Nikon , USA , MRD77220 ) are employed . A multi-band pass filter ( Semrock , FF01-446/523/600/677-25 ) is placed before a CMOS camera ( ORCA-flash4 . 0 , Hamamatsu , Japan ) to filter the excitation wavelengths . Single molecule imaging of individual cells was performed by serially scanning the entire cell nucleus through the light sheet at 20–50 ms exposure per 2D image and 300 nm z-steps resulting in a 3D imaging rate of 3 s per volume . Although significantly faster imaging rates are possible , these conditions were chosen to minimize photo-bleaching and phototoxicity , while specifically selecting stably bound ( >3 s ) molecules . Correlation of stable binding sites with heterochromatin regions was performed by first acquiring a single 3D volume of GFP-HP1 followed by single molecule imaging as described above . 3D localization ( x , y , z ) was conducted using FISH-QUANT software ( Mueller et al . , 2013 ) . The PSF model can be described by the following equation: ( 1 ) I ( x , y , z ) = ( A0e− ( x−x0 ) 22σxy2e− ( y−y0 ) 22σxy2e− ( z−z0 ) 22σz2 ) PSF+B , where A0 is the signal amplitude; σ is the Standard Deviation ( S . D . ) of the Gaussian fit in the indicated direction , in our case S . D . of the x , y direction is the same; B is the number of background photon count . Image registration and drift correction were performed by calculating the centroid displacement of total localization events from every 50 time points ( 2 . 5 min ) and the resulting transformation matrix over time was applied to the data accordingly . We found that this method can efficiently correct drifts which were not significant ( 0–800 nm per minutes ) within the correction time window . Any significantly drifted dataset was not used for later tracking analysis . Localization uncertainty can be calculated by the estimator below ( Rieger and Stallinga , 2014 ) . ( 2 ) Δ2=σ2+a212N ( 169+4τ ) With τ roughly equal to the ratio between the background intensity and the peak signal intensity , which can be directly obtained from the FISH-quant localization program . a , the voxel size in the selected direction . N , total photo count was calculated by integrating voxel photon counts covered by each Gaussian spot . U-track algorithm ( Jaqaman et al . , 2008 ) was used for 3D single particle tracking . For mapping Sox2 stable binding site in live cells , we only reconstructed the first events of track fragments which have step and end-to-end displacements less than 50 nm and have lengths longer than the indicated cutoff time . The final 3D image representation was performed by either ViSP ( El Beheiry and Dahan , 2013 ) or Imaris . 2D single molecule experiments were conducted on a Nikon Eclipse Ti microscope equipped with a 100× oil-immersion objective lens ( Nikon , N . A . = 1 . 4 ) , a lumencor light source , two filter wheels ( Lambda 10-3 , Sutter Instrument , Novato , CA ) , perfect focusing systems , and EMCCD ( iXon3 , Andor , UK ) . Proper emission filters ( Semrock , Rochester , NY ) was switched in front of the cameras for GFP , JF549 , or JF646 emission and a band mirror ( 405/488/561/633 BrightLine quad-band bandpass filter , Semrock , Rochester , NY ) was used to reflect the laser into the objective . For two color single-molecule experiments with JF646 and JF594 labeled HaloTag-Sox2 , we used a 630-nm laser ( Vortran Laser Technology , Inc . ) of excitation intensity ∼60 W cm−2 and a 561-nm laser ( MPB Lasertech , Edmonton , AB ) of excitation intensity ∼800 W cm−2 and the acquisition times are 500 ms ( 630 nm ) and 10 ms ( 561 nm ) . For two color experiments mapping the spatial relationship of heterochromatin and enhancer clusters , we used a SOLA light engine ( Lumencor , Beaverton , OR ) and a 561-nm laser ( MPB Lasertech , Edmonton , AB ) of excitation intensity ∼50 W cm−2 and the acquisition times are 100 ms ( GFP ) and 500 ms ( 561 nm ) . After mapping stable Sox2 binding sites by using the JF646 dye , Dendra2-Rpb1 PALM experiment was performed using the 560-nm laser ( MPB Lasertech , Edmonton , AB ) of excitation intensity ∼1000 W cm−2 for single-molecule detection and a 405-nm laser ( Coherent , Santa Clara , CA ) of excitation intensity of 40 W cm−2 for photo-switching of Dendra2-Rpb1 . The acquisition time is 30 ms . Total ∼10 , 000 frames were recorded . ∼20k localized events were used for the final imaging reconstruction . For two color experiments probing the Sox2 diffusion properties in heterochromatin regions , we used a SOLA light engine ( Lumencor , Beaverton , OR ) and a 561-nm laser ( MPB Lasertech , Edmonton , AB ) of excitation intensity ∼800 W cm−2 and the acquisition times are 100 ms ( GFP ) and 10 ms ( 561 nm ) . The microscopy , lasers , the SOLA light engine , and the cameras were controlled through NIS-Elements ( Nikon , USA ) . For 2D single molecule tracking , the spot localization ( x , y ) was obtained through 2D Gaussian fitting based on MTT algorithms ( Serge et al . , 2008 ) using home-built Matlab program . The localization and tracking parameters in SPT experiments are listed in the Supplementary file 1 . To map stable bound sites in the low excitation , slow acquisition ( 500 ms ) condition , 0 . 05 µm2/s was set as maximum diffusion coefficient ( Dmax ) for the tracking . The Dmax works as a limit constraining the maximum distance ( rmax ) between two frames for a particle random diffusing during reconnection . Therefore , for events lasted more than one frames , only molecules localized within rmax for at least two consecutive frames will be considered as bound molecules . Since we used relatively long acquisition time ( 500 ms ) to blur the image of fast diffusing molecules , events that appeared in single frames were also taken into consideration as bound molecules to have a track length of 0 . 5 s . The duration of individual tracks ( dwell time ) was directly calculated based on the track length . We used 2 s as the time cutoff for mapping stable binding events . MTT algorithm was used to track fast TF dynamics in the high excitation , fast acquisition ( 10 ms ) condition . The resulting tracks were inspected manually by a homemade Matlab program . Tracks with incorrect linking events were discarded . We took GFP-HP1 images before and after SMT experiment to make sure the cell nucleus and HC regions have not moved during the 5–6 min of single-molecule imaging . For experiment investigating co-localization of Pol II-enriched regions and Sox2 EnCs , image registration was performed by calculating and aligning nucleus outlines from both datasets . After background subtraction , the intensity map for heterochromatin regions in single cells was directly calculated by normalizing pixel intensity in the GFP-HP1 channel with the highest pixel intensity in the image . The intensity map for Dendra2 Pol II or stable Sox2 binding sites was calculated by 2D Gaussian kernel density function implemented by Matlab . Specifically , the density probability of X , Y localizations of stable binding events was evaluated in a 100 × 100 matrix with arbitrary units . The bandwidth for density estimation is 2 units . The resulting probability map was rescaled to the original image size . Composite images were constructed by superimposing the two intensity maps as two independent color channels . Binary mask for heterochromatin regions or enhancer clusters was calculated by applying a threshold cutoff of 0 . 2 to the intensity map . 2D single-molecule tracks were divided to track segments resided in the mask and outside of the mask . Track segments from each catalog were pooled . Diffusion coefficients were calculated from tracks with at least eight consecutive frames by the MSDanalyzer ( Tarantino et al . , 2014 ) with a minimal fitting R2 of 0 . 8 . According to Peebles and Hauser ( 1974 ) , we define the pair correlation function g ( r ) measures the probability dP of finding an enhancer site in a volume element dV at a separation r from another enhancer site . ( 3 ) ΔP=ng ( r ) ΔV , where n is the mean number density of the enhancers in the nucleus . In practice , the pair correlation function can be estimated from a sample of objects counting the pairs of objects with different separations r [Peebles & Hauser [4] estimator]: ( 4 ) g ( r ) =NRNDD ( r ) RR ( r ) , where DD ( r ) and RR ( r ) are counts of pairs of enhancers ( in bins of separation ) in the data catalog and in the random catalog , respectively . The random catalog consists of uniformly distributed positions in the same volume defined by data catalog 3D convex hull . To reduce the noise , we computationally generate the random catalog that has a size 10 times greater than that of the data catalog . The normalizing coefficients containing the numbers of points in the initial ( N ) and random ( Tarantino et al . ) catalogs are included in the estimator . Here , non-redundant pair wise Euclidean distance set within each catalog can be constructed by ( 5 ) dst ( i , j ) ( i≠j ) =‖ri→−rj→‖ . We define: ( 6 ) C ( dst ( i , j ) , r ) ={0 ( dst ( i , j ) >r ) 1 ( dst ( i , j ) ≤r ) . The bin size of the g ( r ) distribution function is Δr . Then , ( 7 ) g ( r ) =NRNDD ( r ) RR ( r ) =NRN∑i , jCDD ( dstDD ( i , j ) , r+Δr2 ) −∑i , jCDD ( dstDD ( i , j ) , r−Δr2 ) ∑i , jCRR ( dstRR ( i , j ) , r+Δr2 ) −∑i , jCRR ( dstRR ( i , j ) , r−Δr2 ) . DD ( r ) and RR ( r ) are calculated by pair wise distance function supplied in Matlab 2013a version with 50 nm as the histogram bin . For investigating the spatial cross-correlation between the localizations of two factors , we first converted the 2D super-resolution localization densities to image intensity maps via a 2D Gaussian kernel density function ( see details in Intensity Map Calculation , Mask Definition , and TF Diffusion Analysis ) . Then , we implemented the Pair Cross-Correlation function using a well-established fast Fourier transform based method ( Veatch et al . , 2012 ) . Specifically , ( 8 ) Cross-Correlation Function , c ( r→ ) =Re{FFT−1 ( FFT ( I1 ) ×conj[FFT ( I2 ) ] ) ρ1ρ2N ( r→ ) } , ( 9 ) N ( r→ ) =FFT−1 ( \|FFT ( Mask ) \|2 ) . The normalization fact N ( r→ ) is the autocorrelation of a mask that has the value of 1 inside the nucleus region of the cell . The cell nucleus mask was obtained from the GFP-HP1 or Dendra2-Pol II wide-field image by intensity thresholding . Here , conj[] indicates a complex conjugate . FFT and FFT−1 were implemented by fft2 ( ) and ifft2 ( ) functions in Matlab . ρ1 and ρ2 are the average surface densities of images I1 and I2 respectively , and Re{} indicates the real part . Autocorrelation was calculated by using identical I1 and I2 . This computation method of tabulating pair cross-correlations is mathematically similar to brute force averaging methods . Correlation functions were angularly averaged using polar coordinates ( Matlab command cart2pol ( ) ) , and then binning by radius . Final values are obtained by averaging within the assigned bins in the radius . Because the intensity map pixel size is 160 nm after the 2D Gaussian kernel density estimation , we only calculated pair cross-correlation function at a range of diffraction limited radii ( r > 160 nm ) . In this regime , over-counting has negligible effects on the final output of auto- or cross-correlation function . Permutation was performed by randomizing pixels spatially within the nucleus mask for both images before calculating the cross-correlation . We extended previously published fluctuation model for measuring two dimensional heterogeneous distribution of membrane proteins to quantify 3D enhancer clustering ( Sengupta et al . , 2011 ) . Specifically , ( 10 ) G ( r ) observed=G ( r ) stoch+G ( r ) enhancer⊗G ( r ) PSF , G ( r ) observed , the observed pair correlation function as calculated in the previous section . G ( r ) stoch , the contribution of multiple appearances of the same molecule at a fixed site to the measured total correlation function . In our case , molecules are sparsely labeled . And , we track TF molecules through time/frames and , for each stable binding event , we only count once with the average localization over multiple frames . Thus , the contribution of G ( r ) stoch is negligible under this condition . An exponential function can be used to approximate the correlation function of enhancers if they are present in randomly distributed clusters of no defined shape . ( 11 ) G ( r ) enhancer=AExp ( −rε ) +1 , A , the fluctuation amplitude which is in proportion to the ratio of the density of enhancers in clusters to the average density across the entire space . ε , the fluctuation range which is in proportion to the size of the clusters . The correlation function of PSF of the imaging method is denoted as G ( r ) PSF and can be approximated by ( 12 ) G ( r ) PSF=18π32σ¯3Exp ( −r24σ¯2 ) , σ¯ , is calculated by σ¯2=s¯2+a¯212 , wherein s¯ is the average s . d . of the PSF and a¯ is the average voxel dimension . Then , the final observed pair-correlation function can be fitted by the equation below: ( 13 ) G ( r ) observed= ( AExp ( −rε ) +1 ) ⊗ ( 18π32σ¯3Exp ( −r24σ¯2 ) ) , ⊗ , denotes convolution operator . For the uniformly distributed , simulated sites , the data were fitted with G ( r ) enhancer directly . Curve fitting is performed using the trust-region method implemented in the Curve Fitting Matlab toolbox . According to Einstein's theory , the mean square displacement of Brownian motion is described as ( 14 ) 〈r2〉=2dDt , d , dimensionality , in our case , d = 3 . D , diffusion coefficient . To computationally simulate Brownian motion in the Cartesian coordinate system , we uncoupled each jump to x , y , z one dimensional steps defined by the equation below . ( 15 ) ( x ( t+δt ) y ( t+δt ) z ( t+δt ) ) = ( x ( t ) y ( t ) z ( t ) ) +2Dδt ( N1N2N3 ) , where Ni are independent random numbers obeying Gaussian distribution with a zero mean and a variance of 1 and dt is the sampling interval . Simulation of target search was performed with MathWorks Matlab 2013a . The target search problem was reduced to random walk trapping problem with boundary and multiple traps . Specifically , we limited the 3D diffusion of the TF in a nucleus with a radius of 5 µm . Considering the average length of nucleosome depleted regions as 100–200 base pairs and the persistent length ( lp ) of naked DNA as about 45 nm ( 135 bps ) ( Williams and Maher , 2011 ) , target site radius was set as 40 nm . Overlaps between targets were not allowed in our simulation experiment . Specifically , the minimal distance ( 80 nm ) allowed between the center of two targets is two times of the target radius ( 40 nm ) . In our simulation experiment , the TF binding probability to target is 1 when the TF reaches individual targets . The number of target sites was 7000 as estimated in our previous work . The mean diffusion coefficient ( D ) of the TF is 10 μm2s−1; the sampling interval ( δt ) is 10 ms . Under this condition , the X , Y , Z step sizes are about 14 nm ( when Ni = 1 ) much smaller than the target size , suggesting that the space is not under-sampled . We computationally manipulated the spatial distribution of target site and injection site position of the TF in the nucleus as indicated in the specific experiment and we recorded the first passage ( 3D ) time and trajectory of each trial before the TF was reaching the first target according to the first-hitting-time model in the survival theory . Relative Fluorescence Intensity ( RFI ) for probing Sox2 levels in heterochromatin . ( 16 ) RFI=IHeterochromatin−IBackgroundISurrounding−IBackground , I , stands for mean gray intensity for the selected region . Single-component exponential fitting of τ3D ( 17 ) Density ( τ ) =e−τt , t , the mean lifetime . Two-component exponential fitting of τ3D ( 18 ) Density ( τ ) = Fe−τt1+ ( 1−F ) e−τt2 , t1 , t2 the mean lifetime for each component . The relationship between enhancer concentrations and 3D time ( τ3D ) . According to the Smoluchowski Equation , ( 19 ) kon=4πRD , R , capture radius . D , diffusion coefficient . Observed on-rates for TFs are defined by the following equation , ( 20 ) kon∗=kon[DNA]=konρ ( r ) =4πRDρ ( r ) , where enhancer concentrations ( [DNA] ) are a function ( ρ ( r ) ) of r relative to the center of the cluster . This is the equation linking enhancer concentrations ( ρ ( r ) ) to τ3D . ( 21 ) τ3D=1kon∗=14πRDρ ( r ) ES cells were treated with 50 nM TSA ( Sigma-Aldrich , USA: T8552 ) for 8 hr . Then , cells were cross-linked by formaldehyde and harvested . Chromatin Immunoprecipitation ( ChIP ) was performed according to Boyer et al . ( 2006 ) with minor modifications . Briefly , cross-linked ESC chromatin was sheared using Covaris S2 system to a size range of 100 bp–400 bp . Immunoprecipitation was conducted with either specific antibody conjugated Protein A Sepharose beads ( GE Healthcare ) . ChIP-exo library was prepared by following the published protocol with minor modifications ( Rhee and Pugh , 2011 ) . Specifically , we adapted the SoLid sequencer adaptors/primers to make the final library compatible with the illumina Tru-seq seq small-RNA system . Anti-Sox2 ( R&D Systems , Minneapolis , MN Cat . # AF2018 , Lot # KOY0112011 ) antibody was used for the ChIP experiment . The detailed primer information is in Supplementary file 1 . We sequenced exo libraries in 60 bp ( Sox2 TSA ) single-end format by using the illumina HiSeq platform . After removal of the 3′ most 24 bp ( Sox2 TSA ) or 14 bp ( Sox2 Wild type: 50 bp reads ) which tend to have higher error rates , we mapped our sequencing data back to the mouse reference genome ( mm10 ) by Bowtie 2 ( Langmead and Salzberg , 2012 ) . After mapping , we normalized the total mapped reads for each factor to 40 million . We further reduced the mapped read regions to single 5′-end point , which reflects the cross-linking point between protein and DNA . The resulting cross-linking point distribution was used to identify peaks on the forward ( Left ) and reverse ( Right ) strand separately using the peak calling algorithm in GeneTrack ( Albert et al . , 2008 ) . For bound-region calculation , we first identified any pairs of left and right peaks that were located within 20 bps to each other . Then , we defined the window between the middle point of the left peak and that of the right peak as the bound-region . Peak-pairing and bound-region calculation were performed with Python programming ( the script is available at https://github . com/Jameszheliu/PeakPairingProgram ) . Sox2 ChIP-exo sequencing data using wild type ES cells were obtained from GEO with the accession number of GSM1308179 ( Chen et al . , 2014b ) . Sox2 ChIP-exo data using TSA treated ES cells were deposited to NCBI GEO with the accession number of GSE62972 .	Stem cells in an embryo have the potential to become any type of cell in the body . When a cell begins to specialize , it loses this ability and can only become a limited number of cell types . These transitions are caused by changes in gene expression . Proteins called transcription factors bind to DNA to switch different genes on or off as cells become specialized . One such transcription factor , called Sox2 , binds to particular DNA sequences in the cell's nucleus to encourage nearby genes to be expressed at the right levels and keep a stem cell unspecialized . However , how these binding sites are positioned throughout the three-dimensional space inside the nucleus was unknown , as was the likelihood of Sox2 finding and binding to these sites . Now Liu et al . have taken advantage of advanced microscopes to observe the interaction between Sox2 and its binding site in the nucleus of living embryonic cells . This three-dimensional imaging technology is powerful enough to capture images of individual molecules; and Liu et al . attached fluorescent tags to Sox2 to make it easier to watch them in action . By making a series of time-lapse movies , it was revealed that instead of being evenly scattered in the nucleus , Sox2's binding sites are grouped together to form individual clusters; these clusters preferably occupy spaces in the nucleus that are likely enriched for active genes . Liu et al . suggest that the clustering of Sox2 binding sites makes it more difficult for a Sox2 protein to find these sites at first , but much easier to find when the Sox2 protein is near to the cluster . Thus , the uneven positioning of the binding sites for transcription factors may provide an additional layer of control over gene expression . In the future , it would be important to map Sox2's binding sites while visualizing the activities of single genes in living cells . This would improve our understanding of how the structural organization of the contents of the nucleus can influence the correct timing of specific patterns of gene expression .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "structural", "biology", "and", "molecular", "biophysics" ]	2014	3D imaging of Sox2 enhancer clusters in embryonic stem cells
Listeria monocytogenes hijacks host actin to promote its intracellular motility and intercellular spread . While L . monocytogenes virulence hinges on cell-to-cell spread , little is known about the dynamics of bacterial spread in epithelia at a population level . Here , we use live microscopy and statistical modeling to demonstrate that L . monocytogenes cell-to-cell spread proceeds anisotropically in an epithelial monolayer in culture . We show that boundaries of infection foci are irregular and dominated by rare pioneer bacteria that spread farther than the rest . We extend our quantitative model for bacterial spread to show that heterogeneous spreading behavior can improve the chances of creating a persistent L . monocytogenes infection in an actively extruding epithelium . Thus , our results indicate that L . monocytogenes cell-to-cell spread is heterogeneous , and that rare pioneer bacteria determine the frontier of infection foci and may promote bacterial infection persistence in dynamic epithelia . Editorial note: This article has been through an editorial process in which the authors decide how to respond to the issues raised during peer review . The Reviewing Editor's assessment is that all the issues have been addressed ( see decision letter ) . The widely studied foodborne pathogen Listeria monocytogenes has served as a model system to study cytoskeletal dynamics ( Theriot et al . , 1992; Welch , 1998 ) , epithelial cell biology ( Pentecost et al . , 2010 ) , and host-pathogen interactions ( Kocks et al . , 1995; Mengaud et al . , 1996 ) . This ubiquitous Gram-positive bacterium can invade and replicate within non-phagocytic cells and , importantly , use a form of actin-based motility to spread directly from the cytoplasm of an infected host cell into the cytoplasm of another host cell without exposure to the extracellular milieu ( Tilney and Portnoy , 1989 ) . This process , known as cell-to-cell spread , enables L . monocytogenes to breach and colonize the intestinal epithelium and to subsequently reach distant organs including the liver and brain in immunocompromised patients ( Ghosh et al . , 2018 ) and the placenta in pregnant women ( Faralla et al . , 2016 ) . Indeed , compared to wild-type L . monocytogenes , mutant strains incapable of undergoing cell-to-cell spread are three orders of magnitude less virulent in murine models ( Domann et al . , 1992 ) . L . monocytogenes infections begin in the intestinal epithelium , a tissue made up of polarized epithelial cells connected to each other by cell-cell junctions ( Hartsock and Nelson , 2008 ) . L . monocytogenes preferentially adheres to and invades an epithelium at the tips of intestinal villi ( Pentecost et al . , 2006 ) , where epithelial cells are actively extruded and shed ( Sancho et al . , 2004 ) . Upon bacterial invasion , L . monocytogenes spreads to neighboring host cells , which can allow bacteria to move away from the tip of a villus before the next host cell extrusion event terminates the infection . Therefore , understanding L . monocytogenes virulence requires a quantitative grasp of the spatiotemporal dynamics of cell-to-cell spread . To initiate cell-to-cell spread , L . monocytogenes uses the protein ActA to polymerize actin at its surface and create an actin comet tail ( Pistor et al . , 1994 ) . Actin polymerization generates a propulsive force that allows the bacterium to move within the host cytoplasm . Upon contact with the donor host cell membrane , the intracellular bacterium creates a protrusion that can extend into the cytoplasm of a recipient host cell ( Robbins et al . , 1999 ) . Although cell-to-cell spread has been primarily studied as a mechanism of bacterial dissemination between adjacent host cells , it is well established that L . monocytogenes can create protrusions more than ten microns long ( Pust et al . , 2005 ) , which could , in principle , mediate bacterial spread between two non-adjacent host cells . To complete cell-to-cell spread , the recipient cell engulfs the bacterium-containing protrusion , thus giving L . monocytogenes access to the recipient host cell’s cytoplasm . After escaping the double-membrane vacuole , L . monocytogenes rebuilds the actin comet tail and restarts intracellular motility ( Gedde et al . , 2000 ) . Particular attention has been paid to bacterial and host cell proteins that mediate cell-to-cell spread . The bacterial protein internalin C helps L . monocytogenes to relax cortical tension and increase the rate of bacterial-mediated protrusion formation to promote spread ( Rajabian et al . , 2009 ) . From the perspective of the host cell , it has been shown that TIM4 allows the host to sense bacterial-mediated membrane damage , which then triggers a repair mechanism that L . monocytogenes exploits to promote spread ( Czuczman et al . , 2014 ) . The diaphanous-related formins ( Fattouh et al . , 2015 ) and members of the ERM protein family ( Pust et al . , 2005 ) have been shown to localize to bacteria-containing protrusions and inhibition of their activity decreases the efficiency of cell-to-cell spread . This cell biological approach has been useful in creating a mechanistic understanding of how individual spreading events occur . However , our larger scale understanding of how a population of bacteria spreads through tissue remains poorly developed . Here , we combine live microscopy and statistical modeling to study the dynamics of a population of L . monocytogenes as it spreads through a polarized epithelial monolayer . We simulate cell-to-cell spread as an isotropic random walk because the movement of L . monocytogenes is directionally persistent over short distances but shows no preferred orientation over long distances . Our experimental and computational results indicate that L . monocytogenes cell-to-cell spread includes a majority of local-spreading bacteria but is dominated by rare pioneers , which determine the shape of infection foci . Importantly , we find that pioneers alter the kinetics of spread in a way that might promote bacterial persistence in a dynamic epithelium where cells are actively extruded , as at the tip of an intestinal villus . To explore the dynamics of L . monocytogenes cell-to-cell spread in an epithelial monolayer , we developed a live video microscopy assay to track the progression of a bacterial infection over tens of hours . As a model host cell , we chose Madin-Darby canine kidney ( MDCK ) epithelial cells because they form polarized and homogeneous monolayers in culture ( Mays et al . , 1995 ) and have been widely used to study L . monocytogenes infection ( Robbins et al . , 1999; Pentecost et al . , 2006; Pentecost et al . , 2010 ) . We infected confluent MDCK monolayers with a wild-type 10403 S L . monocytogenes strain that contains an mTagRFP open reading frame under the actA promoter , which becomes transcriptionally active when the bacterium enters the host cell cytosol ( Moors et al . , 1999; Zeldovich et al . , 2011 ) . We then imaged the progression of the infection as described in Materials and Methods . The presence of gentamicin , a bacteriostatic antibiotic that cannot cross the host cell plasma membrane ( Portnoy et al . , 1988 ) , during live imaging ensured that only intracellular bacteria contributed to the growth and spread of the infection focus . Starting at approximately 6 hr post-infection , the earliest time point at which we could detect mTagRFP protein expression , we imaged bacterial foci for up to 22 hr post-infection ( first three panels of Figure 1A , and Video 1 ) . Given that bacterial invasion of a polarized MDCK monolayer is a rare event ( Pentecost et al . , 2006 ) , each infection focus most likely began with a single bacterium entering a host cell’s cytosol . Due to the clonal nature of the replicating bacteria , and the homogeneity of the host monolayer , we were surprised to find behavioral heterogeneity within the bacterial population; the edges of the boundary of the infection focus , determined by the smallest boundary that completely encloses all bacteria , were dominated by a small number of bacteria that spread farther than the rest ( Figure 1A , white arrows in third panel ) . Indeed , this was a common phenomenon that could be observed in most infection foci . Although each focus may have started out roughly circular , far-spreading bacteria , which we refer to as ‘pioneers’ , nearly always created irregular boundaries by the end of the experiment ( Figure 1B ) . Despite boundary irregularity , intracellular bacterial replication was approximately exponential ( Figure 1C ) , and the growth rate could be modeled with a one-term exponential function ( Figure 1—figure supplement 1A ) with an average doubling time of approximately 180 min ( Figure 1—figure supplement 1C ) . This doubling time is comparable to what has been previously reported in other epithelial host cell types using gentamicin protection assays ( Gaillard et al . , 1987 ) . Between 360 and 960 min post-infection , the mean squared displacement ( MSD ) of the bacterial positions ( defined here as the second moment of the fluorescence intensity distribution ) appeared linear ( Figure 1C ) , which is consistent with a random walk ( Berg , 1993 ) . However , the slope of the MSD was not always constant , but instead increased with time ( Figure 1—figure supplement 1B ) , which is consistent with the appearance of fast-spreading organisms within a migrating population ( Shigesada et al . , 1995 ) . We found no correlation between bacterial growth rate and MSD ( Figure 1—figure supplement 1C ) , enabling us to treat these two parameters as independent in the quantitative model described below . What then is the expected range of shapes resulting from the random movement and exponential growth seen in L . monocytogenes cell-to-cell spread ? From the literature , it is expected that , when starting from a point source , random movement and growth should yield isotropic shapes ( Holmes et al . , 1994 ) . To formalize this null hypothesis , we solved the reaction-diffusion equation ( Equation 1 ) : ( 1 ) ∂ϕ∂t=D∂2ϕ∂r2+kϕwhere Φ represents the bacterial concentration as a function of position and time , t refers to time , r refers to the position of the bacteria in polar coordinates , D is the effective diffusion coefficient , and k is the exponential growth rate . Variations of this partial differential equation have been used to model dynamic biological processes such as morphogen pattern formation ( Gordon et al . , 2011 ) and animal migration ( Skellam , 1951 ) . Because Equation 1 is radially isotropic , its solutions correspond to circular infection foci that grow in size and intensity over time ( Figure 2A and Video 2 ) . Such a continuum model cannot account for the experimentally observed heterogeneous focus shapes . It is important to note that treating the bacterial concentration Φ as a continuous variable constitutes a mean-field approximation , which is valid only in the limit of high bacterial counts , and which neglects correlations in the positions of individual bacteria . However , because each bacterium behaves as a discrete entity and because the number of bacteria at the start of each infection focus is very small , the mean-field model breaks down in describing the shape of individual foci . Stochastic variation in the trajectories of individual bacteria , amplified by exponential growth , could in principle lead to more irregular focus shapes such as those observed in Figure 1A–B . We thus turned to simulations with finite numbers of discrete bacterial agents to examine the effect of such stochastic fluctuations . Agent-based simulations have been used to study discrete biological phenomena such as the spread of infectious endemic agents throughout populations ( Juher et al . , 2009 ) and the diversification of lymphocyte antigen-receptor repertoires ( Castiglione , 2011 ) . The benefit of using this method to model L . monocytogenes cell-to-cell spread is that it allows simulation of individual bacteria as discrete particles and avoids the continuum assumption imposed by Equation 1 . We match the simulation run time to experimental conditions , proceeding until 105 bacteria are accumulated . The primary goal was to determine whether fluctuations arising from random trajectory sampling were sufficient to account for the observed boundary anisotropy . In these simulations , individual bacteria execute an isotropic random walk in two dimensions ( Video 3 ) , with the step in each dimension selected from a normal distribution with mean zero and variance 2D∆t where ∆t is the simulation time-step . Each bacterium replicates at a preset time interval after its initial birth , resulting in an overall replication rate k ( Figure 2B and Video 4 ) . The MSD and total counts of simulated bacteria accurately reflect the input parameters of diffusivity D and replication rate k ( Figure 2—figure supplement 1 ) . As expected , the speed of the infection focus boundary , defined as the square root of the area of the boundary , approaches the theoretical limit of 2 times the square root of Dk ( Liebhold and Tobin , 2008 ) at long times ( Materials and Methods; Figure 2—figure supplement 2 ) . While not perfectly isotropic , the stochastic simulations generated foci that were approximately circular and thus differed significantly from the experimental foci ( Figure 2C ) . To quantify the circularity of the experimental and simulated foci , we calculated the ratio of the area of a focus over the area of the smallest circle that fully encloses the focus ( Figure 2D ) . For a perfect circle , this metric would be equal to 1 , and for a square , this metric would be equal to 2/π ( Zheng and Hryciw , 2015 ) . Importantly , this metric is not dependent on the focus size ( Figure 2—figure supplement 3 ) . For all measurements , simulated foci were convolved with the point spread function of individual bacterial cells to match the empirically determined resolution of our microscope system , so that simulation outputs could be directly compared to experimental observations ( Materials and Methods ) . The data showed that simulated foci are substantially more circular than experimental foci ( Figure 2E ) . It is known that intracellular L . monocytogenes does not undergo truly uncorrelated random walks as was assumed in our simulations . Instead , intracellular L . monocytogenes motility , aided by ActA-dependent actin comet tails , exhibits directional persistence over time-scales of a few minutes ( Lacayo and Theriot , 2004; Soo and Theriot , 2005 ) . In addition , our initial simulations ignored the presence of host cell boundaries , which L . monocytogenes encounters as they spread from cell to cell . In fact , it has been shown that L . monocytogenes can ricochet off MDCK host cell boundaries at a frequency dependent on monolayer age ( Robbins et al . , 1999 ) . To test the possibility that bacterial motility persistence and the presence of host cell boundaries could affect the circularity of the simulated foci , we updated our simulations to include both of these effects ( Materials and Methods , and Video 5 ) . We found that neither of these two conditions affects circularity significantly; specifically , foci simulated with these features are only about 3% less circular than foci simulated by a random walk alone ( Figure 2—figure supplement 4A ) . Overall , changing the probability with which the bacteria cross host cell boundaries had a minimal effect on the circularity of the simulated foci ( Figure 2—figure supplement 4B ) . Taken together , our experimental and simulated data show that L . monocytogenes cell-to-cell spread cannot be modeled with only a random walk and exponential growth , and that the presence of host cell boundaries and the persistence of bacterial motility do not have a significant effect on the circularity of infection foci . We therefore decided to look more closely at the influence and significance of pioneer bacteria . Pioneer bacteria , which spread unusually far compared to the overall bacterial population ( Figure 3A and Video 6 ) , have the potential to substantially alter the shape and isotropy of infection foci . For our experiments performed in the presence of extracellular gentamicin , L . monocytogenes only replicates in a host cell’s cytoplasm . We also know from direct observation that at least one round of division must take place before bacteria can resume actin-based motility in the recipient cell ( Robbins et al . , 1999 ) . Therefore , the simplest explanation consistent with our direct observation of events such as the one shown in Figure 3A , is that the bacterium travels from a donor cell directly to a non-adjacent recipient through a long protrusion ( approximately 15–20 µm in the example shown ) . Once inside the cytoplasm of this non-adjacent recipient cell , the bacterium replicates . Importantly , this bacterium can reach a non-adjacent recipient host cell in less than 30 min even though it takes an MDCK cell approximately 45 min to complete the process of taking up a bacterium-containing protrusion ( Robbins et al . , 1999 ) . In contrast , non-pioneer bacteria typically move about 1–2 µm in 5-min intervals in our assay ( Video 6 ) . For a few of these pioneer events , the pioneer bacterium went transiently out of focus in our widefield imaging setup , consistent with the possibility that this long protrusion extended above the apical surface of the monolayer ( Video 7 ) . However , such long protrusions reaching non-adjacent cells could in principle also extend beneath the basal surface of the monolayer or indeed even between cell-cell junctions . For MDCK cells , the tight junctions which presumably would occlude lateral extension of long protrusions between neighboring cells only comprise the top 5–10% of the lateral face of the cells in culture ( Nelson and Veshnock , 1986 ) , so there is ample space for long protrusions to extend between cells in the monolayer prior to protrusion uptake by a non-adjacent recipient host . Pioneers , which appear to determine the frontier of the infection focus boundary ( Figure 1A–B ) , can be incorporated into the stochastic simulation by allowing bacteria to sample from an alternate distribution of step sizes . For simplicity , we thus include pioneers in our model by allowing all bacteria to move in a purely diffusive fashion , with either a slow ( non-pioneer ) diffusivity Dslow or a fast ( pioneer ) diffusivity Dfast . Pioneer behavior in the simulations is then characterized by the ratio of Dfast/Dslow ( i . e . how much further pioneers spread as compared to non-pioneers ) and the probability with which a bacterium becomes a pioneer . When a bacterium replicates , each daughter has a probability P of spreading according to Dfast and probability 1–P of spreading according to Dslow ( Video 8 ) . We assume the assignment of each individual bacterium as either a pioneer or non-pioneer persists until a bacterium’s next replication event . We first simulated cell-to-cell spread by setting the probability of becoming a pioneer to 0 . 10 and the Dfast/Dslow ratio to 100 . These are reasonable parameters because ( 1 ) a relatively small number of bacteria spread much farther than the rest throughout a live microscopy assay , and ( 2 ) an effective diffusion coefficient ratio of 100 translates to pioneer steps that are an order of magnitude longer than non-pioneer steps , which is consistent with what we observe experimentally ( Figure 3A ) . As expected , the presence of pioneers caused the simulated boundaries to become anisotropic , particularly in the early steps of the simulation ( Figure 3B ) . Additionally , these simulations recapitulated the increase in the MSD slope during the later time points of the experimental data ( Figure 3—figure supplement 1A ) . The transition to a larger MSD occurs at a time when the bacterial population stabilizes to contain a larger fraction of pioneers . A probability of 0 . 10 allowed , on average , approximately 75% of bacteria to have a pioneer ancestor or be pioneers themselves by the end of the simulation ( Figure 3—figure supplement 1B ) . It is likely that the approach towards a pioneer majority explains why simulated foci boundaries tended to be more anisotropic during earlier steps of the simulation , and why they became more circular as simulation time increased ( Figure 3—figure supplement 1C ) . Circularity of simulated foci would sometimes drop precipitously if the probability of becoming a pioneer was less than 0 . 001 ( Figure 3—figure supplement 2A ) . We also observed a general time-dependent increase in infection focus circularity in experimental data , which also sometimes exhibited rapid decreases in infection focus circularity ( Figure 3—figure supplement 2B ) . The observed kinetics of changes in circularity over time for both the experimental data and the simulations are consistent with the proposition that the overall focus size and shape become more strongly dominated by the pioneers at later time points , as also illustrated by the transition in MSD slope described above . Overall , less circular foci shapes were observed when the pioneer probability was low enough so that only a few bacteria in each focus exhibited pioneer behavior ( Figure 3C ) . We confirmed these findings quantitatively and showed that experimental circularity is equivalent to the circularity seen for simulations with pioneer probabilities of 10−3 and 10−2 ( Figure 3D ) , which suggests that during our cell-to-cell spread experimental assay , approximately 1 . 4% to 12% of bacteria have pioneer ancestors or are pioneers themselves ( Figure 3—figure supplement 1B ) . We note that these results are dependent on the total simulation time , as higher overall bacterial counts ( longer simulation times ) result in more circular foci for the same value of pioneer probability . We also tested the effect on circularity of changing the ratio of Dfast/Dslow . As expected , a larger ratio , that is pioneers taking longer steps than non-pioneers , decreases circularity significantly more than a smaller ratio ( Figure 3E ) . Together our findings suggest that L . monocytogenes cell-to-cell spread is consistent with individual bacteria having a low but non-zero probability of becoming pioneers , while the majority of the bacteria spread locally . We refer to this form of L . monocytogenes dissemination as heterogeneous cell-to-cell spread following terminology from the ecological study of animal dispersion ( Shigesada et al . , 1986 ) . We next investigated whether L . monocytogenes entering straight long protrusions could form the basis of heterogeneous spread . During L . monocytogenes cell-to-cell spread , it is known that intracellular bacteria create protrusions that can be taken up by a recipient cell directly adjacent to the donor cell . In this case , donor and recipient cells are connected to one another by the protein-protein interactions of constituents of adherens and tight junctions ( Hartsock and Nelson , 2008 ) . However , to explain the pioneer phenomenon , we propose that a few bacteria will create longer protrusions that will allow them to reach a more distant recipient cell that is not adjacent to the donor cell . In other words , this spreading event takes place between two cells that do not form junctions directly with each other ( Figure 4A ) . This is a reasonable hypothesis because L . monocytogenes can form long protrusions that are tens of microns in length , sufficient to allow them to reach non-adjacent host cells ( Pust et al . , 2005 ) . L . monocytogenes’ ability to create long , pioneer-containing protrusions thus would be critical for the complex , non-circular boundaries observed in experimental data . To test this model , we infected confluent MDCK cell monolayers with either wild-type L . monocytogenes or an L . monocytogenes strain where the proline residues in three proline-rich regions of the ActA protein have been mutated to glycine ( Skoble et al . , 2001 ) . This mutant , known as the glycine-rich repeat ( GRR ) mutant , is less persistent than wild-type bacteria , which means that it loses its original direction more quickly than wild-type bacteria . The GRR mutant is also characterized by two-fold shorter actin comet tails ( Auerbuch et al . , 2003 ) . These characteristics make the GRR mutant likely to enter protrusions at a lower frequency than wild-type bacteria and to form protrusions that are less straight . Upon quantifying the circularity of GRR foci , we found that they were significantly more circular than foci created by wild-type L . monocytogenes ( Figure 4B ) . This was likely a consequence of a decrease in the probability of forming long , straight protrusions , which then decreased the probability of bacteria exhibiting pioneer behavior . Indeed , changing the directional persistence in simulations including pioneers had little effect in circularity ( Figure 4—figure supplement 1 ) , thus supporting the idea that pioneer behavior , that is making long straight protrusions , has a stronger effect in circularity than intracellular directional persistence . Because pioneer behavior in wild-type bacteria is expected to occur quite rarely , decreasing the pioneer probability still further should result in more circular infection foci as very few pioneering events occur over the observation period ( Figure 3D ) . Our findings suggest that L . monocytogenes cell-to-cell spread is heterogeneous as it proceeds via local non-pioneers and far-spreading pioneers , each of which can be modeled with a random walk . In addition , we have shown that pioneer behavior is probably based on L . monocytogenes’ ability to spread directly to non-adjacent host cells via long extracellular protrusions . However , because of the intrinsic limitations of our wide-field imaging methodology , we cannot tell whether these long , straight protrusions extend above , below , or between host cells as they are reaching their destination . In considering the possible biological significance of pioneer behavior , we next asked whether heterogeneous cell-to-cell spread would promote L . monocytogenes intracellular survival and growth in a more physiological setting . To answer this question , we updated our simulations to more closely mimic the physiology of the tip of an intestinal villus by including host cell extrusion events , which could terminate bacterial infections in vivo ( Figure 5A ) . Given L . monocytogenes’ ability to spread away from an actively extruding villus tip , the rate of host cell extrusion and the rate of L . monocytogenes cell-to-cell spread together determine the fate of an intestinal infection . In the updated simulations , after a pre-determined period of time , a circular host cell at the center of the simulated monolayer is removed ( extruded ) , taking with it the bacteria found inside . The monolayer then contracts to replace the extruded host cell and moves all other bacteria radially inward ( Figure 5B ) . As before , simulated bacteria spread via a random walk and replicate exponentially . The simulation keeps track of both the number of bacteria in the monolayer and the number of bacteria that have been extruded . Unlike previous simulations , which we terminated at the point at which 105 total bacteria had accumulated , we ended host cell extrusion simulations in one of three ways: ( 1 ) no bacteria left in the monolayer , called bacterial clearance ( Video 9 , left ) ; ( 2 ) too many bacteria , for example 105 , have accumulated in the monolayer , called uncontrolled growth ( Video 9 , right ) ; ( 3 ) the number of bacteria extruded from the monolayer has reached a pre-determined threshold , for example 2 × 105 , without accumulating too many bacteria in the monolayer . This third outcome , which we term a stable steady state , is equivalent to a persistent infection that allows L . monocytogenes to actively replicate and spread in the epithelium while being kept in check by the animal’s host cell extrusion ( Video 9 , center ) . Stable steady state does not harm the host , and it allows L . monocytogenes to exit the animal via feces and infect other animals , which benefits the pathogen ( Begley et al . , 2005; Roldgaard et al . , 2009 ) . To learn about the relationship between the rate of L . monocytogenes cell-to-cell spread and the rate of host cell extrusion , we ran random walk simulations and varied both the effective diffusion coefficient ( D ) and the host cell extrusion period ( E ) . Specifically , we ran 100 independent simulations for each combination of D and E and quantified the outcomes . We found that small values of E , indicative of an actively extruding monolayer , favored bacterial clearance , and that large values of E , indicative of a more quiescent monolayer , favored uncontrolled growth , as expected . Similarly , small values of D favored bacterial clearance and large values of D favored uncontrolled growth . Stable steady state , on the other hand , was only reached by a narrow set of intermediate values of D and E ( Figure 5C ) , corresponding to parameters where the rate of bacterial removal by extrusion was precisely balanced by the rate of replication ( as derived in Materials and Methods ) . We were next interested in asking whether heterogeneous cell-to-cell spread would increase the chances that L . monocytogenes could attain a stable steady state in an actively extruding epithelium . We first chose conditions that produced 100% bacterial clearance outcomes in the case of a random walk , by setting D = 2 and E = 0 . 15 ( Figure 5—figure supplement 1 ) . Next , to simulate heterogeneous spread , we set Dslow = D , kept E the same , varied the value of Dfast , and set P = 0 . 01 , where p is the probability of becoming a pioneer at the time of birth . Interestingly , values of Dfast that were 60- to 90-fold higher than Dslow allowed L . monocytogenes to reach a stable steady state ( Figure 5D ) . A Dfast/Dslow ratio in this range translates to pioneer bacteria taking steps 7 . 4- to 9 . 5-fold longer as compared to non-pioneer bacteria , which is consistent with our experimental observations ( Figure 3A ) . In addition , many Dfast/Dslow ratios , for several host cell extrusion periods , allowed L . monocytogenes to attain a stable steady state ( Figure 5—figure supplement 2 ) . Together , our findings argue that L . monocytogenes heterogeneous cell-to-cell spread improves the chances of the pathogen reaching a stable steady state in vivo as compared to bacteria spreading via a random walk alone . The combination of these outcomes would prevent damage to the host animal tissue , facilitate bacterial dissemination to other host animals , and allow L . monocytogenes to thrive in the actively extruding and ever-changing intestinal epithelium . Listeria monocytogenes cell-to-cell spread has been primarily studied in two ways . First , plaque assays have been used to study late stages of infection where a few millions of bacteria have created plaques—sites of host cell death in cultured epithelial monolayers . The size of the plaque correlates to the efficiency of spread ( Van Langendonck et al . , 1998 ) . Second , individual bacteria have been carefully observed by light and electron microscopy to provide information about the kinetics of protrusion formation and uptake ( Robbins et al . , 1999 ) . In both cases , the identification of host and bacterial proteins has helped elucidate possible molecular mechanisms that facilitate L . monocytogenes spread ( Rajabian et al . , 2009; Chong et al . , 2011; Czuczman et al . , 2014 ) . We were interested in bridging the gap between millions of bacteria creating millimeter-sized plaques and single bacteria creating micron-sized protrusions by studying cell-to-cell spread at a population level , while tracking individual bacteria at the frontier of the infection focus , with the goal of learning about both the collective and single-cell intercellular spreading behavior of L . monocytogenes . We initially predicted that the spatial distribution of bacteria as a function of time would follow that of a random walk , a model characterized by isotropic , uncorrelated directions and normally distributed displacements ( Berg , 1993 ) . We developed this null hypothesis because ( 1 ) there is no evidence in the literature to suggest that intracellular L . monocytogenes motility has directionality , ( 2 ) late stages of L . monocytogenes cell-to-cell spread create circular plaques ( Van Langendonck et al . , 1998 ) , and ( 3 ) MDCK cells form compact and relatively homogeneous monolayers in culture ( Mays et al . , 1995 ) . Our high-resolution video microscopy assay , however , showed that a small number of bacteria spread farther than the rest and caused the infection focus boundary to become irregular . We propose that these pioneer bacteria spread by creating extracellular protrusions ( Pust et al . , 2005 ) , that can reach and be taken up by recipient host cells that are not in direct contact with the donor cell . Through simulations , we have found that allowing each bacterium to choose between two behaviors , far-reaching pioneer and local non-pioneer , approximated the shape of the experimental data better than simulating a single behavior of spreading bacteria ( Figure 3 ) . Simulated foci became less circular when the ratio Dfast/Dslow was high and the total number of pioneer bacteria in the infection focus was very small ( yet non-zero ) . While high numbers of pioneers are expected to increase the overall spreading rate of an infection , the anisotropic non-circular shapes of observed foci imply that the infection boundary is determined by rare events . Because pioneer bacteria themselves are assumed to not have a preferred spreading direction , small absolute numbers of pioneers are required in order to generate non-circular foci shapes as a result of stochastic fluctuations . While we cannot rule out the possibility that infection focus anisotropy can also be attributed in part to host cell heterogeneity , the L . monocytogenes GRR mutant data suggest that properties of the bacteria themselves contribute significantly to this phenomenon . L . monocytogenes GRR mutants produce infection foci that are more circular than those produced by wild-type bacteria . Given that changing directional persistence in our simulations had little to no effect on circularity , but that changing Dfast/Dslow did , it is likely that the GRR mutant L . monocytogenes creates shorter extracellular protrusions , that is GRR pioneers take shorter steps than wild-type pioneers . However , with our current experimental setup , we cannot determine whether these long protrusions seen in cultured cells ( Pust et al . , 2005 ) occur in the apical side of the monolayer , the basal side , or between cell-cell junctions in MDCK monolayers . Whereas at least a few of the pioneer events that we directly observed in MDCK cells appear to involve long apical protrusions , in the intestinal epithelium , which is characterized by a dense and highly organized apical brush border ( Crawley et al . , 2014 ) , it is probably more likely that pioneers would spread either laterally between cells in the epithelium or basally at the junction between the enterocytes and the subjacent basement membrane . Indeed , L . monocytogenes’ ability to cross the basal membrane of an epithelium via an actin-dependent process has been well-characterized ( Faralla et al . , 2018 ) . The key feature of these protrusions , however , is that they enable an L . monocytogenes bacterium to bypass several host cells on its way to the more distant recipient cell . This model explains L . monocytogenes’ ability to seemingly spread across two host cells in less than 30 min , even though the formation , uptake , and resolution of a single intercellular protrusion can take up to 45 min in MDCK cells ( Robbins et al . , 1999 ) . An extension of our model then argues that intracellular pathogens that spread from cell to cell without making extracellular protrusions would be expected to spread via a process resembling a random walk . The Gram-negative bacillus Burkholderia thailandensis is an example of a bacterial pathogen that spreads intercellularly primarily by inducing the cytoplasmic fusion of two neighboring host cells . Indeed , consistent with our pioneer model , infection foci created by B . thailandensis in mammalian host cells are significantly more isotropic than those created by wild-type L . monocytogenes ( French et al . , 2011 ) . This type of dual spreading behavior is not uncommon in other organisms . For example , the rice-water weevil , Lissorhoptrus orzyzophilus , migrates by both crawling and flying ( Shigesada et al . , 1995 ) . If relatively few beetles migrate by flying , then early migration would be dominated by short steps and later migration would be dominated by longer steps . This ecological model parallels our heterogeneous L . monocytogenes cell-to-cell spread model given that: ( 1 ) the mean squared displacement accelerates with time in both rice-water weevil migration pattern data and L . monocytogenes cell-to-cell spread live microscopy assays ( Shigesada et al . , 1995 ) , ( 2 ) the migration boundary of this organism deviates from a circle similar to bacterial infection foci ( Andow et al . , 1990 ) , and ( 3 ) the rice-water weevil bimodal migration mechanism resembles that of bacterial local spread versus pioneer spread in long protrusions . In another ecological example , the population of European starlings , Sturnus vulgaris , is made up of short-distance and long-distance migrants . The latter of the two groups was able to establish colonies that helped to promote survival of the species ( Shigesada et al . , 1995 ) . Indeed , we argue that heterogeneous spread increases the chance of L . monocytogenes survival in an actively extruding tissue ( Figure 5 and discussed below ) . Dual spreading behavior can also occur in a variety of other microbial pathogens . Vaccinia virus undergoes intercellular cell-to-cell spread via an actin-mediated process resembling that of L . monocytogenes . In addition , vaccinia virus can accelerate its own rate of spread by inducing the expression of viral proteins on the surface of an infected host cell . Upon encountering those proteins , new incoming viral particles are repelled from the already-infected host cells by actin projections and encouraged to infect virus-free host cells . The repulsion of superinfecting virions thus creates ‘viral superspreaders , ’ whose spreading behavior resembles that of L . monocytogenes pioneers . Both viral superspreaders and bacterial pioneers skip host cells on their way to a recipient uninfected host cell , create anisotropic infection foci , and accelerate the pathogen’s rate of spread ( Doceul et al . , 2010 ) . Even though mixing two random walks recapitulated the decrease in circularity seen in experimental data , we cannot rule out alternative spread models . For example , a well-characterized mathematical model is the Lévy flight , a random walk model where the step sizes are drawn from a heavy-tailed distribution instead of a normal distribution , thus ensuring a non-trivial fraction of arbitrary long steps ( Dubkov et al . , 2008 ) . Lévy flights are used to model animals foraging for food: animals will take short steps as they are feeding and long steps as they are searching for the next feeding ground ( Viswanathan et al . , 2008 ) . For L . monocytogenes cell-to-cell spread , a Lévy flight would indicate that at any given point , all bacteria have the ability to spread as either a pioneer or as a non-pioneer . With a Lévy flight , however , it is more difficult to mechanistically explain what allows a bacterium to become a pioneer . On the other hand , in our heterogeneous spread model , bacteria interconvert between two spread behaviors and retain that behavior until their next replication event . We designed the simulation this way to resemble a bacterium creating either a short protrusion or a straight , long protrusion , and replicating once they have broken out of the double-membrane vacuole in the cytoplasm of the recipient cell . The increase in directional persistence in pioneers is equivalent to a larger diffusion coefficient at long times . Given that it is a foodborne pathogen , L . monocytogenes infections begin in the host’s alimentary canal . In this work , we propose that L . monocytogenes may have evolved the ability to spread via host cell skipping to maximize its chances of surviving , replicating , and spreading in a host’s actively extruding intestinal epithelium . Under the conditions set by our single random walk simulations , L . monocytogenes’ ability to establish a stable steady state was attained by only a narrow set of effective diffusion coefficients ( Figure 5C ) . Also , in this model , an individual effective diffusion coefficient usually led to an all-or-nothing outcome . If the step sizes were too small , then the infection was cleared 100% of the time . If the step sizes were too large , then the infection got out of control and caused uncontrolled growth 100% of the time . If the step sizes fell in a narrow range in between , the bacterium was able to successfully extrude many bacteria while sustaining the infection 100% of the time . This stable steady state is a desirable outcome for both bacteria and host: L . monocytogenes can promote the extrusion of its offspring , which can either ( 1 ) exit the animal via feces and infect other host animals , or ( 2 ) escape the extruded host cell and try to re-invade a different villus . This second point is important because our simulation considers a single villus only , even though the mammalian small intestine contains millions of individual villi ( Guyton and Hall , 2006 ) , each of which is a potential site of infection . Given our findings , it was important to include pioneers in the host extrusion simulations . Simulating an effective diffusion coefficient that previously led to bacterial clearance and mixing it with larger effective diffusion coefficients allowed L . monocytogenes to attain a stable steady state ( Figure 5D ) . Even though heterogeneous spread did not lead to 100% stable steady state , 10–15% of stable steady state infections become significant in the context of the millions of villi that make up the intestinal epithelium . Under these conditions , bacterial clearance was the most likely outcome , and uncontrolled growth remained low , unlike in the case of the random walk ( Figure 5C ) . It is critical for bacteria to avoid uncontrolled growth in any single villus site as this could result in death of the host animal , which harms both the host and the pathogen since the pathogen can no longer replicate and spread to other hosts ( Falkow , 2006 ) . Importantly , high but not 100% of bacterial clearance allows L . monocytogenes to extrude more offspring while being able to achieve a stable steady state in a smaller fraction of villi . Finally , given that step size is a function of nutrient availability , temperature , and monolayer age , among other factors , our model predicts that heterogeneous spread widens the range of biological conditions that L . monocytogenes can explore to create a stable steady state . This is an import host-pathogen relationship because it does not harm the host and promotes pathogenic success . Beyond the gut , it is also possible that pioneers may be more successful at reaching distant organs within the host animal . In fact , it has been shown that a very small number of founder L . monocytogenes bacteria can spread from the gut to organs such as the spleen and gall bladder , a process that leads to bottlenecking ( Zhang et al . , 2017 ) . In our current model of an actively extruding epithelium , host cell extrusion occurs at regular intervals and is not influenced by bacterial load . An alternate mechanism that would also be expected to lead to a stable steady state would be forcing extrusion to occur after a preset number of bacteria is reached . Uncontrolled growth is inhibited since bacteria are extruded as soon as the number gets too high , and bacterial clearance does not occur since extrusion stops if bacterial counts get low . However , this alternative mechanism would require that the host cell be able to sense the number of intracellular bacteria and specifically alter its behavior accordingly . Our model , in contrast , presents a simple physical mechanism by which steady state can be achieved without additional sensing capabilities on the part of the host cell . In addition to L . monocytogenes , other pathogens have evolved strategies to create persistent infections in their hosts . For example , the lambda phage induces expression of the λ repressor to change its gene expression profile from an active host-killing lytic state to a dormant lysogenic state . During the lysogenic state , the lambda phage integrates its genome into the bacterial chromosome , which is then inconspicuously replicated by the host’s DNA replication machinery ( Ptashne , 2006 ) . The lambda phage stays dormant until environmental conditions , such as host bacteria availability , indicate that it is safe to kill the donor and spread to recipient hosts . Just like pioneer L . monocytogenes behavior , spontaneous induction from a lysogenic to lytic state is rare , a characteristic that promotes phage replication ( Little et al . , 1999 ) . A continuous lytic state , similar to L . monocytogenes taking large steps in an extruding monolayer , would cause indiscriminate host death , thus harming both host and pathogen . Indeed , strategies that help establish persistent infections are critical in creating stable host pathogen interactions that have evolved over millions of years . All 10403S Listeria monocytogenes strains used in this study are summarized in Table 1 . The plasmid pMP74RFP ( Ortega et al . , 2017 ) was stably integrated into the genome of GRR L . monocytogenes via conjugation with E . coli SM10 λpir as previously described ( Lauer et al . , 2002 ) . Three days before carrying out infection assays , bacteria were streaked out onto BHI agar plates containing 200 µg/mL streptomycin and 7 . 5 µg/mL chloramphenicol . Bacteria were inoculated and grown in liquid cultures overnight as previously described ( Ortega et al . , 2017 ) . Madin-Darby canine kidney ( MDCK ) type II G cells ( Mays et al . , 1995 ) were grown in DMEM with low glucose and no phenol red ( Sigma D5921 ) and low sodium bicarbonate ( 1 . 0 g/L ) in the presence of 10% fetal bovine serum ( FBS ) and 1% penicillin-streptomycin . For live microscopy assays , 24-well plastic-bottom plates ( Ibidi 82406 ) were coated with 50 µg/mL rat-tail collagen-I ( Thermo Fisher A1048301 ) , diluted in 0 . 2 N acetic acid , for 2 hr at 37°C and air-dried for 24 hr . Wells were washed with DPBS once before seeding . MDCK cells were cultured and seeded as instant-confluent monolayers as previously described ( Ortega et al . , 2017 ) . Flagellated bacteria ( OD600 of 0 . 8 ) were washed twice with DPBS and diluted in DMEM . Host cells were washed once with DMEM , and bacteria were added at a multiplicity of infection ( MOI ) of 200–300 bacteria per host cell in a volume of 500 µL/well . Bacteria and host cells were incubated together at 37°C for 10 min . Host cells were washed three times with DMEM to remove non-adherent bacteria and were incubated at 37°C for 15–20 min to allow a small number of adherent bacteria to invade host cells . It was important to keep the number of invading bacteria low because it prevents foci from merging with others . Media was replaced for DMEM +10% FBS+50 µg/mL gentamicin , and host cells were incubated at 37°C for 20 min to kill adherent bacteria . Media was replaced for DMEM +10% FBS+10 µg/mL gentamicin , and host cells were incubated for approximately 4 hr . The total time starting with the three DMEM washes until the end of the incubation is 5 hr . For live microscopy assays , MDCK cells were cultured on rat-tail collagen-I-coated 24-well plates ( Ibidi 82406 ) for 48 hr as described above . Five hours post-infection , host cells were washed with Leibowitz’s L-15 once and incubated with 1 µg/mL Hoechst , diluted in L-15 , for 10 min at 37°C . Cells were washed with L-15 three times and media was replaced with L-15 +10% FBS+10 µg/mL gentamicin . MDCK cells and L . monocytogenes were imaged every 5 min with a 20X air objective ( NA = 0 . 75 ) in an inverted Eclipse Ti-E microscope using µManager’s autofocus feature . Red channel ( bacteria ) , blue channel ( nuclei ) , and phase ( MDCK monolayers ) were imaged . Environmental chamber was equilibrated to 37°C for at least 2 hr prior to imaging . For fixed microscopy assays , MDCK cells and L . monocytogenes were co-incubated for 22 hr after the addition of gentamicin . Host cells were washed once with DPBS and fixed with 4% formaldehyde for 10 min at room temperature . Paraformaldehyde was removed and quenched with 50 mM NH4Cl for 10 min . Membranes were permeabilized with 0 . 03% Triton-X100 , diluted in DPBS , for 7 min . Samples were incubated with 0 . 2 µM AlexaFluor488 phalloidin , diluted in DPBS , for 20 min at room temperature . All image TIFF files were imported into MATLAB and processed with the image processing toolbox ( MathWorks ) . To process experimental microscopy data , images were read in as 1024 × 1024 matrices , converted to double-precision numbers , and normalized to intensities ranging from 0 to 1 . Images were thresholded using Otsu’s method ( Sezgin and Sankur , 2004 ) . Bacterial debris was excluded from the thresholded mask by inspection . To quantify the total fluorescence intensity for a given time point , the thresholded mask was dilated until the infection focus was represented as a single continuous round shape . The median of the intensity values found outside of the thresholded mask was set as the image’s background , which was then subtracted from every value in the matrix . Finally , background-subtracted intensity values were summed . For a full time-lapse movie , total fluorescence intensity values were fit to an exponential function , which provided an estimated value for growth rate . Doubling time was calculated by dividing the natural log of 2 by the growth rate . To quantify the mean squared displacement ( MSD ) for a given time point , the distance squared to each pixel of the thresholded mask was normalized by that pixel’s fluorescence intensity . All normalized squared distances were averaged . For a full time-lapse movie , MSD values were fit to a linear function . The slope of this line was divided by four to estimate an effective diffusion coefficient . To quantify the area of an intracellular bacterial focus for a given time point , the x y coordinates of the thresholded mask were calculated . MATLAB’s boundary ( ) function , using x y coordinates as input , was then used to calculate a boundary that fully encompasses all of the points while shrinking towards them . This function also returns the area contained inside the boundary . The radial speed of the focus is equivalent to the slope of the square root of the area divided by π plotted as a function of time ( Liebhold and Tobin , 2008 ) . To quantify circularity , the boundary of the infection focus was used to calculate the smallest circle that fully encompasses the boundary , as described previously ( Zheng and Hryciw , 2015 ) . Then , the area of the boundary was divided by the area of a circle . A perfect circle thus has a circularity of 1 . To use the Voronoi tessellation to estimate the position of host cell boundaries , nuclei were thresholded as described above and segmented using a watershed transform . The center of mass of each nucleus was calculated and used as the input for MATLAB’s Voronoi ( ) function . To calculate the L . monocytogenes point spread function ( PSF ) , twenty 9 × 9 pixeled images containing individual bacterial cells , obtained from live microscopy experiments , were interpolated and aligned at subpixel resolution according to their center of mass . Images were averaged to create the PSF . Simulated data in Cartesian coordinates were binned in a 1024 × 1024 matrix and convolved with the PSF to generate data that matched the resolution of our microscope system . All data were generated from simulations written in MATLAB ( Ortega , 2018; copy archived at https://github . com/elifesciences-publications/Listeria_spread_simulations ) . At the beginning of the random walk simulations , several parameters are set: the effective diffusion coefficient ( D ) , the replication rate ( krep ) , the maximum bacteria to be accumulated ( maxnbact ) , and the time-step ( delt ) . Every run of the for loop is equivalent to a single time step during which ( 1 ) bacteria age , ( 2 ) bacteria replicate , and ( 3 ) bacteria move . For bacterial aging , a vector called bacthist keeps track of each bacterium’s age . These numbers increase monotonically until a particular number reaches that bacterium’s replication time , drawn from a normal distribution with mean ln ( 2 ) /krep and variance mean/5 . For replication , the positions of those bacteria whose age has reached their replication time are duplicated . Both daughters are assigned a new replication time from the same distribution and their ages are set to 0 . Finally , the bacteria move in the x and y dimensions by sampling random numbers from the standard normal distribution scaled by the square root of 2Ddelt , where D is the effective diffusion coefficient and delt = 0 . 01 is the time step . At every time step , the number of bacteria , the MSD , the area of the boundary , and the circularity of the boundary , calculated as above , are recorded . For heterogeneous spread simulations , the values of Dslow , Dfast , and P were set prior to the start of the simulation . P refers to the probability of becoming a pioneer . At the start of a bacterium’s life , it chooses whether to spread according to Dslow with probability P or according to Dfast with probability 1–P . To add persistence to the bacterial cell motility , two new parameters , θ and β , were included . In these simulations , the angle of movement is sampled from a normal distribution with mean θ ( the angle associated with the previous step ) and standard deviation β . For a random walk , β >> 2π , which means that any angle between 0 and 2π is equally possible . For a persistent random walk , β limits the angle of movement to values close to the angle of the previous step . When β = 0 , the angle of movement is constant over time , and the bacteria will be perfectly persistent . To plot circularity as a function of persistence , one thousand random angles were generated for each value of β , and the cosine values of the angles were averaged . For β = 0 , persistence was close to 1 . As β increased , persistence was close to 0 . To add host cell boundaries , a Cartesian lattice was used to define boundaries between host cells . The parameter γ defines the probability with which simulated bacteria will cross the boundaries . In these simulations , the new bacterial positions are calculated , and those bacteria that do not cross a boundary are moved to the new positions . Those new positions that require boundary crossing are attained with probability γ . The remaining bacteria reflect from the boundary , remaining in the same cell . For host cell extrusion simulations , a circular host cell ( of size R = 1 ) is created in the center of the monolayer and extrudes after every fixed period of time . At this point , the simulated bacteria found inside the host cell were eliminated from the monolayer and cumulatively summed over the entire simulation . After extrusion , remaining bacteria are radially moved inwards by a distance equal to the radius of the extruded host cell . The number of bacteria at the beginning of the simulation is 100 . If the number of bacteria goes to 0 , then bacterial clearance is triggered . If the number of bacteria in the monolayer reaches 1 × 105 , then uncontrolled growth is triggered . If the number of extruded bacteria reaches 2 × 105 , stable steady state is triggered . Triggering any of these three outcomes causes the simulation to end . Any combination of the above parameters ( simple random walk , two effective diffusion coefficients , persistence , host cell boundaries , and host cell extrusion ) can be used for any given simulation . The reaction-diffusion equation we used to formalize the null hypothesis of an isotropic random walk is defined as follows: ( 1 ) ∂ϕ∂t=D∂2ϕ∂r2+kϕis a differential equation where Φ represents the bacterial concentration as a function of position and time , t refers to time , r refers to the position of the bacteria in polar coordinates or their radial distance to the center of mass , D is the effective diffusion coefficient , and k is the exponential growth rate . Its analytical solution ( Shigesada et al . , 1995 ) is: ( 2 ) ϕ ( r , t ) =ϕ04πDtexp ( −r24Dt+kt ) where ϕ0 represents the initial concentration of bacteria located at the source ( 0 , 0 ) . To calculate the radial speed , that is how fast the focus grows after long periods of time , we set the above equation equal to some threshold concentration Φ and solved for the radial distance r at which this threshold concentration is reached , as a function of time . We take the derivative with respect to time , and solve for the limit of dr/dt as time approaches infinity to obtain: ( 3 ) limt→∞drdt=2Dk The step-by-step derivation has been previously described ( Andow et al . , 1990 ) . Equation 3 then predicts that stochastic simulations where D = 1 and k = 1 will generate infection foci that move a constant speed of 2 at long times ( Figure 2—figure supplement 2 ) . We calculated the radial speed of simulated data by assuming that the area of the boundary is circular and thus can be approximated by πr2 . We divided the area of the boundary by π and took the square root to obtain the average value of the radial distance , r , from the boundary to the origin of the simulation at ( 0 , 0 ) . We plotted r as a function of time and took the slope of the linear fit to approximate dr/dt ( Liebhold and Tobin , 2008 ) . Here we derive an approximate relation between diffusivity and extrusion rate that yields a steady state in the case of bacteria spreading homogeneously as a random walk . The existence of a steady state requires that the rate at which new bacteria appear through replication equals the average rate at which bacteria are removed by extrusion , and that the spatial spreading of the focus in each extrusion period is balanced by contraction of the monolayer after removal of the extruded cell . We assume that in steady state , the bacterial distribution just before an extrusion event can be approximated by a Gaussian distribution with variance σ2 . Each bacterium replicates at a rate krep , and is extruded at an effective rate 1/E*[1 − exp ( −R2/σ2 ) ] corresponding to the extrusion rate times the probability of the bacterium being found within radius R of the center . In order for these two rates to be precisely balanced , we must have: ( 4 ) σ2=R2−log ( 1−Ekrep ) When a contraction operation corresponding to the extrusion event is performed on the steady-state Gaussian distribution , the new radial bacterial distribution is given by: ( 5 ) P ( r ) =1𝒩 ( r+R ) e− ( r+R ) 2σ2where 𝒩 is a normalization constant . The mean squared radial displacement for such a distribution can be calculated as: ( 6 ) ⟨r2⟩post−ext=σ2−πRσeR2σ2erfc ( Rσ ) To achieve steady state , we must have ⟨r2⟩post-ext +4 DE = σ2 , as additional spreading during the extrusion period should return the assumed variance σ2 before the next extrusion event . This allows a solution for D in terms of σ which , together with Equation 4 , yields the following equality for maintaining steady state: ( 7 ) D=R2πerfc ( −log ( 1−Ekrep ) ) 4E ( 1−Ekrep ) −log ( 1−Ekrep )	Eating food that has been contaminated with bacteria called Listeria monocytogenes can result in life-threatening infections . The bacteria first invade the epithelial cells that line the small intestine . After this , L . monocytogenes can move from one host cell to another , which allows the infection to reach other organs . Most studies into how L . monocytogenes infections spread have focused either on how single bacterial cells move from one host cell to the next , or on how millions of bacteria damage host tissues . Little was known about the intermediate steps of an infection , where the bacteria start to colonize the small intestine . To investigate , Ortega et al . recorded videos of L . monocytogenes spreading between epithelial cells grown on a glass coverslip , and developed computer simulations to try to reproduce how the bacteria spread . This revealed that the bacteria do not all move in the same way . Instead , less than 1% of the bacteria move in ‘steps’ that are up to 10 times longer than those taken by the others . Ortega et al . named these bacteria ‘pioneers’ . Ortega et al . propose that the pioneers form long protrusions that allow them to spread directly from an infected cell to a non-neighboring cell . By taking these large steps , the pioneers may increase the chances that the bacteria will cause a long-lasting infection . Future research will be needed to answer further questions about the pioneers . For example , how do the pioneer bacteria differ from the majority of bacterial cells ? Would targeting anti-bacterial treatments at pioneers make it easier to treat infections ? It also remains to be seen if other types of bacteria also show this pioneer behavior .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "research", "communication", "cell", "biology", "microbiology", "and", "infectious", "disease" ]	2019	Listeria monocytogenes cell-to-cell spread in epithelia is heterogeneous and dominated by rare pioneer bacteria
Intracellular protein gradients are significant determinants of spatial organization . However , little is known about how protein patterns are established , and how their positional information directs downstream processes . We have accomplished the reconstitution of a protein concentration gradient that directs the assembly of the cell division machinery in E . coli from the bottom-up . Reconstituting self-organized oscillations of MinCDE proteins in membrane-clad soft-polymer compartments , we demonstrate that distinct time-averaged protein concentration gradients are established . Our minimal system allows to study complex organizational principles , such as spatial control of division site placement by intracellular protein gradients , under simplified conditions . In particular , we demonstrate that FtsZ , which marks the cell division site in many bacteria , can be targeted to the middle of a cell-like compartment . Moreover , we show that compartment geometry plays a major role in Min gradient establishment , and provide evidence for a geometry-mediated mechanism to partition Min proteins during bacterial development . Intracellular concentration gradients of proteins in micrometer-sized cells have long been thought to be unsustainable due to diffusion . However , the past decade revealed the existence of multiple intracellular gradients in eukaryotes and prokaryotes and their significance in providing positional information within the cellular compartment ( Kiekebusch and Thanbichler , 2014 ) . Although these protein gradients are now emerging as significant general motifs to spatially organize cells , little is known about how protein patterns are established by local unmixing , and maintained on a molecular level . What are the ultimate cues and mechanisms to localize gradient forming proteins ? Are protein gradients modulated by boundary conditions of the cell and how are downstream proteins directed by the positional information of protein gradients ? Cellular reconstitution methods enable us to study biological processes under defined conditions , and have significantly contributed to our understanding of molecular interactions and kinetics of proteins in simplified environments . To systematically investigate the mutual dependence between biochemical networks and three-dimensional cellular organization several techniques to confine proteins in micro compartments have recently been devised . One approach to achieve three-dimensional confinement of proteins and cytoplasmic extracts is by their encapsulation in water-oil droplets or vesicles . ( Pinot et al . , 2009 , 2012; Good et al . , 2013; Hazel et al . , 2013 ) These systems add tremendously to our understanding of how a constrained reaction space , and varying surface-to-volume ratios affect ( bio ) chemical reactions . Applying subtle mechanical forces to these systems , they may be also deformed into non-spherical shapes . However , if more complex geometries need to be realized , or if the compartments need to be mechanically stabilized , biochemical systems may also be reconstituted on and in spatially tailored microfabricated environments ( Garner et al . , 2007; Laan et al . , 2012 ) . In particular , combinations of two- and three-dimensionally engineered substrates with lipid membranes have been applied to study the role of simple membrane geometries on membrane interacting protein networks in vitro ( Schweizer et al . , 2012; Zieske and Schwille , 2013 ) . However , although such in vitro experiments are now being successfully applied to identify the minimal components and interactions of protein networks , we are still far from understanding the mutual interdependence of protein functionalities and physical parameters in generating nonhomogeneous protein distributions for providing positional information within the cellular compartment . Therefore , further development of bottom-up approaches are required to tackle the minimal requirements for nonhomogeneous protein localization . Studying gradient formation in a simplified and controlled environment , detached from the complexity of living cells , should shed light on the role of individual biochemical and physical parameters for regulated protein localization in space and time . Among the most fundamental and significant gradients in cells are those that set the spatial cues for cytokinesis . In many bacteria , and also in eukaryotes that undergo symmetric division , protein gradients with the lowest concentration in the middle of the cell and the highest concentration at the cell poles have been identified . These gradients target the division site through polar inhibition of the division machinery assembly . For instance , in fission yeast , a Pom1 gradient regulates the onset of mitosis ( Celton-Morizur et al . , 2006; Padte et al . , 2006; Martin and Berthelot-Grosjean , 2009; Moseley et al . , 2009 ) . Similarly , in many bacteria , the assembly of the conserved ring-forming cell division protein FtsZ is inhibited at the cell poles by protein gradients of FtsZ inhibitors . In Caulobacter crescentus , a gradient of the cytokinesis inhibiting protein MipZ is established , ( Thanbichler and Shapiro , 2006 ) whereas in Bacillus subtilis , an inhibitory complex of MinC and MinD is attached to the cell poles by MinJ and DivIVA . ( Edwards and Errington , 1997; Bramkamp et al . , 2008; Patrick and Kearns , 2008 ) Comparably , Escherichia coli employs the membrane-targeted MinC/MinD complex to inhibit FtsZ assembly at the cell poles . ( Lutkenhaus , 2007 ) Interestingly however , MinC and MinD in E . coli do not form a static gradient , but the Min protein patterns oscillate from pole to pole . ( Raskin and de Boer , 1999b; Raskin and de Boer , 1999a ) These dynamic oscillations require ATP as an energy source for the ATPase MinD and MinE , which accelerates hydrolysis of ATP by MinD . ( de Boer et al . , 1991; Hu and Lutkenhaus , 2001 ) On time-average , the oscillations result in an effective concentration gradient of the MinC/MinD complex . While MinD and MinE have been identified as a minimal set of proteins that is able to self-organize into dynamic pattern on flat supported membranes , ( Loose et al . , 2008 ) the minimal requirements for gradient formation , and the mechanisms of forwarding positional information to downstream processes are still controversial and only begun to be elucidated . Labeling MinE and co-reconstituting MinE with MinD in cell-shaped compartments , we demonstrated that pole-to-pole oscillations of the Min system can be reconstituted in vitro ( Zieske and Schwille , 2013 ) . However , time-averaged Min protein gradients in vitro have not been reported to date and thus , direct evidence for the establishment of steady protein gradients by self-organizing oscillations of MinD and MinE without additional proteins is still missing . Moreover , it is still ambiguous how cell geometry affects gradient formation . On the one hand , live cell experiments with cell shape mutants , as well as in vitro reconstitutions , have shown that compartment geometry affects pattern formation of the Min proteins ( Corbin et al . , 2002; Varma et al . , 2008; Raskin and de Boer , 1999b; Zieske and Schwille , 2013 ) . On the other hand , protein gradients should be robust against morphological changes during cell development and growth . Living bacteria typically double their length during one live cycle and gradually constrict during cell division . Thus , they undergo significant changes in cell geometry , which should not disturb the establishment and functional role of Min protein gradients . Whether cell shape affects pattern formation only if the geometry of a cell is artificially perturbed , or whether a cell might use its own geometry as a readout and control parameter to shape protein gradients , still has to be determined . Thus , the identification of principal cues for gradient establishment , and their modification by spatial parameters , such as the three-dimensional geometry of a cell , remains an outstanding challenge . In this study , we were able to establish , from a system of soluble and initially well-mixed proteins , an effective steady protein concentration gradient , potent of targeting the assembly of cell division proteins to the middle of artificial membrane-clad compartments . In particular , FtsZ , which marks the cell division site and represents the cytoskeletal backbone for the divisome machinery in many bacteria , can be targeted to the middle of a cell-like compartment by a gradient of Min proteins . Moreover , we demonstrate that orientation and partitioning of Min protein gradients is controlled by compartment geometry . Notably , our synthetic system opens a new way to study complex organization principles in a simplified environment and provides novel insights into the basic biophysical and biochemical requirements for gradient formation . To determine the minimal requirements for gradient formation , we investigated the spatially self-organizing Min system as a prototype for gradient formation in a synthetic environment ( Figure 1A ) . We first enclosed purified MinD and MinE proteins from the bacterium E . coli and ATP in membrane clad soft polymer compartments of picoliter sample volume and cell-shaped geometry ( Figure 1—figure supplement 1 , Figure 1—figure supplement 2 ) . Consistent with previous results and live-cell studies , MinE and MinD dynamically oscillate from pole-to-pole . Previous controls with purified Min proteins on flat bilayers and the non-hydrolysable ATP analog ATPγS confirmed that the dynamic protein patterns were thereby powered by ATP ( Loose et al . , 2008 ) . 10 . 7554/eLife . 03949 . 003Figure 1 . Protein gradients self-organize in soft-polymer containers . ( A ) Experimental setup . Purified proteins of the MinC/D/E system were reconstituted in membrane-clad soft-polymer compartments and imaged by confocal microscopy . Profiles and electron micrographs of the compartments are shown in Figure 1—figure supplement 1 and a more detailed description of the assay to reconstitute Min protein oscillations is presented in Figure 1—figure supplement 2 . Figure 1—figure supplement 3 demonstrates how the intensity profiles of MinD and MinE along the length axis of the compartments are modulated with time . To mimic a bacterial membrane we used E . coli polar lipids to generate supported lipid membranes . A more detailed description of how lipid composition influences pattern formation of Min proteins is presented in Figure 1—figure supplement 4 . ( B ) In cell-shaped compartments , eGFP-MinC ( yellow ) follows the oscillations of MinD and MinE ( red ) . Comparable results were obtained in more than a hundred soft-polymer compartments . Confocal time-lapse images with the image plane at the bottom of the compartment . Protein concentrations: 1 µM MinD , 0 . 9 µM MinE , 0 . 1 µM MinE . Atto655 ( red ) , 0 . 05 µM eGFP-MinC ( yellow ) . Time between individual frames: 30 s . Scale bar: 5 µm . ( C ) The time-averaged concentration profile of MinC along the long axis of a compartment has a distinct concentration minimum in the middle of the compartment . The time averaged distribution of protein concentrations was calculated by acquiring time-lapse-images with the focal plane at the middle of the compartment and averaging the intensity of the acquired frames . ( D ) Time-averaged fluorescent signal of eGFP-MinD with image plane in the middle of a compartment . An intensity offset is subtracted to better visualize the gradient along the boundary of the compartment . 0 . 9 µM MinD , 0 . 1 µM eGFP-MinD , 1 µM MinE . Scale bar: 5 µm . Stable pole-to-pole oscillations which result in the time-averaged gradient of MinD are severely affected if the membrane targeting sequence of MinE is deleted ( Figure 1—figure supplement 5 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03949 . 00310 . 7554/eLife . 03949 . 004Figure 1—figure supplement 1 . Micro compartments in a soft-polymer chip . ( A ) Scanning electron micrographs of a multi-compartment PDMS chip . Arrays of compartments with different geometries , for example , different lengths , provide the possibility to study geometry dependent effects of protein self-organization on one chip . Each compartment serves as reaction container to reconstitute protein networks . Scale bar: 20 µm . ( Note that the scale bars in the electron micrographs only applies to the x-direction of the picture . The sample was tilted in the other direction to create the three-dimensional impression of electron micrographs . ) ( B ) Same sample as descibed in ( A ) , with higher magnification . Scale bar: 1 µm . ( C ) Confocal x/z-scan of a membrane clad compartments with a flat bottom ( left ) vs a more tapered compartment ( right ) . The membrane was labeled with DiI to visualize the profile of the membrane compartments . The width of the compartments is about 10 µm at the upper region . Small variations in the corresponding shapes of the compartments occurred during the fabrication process , but still supported oscillations and gradient formation of the Min proteins . ( D ) Confocal images of FtsZ-mts in compartments with a flat bottom ( left ) and in a more tapered compartment . Scale bar: 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03949 . 00410 . 7554/eLife . 03949 . 005Figure 1—figure supplement 2 . Experimental setup . ( A ) A thick layer of PDMS was poured on top of a wafer with about 10 µm high resist microstructures . Glass cover slips were manually pressed in the liquid PDMS , such that an about 30 µm thin layer of PDMS remained below the glass surface and a thick layer of PDMS covered the glass surface on top . After curing the PDMS at 80°C the thick layer of PDMS on top was peeled off . Then the glass cover slip with the thin , micro structured layer of PDMS attached was peeled off together with the help of a razor blade . ( B ) After cleaning and plasma treading , a plastic ring was glued on top of the PDMS microstructures . Supported lipid membranes , which adapt to the topography of the underlying PDMS support , were produced using a vesicle fusion technique . Then proteins , ATP and/or GTP ( depending on experiment ) of defined concentration were added to a buffer volume of 200 µl on top of the supported lipid membranes . At this step of the protocol dynamic Min patterns formed on top of the membrane . These patterns did not oscillate but formed travelling waves . A Confocal image of Min waves on the upper level of the membrane support is shown in ( C ) . Yellow: eGFP-MinD , red: MinE . Atto655 . ( D ) Subsequently the buffer reservoir was manually reduced using a pipette , such that the buffer remained only in the micro fabricated chambers . Due to the confinement in small sample volumes with cell shaped geometry , the Min protein patterns oscillated within these chambers . The top of the chambers was open with a buffer/air interface . To limit evaporation of the buffer a lid was placed on top of the plastic ring . The upper level of the PDMS supports ( outside of the compartments ) was dry and residual biomolecules , such as remaining lipids and Min proteins were immobile , as determined by FRAP experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 03949 . 00510 . 7554/eLife . 03949 . 006Figure 1—figure supplement 3 . Protein gradients in vitro are established by dynamic redistribution of proteins . ( A–D ) Concentration profiles of MinD along the length axis of a cell-shaped container were measured in intervals of 5 s . The whole oscillation cycle takes about 1 min . ( A ) Initially MinD forms a cap at the left pole of the compartment ( light blue ) . Then the concentration of the MinD at the left pol decreases whereas it increases at the opposite pole . ( B ) The trail of the MinD zone at the right pol moves towards the right pole . ( C ) The concentration of MinD at the right pol decreases whereas it increases at the left pole . ( D ) The trail of the MinD cap at the left pole moves towards the left pole . ( E ) MinD ( black ) and MinE ( red ) concentration profiles during an oscillation from the left to the right pole . Dotted line: reference line to better visualize the movement of the MinE peak and MinD tail to the right pole . DOI: http://dx . doi . org/10 . 7554/eLife . 03949 . 00610 . 7554/eLife . 03949 . 007Figure 1—figure supplement 4 . Lipid composition dependent formation of protein gradients . ( A ) Pattern of MinD and MinE ( labeled with Atto655 ) on supported lipid membranes with different amounts of cardiolipin . Scale bar: 50 µm . ( B ) Dependence of the Min pattern wavelength on cardiolipin concentration . ( C ) Dependence of temporal period on cardiolipin concentration . ( D ) Dependence of wave velocity on cardiolipin concentration . ( E ) Pole-to-pole oscillations emerge both in DOPC-membranes supplemented with cardiolipin and in DOPC-membranes supplemented with PG instead . Scale bar: 5 µm , DOPC: 1 , 2-dioleoyl-sn-glycero-3-phosphocholine , CA: E . coli cardiolipin , PG: E . coli L-α-phosphatidylglycerol ( Avanti polar lipids , Alabaster , AL ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03949 . 00710 . 7554/eLife . 03949 . 008Figure 1—figure supplement 5 . Deletion of the membrane targeting sequence of MinE affects pattern formation in micro compartments . 1 µM MinD and 1 µM MinE ( Δ3–8 ) ( 5% labeled with Alexa Fluor 647 ) were reconstituted in micro compartments . ( A ) Snap shots of Min protein pattern at the bottom of the compartment ( upper panel ) and at the walls of the compartments ( lower panel focus level in the middle of the compartment ) . ( B ) Time-lapse series of Min pattern within a compartment . Scale bar: 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03949 . 008 Compared to living cells , the temporal scale for the oscillation period is conserved in the synthetic system . However , both the compartments and the spatial scales of the protein waves are about ten times bigger as compared to living cells . Why the Min patterns in vivo are about ten times larger than in vitro is quantitatively not fully understood . Different protein ratios , ionic strength of the buffer and membrane composition have been shown to affect the spatial scales ( Loose et al . , 2008; Vecchiarelli et al . , 2014 ) . Furthermore the membrane potential ( Strahl and Hamoen , 2010 ) and molecular crowding in a living cell could influence Min protein patterns . An advantage of the larger spatial scales in vitro is the possibility to study with much greater detail how the concentration profiles of MinD and MinE , which result in the formation of MinD gradients , are modulated during an oscillation period . Following the MinD and MinE concentration profiles along the compartment length over time , we found that the dynamic profiles are characterized by the following succession of events: First , attachment and increase in concentration of MinD at one site of the compartment , then delayed attachment of MinE to MinD and increase in MinE intensity at the trail of MinD ( Figure 1—figure supplement 3 ) . Although the size of bacterial cells and photo bleaching is limiting for characterizing the Min profiles in vivo in such detail as it is possible in vitro , the intensity peak of MinE at the trail of the Min pattern was also observed in E . coli and is generally referred to as ‘MinE ring’ . ( Raskin and de Boer , 1997; Fu et al . , 2001 ) After formation of the MinE peak MinD/E were disassembled from the membrane at the trail of the MinD zone , which was accompanied by assembly of a new MinD/E zone at the other pole of the compartment ( Figure 1—figure supplement 3 ) . This disassembly of Min protein patterns at the trailing zone during each oscillation is comparable to the disassembly at the trailing edge of travelling Min patterns on flat membranes . The main difference of Min proteins profiles on flat membranes and in microcompartments is that the concentration profiles on flat membranes only change their localization while traveling across the membrane . In contrast , during oscillations in compartments new membrane attached Min protein patterns need to be repeatedly established and disassembled every half oscillation period , which results in a remarkable remodelling of the concentration profiles during the oscillation cycle and an asymmetry in their rise and decay phases at the poles . To determine the mean spatial distribution of MinD , we averaged the fluorescence signal of MinD within the cell-shaped compartments over time using confocal time-lapse movies . Remarkably , MinD formed indeed a clear nonhomogeneous concentration profile , with the lowest concentration in the middle of the compartment ( Figure 1D ) . Thus , we conclude that using the Min system , the establishment of an effective protein gradient indeed requires only a lipid membrane , an appropriate reaction space , and two proteins that self-organize under the consumption of energy , in the form of ATP . While the requirements for Min gradient formation are now understood to a point , at which these gradients can be reconstituted in vitro , less is known about the importance of the structural properties of the gradient forming proteins . Although MinD has been shown to interact with membranes in vitro and MinE interacts with MinD ( Hu et al . , 2002 ) , MinE also harbours a short membrane targeting sequence at its N-terminus ( Park et al . , 2011 ) . Mutations in the membrane targeting domain of MinE result in defects in cell division and in distinct phenotypes , such as filamentous cells or a minicelling phenotype , respectively ( Park et al . , 2011 ) . However how the membrane targeting of MinE contributes to a normally functioning Min system is not fully understood . To gain a deeper insight in how the membrane targeting sequence of MinE contributes to gradient formation of the Min system we purified and labelled a MinE mutant without its membrane targeting sequence ( MinE without amino acids three to eight , herein referred to as MinE ( Δ3–8 ) ) and reconstituted it with MinD in microcompartments . Interestingly , MinE ( Δ3–8 ) resulted in a severe phenotype in the bottom-up system . The Min system was still able to self-organize into dynamic pattern , but instead of a stable pole-to-pole oscillation , the Min system formed dynamic patches with irregular shape and without a constant axis of movement ( Figure 1—figure supplement 5 ) . Thus , although MinE ( Δ3–8 ) still counteracted MinD , no stable MinD gradients were generated . Depending on the degree of affecting the membrane targeting sequence of MinE , the effective generation of a gradient and its localization along the long axis of the cell might therefore be altered to different degrees , which might account for the distinct phenotypes in vivo , such as filamentous cells or mini cells . While the structural properties and the diffusion-reaction driven self-organization of MinD and MinE determine the non-equilibrium distribution of the Min system , a third protein , MinC , is proposed to be the actual mediator to forward positional information of the Min gradient to downstream targets . In particular , MinC is proposed to inhibit FtsZ–the cytoskeletal framework of divisome assembly ( Hu et al . , 1999; Hu and Lutkenhaus , 2000; Shen and Lutkenhaus , 2009 , 2010 ) . To investigate the spatial distribution of eGFP-MinC in our minimal system , we co-reconstituted MinD and MinE oscillations with MinC in micro compartments . We found that eGFP-MinC oscillated from pole-to-pole by coupling to MinD/E patterns , which strongly resembles the dynamics of MinC in vivo ( Figure 1B ) . Based on time-lapse movies , we then analyzed the time-averaged distribution of eGFP-MinC and found the concentration profile of MinC also exhibit a pronounced minimum in the middle of the compartment ( Figure 1C ) . This experiment revealed that the effective non-equilibrium distribution of the complete Min system , with the highest concentration of MinC at the cell poles , is fully recapitulated , in our simple cell-free setting . Note that the E . coli lipid extract for generating membranes in these experiments is a mixture of mainly three different lipids , ( phosphatidylethanolamine [PE , neutral] , phosphatidylglycerol [PG , negatively charged] and cardiolipin [CA , negatively charged] ) and that the membranes have a net negative charge . Recently it has been shown that a minimal lipid bilayer for pattern Min formation on flat supports only requires two different lipids . Moreover , cardiolipin , which has previously been proposed as a structural polar cue , which might help in triggering the pole to-pole oscillations of Min proteins , was not required for Min pattern formation on planar supports . Instead , the negative charge has been suggested to play a primary role in formation of surface waves on planar membranes . ( Vecchiarelli et al . , 2014 ) To address whether membrane charge and not the chemical membrane composition is also sufficient for gradient forming pole-to-pole oscillations , we established the minimal membrane composition for Min protein oscillations in our synthetic system . First , we confirmed the aforementioned results that membrane charge is a strong determinant for Min pattern formation on flat supported membranes ( Figure 1—figure supplement 4A–D ) . Then we reconstituted the Min proteins in compartments with two different minimal membrane mixtures: First , with 70% neutrally charged DOPC and 30% negatively charged cardiolipin , and then with 70% DOPC and 30% negatively charged PG . In both cases , the proteins retained their ability of pattern formation and pole-to-pole oscillations ( Figure 1—figure supplement 4E ) . Thus , we conclude that Min gradients can be established without any structural cues in the membrane . In contrast , simple physical parameters , such as electrostatic interactions between the membrane and the proteins , are strong determinants for gradient formation . For high amounts of negatively charged lipids , the length scales of the protein patterns decrease ( Figure 1—figure supplement 4B and [Vecchiarelli et al . , 2014] ) . Thus , the spatial scales of the Min proteins on highly negatively charged membranes would be too small to form stable pole-to-pole oscillations in compartments , which were engineered for protein length scales on E . coli membranes . The implication that Min protein oscillations are perturbed if the membrane charge increases is in agreement with live cell studies of GFP-MinD in PE-lacking E . coli cells , which demonstrate that Min protein pattern are affected if the membrane charge is increased ( Mileykovskaya et al . , 2003 ) and provides further evidence that a balanced ratio of charged and non-charged lipids is of critical importance for the live cycle of E . coli . Intracellular protein gradients play a pivotal role in cellular organization through providing positional information for downstream proteins . However , although protein gradients are now emerging as general motives to organize cells , tools to study gradient-mediated organization of three-dimensional space are limited . To directly investigate whether the established Min gradient indeed provides an efficient cue to position downstream targets , we attempted the assembly and positioning of FtsZ to the middle of a synthetic compartment in a co-reconstitution assay . To keep the system as minimal as possible , we avoided the native membrane adaptors of FtsZ , FtsA and ZipA , and used a FtsZ hybrid protein , fused to YFP and a membrane targeting sequence ( FtsZ-mts ) ( Osawa et al . , 2008; Arumugam et al . , 2014 ) . In spite of this simplification , we found that highly FtsZ-mts enriched regions were clearly visible in the center of the compartments , while FtsZ-mts localization was strongly reduced at the polar regions ( Figure 2 ) . Negative controls with FtsZ-mts bundles on flat supported bilayers confirmed that all thee Min proteins are required to deplete FtsZ-mts from the membrane ( Figure 2—figure supplement 1 ) ( Arumugam et al . , 2014 ) . This unequivocally demonstrates that an effective time-averaged protein gradient can indeed regulate the localization of down-stream targets to predefined localizations , such as the middle of a compartment . 10 . 7554/eLife . 03949 . 009Figure 2 . Reconstituted protein gradients coordinate localization of downstream targets in cell-shaped compartments . ( A ) Confocal image of the tubulin homolog FtsZ-mts on the bottom of membrane-clad compartments . FtsZ-mts assembles into bundles which are aligned perpendicular to the length axis of the compartments . Scale bar: 10 µm . ( B ) Images with the focal plane at the bottom , middle and top of the compartment , respectively . The buffer was not yet reduced to a level below the upper level of the chip . Therefore membrane and FtsZ-mts network are still intact on the upper level of the chip . Scale bar: 5 µm . ( C and D ) When the gradient forming Min system ( red: MinE . Atto655 ) and FtsZ-mts ( blue ) are reconstituted together , the Min gradient coordinates the localization of FtsZ-mts to the middle of the compartments . Scale bar: 5 µm . ( C ) Time between frames: 20 s . Negative controls in Figure 2—figure supplement 1 confirm that MinD and MinE are required to form dynamic patterns and that MinC is needed in addition to deplete FtsZ-mts from the membrane . ( E ) Schematic image of reconstituted FtsZ-mts positioning . DOI: http://dx . doi . org/10 . 7554/eLife . 03949 . 00910 . 7554/eLife . 03949 . 010Figure 2—figure supplement 1 . MinD MinE and ATP are required to form dynamic patterns and MinC is needed to deplete FtsZ from the membrane . ( A ) Flat supported lipid membrane with reconstituted FtsZ-mts . Scale: 5 µm . ( B ) FtsZ-mts and MinE . Scale: 5 µm . ( C ) FtsZ-mts , MinE and MinD scale ( overview ) : 20 µm , scale ( zoom in ) : 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03949 . 010 In contrast to the living cell , the FtsZ-mts distribution in the middle of synthetic compartments appears to be rather broad . The major reason for the wide FtsZ-mts distribution is most likely the ten times larger Min protein pattern in vitro , as compared to living cells . The wide MinC minimum allows multiple FtsZ-mts bundles to localize in a wide region in the middle of the compartment . Note that FtsZ-mts bundles in the central region aligned perpendicular to the long axis of the compartments ( Figure 2A , B , D ) . This finding is in agreement with previous results that demonstrated a preferential alignment of FtsZ-mts along negatively curved membranes on grooved glass as membrane support . ( Arumugam et al . , 2012 ) . Positional information of the MinD gradient in vivo and in our synthetic system is mediated by the FtsZ inhibitor MinC which directly interacts with FtsZ ( Hu et al . , 1999 ) . MinC has a C-terminal and an N-terminal domain , ( Hu and Lutkenhaus , 2000; Cordell et al . , 2001 ) which are both involved in FtsZ inhibition . The C-terminal domain of MinC is responsible for binding to MinD and dimerization of MinC . Furthermore , it binds to the conserved C-terminus of FtsZ and was suggested to compete with FtsA and ZipA to inhibit FtsZ ring formation . ( Hu and Lutkenhaus , 2000; Shen and Lutkenhaus , 2009 ) . Note , that FtsZ-mts comprises amino acids 1–366 of E . coli FtsZ , but instead of the C-terminally conserved tail , it comprises YFP and a membrane targeting sequence . The proposed interference of the C-terminal domain with FtsZ is therefore not involved in the system described above , which suggest that the inhibitory activity of the C-terminal domain is not required in a minimal system to inhibit FtsZ at compartment poles . In the context of a living cell , the inhibitory activity of the C-terminal domain of MinC might provide an additional mechanism for disrupting FtsZ , rendering the Min system more robust . The N-terminal domain of MinC inhibits FtsZ polymerization by binding to an alpha-helix of FtsZ at the interface of FtsZ subunits in a filament ( Hu et al . , 1999; Shen and Lutkenhaus , 2010 ) . Since the FtsZ-binding region for the C-terminal domain of MinC is not present in FtsZ-mts , the major inhibitory activity contributing to FtsZ-inhibition in our synthetic system should originate from the N-terminal domain of MinC . To test this hypothesis , we used a MinC mutant ( MinC-G10D ) with lower inhibitory activity of its N-terminal domain ( Hu et al . , 1999; Shen and Lutkenhaus , 2009 , 2010 ) . The MinC-G10D mutant has reduced ability to inhibit FtsZ in vivo and in vitro , but interacts with MinD ( Hu et al . , 1999 ) . We first characterized MinC-G10D by co-reconstituting it at different concentrations with FtsZ-mts , MinD and fluorescently labeled MinE on flat supported membranes . Consistent with the behavior of wild type MinC , we found that the self-organizing Min protein patterns were unperturbed at low concentrations of MinC-G10D , and that the Min patterns were weakened by high concentrations of MinC-G10D ( Figure 3A ) . The weakening of Min patterns for high concentrations of MinC is likely due to an overlap of the binding sites for MinC and MinE on MinD ( Ma et al . , 2004; Wu et al . , 2011 ) , which might result in a competition of MinC with MinE for binding sites on MinD at large MinC concentrations . The weakening of Min Protein patterns therefore suggests that MinC-G10D binds to MinD . However , in contrast to MinC the MinC-G10D mutant was inefficient in depolymerizing FtsZ-mts on flat membranes , as well as at the poles , when co-reconstituted in microcompartments ( Figure 3B , C , Figure 3—figure supplement 1 ) . These observations confirm that the inhibitory activity in the synthetic system originates from the N-terminal domain of MinC and that the inhibitory activity of the N-terminal domain is sufficient to inhibit FtsZ-polymerization at compartment poles in the context of a minimal system . 10 . 7554/eLife . 03949 . 011Figure 3 . MinC-G10D does not inhibit polymerization of FtsZ-mts . ( A–C ) FtsZ-mts ( blue ) , MinD , MinE ( 5% labeled with Atto655: red ) were reconstituted with the MinC-G10D mutant . Concentrations of MinD , MinE and FtsZ-mts are constant for all images . ( A ) Confocal images of Min protein pattern ( MinE . Atto655: red ) on flat supported membranes at different concentrations of MinC-G10D . High concentrations of MinC-G10D disturb Min protein patterns . The contrast between images is not comparable and is increased for higher concentrations of MinC-G10D , because the MinE intensity at the membrane decreased with higher concentrations of MinC-G10D . Scale bar: 50 µm . ( B ) Confocal images of the Min system ( MinE . Atto655:red ) and FtsZ-mts ( blue ) on flat membranes at different concentrations of MinC-G10D demonstrate that MinC-G10D is inefficient in disturbing FtsZ-mts networks . Scale bar: 2 µm . ( C , Figure 3—figure supplement 1 ) In cell-shaped micro compartments with MinD , MinE , FtsZ-mts and 50 nM MinC-G10D the Min system ( MinE . Atto655: red ) oscillates from pole-to-pole . However MinC-G10D does not inhibit polymerization of FtsZ-mts ( blue ) at the compartment poles . Time between frames: 90 s , scale bar: 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03949 . 01110 . 7554/eLife . 03949 . 012Figure 3—figure supplement 1 . MinC-G10D does not inhibit polymerization of FtsZ-mts at compartment poles . Confocal images of three compartments with FtsZ-mts ( blue ) , MinD , MinE ( 5% labeled with Atto655: red ) and 50 nM MinC-G10D in microcompartments . The FtsZ-mts bundels are not inhibited at compartment poles . Scale bar: 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03949 . 012 We demonstrated that coordinated spatial control by intracellular protein gradients can be reconstituted in cell-free systems , and that these gradients are modulated by geometric parameters . The reconstitution of self-organizing protein systems in membrane clad soft-polymer compartments represents a simple approach to study membrane interacting proteins in a defined sample volume and geometry . Applying this system , we particularly tackled three major questions . First , what are the minimal requirements for protein gradient formation ? Second , can the organized localization to predefined sites be achieved by reading out the positional information of a minimal protein gradient system ? And finally , what are the geometric boundary conditions to generate and modulate protein gradients ? To address the minimal requirements for gradient formation , we considered the Min protein system as a prototype for a gradient forming system and accomplished the reconstitution of a time-averaged nonhomogeneous concentration profile with the lowest concentration in the middle of the compartment . We demonstrated that the minimal requirements for Min gradient formation comprised only a negatively charged membrane , a cell-shaped compartment , ATP as an energy source , and the two antagonistic proteins MinD and MinE . To investigate whether the reconstituted gradient provides an efficient cue to organize cellular space by positioning down-stream targets , we co-reconstituted the MinD/E system with the cell division protein FtsZ-mts and the FtsZ inhibitor MinC . This cell-free bottom-up system recapitulated remarkably cell-like properties , such as coordinated spatial coordinated localization of FtsZ-mts to the middle of a cell-like compartment . Interestingly , although it is proposed that MinC has two inhibitory domains which act on FtsZ–FtsZ interactions and on the ternary system FtsZ-FtsA/ZipA , respectively ( Shen and Lutkenhaus , 2009 , 2010 ) , our data suggest that MinC can already inhibit FtsZ-localization at compartment poles by only disrupting FtsZ–FtsZ interactions . While we do not exclude that the interaction of MinC with the conserved C-terminal domain , which interacts with FtsA and ZipA , is a parallel mechanism to inhibit FtsZ polymerization at cell poles and possibly make FtsZ structures more sensitive to the action of the Min system , our results show that the minimal biochemical requirements to displace FtsZ from compartment poles by the Min system do not involve the disruption of FtsZ-membrane binding sites . Moreover , we demonstrated that very long compartments harbored a MinC protein concentration profile with multiple minima . While a normal growing wild type cells typically comprises a Min concentration gradient with only one minimum for division protein assembly , multiple minima resulting in more than one sides for cell division , are potentially advantageous in filamentous cells . If transient unfavorable environmental conditions inhibit cell division but not cell elongation , divisions at multiple sides could ensure the re-establishment of a normal cell length when conditions , which support division , reoccur . To systematically address how compartment geometry modulates Min protein gradients , we reconstituted Min protein patterns in compartments with systematically varying shape . Varying the length of the compartments , we found that robust pole-to-pole oscillations occur in compartments , even if the length of these compartments was varied by a factor of 2 . 3 . In a live cell , this robustness of the Min oscillations with respect to cell length should be highly significant , because the length of living cells is not constant over time , and during the life cycle of E . coli , the cell doubles in length . The observation that the Min protein patterns in vitro deviate from a stable axis in very short compartments , and that higher order oscillations systematically occur the longer the compartment is , confirm that the MinD/E patterns in synthetic compartments are strikingly similar to the Min pattern of round , wild type and filamentous cells in vivo ( Fu et al . , 2001; Hale et al . , 2001; Corbin et al . , 2002; Shih et al . , 2005; Raskin and de Boer , 1999b ) . Interestingly , when we increased the width of the compartments , the Min proteins oscillated along the short axis of the compartment . The cause of oscillations switching along the length axis to oscillations along the short axis is not known . However , the width at which this switching occurs might be correlated with the wavelength of the Min protein patterns and therefore related to the mechanism for determining the wavelength of the Min system . It is also unknown whether a similar gradient along the short axis of the cell is used in living cells , and which factors might contribute on the bacterial scale to regulate gradients along the short axis . However , note that FtsZ involving cell-division in certain bacteria , such as Laxus oneistus , occurs along the long axis . ( Polz et al . , 1992; Leisch et al . , 2012 ) Although the spatial cues to determine the position of their division plane are unknown , it is intriguing to speculate that gradients along the short axis might also provide an effective cue to spatially organize space along the short axis Thus , protein gradients might also orient the cell division plane in cells which divide along the length axis of the cell . In line with our in vitro results , Min oscillations along the short axis of a bacterial cell might be accomplished by an increased width as compared to an E . coli cell , or by smaller scales of Min protein patterns through modified reaction and/or diffusion rates of the proteins . However , also other systems which generate gradients along the length axis of a bacterial cell , such as activity gradients of phosphorylated proteins ( Chen et al . , 2011 ) , might provide mechanisms to generate gradients along the short axis . In addition to varying the overall width of a compartment , we mimicked septum closure in the middle of the compartments and thereby simulated morphologically different stages of bacterial cell division . While it has been controversial how the pool of Min proteins is distributed to the two daughter cells after cell division , we provide experimental evidence for the theoretical model that Min protein partitioning is a result of geometric variation during septum constriction ( Di Ventura and Sourjik , 2011 ) . In summary , we reconstituted a cell-free bottom-up system which recapitulates cell-like properties , such as coordinated spatial control for division site placement by intracellular protein gradients . In particular , our results unravel the minimal requirements for localization of the cell division protein FtsZ . Moreover , our findings demonstrate that the interplay of geometrical cell boundaries and self-organizing Min proteins determine the orientation of Min gradients , the number of potential FtsZ assembly sites , as well as Min protein partitioning during cell division . Thus , our synthetic system opens a way to study complex organization principles in a simplified environment , and provides novel insights into the basic biophysical and biochemical requirements for gradient formation . Using photolithographic techniques , resist micro-patterns were produced on Si wafers mainly as described in reference ( Zieske and Schwille , 2013 ) . Photoresist patterns ( ma-P 1275 , micro resist technology GmbH , Germany ) of about 10 µm height on top of Si wafers ( Si-Mat , Kaufering , Germany ) were produced by photolithography . The chrome mask was purchased from Compugraphics Jena GmbH . To better mimic the curvature of the cellular membrane , we left a small gap between photoresist and mask during exposure of the mask to UV light , resulting in more tilted walls as compared to the contact mode . For reconstitution of Min protein oscillations along the length axis of a compartment , the mask patterns had a width of 10 µm . The Si wafers were coated with chlorotrimethylsilane ( Sigma–Aldrich , St . Louis , MO ) to prevent sticking of PDMS to the wafer . PDMS ( Sylgard184 , Dow Corning , Midland , MI ) was mixed at a ratio ( monomer to cross-linker ) of 10:1 and degased under vacuum . The PDMS mixture was then poured on top of the wafer . Standard glass coverslides where then manually pressed into the liquid PDMS on top of the wafer , leaving a PDMS layer of about 30 µm between the wafer and the glass coverslide . After curing the PDMS at 80°C overnight , the bulk PDMS layer was peeled off . With the help of a razor blade , the glass coverslide with the about 30 µm thin , micro structured PDMS layer was carefully separated from the wafer and used as sample support ( Figure 1—figure supplement 2 ) . The microstructured PDMS/glass devices were stored at room temperature until further use . Before the microstructured PDMS was used as a membrane support , it was sonicated for 5 min in ethanol , washed with water , air dried , and treated with air plasma . The two-dimensional geometry at the upper level of the compartment is determined by the patterns on a chrome mask . The profiles of the micro compartments can be designed with a flat bottom or a tapered bottom ( Figure 1—figure supplement 1C ) . The Min protein oscillations and Min gradient formation were supported by both profiles . However , the image quality was better when the bottom of the compartments was flat . The advantage of a tapered bottom was that these structures allowed FtsZ-mts to align perpendicularly to the length axis of the compartments ( Figure 1—figure supplement 1D ) . Thus , all compartments containing FtsZ-mts ( Figures 2 , 3 , 5 ) had a tapered bottom . To keep the boundary conditions comparable , the MinCDE oscillations and gradients in Figure 1 were reconstituted in tapered compartments as well . To better visualize the oscillation mode and measure the oscillation period of MinD/E pattern , compartments with a flat bottom were employed ( Figure 4 , Video 1 and 2 ) . Note that the oscillation mode in flat compartments ( Figure 4A ) is in agreement with FtsZ localization in tapered compartments ( Figure 5B ) . Compartments with constrictions and compartments for determining width dependent Min patterns had a flat bottom , as well ( Figure 6 , Video 3–5 ) . Experiments with MinD/E pattern in compartments with constricting septum ( Figure 6 and Figure 6—figure supplement 1–3 ) : The walls of the compartments were slightly tilted and the bottom of the compartment was flat . The width at the unconstricted region was about 12 µm at the top and 10 µm at the bottom of the compartments . The septal region of compartments , in which the switch of the oscillation modes occurred , had a width of about 2 µm at the top . The resolution limit in z-direction and the tilted PDMS-surface through which the laser beam had to pass , render an exact description of the profile at the septum challenging , However a confocal x/z-scan of the septum is depicted in Figure 6—figure supplement 3C . Also note that a meniscus due to the surface tension of the buffer might render the actual cross-sectional area of the buffer smaller than the profile of the structure suggests . E . coli lipid membrane within the micro compartment was prepared as described previously ( Zieske and Schwille , 2013 ) . Two-dimensional motility of the lipids within the membranes was confirmed by labelling the membrane with 0 . 1% DiI ( FAST DiI , Invitrogen , Carlsbad , CA ) and performing FRAP experiments . DiI labeled membranes were also applied to acquire compartment profiles . Proteins , ATP and/or GTP ( depending on experiment ) of defined concentration were added to a buffer reservoir of 200 µl on top of membrane clad supports . Dynamic Min patterns and/or FtsZ-mts bundles formed by self-organization on top of the membrane . At this step the Min patterns did not oscillate but formed travelling waves . Afterwards the buffer above the microstructures was removed by pipetting . Restricted to a small buffer compartments within the micro structures the Min protein patterns start to oscillate in cell-shaped geometries . To limit evaporation a lid was placed on top of the plastic ring surrounding numerous microstructures . The top of individual chambers was open with a buffer/air interface ( Figure 1—figure supplement 2 ) . MinC , eGFP-MinC , MinD , eGFP-MinD and MinE were purified with N-terminal His-tags as previously described ( Loose et al . , 2008 , 2011 ) . Purification of FtsZ-mts was originally described by the Erickson-group ( Osawa et al . , 2008 ) . MinE was labeled with Alexa Fluor 488 C5 Maleimide ( Molecular Probes , Carlsbad , CA ) or ATTO655 Maleimide ( ATTO-TEC , Siegen , Germany ) according to the manufactures manual . The plasmid for overexpression of MinC codes for an open reading frame for MinC connected to an N-terminal hexahistidine tag by a linker ( Loose et al . , 2011 ) . The MinC-G10D mutant was obtained by site-directed mutagenesis ( QuikChange II , Agilent Technologies , Santa Clara , CA ) of this plasmid to obtain pET28a-MinC ( G10D ) . MinE without membrane targeting sequence ( MinE ( Δ3–8 ) ) was generated by deleting the coding sequence for aminoacids 3–8 of MinE of the overexpression plasmid for MinE ( Loose et al . , 2008 ) using a ‘GeneArt , Seamless Cloning and Assembly Enzyme Mix’ ( Invitrogen , Life Technologies , Carlsbad , CA ) and primers TTCCGCGATGCGAATTCGGATCCGCGACC and AATTCGCATCGCGGAAGAAAAACACAGCCAACA to obtain pET28a-MinE ( Δ3–8 ) . MinC-G10D and MinE ( Δ3–8 ) were purified like MinE . Purified MinE ( Δ3–8 ) was labeled with Alexa647 maleimide ( Molecular Probes , Carlsbad , CA ) according to the manufactures manual . The sample buffer in all experiments contained 25 mM Tris–HCl pH 7 . 5 , 150 mM KCl , 5–15 mM MgCl2 . ( 5 mM MgCl2 if only the MinD and MinE were reconstituted , 15 mM MgCl2 if FtsZ-mts was reconstituted ) When Min proteins where included in the experiments , the buffer was supplemented with 2 . 5 mM ATP . When FtsZ-mts was used 3 mM GTP was added to the buffer . Protein concentrations used in this study were as follows . MinC or eGFP-MinC: 0 . 05 µM , MinE: 1 µM , MinD: 1 µM ( depending on the experiment MinD was supplemented with 10% eGFP-MinD ) , FtsZ-mts: 1 µM . When MinE was imaged 10% of MinE were substituted with the labeled version of MinE . Note that these concentrations refer to the average concentrations in the sample buffer before the buffer volume was reduced below the upper level of the micro compartments . Microscopy: Image acquisition was performed on a ZEISS ( Jena , Germany ) LSM780 confocal laser scanning microscope ( inverted microscope ) equipped with a ZEISS Plan-APO 25×/NA 0 . 8 objective and ZeissC-Apochromat 40×/1 . 20 objective . For acquiring images of Min pattern and FtsZ bundles the focal plane was typically adjusted to the bottom of the compartments ( if not stated otherwise ) . Note that the bottom area of the microstructures is smaller than the top area . Therefore the sizes of structures on the pictures might appear smaller than indicated in the text ( text always refers to upper rim of microstructures ) . The relative intensity profiles along the length axis of compartments can be more accurately determined if the focus plane is in the middle of the compartment , because at this plane the relative intensity profiles at different time steps are less prone to errors due to slight focal shift . Thus , data in Figure 1D and Figure 1—figure supplement 3 were extracted from time-lapse images with the focal plane in the middle of the compartments .	When a cell divides , it is important that its contents are separated in the right place to make sure that both daughter cells have everything that they need to survive . To do this , the molecular ‘machinery’ that physically divides the cell needs to know where to assemble . In the bacterium E . coli , the location of cell division depends on a group of proteins called the Min proteins . These proteins are not evenly distributed over the cell . Instead , they oscillate back and forth to set up concentration gradients , with the concentration of Min proteins being lowest in the middle of the cell and highest at the ends . The machinery that divides the cell assembles at the point where the concentration of Min proteins is lowest . However , it is not clear exactly how the protein gradients are set up in the cell , and whether these gradients are indeed sufficient to position the cell division machinery . To explore this process , Zieske et al . engineered artificial cells that mimicked some of the basic properties of living cells . In these artificial cells , the Min proteins organized themselves into gradients that were similar to those found in living cells . This gradient then caused another protein called FtsZ—which is involved in cell division—to accumulate in the middle of the artificial cell . Zieske et al . also showed that the shape of the artificial cell influenced the shape of the protein gradient . This research shows that the interplay between the shape of a bacterial cell and a defined set of proteins could control the position of cell division . The simplified system that Zieske et al . have developed could also be used to study other aspects of cell organization and cell division .	[ "Abstract", "Introduction", "Results", "Discussion", "Material", "and", "methods" ]	[ "cell", "biology" ]	2014	Reconstitution of self-organizing protein gradients as spatial cues in cell-free systems
In plants , a complex mixture of solutes and macromolecules is transported by the phloem . Here , we examined how solutes and macromolecules are separated when they exit the phloem during the unloading process . We used a combination of approaches ( non-invasive imaging , 3D-electron microscopy , and mathematical modelling ) to show that phloem unloading of solutes in Arabidopsis roots occurs through plasmodesmata by a combination of mass flow and diffusion ( convective phloem unloading ) . During unloading , solutes and proteins are diverted into the phloem-pole pericycle , a tissue connected to the protophloem by a unique class of ‘funnel plasmodesmata’ . While solutes are unloaded without restriction , large proteins are released through funnel plasmodesmata in discrete pulses , a phenomenon we refer to as ‘batch unloading’ . Unlike solutes , these proteins remain restricted to the phloem-pole pericycle . Our data demonstrate a major role for the phloem-pole pericycle in regulating phloem unloading in roots . In plants , the products of photosynthesis in green tissues are delivered by the phloem to distant organs where they are utilized in growth or storage ( Turgeon and Wolf , 2009; De Schepper et al . , 2013 ) . Modern agriculture aims to maximize the amounts of carbon-based products allocated to storage organs such as grains and tubers , structures that act as carbon ‘sinks’ for the assimilates delivered by the phloem . The process by which solutes exit the phloem is termed phloem unloading , and is a central target for regulating the flux of carbon into sink tissues ( Ham and Lucas , 2014 ) . In addition to assimilates , phloem sap also contains numerous proteins and RNAs ( Kehr , 2006; Atkins et al . , 2011; Batailler et al . , 2012; Turnbull and Lopez-Cobollo , 2013 ) . Unloading must therefore combine the seemingly antagonist functions of high selectivity with large permeability in order to allocate assimilates to growth zones while controlling the movement of macromolecules . The mechanisms that enable the precise and coordinated unloading of phloem-mobile compounds remain unknown . Solute flow occurs through sieve elements ( SEs ) , elongated cells connected to each other by perforated end walls called sieve plates . During differentiation SEs lose most of their cellular components , including their nucleus ( Oparka and Turgeon , 1999; Furuta et al . , 2014 ) . Ontogenetically related , metabolically active companion cells ( CCs ) support the adjacent enucleate SEs throughout their lifespan ( Lucas et al . , 1996; Pritchard , 1996; van Bel and Knoblauch , 2000; Lalonde et al . , 2001; Otero and Helariutta , 2017 ) . The pressure flow model of phloem transport , originally proposed by Münch ( 1930 ) , envisages that an osmotically generated pressure differential drives the bulk flow through the SEs that connect photosynthetic tissues ( sources ) with those in which carbon consumption occurs ( sinks ) . The loading of solutes in source tissues results in a high osmotic potential within SEs . The reduction of turgor pressure in sink organs due to carbon consumption leads to a pressure gradient that provides the energy to overcome viscous resistance within the SEs , resulting in a passive phloem flow from source to sink ( Münch , 1930 ) . Phloem sap collected from excised aphid stylets contains a complex mixture of macromolecules and low-molecular weight solutes ( Atkins et al . , 2011 ) . In a recent study , Paultre et al . ( 2016 ) showed that many proteins , including those with targeting sequences , can move across a graft union and be unloaded near the root tip . Significantly , these proteins entered a post-phloem domain beyond which their movement was restricted . This observation begs the question as to how the phloem is able to discriminate between macromolecules and solutes during unloading . Besides the central function in resource allocation , it is now well established that the phloem also serves as network for transmission of chemical ( Kramer and Boyer , 1995a; Mullendore et al . , 2015 ) and electrical ( Hedrich et al . , 2016 ) signals . Phloem unloading in actively growing tissues such as the shoot or root apex occurs through the protophloem , a transient tissue that connects the conducting phloem with the receiver cells in sink tissues ( Oparka et al . , 1994 ) . Initial investigations of phloem unloading in the root tip of Arabidopsis ( Oparka et al . , 1994 , 1995; Wright and Oparka , 1996 ) provided evidence that unloading occurs through plasmodesmata ( PD ) , the specialized pores that connect plant cells . Due to cell division and growth in the apical region of the root the demand for assimilates is high . Protophloem sieve elements ( PSEs ) become mature in such regions to accommodate this demand while the neighboring cells are still differentiating ( Furuta et al . , 2014 ) . Because PSEs lose their nucleus , they cannot divide to keep pace with the growth of the neighboring cells . Therefore , they become increasingly elongated until they become inactive in transport and eventually obliterated ( Erwin and Evert , 1967; Eleftheriou and Tsekos , 1982; Furuta et al . , 2014 ) . In the elongation zone of the root , solutes are transferred laterally from the metaphloem SEs ( MSEs ) to the PSEs , allowing phloem continuity between source and sink tissues ( Stadler et al . , 2005; Winter et al . , 1992 ) . A currently favored hypothesis of phloem unloading is the ‘high-pressure manifold model’ proposed by Fisher ( 2000 ) , recently evaluated by Patrick ( 2013 ) . A central element of this model is that a low pressure gradient occurs along the flow path , with a steep drop in pressure between the PSEs and surrounding cells in the phloem unloading zone . In this scenario , allocation of carbon is controlled by the lateral hydraulic conductance in the unloading zone . Recent phloem turgor measurements in morning glory , however , did not support this model , as the bulk of the pressure is consumed by friction within the SEs , and only small pressure gradients are available for unloading ( Knoblauch et al . , 2016 ) . The Fisher ( 2000 ) model also does not explain how both small solutes and macromolecules can leave the phloem simultaneously . Paultre et al . , 2016 suggested recently that the removal of mobile proteins into a post-phloem domain may be necessary to prevent the terminal PSEs from becoming occluded , an event that would lead to dissipation of the turgor gradient between source and sink . Unfortunately , many of the factors that determine phloem unloading are not well studied . This is because the phloem in most sinks is difficult to access as it is embedded in an opaque layer of tissue . Phloem transport ceases immediately when the source is detached from the sink . Therefore , the function of the phloem can only be studied in situ , requiring new approaches for dissecting the factors that regulate phloem unloading . The many unknowns surrounding phloem unloading in plants prompted us to conduct a detailed structure/function study of the terminal PSEs in the phloem unloading zone which , in Arabidopsis , is amenable to non-invasive imaging ( Oparka et al . , 1994; Knoblauch et al . , 2015 ) . We combined a detailed ultrastructural analysis of the cellular interfaces in the phloem unloading zone with the kinetics of phloem unloading of fluorescent solutes and macromolecules obtained by real-time imaging of growing roots . This analysis allowed us to derive new quantitative data on the factors that regulate phloem unloading in Arabidopsis . We report on the presence of a unique class of ‘funnel plasmodesmata’ that are involved specifically in the unloading of molecules into the phloem-pole pericycle . We show , by mathematical modelling , that phloem unloading of small solutes from PSEs is convective , i . e . it occurs continuously by a combination of mass flow and diffusion . In contrast , we find that macromolecules are unloaded in discrete pulses , a phenomenon we refer to as ‘batch unloading’ . These macromolecules are diverted specifically into the two phloem-pole pericycle cells that abut each PSE where they are filtered out from the unloaded solutes . In Arabidopsis roots , phloem unloading occurs exclusively from the protophloem , a short-lived tissue that is functional in the zone of root elongation ( Oparka et al . , 1994 ) . Published images of phloem unloading in Arabidopsis give the impression that the unloading zone is a relatively broad region ( Oparka et al . , 1994; Wright and Oparka , 1996; Knoblauch et al . , 2015 ) . However , such images represent a ‘snapshot’ of unloading , taken at a defined time point following the application of fluorescent solutes to the leaf . Carboxyfluorescein diacetate ( CFDA ) is the most widely utilized phloem-mobile probe ( Oparka et al . , 1994; Wright and Oparka , 1996; Knoblauch and van Bel , 1998 ) . This probe is non-fluorescent when applied to source leaves but is subsequently cleaved by endogenous esterases to produce fluorescent , membrane-impermeant , carboxyfluorescein ( CF; Knoblauch et al . , 2015 ) . The dye travels with the phloem translocation stream to sink tissues , where it can be visualized ( Figure 1A ) . When studied in real time , the dye characteristically shows preferential movement outwards into the cortex relative to the stele ( Figure 1A; see also Oparka et al . , 1994 ) . Unloading of the dye in the root tip indicates a symplastic ( plasmodesmata-mediated ) pathway ( Oparka et al . , 1994; Wright and Oparka , 1996; Knoblauch et al . , 2015 ) . The cells in the unloading zone sequester the dye rapidly into their vacuoles where it becomes trapped . This results in an increasing fluorescence within cells over time ( Wright and Oparka , 1996 ) . Due to root growth , the cells initially involved in unloading move out of the phloem-unloading zone basipetally , but maintain their fluorescence due to the presence of dye in their vacuoles . Therefore , the apparent unloading zone broadens over time as root growth progresses , obscuring the current site of unloading . We therefore sought to define precisely the dimensions of the true phloem-unloading zone . 10 . 7554/eLife . 24125 . 003Figure 1 . Symplastic unloading of phloem mobile probes . ( A ) 2D optical section of unloading of CFDA in the root tip . The two protophloem files leading into the root tip are shown ( solid arrows ) and sequestration of CFDA into the vacuoles is apparent ( dashed arrows ) . ( B ) Unloading of esculin ( blue ) in the root tip of a transgenic Arabidopsis line expressing GFP ( green ) targeted to the ER lumen of the PSE ( pMtSEO2::GFP5-ER ) . Esculin escapes the protophloem file ( solid arrow ) into the cytoplasm of neighboring cells ( open arrow ) . In contrast to CFDA , esculin is only sequestered in the vacuoles at high concentrations ( dashed arrow ) . ( C–E ) Three frames extracted from Video 1 . ( C ) GFP targeted to the ER lumen of PSEs demarcates the nuclear membrane of young sieve elements that have not yet been integrated into the unloading zone ( solid arrows ) . Dashed arrows indicate two degrading nuclei in cells that are already filled with esculin ( blue ) ( also for D and E ) . ( D ) Degradation of the nucleus ( yellow arrow ) coincides with the opening of the sieve-plate pores , allowing esculin ( blue ) to enter the cell . This defines the new PSE zero . ( E ) As nuclear degradation continues , the sieve element becomes an integral member of the phloem unloading zone . DOI: http://dx . doi . org/10 . 7554/eLife . 24125 . 003 Esculin is a naturally fluorescent , glucosylated coumarin derivative recently introduced as phloem-mobile probe ( Knoblauch et al . , 2015 ) . Unlike CFDA , it is loaded into the phloem by the sucrose transporter , SUC2 , in the CCs of source tissues ( Gora et al . , 2012; Knoblauch et al . , 2015 ) . Sequestration of this probe is minimal and occurs only when high concentrations of the probe are applied ( Figure 1B ) . This feature allows for extended acquisition of time-lapse movies . Unlike CF , esculin can also be detected clearly in lines expressing GFP ( Knoblauch et al . , 2015 ) . To define the developmental stage at which differentiating PSEs become integrated into the unloading zone , we used transgenic Arabidopsis lines expressing GFP targeted to the sieve-element ER ( HDEL-GFP ) under control of the MtSEO2 promoter , a SE-specific promoter ( Froelich et al . , 2011; Knoblauch and Peters , 2010 ) ; Figure 1B–E ) . This transgenic line clearly demarcates PSEs in the early stages of differentiation and is an excellent marker of the nuclear membrane ( Figure 1C–E ) . We loaded source leaves of this line with esculin and acquired time-lapse movies of unloading in the terminal PSEs . Esculin did not enter differentiating PSEs that were symplastically isolated from the rest of the protophloem file . These cells still had a fully intact nuclear membrane ( Figure 1C , Video 1 ) . Once the nuclear membrane started to degenerate the sieve-plate pores opened rapidly , allowing esculin to enter the cell , which then became an integral member of the protophloem-unloading zone ( Figure 1D and E ) . For convenience of reference , we refer to this protophloem cell as ‘PSE zero’ ( yellow dashed arrow in Figure 1D ) . Remnants of the nuclear membrane remained for some time in PSEs that had been newly incorporated into the phloem-unloading zone ( Figure 1C–E ) . These observations can be seen in Video 1 . Our data suggest that nucleate , differentiating PSEs are isolated from the translocation stream . However , the degeneration of the nucleus and the opening of the sieve-plate pores are closely related events that lead to incorporation of PSE zero into the phloem-unloading zone . At this point the cell becomes competent to unload solutes ( Figure 1C–E; Video 1 ) . 10 . 7554/eLife . 24125 . 004Video 1 . Visualization of the development of PSE zero . An Arabidopsis line with GFP tagged ER ( green ) in the protophloem sieve elements is loaded with Esculin ( blue ) . When the nucleus in the sieve element degrades , sieve plate pores open and the blue Esculin enters the cell . This integrates the cell into the unloading zone and defines a new PSE zero . DOI: http://dx . doi . org/10 . 7554/eLife . 24125 . 004 In order to acquire functional data on the dimensions of the unloading zone , we conducted flow velocity measurements in individual PSE files using fluorescence recovery after photobleaching ( FRAP; Froelich et al . , 2011 ) . Phloem flow velocities in the terminal region of the root ( basipetal to the unloading zone ) are in the range of 25 µm/s ( Froelich et al . , 2011 ) . In this study , we photobleached CF after it had arrived in the phloem unloading zone of the root . In a tube of constant diameter and impermeable walls , the flow velocity is constant . When the walls become leaky , the flow velocity decreases because of loss of fluid . In the case of the protophloem , the tube becomes leaky when unloading occurs . To define the size of the unloading zone , we measured flow velocity along the phloem files and found that deceleration started at about 300–400 µm behind PSE zero ( this dimension varied slightly , even within the two sieve tube files in the same root tip; Figure 2 ) . Our data revealed that the phloem of the root is subdivided into distinct structural and functional domains . Translocation into the main root occurs through the metaphloem . In the elongation zone of the root , the metaphloem overlaps with the mature protophloem file , at which point solutes are transferred laterally from the metaphloem to the protophloem ( ‘transfer zone’; Figure 2 ) . Movement then occurs through the protophloem towards the root tip ( ‘protophloem translocation zone’; Figure 2 ) and subsequently into the terminal functional PSEs ( ‘protophloem unloading zone’ ) , at which point solutes are distributed laterally into the root tip . Apical to this zone lies the ‘protophloem differentiation zone’ that , as described above , plays no role in phloem unloading ( Figure 2 ) . 10 . 7554/eLife . 24125 . 005Figure 2 . Schematic diagram illustrating the organization of phloem cells in specific zones in the root tip of Arabidopsis . The graph represents experimentally derived velocities at defined points relative to the terminal sieve element ( PSE zero ) in the protophloem unloading zone . Error bars show standard deviation of the mean ( n = 8 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24125 . 005 Arabidopsis roots grow at speeds of about 100–150 µm per hour ( Beemster and Baskin , 1998 ) . Consequently , after the sieve-plate pores open , a single PSE in the phloem unloading zone is active in unloading for only about 2–4 hr . Initially , cellular remnants such as the nuclear membrane , ribosomes and tonoplast are degraded and removed from the PSE . Subsequently , the PSEs become physically stretched in the elongation zone and finally move into an area that is active in translocation but inactive in unloading ( ‘protophloem translocation zone’; Figure 2 ) . The PSEs that progress basipetally from the phloem unloading zone into the protophloem translocation zone must undergo a rapid cellular transformation , particularly in the PD on their lateral walls . However , the structural alterations that control the cessation of unloading are not known . Callose is deposited in the neck region of plasmodesmata and restricts the size exclusion limit of the pore in response to various stimuli ( Luna et al . , 2011; Nakashima et al . , 2003; Radford et al . , 1998 ) . To investigate if callose was involved in occluding PD in the translocation zone , we stained roots of intact seedlings with Sirofluor ( Evans et al . , 1984; Vatén et al . , 2011 ) . This fluorochrome has a strong affinity for callose and demarcates sieve plates and PD in developing cell walls ( Stone et al . , 1984; Vatén et al . , 2011 ) ; Figure 3 ) . The two protophloem files could be visualized easily with this stain ( Figure 3A and B ) and individual lateral PD were identified in the phloem unloading zone ( Figure 3C and D ) . At the junction of the phloem unloading zone with the phloem translocation zone , callose deposits on the PSE wall increased in number ( Figure 3E ) until significant parts of the PSE walls were covered in callose ( Figure 3F ) . This deposition of callose along the lateral walls of PSEs in the protophloem translocation zone correlated with the reduction in flow we observed in dye-loading experiments and may provide a structural basis for the functional isolation of this zone . 10 . 7554/eLife . 24125 . 006Figure 3 . Confocal micrographs of the unloading zone in the Arabidopsis root tip stained with Sirofluor . Low magnification images showing the relatively strong fluorescence at sieve plates ( arrows ) . ( C–F ) Higher magnification images at the locations indicated by boxes in ( B ) . Individual plasmodesmata are resolved in the unloading zone ( C , D ) . In the translocation zone , large deposits of callose are abundant ( E , F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24125 . 006 As well as sugars and amino acids , a large number of macromolecules including nucleic acids , and proteins occur in various amounts in the translocation stream ( Fukumorita and Chino , 1982; Kehr , 2006; Turnbull and Lopez-Cobollo , 2013 ) . Many of these may access the phloem sap via companion cells ( CCs; Paultre et al . , 2016 ) . In addition , when new SEs develop in source regions of the plant , remnants of the nucleus , ribosomes , and tonoplast pass into the translocation stream . In both these scenarios , macromolecules must be removed from the terminal PSEs involved in unloading or they would impede the process of phloem unloading ( Paultre et al . , 2016 ) . While small solutes and GFP ( Stokes radius 2 . 82 nm; Terry et al . , 1995 ) are unloaded from the PSEs into surrounding cells ( Imlau et al . , 1999 ) , larger proteins unloaded from the PSE enter a unique post-phloem domain ( Stadler et al . , 2005; Paultre et al . , 2016 ) and cannot cross the interface between the pericycle and endodermis ( Paultre et al . , 2016 ) . We therefore sought to identify this domain as well as the routes taken by solutes and macromolecules during the unloading process . In Arabidopsis , the root protophloem file is surrounded by five distinct cell files ( Figure 4A and B ) . One immature metaphloem sieve element ( MSE ) , located towards the center of the root , is in contact with the PSE . Two CCs , one to the left and one to the right , share a common cell wall area with the PSE and MSE , respectively . The complex is capped by two phloem pole pericycle ( PPP ) cell files ( Figure 4A and B ) that share cell walls with both the PSE and CCs . Small probes of the dimensions of sugars and amino acids ( CFDA , esculin ) , and small proteins ( GFP ) move from cell to cell throughout the entire root tip without noticeable barriers ( Figure 1; Oparka et al . , 1994; Stadler et al . , 2005; Wright and Oparka , 1996 ) . However , translocated proteins of CC origin , are restricted to cells immediately adjacent to the mature protophloem file ( Stadler et al . , 2005; Paultre et al . , 2016 ) . 10 . 7554/eLife . 24125 . 007Figure 4 . Functional organization in the root unloading zone . ( A , B ) TEM images showing a cross section of an Arabidopsis root unloading zone . ( A ) An overview of the central cylinder with phloem pole pericycle cells ( PPP ) , endodermis ( EN ) , companion cells ( CC ) , metaphloem sieve element ( MSE ) , and protophloem sieve element ( PSE ) . ( B ) TEM image of the pentagonal organization of cells surrounding the protophloem file . ( C ) Confocal micrograph of a transgenic Arabidopsis line expressing SEOR-YFP protein ( yellow ) and GFP targeted to the sieve element ER ( green ) , both under control under a sieve element specific promoter . While GFP is restricted to the ER of the PSE ( solid arrow ) , SEOR-YFP expressed into the cytoplasm escapes into two neighboring cell files ( dashed arrow ) . ( D , E ) Root cross section of a fixed and embedded transgenic Arabidopsis plant expressing SEOR-YFP protein . The micrograph identifies the two cell files into which SEOR-YFP escapes as the PPP F , ( G ) Two confocal micrographs extracted from Video 2 showing SEOR-YFP protein ( yellow ) in the PPP and PSE . New PPP cells become fluorescent as unloading progresses . Note that small aggregates of SEOR-YFP become increasingly larger basipetal to the unloading zone . ( H ) Confocal micrograph of a transgenic Arabidopsis line expressing GFP ( green ) in the nuclei of companion cells ( solid arrows ) and SEOR-YFP . The nuclei in the CCs do not match the location of the nuclei in the cell files containing SEOR-YFP ( dashed arrows ) , providing further evidence that the two files are the PPP . ( I , J ) Root tip of a grafted Arabidopsis plant in which the rootstock was wildtype and the scion expressed SEOR-YFP in the shoot . The root was imaged at 10 days after grafting and shows clearly that SEOR-YFP protein has moved from shoot to root , with subsequent unloading into the PPP . Scale Bars; B = 1 µm; F , G = 5 µm; C , D , E , H = 10 µm; I , J = 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 24125 . 007 To monitor the fate of different phloem cargos , we compared the phloem-unloading pathway of small fluorescent probes ( Knoblauch et al . , 2015 ) with fluorescent fusion proteins ( Froelich et al . , 2011; Stadler et al . , 2005 ) , covering a molecular mass range of 340 Da to more than 100 kDa . Experiments were conducted in situ on living roots . As shown before , small probes were unloaded laterally towards the cortex and subsequently distributed amongst all root cells ( Figure 1 ) . Large probes of CC origin , however , were restricted to a specific domain ( Figures 4 and 5 ) . We wished to determine if proteins synthesized within SEs , rather than CCs , would show a similar pattern of unloading . Accordingly , we generated a transgenic Arabidopsis line expressing the Sieve Element Occlusion Related ( SEOR ) protein ( Froelich et al . , 2011; Pélissier et al . , 2008 ) fused to YFP ( SEOR-YFP ) . This protein is generated exclusively in young sieve elements and remains a structural component after maturation of the cell and its integration into the translocating sieve tube file . We crossed this line with a second reporter line in which HDEL-GFP was targeted to the ER ( Figure 1 ) . Both these fluorescent fusions were generated under the sieve-element specific promoters pSEOR; pMtSEO2 . The promoters are active in young sieve elements only , but the gene products remain as structural components in mature sieve tubes ( Pélissier et al . , 2008; Froelich et al . , 2011 ) . GFP demarcated clearly the ER of the PSEs ( Figures 1 and 4 ) . Unlike HDEL-GFP , the signal from the SEOR-YFP fusion ( 112 kDa ) , which is expressed in the cytoplasm of immature SEs , was not restricted to SEs but entered two specific cell files significantly larger than the adjacent PSEs , and did not travel outward beyond this domain ( Figure 4C ) . We used resin embedding of the phloem-unloading zone ( Bell et al . , 2013 ) to localize the YFP signal in transverse sections of the root . The YFP signal was confined to the two large PPP cells that abut the PSE files ( Figure 4D and E ) . Time-lapse movies of the phloem unloading zone supported this observation ( Figure 4F and G; Video 2 ) . Initially , SEOR-YFP was detected only in the immature PSEs as small aggregates . Subsequently , groups of new PPP cells became fluorescent adjacent to the PSEs , presumably when PD connections opened up between the two cell types ( Figure 4F and G ) . Using a fluorescent marker for CC nuclei ( Zhang et al . , 2008 ) we were unable to detect phloem unloading of SEOR-YFP into CCs , only into the PPP ( Figure 4H , I ) . This result was unexpected as most models of phloem unloading assume that the exit of macromolecules occurs through CCs ( Patrick , 1997; Thorne , 1985 ) . 10 . 7554/eLife . 24125 . 008Video 2 . YFP tagged SEOR-protein in the PSE and two neighboring PPP cell files . During root growth , new PPP cells are integrated into the unloading domain as indicated by tagged protein entering the cells , presumably due to the opening of connecting plasmodesmata . DOI: http://dx . doi . org/10 . 7554/eLife . 24125 . 00810 . 7554/eLife . 24125 . 009Figure 5 . Batch unloading of proteins . ( A–F ) six frames taken from Video 3 . A ) The unloading zone was photobleached ( boxed region ) . Refilling of the unloading zone shows that GFP exits the PSE in discrete batches ( arrows in B ) . Over time , all cells in the root transported GFP until an even distribution of the fluorescent protein was reinstated ( C–F ) . ( G–J ) Compared to GFP ( 27 kDa ) , aequorin-GFP ( 48 kDa ) was batch unloaded but did not move beyond the PPP . ( K–N ) Four frames extracted from Video 4 showing batch unloading of SEOR-YFP ( 112 kDa ) . In contrast to the CC-expressed GFP probes , SEOR-YFP was expressed in young sieve elements and entered the translocation stream when the sieve-plate pores opened . The immature PSEs are indicated ( dashed arrow ) and PPP cells are visible ( open arrow ) . When SEOR-YFP aggregates arrive in the phloem unloading zone , they are batch unloaded from the terminal PSEs ( solid arrows ) . As the root continues to extend , the aggregates enlarge and eventually disappear ( see also Figure 4F , G and Videos 2 and 4 ) , probably due to their breakdown in the older PPP cells . DOI: http://dx . doi . org/10 . 7554/eLife . 24125 . 009 It was possible that SEOR-YFP was delivered directly to the PPP from neighboring immature PSEs in the root , rather than by long-distance transport of the protein from the shoot . To test if the latter occurred , we grafted SEOR-YFP scions onto non-transgenic rootstocks . 10 days after grafting , the roots of the grafted plants were indistinguishable from those of the native SEOR-YFP line ( Figure 4I , J ) , demonstrating that SEOR-YFP protein , synthesized in SEs of the shoot , is unloaded from PSEs into the PPP . To monitor the unloading of large proteins in real time , we used the transgenic lines pAtSUC2-GFP ( 27 kDa ) , pAtSUC2-ubiquitin-gfp ( 36 kDa ) , pAtSUC2-aequorin-gfp ( 48 kDa ) , , and pAtSEOR-AtSEOR-yfp ( 112 kDa; described above ) and conducted fluorescence recovery after photobleaching ( FRAP ) experiments . Time-lapse movies revealed that unloading of large probes ( GFP 27 kDa and above ) did not occur at a constant rate , as observed with small solutes . Rather , these proteins were suddenly released into the adjoining PPP cells in distinct pulses ( Figure 5 , Video 3 ) . We refer to this phenomenon as ‘batch unloading’ , as the proteins are delivered in discrete pulses into individual cells . These cells then became highly fluorescent relative to their neighbors . Batch unloading did not occur at the same time into all cells , but independently into individual cells ( Figure 5; Video 3 ) . It appears that specific domains exist in the root tip that confer a size-dependent filtration of the macromolecules that arrive in the unloading zone . Small proteins such as GFP were batch unloaded and subsequently moved freely throughout the root ( Figure 5A–F ) . However , large probes such as aequorin-GFP and SEOR-YFP became trapped inside the PPP after batch unloading ( Figure 5G–J; Video 4 ) . 10 . 7554/eLife . 24125 . 010Video 3 . Time-lapse movie of batch unloading of free GFP ( 27 kDa ) . In situ time lapse movie of a transgenic line constantly supplying GFP into SEs via leakage from CCs where GFP is expressed under control of the SUC2 promoter . After photobleaching of GFP in the unloading zone , refilling reveals that GFP is unloading in batches into individual cells from where it diffuses into the post unloading zone . DOI: http://dx . doi . org/10 . 7554/eLife . 24125 . 01010 . 7554/eLife . 24125 . 011Video 4 . Batch unloading of YFP tagged SEOR protein ( 112 kDa ) . SEOR-YFP protein is expressed in young sieve elements and remains as aggregates in the sieve elements after degradation of the nucleus . The time-lapse movie shows batch unloading of this large protein into the PPP . DOI: http://dx . doi . org/10 . 7554/eLife . 24125 . 011 To further test the role of the PPP in phloem unloading we inhibited the lateral movement of solutes into this cell layer using an inducible callose synthase system ( Vatén et al . , 2011 ) . In this approach a modified , PD-specific , callose synthase ( icals3m ) is induced under estradiol treatment , inhibiting cell-cell movement into the induced cell layer of the root ( Vatén et al . , 2011 ) . In Arabidopsis there are twelve CalS proteins , with diverse roles in callose production ( Chen and Kim , 2009 ) . During the course of this study , we found that CalS8 is expressed specifically in the PPP ( Figure 6A , B ) , allowing us to generate an estradiol-inducible line expressing icals3m under the CalS8 promoter ( pCALS8::icals3m ) . This line was generated specifically to block the connection between PSEs and the PPP . After callose induction we observed a significant arrest in primary root growth relative to roots that were transferred to a non-inducing medium ( Figure 6—figure supplement 1 ) . Next we labelled the cotyledons of induced and non-induced pCALS8::icals3m roots with CF at 8 hr and 24 hr after callose induction . Control roots showed the characteristic unloading pattern of wild-type plants ( Figure 6C , n = 22 ) while induced plants showed a severe restriction in phloem unloading . At 8 hr induction , CF was restricted to PSE files with minimal lateral unloading into the root ( Figure 6D , n = 20 ) . At 24 hr induction , this effect was even stronger . CF did not enter the unloading zone and remained completely restricted to PSEs distal to the unloading zone ( Figure 6E , n = 9 ) . To determine the potential role played by CCs in phloem unloading , we used the sister of apple ( sAPL ) promoter to drive the production of the mutant CALS3 protein ( psAPL::icals3m ) . We chose this promoter as other CC-specific promoters ( e . g . SUC2 ) were expressed only weakly within the phloem-unloading zone ( data not shown ) . Within the phloem unloading zone the psAPL promoter was expressed strongly in CCs and MSE , but not in the PPP ( Figure 6F , G ) . As above , CF transport in these seedlings was monitored at 8 hr and 24 hr after estradiol induction . At both time points CF was unloaded without restriction , and induced and non-induced seedlings showed an identical pattern of unloading ( Figure 6H–J , n = 3 , n = 3 , n = 2 , respectively ) . Furthermore , we found that callose induction in the psAPL::icals3m seedlings did not significantly affect root growth relative to the uninduced control plants ( Figure 6—figure supplement 1 ) . To confirm that callose synthase was overexpressed in the predicted cell layers , roots were fixed , embedded and stained for callose after the CF transport assays . Control roots showed the characteristic callose labelling associated with PSE files ( Figure 6K; see also Figure 3 ) . Confocal imaging of pCALS8::icals3m seedlings confirmed that callose was deposited strongly in the PPP relative to other cells surrounding the PSE files ( Figure 6L–O ) , while psAPL::icals3m roots showed strong callose labelling in the CC files , as predicted ( Figure 6—figure supplement 6P–S ) . Collectively , these data provide strong evidence that phloem unloading occurs predominantly via the PPP , not CCs , and suggest that the PD between PSE and PPP are central players in the unloading process . 10 . 7554/eLife . 24125 . 012Figure 6 . Callose induction in the PPP , but not CCs , blocks phloem unloading . ( A ) pCALS8::ER-YFP is expressed exclusively in the PPP . ( B ) Transverse optical section of A . ( C ) CF unloading in a control root expressing pCALS8::icals3m transferred to non-inducing medium . Unloading progresses as in wild-type roots . ( D ) CF unloading is restricted to the PSE files in pCALS8::icals3m roots at 8 hr after callose induction in the PPP . ( E ) As D but at 24 hr post-callose induction in the PPP . ( F ) psAPL promoter expression ( psAPL-GFP ) is restricted to CCs and MSE . ( G ) Transverse optical section of F . ( H ) CF unloading in a control root expressing psAPL::icals3m transferred to non-inducing medium . ( I ) CF unloading is not restricted in psAPL::icals3m roots at 8 hr after callose induction in CCs . J ) As I but at 24 hr post-callose induction in CCs . ( K ) Sirofluor staining of a control root showing general background staining of PD around PSE files . L ) Sirofluor staining of a pCALS8::icals3m root at 8 hr after callose induction in the PPP . ( M ) Sirofluor staining of a pCALS8::icals3m root at 24 hr after callose induction . In both L and M the roots were stained immediately after CF transport . ( N ) Callose immunolabelling ( green ) of a pCALS8::icals3m root at 8 hr after callose induction in the PPP . Cell walls are labelled red . ( O ) As N but at 24 hr after callose induction . ( In addition to the PPP , sometimes callose staining is also observed in the CC ) . ( P ) Sirofluor staining of a psAPL::icals3m root at 8 hr after callose induction in CCs . ( Q ) As P but at 24 hr after callose induction . ( R ) Callose immunolabelling ( green ) of a psAPL::icals3m root at 8 hr after callose induction in CCs . ( In addition to the CC , callose staining was sometimes observed in the MSE ) . Cell walls were counterstained with calcofluor white ( labelled red ) . S ) As R but at 24 hr after callose induction . Scale bars: N , O , R , S: 5 um . A , B , F , G , K , L , M , P , Q: 10 um . C , D , H , I , J: 50 um . E , J: 100 um . Abbreviations as in Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 24125 . 01210 . 7554/eLife . 24125 . 013Figure 6—figure supplement 1 . Growth of pCALS8::icals3m and psAPL::icals3m seedlings on 5 uM beta estradiol relative to uninduced controls ( mock DMSO ) . Seedlings were transferred to inducing or non-inducing media at 4 days post germination . For pCALS8::icals3m each time point is the average of 82 independent measurements . For psAPL::icals3m each time point is the average of 67 independent measurements . Bars show standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 24125 . 013 In order to understand the structural basis for the differential filtration of molecules in the unloading zone , we investigated the ultrastructure of PD at all cell interfaces between the protophloem file and its five interconnecting cells ( See Figure 4A and B ) . We found different PD types in the unloading zone ( Figure 7A ) . PD between the PSE and MP were simple , or occasionally branched ( Figure 7B ) . Between PSEs and CCs , PD displayed the typical single pore on the PSE side and multiple branches towards the CC ( Figure 7C ) , as shown often for connections between SEs and CCs in other tissues ( Esau and Thorsch , 1985; Oparka and Turgeon , 1999 ) . The structure of most of the PD between PSE and PPP , however , was quite different and showed an architecture not previously described . In addition to a few simple PD ( 10%; n = 20 ) , funnel-shaped PD with apertures of up to 300 nm diameter were found at the PSE entrance , tapering towards the PPP entrance ( Figure 7D–H ) . 10 . 7554/eLife . 24125 . 014Figure 7 . Types of plasmodesmata connecting different cell interfaces . ( A ) Schematic diagrams of the different plasmodesmata connecting protophloem sieve elements to surrounding cell types . ( B ) Image of the PSE-MSE interface showing a cell wall with two simple plasmodesmata . ( C ) A pore-plasmodesma in the cell wall between PSE and CC . ( D–I ) Plasmodesmata connecting PSE with PPP . ( D ) Simple plasmodesmata , found rarely . ( E–H ) Funnel plasmodesmata . These showed a wide opening on the PSE entrance tapering towards the PPP . ( H ) Electron-dense components ( white arrow ) of unknown composition were often observed within funnel plasmodesmata ( black arrows ) . DT = desmotubule , CW = cell wall , P = pore , F = funnel , CS = cytoplasmic sleeve . Scale bars: B , C , G , H = 200 nm; D , E , F = 500 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 24125 . 014 In order to extract the number of PD available for unloading , we conducted serial block-face scanning electron microscopy followed by 3D reconstruction of the interface between PSEs and the adjacent cells ( Denk and Horstmann , 2004; Furuta et al . , 2014; Video 5 ) . Using this method , we collected 2100 serial transverse sections that spanned six complete SEs ( Figure 8A ) , including the junction between PSE zero and the adjacent immature PSE ( Figure 8B ) . During reconstruction , we attributed the exact position of each PD to the different wall interfaces shared by the PSE , color-coded to ease identification ( Video 5 ) . Using 3D reconstruction , we could image PD on both the outer ( Figure 8C ) and inner ( Figure 8D ) walls of the PSE files . A movie showing a ‘fly through’ of the interior of the PSE file is shown in Video 5 . PD relative distribution was 45 . 3% , 40 . 8% , and 13 . 9% between PSE-PPPs , PSE-CCs , and PSE-MSE , respectively , when corrected for the total interface between the cell types . Thus , PD leading from the PSE to PPP and CCs are equally abundant while those to the MSE are significantly lower . Next , we acquired data for the total numbers of PD along the entire phloem-unloading zone . We found that the total number averaged 527 ± 58 ( n = 4 ) per protophloem unloading domain ( since Arabidopsis roots are diarch , twice the number is available for unloading within the entire root tip ) . Based on the percentage distributions of PD obtained using serial sectioning , we calculated that in the phloem-unloading zone of the root there are approximately 240 plasmodesmata at the PSE-PPP interface , 215 plasmodesmata at the PSE-CC interface , and 73 plasmodesmata at the PSE-MP interface potentially available for unloading . 10 . 7554/eLife . 24125 . 015Video 5 . 3D volume reconstruction of serial block face data . The movie shows the volume reconstruction of the PSE file with highlighted cell walls . Color coding reveals the location of plasmodesmata connecting the PSE to the neighboring cell types . DOI: http://dx . doi . org/10 . 7554/eLife . 24125 . 01510 . 7554/eLife . 24125 . 016Figure 8 . 3D overview of protophloem sieve elements in the root tip obtained by serial block-face scanning electron microscopy . ( A ) Cross-section of one phloem pole in the unloading zone . PD are indicated by darts . ( B ) Longitudinal section of the protophloem unloading zone . PSE zero ( X ) is connected to a neighboring immature protophloem sieve element . ( C , D ) 3D longitudinal view of the protophloem unloading zone . Serial sections were used to reconstruct the unloading zone and quantify PD connections from PSEs to adjacent cells . ( C ) shows PD on the outer face of the PSE . ( D ) is derived from Video 4 and shows the PD on the inner faces of the PSE . In the images , PD are color coded ( blue/cyan PSE-PPP , red/green PSE-CC , and yellow PSE-MSE ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24125 . 016 Current models of phloem unloading , such as the high-pressure-manifold model ( Fisher , 2000; Patrick , 2013 ) assume high pressure differentials between PSEs and the unloading zone in order to drive fluids and solutes through the PD into the neighboring cells . Recently , direct sieve tube pressure , viscosity , flow velocity , and tube geometry measurements in morning glory showed , however , that the majority of energy provided by the pressure differential between source and sink is consumed to drive flow , and that a high pressure differential in the root system is unlikely to exist , calling the current model into question ( Knoblauch et al . , 2016 ) . In principle , there are two possibilities of how the phloem sap might escape the PSEs . Bulk flow could move the entire volume of solvent and solutes through the PD into neighboring cells . Alternatively , the solutes could diffuse through PD while the solvent ( water ) could be removed from PSEs via membrane leakage , potentially via aquaporins ( Doering-Saad et al . , 2002 ) . Finally , a combination of diffusion and bulk flow could account for the observed transport . In order to evaluate the feasibility of the different routes , we gathered the necessary parameters to model flow in the phloem-unloading zone ( Table 1 ) . We measured the average phloem flow velocity in the protophloem translocation zone to be u = 22 . 6 ± 5 . 1 µm/s ( n = 11 ) and the average sieve tube diameter to be: d = 3 . 6 ± 0 . 44 µm ( n = 11 ) resulting in a volume flow rate of Q=π4d2u= 230 femtoliter/s ( 230 µm3/s ) to be unloaded from a single PSE file . Literature values on phloem sucrose concentration average about 500 mM ( Hall and Baker , 1972; Kallarackal et al . , 1989 ) ; Mendoza-Cózatl et al . , 2008; Winter et al . , 1992 ) , hence the total amount of sugar to be unloaded is approximately I=Qc=1 . 2×10−13 mol/s . The solutes must leave the phloem through the PD , while some of the solvent ( water ) may be removed via membrane leakage . To facilitate transport , an average of 240 PD are available at the PSE/PPP interface , while the water may leak anywhere along the 350 µm-length of the unloading zone . For simple PD we assumed a cylinder of constant dimensions with a desmotubule of 15 nm diameter and a cytoplasmic sleeve width of 2 . 8 nm , corresponding to the hydrodynamic radius of GFP . Based on our observations that GFP is batch unloaded , but diffuses relatively freely throughout the root , this molecule appears to be slightly larger than the size exclusion limit imposed by the neck region of funnel PD . For funnel PD we assumed the same dimensions on the PPP side but an average funnel opening of 150 nm ( Figure 7 ) towards the PSE . The location of the desmotubule in funnel PDs impacts the resistance which led us to assume two extremes which are discussed in detail in the appendix . The cell wall thickness ( length of the simple and funnel PDs ) was taken to be 500 nm . 10 . 7554/eLife . 24125 . 017Table 1 . Base parameters used to model phloem unloading . DOI: http://dx . doi . org/10 . 7554/eLife . 24125 . 017Assuming transport through simple PD at PSE/PPP interfaceTransport through PD at PSE/PPP interfaceLength of unloading zone350 µm350 µmDesmotubule Diameter15 nm15 nmCytoplasmic Sleeve Diameter2 . 8 nm2 . 8 nmCell Wall Thickness500 nm500 nmPhloem Sap Osmotic Potential500 mM500 mMFunnel opening towards PSE–150 nm# of PD available for Unloading240 simple PD24 simple PD , 216 funnel PDTotal Sap Volume230 fl/s230 fl/sRequired Pressure Differential8 . 14 MPa0 . 05–0 . 2 MPa To elucidate the role of funnel PD in unloading , we first considered the conditions necessary to facilitate unloading by simple PD . We found that the required pressure differential to drive unloading solely by bulk flow through simple PD would be 8 . 14 MPa ( see Appendix for detailed calculations ) . This pressure differential has neither been measured in SEs , nor can it be considered as feasible . However , assuming funnel PD instead of simple PD , the required pressure would be as low as 0 . 05–0 . 2 MPa . Such a relatively low pressure differential would be facile to maintain between plant cells . Because the viscous resistance in wide pores is greatly reduced ( Appendix ) , funnel PD are much more efficient in unloading compared to simple PD . To estimate the solute concentration difference that would be required to account for unloading by pure diffusion , we assumed a length of the unloading zone of 350 µm . For the geometrical and physiological values listed above , a concentration difference of 276 mM would be required for diffusive unloading through funnel PDs . In this case , however , a second route for unloading of the solvent ( water ) would be necessary . The permeability of the plasma membrane varies significantly between cell types and values of 10−14 to 10−12 m/s/Pa have been reported ( Kramer and Boyer , 1995a ) . Considering the parameters outlined above , a pressure of 0 . 059 to 5 . 9 MPa would be needed to remove the solvent from the PSE . Thus , at a high membrane permeability removal of the solvent over the membrane would be feasible . Our analysis of unloading kinetics leads us to conclude that the distinctive funnel shape of the SE-to-PPP PD is crucial to enabling efficient unloading . In this case , convective phloem unloading , i . e . a combination of diffusion and bulk flow , is feasible , and that relatively moderate pressure and concentration differentials are necessary to drive transport . In our calculations ( Appendix ) we have assumed that PD connecting PSEs with CCs and MSEs play little or no direct role in the unloading process , consistent with our results using the icals3m system ( Figure 6 ) . In this scenario , the unique architecture of funnel PD could still accommodate the calculated unloading rates ( Appendix ) Based on our structural and physiological data bulk flow is likely to be the dominant mechanism of unloading . Bulk flow would require a pressure differential of about 0 . 05–0 . 2 MPa , equal to an osmotic potential difference of about 20–80 mM . Solute unloading by diffusion alone , on the other hand , would require a concentration difference of 276 mM . While bulk flow predominates , the above mechanisms are not exclusive , and both may contribute to unloading . Bulk flow requires a pressure differential generated by osmosis which would lead to a concentration difference between PSEs and PPPs . This would induce diffusion , in parallel with bulk flow , even if its contribution is lower . The term for the combination of diffusion and bulk flow is convection , which leads us to propose a new model of Convective Phloem Unloading . We have identified different domains within the root tip that provide a size dependent filtration of molecules . Cross sections of Arabidopsis roots ( Figure 9A ) show the typical diarch structure . The pentagonal architecture of the cell complex surrounding the PSEs and the different PD connecting the cell types is depicted in Figure 9B and C . Our results indicate that the PSE-PPP interface is the principal route for all solutes to be unloaded . Diffusion will , however , lead to a relatively quick redistribution of the solutes through simple PD connecting the cells within the post-phloem unloading zone , along a solute concentration gradient ( Figure 9D , E ) . This gradient will also direct solutes to areas of highest demand , while higher consumption will lead to steeper gradients . 10 . 7554/eLife . 24125 . 018Figure 9 . Cross sectional overviews of the Arabidopsis roots showing PD connections and size-dependent phloem unloading of solutes and macromolecules . ( A ) Standard light micrograph showing a cross section of the Arabidopsis root . ( B ) A false colored cross section of the Arabidopsis root highlighting the two phloem poles in the unloading zone . ( C ) Diagram of the cells in the phloem pole and the types of PD that connect the PSEs to each adjacent cell . PPP cells are connected to the PSE by funnel PD , CCs are connected by pore-PD , and MSEs are connected by simple PD . ( D and E ) Diagram showing the location of various solutes and macromolecules depending on molecular mass . Once unloaded via the PPP , sucrose ( blue dots ) and GFP ( green dots ) are able subsequently to enter all cell types via PD . However , larger macromolecules such as SEOR-YFP ( yellow/orange dots ) are unloaded only into PPP cells . EN = endodermis . DOI: http://dx . doi . org/10 . 7554/eLife . 24125 . 018 Unloading of small proteins such as GFP ( 27 kDa ) occurs by batch unloading through funnel PD into PPPs . Once unloaded , GFP enters the post unloading zone and is evenly distributed throughout the root tip ( Stadler et al . , 2005 ) . Large proteins of the size of aequorin-GFP ( 48 kDa ) or SEOR-YFP ( 112 kDa ) , however , remain trapped within PPPs and cannot enter the post-phloem pathway to the meristem . Clearly , the different types of PD between PSEs and neighboring cells have major functional impacts on molecular flow in the root tip . Many proteins have been found in extracted phloem sap , most showing no obvious function in long-distance signaling ( Turnbull and Lopez-Cobollo , 2013; Batailler et al . , 2012; Paultre , et al . , 2016 ) . Thus , the question arises as to how many of these molecules are components of the phloem sap by default rather than by design ( Paultre , et al . , 2016 ) . To address this point we need to consider where such macromolecules originate . Many systemic macromolecules are thought to arise in CCs and pass into SEs through the pore-PD that connect SEs with CCs ( Fisher et al . , 1992; Lucas et al . , 1996; Sjolund , 1997; Oparka and Turgeon , 1999 ) . There appears to be selectivity to this movement , large proteins ( above 70 kDa ) are restricted from entering the translocation stream while those below this cutoff are not ( Paultre , et al . , 2016 ) . Interestingly , in our current study SEOR-YFP ( 112 kDa ) , which is translated in immature SEs rather than CCs , was able to unload into the PPP , suggesting that the size exclusion limit that regulates macromolecular exchange between PPP and SE in sinks may be larger than between SEs and CCs in source phloem tissues . Another potential origin of macromolecules in phloem sap is young developing sieve elements . As shown in Figure 1 , PSEs start to conduct when their nucleus degenerates . At this point the cytoplasmic content in the PSEs is still very dense . After integration into the mature PSE file , there is no other route left for degraded structures but phloem unloading , most likely into the PPP . Unlike the xylem pole pericycle ( XPP ) , which is involved in lateral-root formation ( Parizot et al . , 2012 ) and apoplastic xylem loading ( Takano et al . , 2002 ) , a specific role for the PPP has not been proposed , although its transcriptome shares similarity to the underlying protophloem ( Parizot et al . , 2012 ) . We have now identified the ‘post-phloem domain’ previously described by Stadler et al . ( 2005 ) and Paultre , et al . , 2016 as the PPP . It appears that while the XPP may be involved in apoplastic xylem loading , the PPP is intimately involved in symplastic phloem unloading , and may function as a repository for degraded PSE contents as well as systemic macromolecules . Although PD connect all cell types surrounding the PSEs , our data suggest that exit via the PPP is the major route of unloading in Arabidopsis roots . In mature source tissues , as well as in secondary sieve elements produced by the cambium , macromolecules will arrive in the phloem-unloading zone . This was demonstrated when we grafted SEOR-YFP scions onto wild-type rootstocks ( Figure 4I , J ) . Without continuous macromolecule unloading , PSEs would fill up rapidly and phloem unloading would be impaired ( Paultre , et al . , 2016 ) . PPPs are the major recipients of systemic macromolecules and it might be expected that they are specialized for the degradation of proteins and RNAs , an interesting area for future research . PPPs may also protect the downstream cells from receiving unwanted , potentially interfering mRNA molecules . How does the root distinguish between true signaling molecules and those moving by default in the phloem ? One possibility is that unloaded macromolecules destined to traffic further than the PPP may have specific sequences that interact with the PD that connect the PPP to the endodermis and beyond . Phloem-mobile viruses are able to move beyond the PPP suggesting that they can break the ‘barrier’ normally imposed by the PPP ( Valentine et al . , 2004 ) . A clear future challenge will be to identify systemic macromolecules that can move beyond the PPP , and to analyze these macromolecules for specific motifs that , like transcription factors ( Xu et al . , 2011 ) , allow them to interact with and modify PD . Arabidopsis thaliana were grown in microROCs ( Advanced Science Tools LLC . Pullman , WA ) as described elsewhere ( Froelich et al . , 2011 ) . Soil was kept saturated by placing the microROCs in 1–2 inches of water daily for 5 min creating a soil water saturation of >80% . Chamber conditions were 16 hr photoperiods with 100–200 μE m−2 s−1 at 18°C to 22°C . Experiments were conducted on a Leica SP8 confocal microscope equipped with a supercontinuum laser ( 470–630 nm ) and a pulsed 405 nm diode laser . 5 ( 6 ) -Carboxyfluorescein diacetate ( CFDA ) stock solution , 5 mg/ml in acetone , was diluted 1:10 ( v/v ) in ddH2O and loaded into the phloem through application to cut cotyledons at 7–10 days post-germination as described earlier ( Knoblauch et al . , 2015 ) . Sieve elements in the most apical root protophloem loaded with CFDA were photo-bleached with 480 nm , 488 nm , and 496 nm lasers concurrently at maximum power and 6 . 5x zoom with a 20x lens . Bleaching started apically and moved toward the hypocotyl until the entirety of CFDA in the sieve tube in the field of view was bleached . Subsequent recording of the refilling occurred immediately following photo-bleaching at 200 Hz and 0 . 75x digital zoom by excitation with the 488 nm laser line at 15% of continuous power . Emission was collected between 505 and 545 nm . Seeds of Arabidopsis wt and the transgenic line pAtSEOR-AtSEOR-yfp were sterilized with 8% bleach and 1% TWEEN-20 . After 5x wash in distilled water seeds were plated on MS with 1 . 2% agar and 0 . 2% sucrose , pH 5 . 7 . After 48 hr stratification in the dark at 4°C seeds were oriented vertically at 23°C with 18 hr photoperiod . After 5–7 days , seedlings were grafted following the hypocotyl-grafting procedure of Turnbull et al . ( 2002 ) consisting of a transverse cut and butt alignment with silicon collars . The seedlings were cut transversely in the upper region of the hypocotyl with ultrafine microknives ( Interfocus , n°10315–12 ) . Scions were grafted onto wild type stocks using a short silicon collar for support on MS agar plates . The grafts were left to grow under LD with the plates still oriented vertically until new lateral roots of the stocks were fully established ( ~10 days ) . The grafts were imaged between 5 and 10 dag . Ten days old seedlings of transgenic Arabidopsis thaliana , expressing AtSEO::SEO-YFP , were embedded in LR White following the method described by Bell et al . ( 2011 ) . Tissue samples were fixed overnight at 4°C in a solution of 4% formaldehyde , 1% glutaraldehyde , 50 mM 1 , 4-piperazinediethanesulofonic acid ( PIPES ) and 1 mM CaCl2 . The samples were then washed in buffer ( 50 mM PIPES , 1 mM CaCl2 ) three times for 10 min before dehydration in a graded ethanol series ( 50% , 70% , 2 × 90% ) . The tissue samples were then infiltrated at 4°C in medium grade LR White at 1:1 , 1:2 , 1:3 , 1:9 ratios of 90% ethanol:resin for 45 min each before two 60 min changes in 100% LR White . The final embedding step was done at ambient temperature . The samples were then polymerised in gelatin capsules ( TAAB ) at 50°C for 24 hr . Esculin and Sirofluor were excited with a pulsed 405 nm diode laser . Emission was collected with a hybrid detector between 420 and 480 nm or 420 and 600 nm respectively . GFP and esculin were imaged with sequential scan for fluorescence emission separation . For GFP , excitation was 488 nm and emission detection with a hybrid detector at 495–535 nm . Esculin excitation was 405 nm and emission collection with a hybrid detector at 420–470 nm . GFP and YFP were sequentially scanned for separation of emission fluorescence . GFP excitation was 484 nm at 50% continuous power . Emission collected with a hybrid detector at 489–505 nm . YFP excitation was 514 nm at 15% continuous power . Emission collected at 519–564 nm with a hybrid detector . Samples were chemically fixed in the microwave with 2% glutaraldehye , 2% paraformaldehyde and 2% DMSO in 0 . 1 M cacodylate buffer . Microwave fixation time was 2 min , followed by a 2 min break and a final 2 min microwave period at maximum temperature of 25°C . Samples were then washed 5x for 10 min in 0 . 1 M cacodylate buffer . Postfixation took place in 1% osmium tetroxide ( 0 . 1 M cacodylate ) overnight at 4°C . Samples were rinsed 3x for 10 min . in 0 . 1 M cacodylate buffer . Dehydration was carried out with methanol in the microwave for one minute followed by 5 min at room temperature for each step in 5% increments from 5% to 100% methanol . 100% methanol was replaced twice and the samples were then transferred to 100% propylene oxide with three exchanges . Samples were embedded in Spurr resin at the following propylene oxide:Spurr ratios: 2:1 for 2 hr , 1:1 for 2 hr , 1:2 for 2 hr , and 100% Spurr overnight 3X . The resin was cured at 65°C for 24 hr . Samples were sectioned and stained with 1% uranyl acetate for 6 min followed by a ddH2O rinse and stained in Reynold’s lead for 6 min . Following Reynold’s lead samples were quickly rinsed in 0 . 1 N NaOH and ddH2O . Samples were imaged with an FEI T20 at 200kv . Material for SBFSM was prepared as described by Furuta et al . ( 2014 ) . Wild-type Columbia roots were fixed with 2 . 5% glutaraldehyde , 2% formaldehyde in 0 . 1 M Na-Cacodylate buffer ( pH 7 . 4 ) supplemented with 2 mM CaCl2 for 2–3 hr at room temperature and embedded in Durcupan ACM resin ( Fluka , Sigma-Aldrich ) . After standard dehydration steps , samples were embedded in silicone holders filled with 100% Durcupan and infiltrated for at least 2 hr before polymerization at 60°C . The roots were trimmed to the desired starting point from the tip using an EM Ultracut UC6i ultramicrotome ( Leica Mikrosysteme GmbH ) . Images were acquired with a FEG-SEM Quanta 250 ( FEI , Hillsboro , OR ) , using a backscattered electron detector ( Gatan Inc . ) . The block faces were sectioned at 40 nm increments . The images were initially processed and segmented using Microscopy Image Browser , a self-developed program written under Matlab environment ( Mathworks Inc . ) and available at http://mib . helsinki . fi/ . A SBFSEM dataset of 2100 images was subsequently segmented and visualized in 3D using Amira 6 . 0 software ( FEI Corp ) . The outer structure cell walls were discriminated in a hybrid method of interpolated interactive masking and grey-level segmentation . The individual plasmodesmata were identified and segmented sequentially by a manual interactive method and color-coded into groups that related to the particular cell wall interface along the protophloem files . The promoters of CALS8 ( At3g14570 ) and sAPL ( At3g12730 ) were amplified using the primers CALS8-AscI-F AAGGCGCGCCCGGCAACATGAAATACGGGA and CALS8-XhoI-R ACAGCTCGAGGTTTTGGGAGAAAATCAATCAGAA , and SAPL-AscI-F AAGGCGCGCCAGCTAATAAGAAAGGGAGATCTCTG and SAPL-XhoI-R ACAGCTCGAGTTAACTAACAAAGTACTAAATGCCGA , respectively ) and cloned into P4P1RpGEMt containing the estrogen receptor XVE ( Zuo et al . 2000 ) . Using the MultiSite Gateway system ( Invitrogen ) , the inducible promoters were combined with the icals3m construct and the nopaline synthase terminator in destination vector pBm43GW . Arabidopsis Col-0 plants were dipped with the different constructs and positive transformants were selected using Basta ( pCALS8 ) and Hygromycin ( psAPL ) . Lines with single insertions were selected in T2 and homozygous plants were obtained in T3 . pCALS8::icals3m and psAPL::icals3m seeds were stratified for 2 days at 4°C in the dark and were sown on media containing 0 . 5 MS and 1 . 2% plant agar , pH 5 . 8 . At 4 days post sowing , seedlings were induced in plates containing the same media plus 5 µM Beta estradiol or the corresponding amount of DMSO ( mock ) . Plates were scanned at times 0 , 3 , 8 and 24 hr after the induction . Roots were measured using ImageJ and seedlings were subsequently used for CFDA loading . 1 mM CFDA in a solution of 72% acetone in water was mixed with 0 . 5% Adigor solution ( Syngenta ) to 10:1 proportions . 0 . 2 µL of this solution was applied to the adaxial side of the leaves , separating the leaves from the media to avoid undesired dye spreading . Approximately 10 roots were used for each treatment . 20 min after loading , roots were mounted in propidium iodide and imaged by confocal laser scanning microscopy ( Zeiss LSM700 ) . Seedlings used for phloem unloading analysis subsequently underwent callose detection experiments through Sirofluor staining and callose immunolocalization allowing three independent experiments on the same biological material . Callose accumulation on the CFDA positive roots used for unloading experiments was confirmed first by Sirofluor staining . Whole seedlings were incubated 2 hr at room temperature in a 25 µg/mL ( 42 . 66 µM ) solution of Sirofluor fluorochrome ( Biosupplies , AU ) in a 50 mM K2HPO4 buffer under vacuum ( 60 mPa ) . Roots were then dissected and mounted in a 1:1 solution of AF1 antifadent ( Citifluor ) and 67 mM K2HPO4 buffer , and subsequently imaged by confocal laser scanning microscopy ( Zeiss LSM700 ) . Following Sirofluor staining , roots were fixed for subsequent immunolocalization on cross sections in a fixative solution containing 4% formaldehyde ( freshly prepared from paraformaldehyde powder , Sigma ) and 0 . 5% glutaraldehyde ( Sigma ) in a 0 . 1 M phosphate buffer pH7 . Fixation , dehydration and resin infiltration steps were done in a micro-wave using a PELCO BioWave Pro ( Ted Pella , Redding , CA ) . Fixation was achieved at 150 W , under vacuum ( 20 Hg ) and for 6 min ( 2’ . Followed by a 2’break and a final 2’ microwave ) . Samples were left in the fixative overnight at 4°C and then washed 3 times 3 min ( 20 Hg , MW 150 W 1’ on – 1’ break – 1’ on ) . Roots were then aligned and embedded in 1% low-melting Agarose ( Calbiochem ) and processed through increasing dehydration steps ( 25% , 50% , 70% , 90% , 96% , 3 × 100% Ethanol , vacuum 20 Hg , MW 150 W 1’ on followed by 5’ break ) . Resin infiltration ( LR White medium grade , Agar scientific ) was achieved through increasing resin concentrations: 33% resin in ethanol 100% , 66% resin in ethanol 100% , and 3 times 100% resin ( 20 Hg , MW 200 W 2’ on – 2’ break – 2’ on ) . Polymerization was conducted overnight at 60°C . Semi-thin sections ( 0 . 5 µm ) were taken with a Leica EM UC7 ultramicrotome . Callose immunolocalization on semi-thin sections was micro-wave assisted and performed as follows: blocking step ( BSA 2% in PBS , 1 mL per slide; primary antibody ( anti ( 1→3 ) -β-glucan ( Biosupplies , AU ) , 1/200 in BSA 2% in PBS , 500 µL per slide , MW 170 W 2’ on – 2’ break – 2’ on ) ; three washes in BSA 2% in PBS; secondary antibody ( Alexa Fluor 488 anti-mouse IGG ThermoFisher Scientific , A-11017 , 1/400 in BSA 2% in PBS ( MW 170 W , 1’ ) , 500 µL per slide; three washes in BSA 2% in PBS . Slides were finally mounted in a 1:1 solution of AF1 antifadent ( Citifluor ) with PBS , containing calcofluor as a cell wall counterstain , and imaged by confocal laser scanning microscopy ( Zeiss LSM700 ) .	A mechanism called photosynthesis allows plants to use energy from sunlight to make sugars from carbon dioxide gas and water . These sugars can then be used as fuel , or as building blocks for wood and other plant structures . Every part of the plant requires sugars , but most photosynthesis happens in the leaves and stems , so the sugars need to be able to move around the plant to wherever they are needed . Phloem tubes form a network that transports sugar , proteins and other molecules around the plant within a fluid known as sap . Because this network is so extensive , it is very difficult to study , which has left researchers with major questions about how it works . For example , it is not clear how the sugar and other molecules leave the phloem when they reach their destination . Ross-Elliot et al . used a combination of microscopy and mathematical modeling to investigate how sugars and other molecules leave the phloem in the roots of a plant called Arabidopsis thaliana . The experiments show that these molecules move directly into cells within a neighboring tissue called the phloem-pole pericycle via pores known as funnel plasmodesmata . Ross-Elliot et al . incorporated the experimental data into a mathematical model of phloem unloading . This model suggests that sugars and other small molecules move freely through the funnel plasmodesmata , but large proteins pass through these pores in pulses . Future challenges include finding out exactly how plants control phloem unloading and to investigate whether it is possible to modify the delivery of specific molecules to particular parts of the plant .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "plant", "biology" ]	2017	Phloem unloading in Arabidopsis roots is convective and regulated by the phloem-pole pericycle
KATP channels are metabolic sensors that couple cell energetics to membrane excitability . In pancreatic β-cells , channels formed by SUR1 and Kir6 . 2 regulate insulin secretion and are the targets of antidiabetic sulfonylureas . Here , we used cryo-EM to elucidate structural basis of channel assembly and gating . The structure , determined in the presence of ATP and the sulfonylurea glibenclamide , at ~6 Å resolution reveals a closed Kir6 . 2 tetrameric core with four peripheral SUR1s each anchored to a Kir6 . 2 by its N-terminal transmembrane domain ( TMD0 ) . Intricate interactions between TMD0 , the loop following TMD0 , and Kir6 . 2 near the proposed PIP2 binding site , and where ATP density is observed , suggest SUR1 may contribute to ATP and PIP2 binding to enhance Kir6 . 2 sensitivity to both . The SUR1-ABC core is found in an unusual inward-facing conformation whereby the two nucleotide binding domains are misaligned along a two-fold symmetry axis , revealing a possible mechanism by which glibenclamide inhibits channel activity . Studies into the electric mechanisms of insulin release of the pancreatic β-cell in the early 1980s led to the discovery and identification of an ATP-sensitive potassium ( KATP ) channel as the key molecular link between glucose metabolism and insulin secretion ( Ashcroft and Rorsman , 1990; Cook and Bryan , 1998 ) . Subsequent cloning and characterization revealed the β-cell KATP channel as a complex of two proteins: a potassium channel Kir6 . 2 of the inwardly rectifying K+ channel family , and a sulfonylurea receptor SUR1 , a member of the ATP binding cassette ( ABC ) transporter protein family ( Inagaki et al . , 1995 ) . Physiological activity of KATP channels is determined primarily by the relative concentrations of ATP and ADP: ATP inhibits , whereas MgADP stimulates channel activity ( Nichols , 2006 ) . As KATP channels set the β-cell membrane potential , this regulation by nucleotides endows them the ability to sense metabolic changes and translate those into changes in membrane excitability , which ultimately initiates or stops insulin secretion ( Ashcroft , 2005 ) . Another key player for KATP function is membrane phosphatidylinositol-4 , 5-bisphosphate ( PIP2 ) ; as in all other Kir family members , PIP2 is required for channel opening and sets the intrinsic open probability ( Po ) of the channel ( Hibino et al . , 2010; Nichols , 2006 ) . Mutations disrupting channel assembly or the above-gating properties result in insulin secretion disorders , with loss- or gain-of-function mutations causing congenital hyperinsulinism ( HI ) or permanent neonatal diabetes mellitus ( PNDM ) , respectively ( Ashcroft , 2005 ) . Importantly , KATP channels are the targets of sulfonylureas , one of the most commonly prescribed treatments for type 2 diabetes , which stimulate insulin secretion by inhibiting channel activity ( Gribble and Reimann , 2003 ) . In particular , glibenclamide ( GBC ) binds the channel with nanomolar affinity and was instrumental for the purification and cloning of SUR1 ( Aguilar-Bryan et al . , 1995 ) . A member of the Kir channel family , Kir6 . 2 consists of two transmembrane helices and N- and C-terminal cytoplasmic domains ( Hibino et al . , 2010 ) . By comparison , SUR1 , a member of the ABC transporter family , is much larger in size . In addition to a characteristic ABC core structure comprising two transmembrane domains ( TMD1 and 2 ) and two cytoplasmic nucleotide binding domains ( NBD1 and 2 ) , it has an N-terminal extension that contains a transmembrane domain ( TMD0 ) followed by a long , cytoplasmic loop ‘L0’ which connects to the ABC core ( Aguilar-Bryan et al . , 1995; Tusnády et al . , 2006 ) . Kir6 . 2 and SUR1 are uniquely dependent on each other for expression and function ( Inagaki et al . , 1995 ) . Interestingly , unlike most ABC transporters such as the cystic fibrosis transmembrane conductance regulator ( CFTR ) and the multidrug-resistant protein P-glycoprotein , SUR1 itself has no known ion channel or transporter activity; instead , its function is to regulate Kir6 . 2 channels ( Aguilar-Bryan et al . , 1995; Inagaki et al . , 1995; Wilkens , 2015 ) . A central question is how the two proteins assemble and function as a complex to sense metabolic signals . Biochemical and biophysical studies have indicated that the KATP channel is an octamer of four Kir6 . 2 and four SUR1 subunits . ATP and PIP2 bind Kir6 . 2 directly to close or open the channel , respectively ( Baukrowitz et al . , 1998; Shyng and Nichols , 1998; Tanabe et al . , 1999; Tucker et al . , 1997 ) . Although Kir6 . 2 alone can be gated by ATP and PIP2 , its sensitivities to both ATP and PIP2 are increased by SUR1 by ~10-fold ( Baukrowitz et al . , 1998; Enkvetchakul et al . , 2000; Shyng and Nichols , 1998; Tucker et al . , 1997 ) . How SUR1 sensitizes Kir6 . 2 to ATP inhibition and PIP2 stimulation remains unclear . In contrast to ATP inhibition of the channel , which does not depend on Mg2+ and ATP hydrolysis , nucleotide stimulation of the channel is conferred by SUR1 and requires Mg2+ ( Ashcroft and Gribble , 1998; Gribble et al . , 1997 , 1998; Nichols , 2006 ) . Evidence suggests that MgATP and MgADP interact with the nucleotide binding domains ( NBDs ) of SUR1 and either through MgATP hydrolysis or through direct MgADP binding at NBD2 , promote NBDs dimerization and channel opening ( de Wet et al . , 2012; Nichols , 2006; Zingman et al . , 2007 ) . Moreover , GBC has been proposed to inhibit KATP channels by preventing Mg-nucleotide stimulation ( de Wet and Proks , 2015 ) , and may do so by stabilizing the ABC core of SUR1 in an inward-facing conformation ( Ortiz et al . , 2012 ) but direct evidence is lacking . In order to understand how the channel functions as a complex to respond to physiological and pharmacological molecules and mechanisms by which channel mutations cause disease , detailed structural information is crucial . Here , we used cryo-EM to elucidate the structural basis of KATP channel assembly and gating . To obtain sufficient quantity of purified channel complexes , we used rat insulinoma INS-1 cells , which naturally express KATP channels , for overexpression . Cells were transduced with recombinant adenoviruses encoding genes for a FLAG-tagged hamster SUR1 and a rat Kir6 . 2 ( Pratt et al . , 2009 ) , which are 95% and 96% identical to the human sequences , respectively . These heterologously expressed channels have gating properties indistinguishable from endogenous KATP channels ( Pratt et al . , 2009 ) . Channel integrity was found to be best preserved when membranes were solubilized in digitonin and channels purified in the presence of 1 µM glibenclamide ( GBC ) and 1 mM ATP ( see Materials and methods ) ( Figure 1 ) , which was the condition used for cryo-EM structure determination . 10 . 7554/eLife . 24149 . 003Figure 1 . Purification and single-particle EM imaging of the SUR1/Kir6 . 2 KATP channel . ( A ) Size exclusion chromatography ( SEC ) profile of affinity purified KATP channels on a Suprose 6 column showing peak elution at ~11 . 5 ml ( the red rectangle ) . ( B ) Left: Blue native gel showing the size of the purified complex at ~1 mDa ( arrow ) corresponding to four SUR1 and four Kir6 . 2 . Input: samples eluted from anti-FLAG M2 agarose beads; void: sample from the SEC void fraction; 11 . 5 ml: sample from the SEC 11 . 5 ml elution fraction . Right: SDS-PAGE of the 11 . 5 ml fraction showing SUR1 ( lower band: core-glycosylated; upper band: complex-glycosylated ) and Kir6 . 2 as the main proteins . A vertical line separates MW markers from the sample lane in the same gel . ( C ) Negative-stain two-dimensional class averages showing topdown views ( 1 , 2 ) and side views ( 3 , 4 ) of the channel complex . ( D ) A representative cryoEM micrograph of KATP channel particles imaged on an UltrAufoil grid . ( E ) Representative two-dimensional class averages of KATP channels . DOI: http://dx . doi . org/10 . 7554/eLife . 24149 . 00310 . 7554/eLife . 24149 . 004Figure 1—figure supplement 1 . Cryo-EM data processing flowchart . DOI: http://dx . doi . org/10 . 7554/eLife . 24149 . 00410 . 7554/eLife . 24149 . 005Figure 1—figure supplement 2 . Cryo-EM density map analysis . ( A ) Euler angle distribution plot of all particles included in the calculation of the final map . ( B and C ) Fourier shell coefficient ( FSC ) curves of unmasked and masked whole complex , as well as masked Kir6 . 2 maps showing resolutions corresponding to FSC = 0 . 143 for the two 3D classes . ( D and E ) 3D density map with colored local resolution viewed from the side ( D ) and the bottom ( E ) . ( F ) Comparison of the cytoplasmic domain of Kir6 . 2 of the two 3D classes showing a counterclockwise rotation of ~14° of class 2 relative to class 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 24149 . 005 Single-particle analysis using RELION identified two three-dimensional ( 3D ) classes of particles with distinct conformations in the cytoplasmic domain of Kir6 . 2 ( see Discussion below ) . The dominant class ( ~60% ) produced a reconstruction which has an overall resolution of 6 . 7 Å ( FSC = 0 . 143 ) with C4 symmetry imposed ( Figure 1—figure supplements 1 and 2; Table 1 ) . With masking the FSC measurement at 0 . 143 reached 5 . 8 Å and the Kir6 . 2 core 5 . 1 Å . The other class yielded a reconstruction with an overall unmasked resolution ~7 . 6 Å , and masked whole channel and Kir6 . 2 core ~7 . 2 Å and 6 . 9 Å , respectively . The higher resolution map was used for model building and structural analysis . All transmembrane ( TM ) helices were clearly resolved in the density map ( 76 total; 17 from each SUR1 , 2 from each Kir6 . 2; Figure 2 ) and contained significant side-chain density which allowed for registration of the models . 10 . 7554/eLife . 24149 . 006Figure 2 . Three-dimensional reconstruction of the KATP channel . ( A ) Cryo-EM density map of the KATP channel complex at an overall resolution of 5 . 8 Å , viewed from the side . The four Kir6 . 2 subunits in the center are colored blue , SUR1 is in orange ( TMD0 ) , lavender ( L0 ) , green ( TMD1/NBD1 ) , and yellow ( TMD2/NBD2 ) . Gray bars indicate approximate positions of the lipid bilayer . ( B ) View of the complex from the cytoplasmic side . ( C and D ) Cross-sections of the density map . The planes where the sections 1 and 2 are made are shown in ( A ) . ( E ) Model of SUR1 and Kir6 . 2 constructed from the EM density map viewed from the side . A Kir6 . 2 tetramer and only two SUR1 subunits are shown for clarity . ( F ) The model viewed from the extracellular side . DOI: http://dx . doi . org/10 . 7554/eLife . 24149 . 00610 . 7554/eLife . 24149 . 007Figure 2—figure supplement 1 . Sequence and structure comparison between Kir6 . 2 and Kir3 . 2 . ( A ) Sequence alignment of rat Kir6 . 2 and mouse Kir3 . 2 . Only the Kir3 . 2 sequence that was used to solve the structure in the PIP2-bound state is shown ( PDB ID: 3SYA; residue 380 in mouse Kir3 . 2 is marked by a red vertical line on the right ) . Transmembrane helices in this and Figure 2—figure supplements 2–4 are colored dark blue . Kir6 . 2 sequence with no corresponding secondary structures shown at the top was not modeled due to lack of density in the map . ( B ) Superposition of the Kir6 . 2 structure and the structure of Kir3 . 2 ( PDB ID: 3SYA ) viewed from different angles . Blue: Kir6 . 2; lavender: Kir3 . 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 24149 . 00710 . 7554/eLife . 24149 . 008Figure 2—figure supplement 2 . Sequence alignment and structure comparison between SUR1 TMD1 and a bacterial peptidase-containing ABC transporter PCAT-1 ( PDB ID: 4RY2 ) . The structure 4RY2 of PCAT-1 was used for homology modeling of SUR1 TMD1 ( a . a . 284–616 ) . ( A ) Alignment of the hamster SUR1 sequence from 1 to 624 and the sequence of PCAT-1 in the crystal structure 4RY2 . ( B ) Superposition of the PCAT-1 structure 4RY2 and the final model of TMD1 of SUR1 . DOI: http://dx . doi . org/10 . 7554/eLife . 24149 . 00810 . 7554/eLife . 24149 . 009Figure 2—figure supplement 3 . Sequence alignment and structure comparison between SUR1 NBD1 and the mouse P-glycoprotein NBD1 ( PDB ID: 4MLM ) . The NBD1 structure of the mouse P-glycoprotein ( mPgp; PDB ID: 4MLM ) was used for homology modeling of SUR1 NBD1 . ( A ) Alignment of the hamster SUR1 sequence from 631–930 and the sequence of mPgp NBD1 in the crystal structure 4MLM . ( B ) Superposition of the mPgp-NBD1 and the final modeled NBD1 structure of SUR1 . DOI: http://dx . doi . org/10 . 7554/eLife . 24149 . 00910 . 7554/eLife . 24149 . 010Figure 2—figure supplement 4 . Sequence alignment and structure comparison between SUR1 TMD2-NBD2 and a bacterial ABC exporter TM287/288 ( PDB ID: 4Q4H ) . The structure 4Q4H of TM287/288 was used for homology modeling of SUR1 TMD2-NBD2 . ( A ) Alignment of the hamster SUR1 sequence from 961 to 1982 and the sequence of TM287/288 in the crystal structure 4Q4H . ( B ) Superposition of the TM287/288 structure 4Q4H and the final modeled TMD2-NBD2 structure of SUR1 . DOI: http://dx . doi . org/10 . 7554/eLife . 24149 . 01010 . 7554/eLife . 24149 . 011Table 1 . Statistics of cryo-EM data collection , 3D reconstruction and model building . DOI: http://dx . doi . org/10 . 7554/eLife . 24149 . 011Data collection/processing MicroscopeKriosVoltage ( kV ) 300CameraGatan K2Camera modeCountingDefocus range ( µm ) 1 . 2 ~ 3 . 5 Exposure time ( s ) 20Dose rate ( e-/pixel/s ) 6Magnified pixel size ( Å ) 1 . 72Total dose ( e-/Å2 ) 40Reconstruction SoftwareRELIONSymmetryC4Particles refined27371 Resolution ( unmasked , Å ) 6 . 7Resolution ( masked , Å ) 5 . 8Resoultion ( Kir6 . 2 masked , Å ) 5 . 1Map sharpening B-factor ( Å2 ) −250 Model Statistics Map CC0 . 95 ( masked ) Resolution ( FSC = 0 . 5 , Å ) 5 Å ( via phenix model-map FSC ) MolProbity score2 . 26Cβ deviations0Ramachandran Outliers0 . 12% Allowed4 . 68% Favored95 . 20% RMS deviations Bond length0 . 005Bond angles1 . 262 Kir6 . 2 is a member of the highly conserved Kir channel family in which several structures have been solved ( Hibino et al . , 2010 ) . By contrast , SUR1 is one of the few ABC transporter proteins which have an N-terminal extension consisting of a transmembrane domain termed TMD0 followed by a long intracellular loop ( the third intracellular loop , ICL3 ) termed L0 , in addition to an ABC core structure comprising two transmembrane domains ( TMD1 and 2 ) and two nucleotide binding domains ( NBD1 and 2 ) ( Tusnády et al . , 2006 ) . The Kir6 . 2 and SUR1 ABC core domain models were built initially from homologous Kir and ABC transporter structures ( sequence and model comparisons with templates shown in Figure 2—figure supplements 1–4 ) and then refined to fit the density . Because there is no known structural template for the TMD0-L0 of SUR1 , this region was modeled de novo . The structure shows that the KATP channel is an octamer built around a Kir6 . 2 tetramer with each subunit complexed to one SUR1 ( Figure 2 ) . The complex is ~200 Å in width in the longest dimension and ~125 Å in height , and is shaped like a propeller with the Kir6 . 2 pore and TMD0 forming a compact central core and the SUR1-ABC core structure forming the blades . A long-standing question has been where TMD0 and L0 are in relation to Kir6 . 2 and the ABC core structure , as this region has been shown to be crucial for channel assembly and gating ( Babenko and Bryan , 2003; Chan et al . , 2003; Schwappach et al . , 2000 ) . An earlier model hypothesized TMD0 to be sandwiched between Kir6 . 2 and the TMDs of the ABC core ( Bryan et al . , 2004 ) , but a later cryo-negative stain single-particle EM study of a channel formed by a SUR1-Kir6 . 2 fusion protein placed TMD0 next to Kir6 . 2 in between two adjacent SUR1-ABC core domains ( Mikhailov et al . , 2005 ) . In our structure , TMD0-L0 sits in between the SUR1 and Kir6 . 2 subunits and is the primary point of contact between the SUR1-ABC core and Kir6 . 2 ( Figure 2 ) . The Kir6 . 2 tetramer is the best resolved region in the complex ( Figure 3A ) . Side-chain density of many residues , in particular those in the two TM helices are visible ( Figure 3B ) . With knowledge of existing Kir channel structures , this allowed for confident model building ( see Materials and methods; sequence comparison with the template is shown in Figure 2—figure supplement 1 ) . 10 . 7554/eLife . 24149 . 012Figure 3 . Kir6 . 2 in a closed conformation . ( A ) Cryo-EM density map of Kir6 . 2 at 5 . 1 Å resolution . ( B ) Density of M1 and M2 . Residues with clear side chain density are labeled . ( C ) A central slice through the density highlighting the ion permeation pathway . ( D ) View of the inner helix gate ( F168 ) looking down the pore from the extracellular side . Kir3 . 2 apo ( yellow , PDB ID: 3SYO ) and Kir3 . 2-R201A+PIP2 ( red , 3SYQ ) structures were aligned to the region surrounding the gate . ( E ) Comparison of G-loop conformations of Kir6 . 2 and Kir3 . 2 ( 3SYO and 3SYQ ) by alignment of the cytoplasmic domain; same coloring as in ( D ) . The distance shown in ( D ) and ( E ) is between the main chains; the constriction should be even narrower due to side chains that should be protruding into the pore , as is seen in homolgous structures . Density depictions contoured to 2 . 5σ in ( B , D , E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24149 . 012 A vertical slice through the middle of the channel highlights the K+ conduction pathway ( Figure 3C ) . The three constriction points correspond to the selectivity filter , inner helix gate , and G-loop gate in other known Kir structures ( Hansen et al . , 2011; Whorton and MacKinnon , 2011 ) . In Kir6 . 2 , the inner helix gate is formed by F168 in M2 just below the central cavity . In our model , there is only ~6 Å between opposing atoms of the gate ( ~3 Å when considering the van der Waals radii ) , which is too narrow to allow passage of a ~8 Å diameter hydrated K+ ion ( Figure 3D ) . The G-loop gate formed at the apex of the cytoplasmic domains is shown in Figure 3E . A comparison of closed ( Kir3 . 2 apostate ) and open ( Kir3 . 2-R201A + PIP2 ) G-loop structures in relation to Kir6 . 2 suggests that this gate is also closed ( Figure 3E ) . Together , these observations indicate a closed channel structure , which is expected since the sample contained saturating concentrations of inhibitory ATP and GBC . Interestingly , 3D classification identified two classes with distinct conformations in the cytoplasmic domain ( CTD ) of Kir6 . 2 . The two classes differ by a rigid-body rotation of the CTD of ~14° ( Figure 1—figure supplement 2F ) . A similar rotation has been observed in multiple Kir channel members and has been associated with channel gating ( Clarke et al . , 2010; Whorton and MacKinnon , 2013 ) . However , the TMD and gates as well as the density corresponding to bound ATP ( see below ) in both classes are largely unaffected , suggesting rotational freedom for the CTD in the closed state . Whether this rotation represents a conformational transition that occurs during gating needs further investigation . A hallmark of the KATP channel is its inhibition by intracellular ATP . Mutagenesis and biochemical studies suggest that ATP binds directly to Kir6 . 2 ( Tanabe et al . , 1999; Tucker et al . , 1997 ) , and that residues in both N- and C-terminal domains are involved ( Antcliff et al . , 2005; Nichols , 2006 ) . However , while Kir6 . 2 is sensitive to ATP in the absence of SUR1 ( IC50 ~100 µM ) , SUR1 increases this sensitivity by ~10-fold ( IC50 ~10 µM ) ( Tucker et al . , 1997 ) . Where ATP binds and how SUR1 enhances the sensitivity to ATP inhibition remain key questions . Since our preparation contained 1 mM ATP , we reasoned that ATP is likely bound to the channel . Indeed , we observed a prominent bulge in the EM density that is too large to be accounted for by the main chain and the surrounding side-chains . The density is about the size of an ATP molecule and is immediately adjacent to K185 , a residue that has been implicated in ATP binding ( John et al . , 2003; Tanabe et al . , 1999; Tucker et al . , 1997 ) . Extensive mutagenesis of the K185 residue assessing the effects of various amino acid substitutions on channel sensitivity to inhibition by ATP , ADP , and AMP has provided strong evidence that this residue is important for binding to the β-phosphate of ATP ( Jöns et al . , 2006 ) . We used this information to guide the initial docking of ATP into the density and then refined with the surrounding protein in RSRef ( Chapman et al . , 2013 ) . An overview of the ATP binding site from the side ( Figure 4A ) , and from the top ( Figure 4B ) , with ATP colored in red , illustrates that the pocket is at the interface of adjacent Kir6 . 2 N and C domains . A close-up view ( Figure 4C ) shows that the docked ATP is surrounded by residues I182 , L205 , Y330 , F333 , and G334 from the same subunit , and R50 from the adjacent subunit . The adenine ring is pointing toward the N-terminus of subunit A , and could be supported by I182 , L205 , Y330 , and F333 of subunit B . R50 in subunit A is in a position that would allow it to interact with the γ-phosphate but may also interact with the adenine ring , which would explain mutagenesis data indicating that the interaction of R50 and ATP is not entirely electrostatic ( John et al . , 2003 ) . K185 is only ~3 Å from the β phosphate , while the α-phosphate is close to the main-chain nitrogen of G334 ( Figure 4C , D ) . Importantly , most residues surrounding the ATP density have been mutated and shown to affect ATP sensitivity ( Antcliff et al . , 2005 ) , providing direct validation of our structure . 10 . 7554/eLife . 24149 . 013Figure 4 . The ATP binding pocket . ( A and B ) Overview of ATP site from the side and from the top . ( C and D ) Difference map calculated from model prior to ATP docking , contoured to 3σ . Residues surrounding the ATP density are labeled . Side chains of residues with supporting density are shown . The N-terminus from Kir6 . 2 subunit A is colored in cyan and R50 is labeled followed by ( A ) . The adjacent subunit is colored in blue , and SUR1-L0 is colored lavender , with the K205 position labeled . DOI: http://dx . doi . org/10 . 7554/eLife . 24149 . 013 In our structure , we see that the density corresponding to ATP is located on the periphery of the Kir6 . 2 cytoplasmic domain , and traversed by the N-terminal segment of L0 of SUR1 immediately following TMD0 ( Figure 4B ) , with the Cα of K205 coming within only ~10 Å of the site ( Figure 4D ) . Interestingly , we have previously shown that mutation of K205 of L0 to alanine or glutamate reduce ATP sensitivity by ~10-fold ( Pratt et al . , 2012 ) . While there is no density in the map to allow placement of the K205 side chain , its Cα position lies directly over the site and is poised to make electrostatic contribution to ATP binding . This finding offers a mechanism by which SUR1 could enhance the ATP-sensitivity of the Kir6 . 2 channel . As shown in Figure 2 , TMD0-L0 is sandwiched between the SUR1-ABC core structure and Kir6 . 2 . In the map , densities corresponding to TMD0 and L0 are clearly seen , particularly TMD0 , with much of this domain reaching 5 Å resolution . This is in contrast to a recent cryo-EM study of another ABC transporter containing a TMD0 , TAP1/2 , where TMD0 could not be resolved ( Oldham et al . , 2016 ) , possibly because SUR1-TMD0 in our structure is stabilized by Kir6 . 2 . Overall , TMD0 is a five helix bundle which contains an extracellular N-terminal segment of 25 residues with a brief helical stretch , and mostly short loops connecting helices 2–3 , 3–4 , and 4–5 , but a longer ICL1 of ~14 residues connecting TM1-2 ( Figure 5A ) . The N-terminus containing the FLAG-peptide was disordered up until residue C6 of SUR1 where a highly conserved disulfide bond is formed with C26 ( Fukuda et al . , 2011 ) at the entrance to TM1 . This region contacts the Kir6 . 2 turret and pore loop ( Figure 5—figure supplement 1A ) , suggesting a role in assembly and functional coupling with the pore . A number of HI-causing mutations in the N-terminal extracellular loop of TMD0 including C6G , G7R , V21D , N24K , and C26S , which disrupt channel biogenesis efficiency or gating have been reported ( Martin et al . , 2016; Yan et al . , 2007 ) , further supporting the significance of this region in channel assembly and gating . 10 . 7554/eLife . 24149 . 014Figure 5 . The interface between TMD0 and the N-terminal segment of L0 with Kir6 . 2 . ( A ) Overall structure of the interface region , with TMD0 in orange , Kir6 . 2 in blue , and L0 in lavender . ECL: extracellular loop; ICL: intracellular loop; IF helix: interfacial ( slide ) helix . ( B and C ) Detailed view of the region boxed in red in ( A ) shown in ribbon ( B ) and surface ( C ) representations . ATP is docked as in Figure 3 and PIP2 was docked hypothetically using PIP2 bound Kir3 . 2 and Kir2 . 2 structures for placement . ( D ) A side view of the ICL2 showing close interactions with the Kir6 . 2 IF helix . E128 and F132 , mutation of which alters channel Po and ATP sensitivity , are highlighted . ( E ) A top-down view of this region with both docked ATP ( in the back ) and PIP2 in view . DOI: http://dx . doi . org/10 . 7554/eLife . 24149 . 01410 . 7554/eLife . 24149 . 015Figure 5—figure supplement 1 . Interactions between TMD0 and Kir6 . 2 . ( A ) Interactions of SUR1 N-terminus with the pore loop and turret of Kir6 . 2 . Note continuous density extending from the pore loop to the N-term/extracellular loop 2 ( ECL2 ) and from the turret to the short helical segment of ECL1 . Map is displayed at 2 . 5σ . ( B ) The Kir6 . 2 M1-SUR1 TM1 interface showing the tight association of these two helices and interaction between ICL2 and the Kir6 . 2 N-terminal interfacial ( IF ) helix . ( C ) Possible hydrophobic interactions between M1 ( blue , Kir6 . 2 ) and TM1 ( orange , SUR1 ) helices . DOI: http://dx . doi . org/10 . 7554/eLife . 24149 . 015 In the transmembrane region , TM1 of TMD0 and the M1 helix of Kir6 . 2 are the primary sites of interaction . These helices make close contact throughout their entire length ( Figure 5A ) and at residue P45 in TM1 , a kink is introduced that places the trajectory of the two helices in alignment ( Figure 5—figure supplement 1B ) . There are many potential hydrophobic interactions between opposing faces of these helices , which may facilitate association of the complex ( Figure 5—figure supplement 1C ) . Indeed , multiple HI-causing mutations in TM1 of TMD0 ( F27S , A30T , L31P , L40R ) have been shown to impair channel assembly and surface expression ( Martin et al . , 2016 ) , likely by disrupting interactions between the two helices . On the cytoplasmic side , there are intimate interactions between the ICLs of TMD0 , the start of L0 , the Kir6 . 2 PIP2 binding pocket ( cytoplasmic ends of M1 and M2 helices ) identified based on other PIP2-bound Kir structures ( Hansen et al . , 2011; Whorton and MacKinnon , 2011 ) , and the Kir6 . 2 ATP binding pocket . As shown in Figure 5B and C , the hypothetically docked PIP2 is surrounded by the cytoplasmic loop connecting TM3 and 4 ( ICL2; E128-P133 ) of TMD0 and the N-terminal stretch of L0 ( K192-K199 ) from one SUR1 subunit , and the cytoplasmic end of TM1 ( K57 ) of TMD0 from the adjacent SUR1 subunit . Previous studies have shown that TMD0 and the N-terminal section of L0 increase the Po of Kir6 . 2 to resemble intact channels ( Babenko and Bryan , 2003; Chan et al . , 2003 ) . As Po is determined by PIP2 interactions , our structure suggests these regions may contribute directly to PIP2 binding to account for the increase in PIP2 sensitivity conferred by SUR1 ( Enkvetchakul et al . , 2000 ) . Below PIP2 and near the periphery of Kir6 . 2 lies ATP , separated from PIP2 by L0 ( Figure 5B , C ) and also ICL2 of TMD0 ( Figure 5B , E ) . The ICL2 sits directly atop the Kir6 . 2 N-terminus , just before the interfacial helix ( i . e . the ‘slide helix’ ) at Q52 ( Figure 5D ) , and simultaneously contacts ICL1 of TMD0 and the most C-terminal portion of TMD0 at TM5 . Mutation of E128 ( E128K , a HI mutation ) and F132 ( F132L , a PNDM mutation ) in ICL2 as well as Q52 in Kir6 . 2 ( Q52R , a PNDM mutation ) is known to disrupt channel gating by ATP and PIP2 ( Pratt et al . , 2009; Proks et al . , 2004 , 2006 ) ( Figure 5C , D ) . Our finding that this region is close to both the ATP and PIP2 sites illustrates that it is well positioned to contribute to gating regulation by both , explaining the effects of these disease mutations . L0 ( i . e . ICL3 ) is nestled between TMD0 and the ABC core of SUR1 , and comprises ~90 amino acids . We have modeled L0 as a polyalanine chain with two helical segments that are strongly supported by the map , one an amphipathic helix from L224-A240 and the other from L260-D277 , which connects to TMD1 . In the model , the N- and C-terminal stretches of L0 make a ‘V , ’ with the intervening sequence ( L213-L260 ) forming a hairpin structure at the apex ( Figure 6A , B ) . This hairpin structure is simultaneously bridging multiple sites within TMD0 with the ABC core structure ( TMs 15 + 16 ) , and may also interact with the Kir6 . 2 N-terminus ( A45-Q52 ) , which would allow L0 to transduce signals from the ABC core to gate the channel . The strategic placement of L0 is consistent with its multiple functional roles reported , including regulation of channel Po , sensitivity to ATP inhibition , and sensitivity to Mg-nucleotide stimulation ( Babenko and Bryan , 2003; Chan et al . , 2003; Masia et al . , 2007 ) . 10 . 7554/eLife . 24149 . 016Figure 6 . The SUR1-L0 connecting TMD0/Kir6 . 2 with the SUR1-ABC core . ( A ) View of the L0 region from the side along the plane of the membrane; Kir6 . 2 density has been removed for clarity . The hairpin structure is outlined . ( B ) Slice through the N- and C-terminal segments of L0 . ( C ) Model of L0 highlighting relation between Y230 and S1238 ( marked red ) in TM16 , which are separated by ~20 Å ( Cα to Cα ) . Side chain of Y230 is shown based on supporting density . The gray dashed line marks the approximate boundary of the inner leaflet of the lipid bilayer . DOI: http://dx . doi . org/10 . 7554/eLife . 24149 . 016 Another role of L0 that has been reported is interaction with GBC ( Winkler et al . , 2012 ) . GBC is a second-generation sulfonylurea containing a sulfonylurea group and a benzamido moiety that binds KATP channels with nanomolar affinity ( KD ~1 nM ) ( Gribble and Reimann , 2003 ) . L0 has been proposed to participate in binding to the benzamido group , with mutation Y230A in L0 reducing GBC binding . We find that the amphipathic helix of L0 containing Y230 sits next to TM16 containing S1238 , a residue which when mutated disrupts binding of the sulfonylurea group ( Ashfield et al . , 1999 ) . The two residues are separated by ~20 Å ( Cα to Cα ) , which explains how the two residues distant in the primary sequence can both contribute to binding . Although at the current resolution , we are unable to discern the density for GBC , it is likely to be bound given its high affinity . The model can now be used to guide future studies to clearly define the GBC binding site . The SUR1 core is built from two homologous halves , TMD1-NBD1 and TMD2-NBD2 . Each of the 12 combined TM helices from both TMD1 and TMD2 are clearly resolved , as well as the short lateral ‘elbow’ helices leading into the first helix of each TMD ( TM6 and TM12 ) ( Figure 7A , B ) . Characteristic of other ABC exporters , there is a domain swap at the extracellular linker between helices 3 and 4 of each TMD ( Jin et al . , 2012; Kim et al . , 2015 ) , such that each ‘half’ of the ABC core is composed of TMs 1–3 , and 6 of one TMD , plus TMs 4 and 5 of the other ( Figure 7A ) . 10 . 7554/eLife . 24149 . 017Figure 7 . SUR1 with a twisted ABC core conformation in saturating concentrations of GBC . ( A ) Model of SUR1 with the various domains colored as in Figure 1 , with each TM helix labeled . On the left , TMD1/NBD1 ( green ) is toward the front and TMD2/NBD2 ( tan ) is toward the back . ( B ) Cross-section of the SUR1 model , showing relative orientation of each of the 17 TM helices and a helix in L0 . ( C ) Comparison of inward-facing ABC transporter structures: From left , C elegans Pgp ( PDB code 4 F4C ) ; mouse Pgp ( 4 M1M ) ; hamSUR1 . For each model , TMD2/NBD2 is colored tan . Lines on the side of the SUR1 NBDs denote the relative orientation of the NBD dimerization interface , demonstrating the observed twisting relative to other inward-facing structures . DOI: http://dx . doi . org/10 . 7554/eLife . 24149 . 017 Overall , the SUR1-ABC core is in an inward-facing conformation , with the NBDs clearly separated ( Figure 7C ) . This is consistent with other ABC exporter structures solved without Mg-nucleotides . However , in contrast to other ABC exporters of known structure whereby transporter halves are related by either a true or a pseudo two-fold symmetry axis , depending on whether the two halves are identical or not ( Wilkens , 2015 ) , we find a clear rotation and a translation of TMD1-NBD1 relative to TMD2-NDB2 , such that TMD1-NBD1 is ~15° off the symmetry axis and is translated by ~10 Å horizontally ( relative to the membrane ) ( Figure 7C ) . In this configuration , the SUR1 NBDs likely could not dimerize without a twisting motion to align the dimerization interface . Dimerization of NBDs in SUR1 has been proposed to follow MgATP hydrolysis or MgADP binding to stimulate channel activity ( Nichols , 2006 ) , and GBC inhibits channel activity by preventing Mg-nucleotide stimulation ( de Wet and Proks , 2015; Gribble and Reimann , 2003 ) . As discussed above , given its high-affinity GBC is likely to be bound in our structure . Thus , an interesting hypothesis is that the twisted conformation is caused by GBC binding , which would suggest that GBC prevents MgADP from stimulating the channel by causing a misalignment of the NBDs dimerization interface . Alternatively , the conformation may be unique to SUR1 and that Mg-nucleotide binding/hydrolysis is required to restore symmetry for dimerization . In this case , GBC may block stimulation by clamping down L0 and preventing it from communicating with Kir6 . 2 . A structure in the absence of GBC will be needed to test these hypotheses . The structure reported here provides the first glimpse of the detailed domain organization of KATP channels and the intricate structural interactions between SUR1 and Kir6 . 2 . These data offer mechanistic insight into how SUR1 and Kir6 . 2 function as a complex to regulate insulin secretion ( Figure 8A ) . We propose that like other ABC transporters ( Wilkens , 2015 ) , the ABC core of SUR1 switches between an inward-facing and outward-facing conformations as MgATP undergoes hydrolysis or as MgADP binds at NBD2 and induces NBD dimerization . The conformational switch at the ABC core causes movement of the L0 and TMD0 , which alters channel interactions with ATP and PIP2 by remodeling the interface formed by the cytoplasmic domain of Kir6 . 2 , the bottom of the Kir6 . 2 transmembrane helices , the intracellular loops of TMD0 and the N-terminal segment of L0 . In this way , the SUR1 ‘transport’ cycle is coupled to Kir6 . 2 opening or closing rather than transport of substrates through SUR1 itself . 10 . 7554/eLife . 24149 . 018Figure 8 . KATP channel gating model . ( A ) Cartoon illustrating how changes in the ATP/ADP ratio upon feeding and fasting alter the equilibrium between the inward-facing and outward-facing states of the SUR1-ABC core and interactions of the channel with ATP and PIP2 to control channel activity . ( B ) Model of the hypothesized mechanism whereby GBC causes misalignment of the NBDs to prevent Mg-nucleotides activation of KATP channels . In both A and B , Kir6 . 2 transmembrane helices: green; Kir6 . 2 cytoplasmic domain: lime green; SUR1-TMD0/L0: magenta; SUR1-TMD1/2: blue; SUR1-NBDs: orange; GBC: yellow; ATP: red; PIP2: cerulean . Note the different states shown are not meant to reflect the actual conformational transitions . DOI: http://dx . doi . org/10 . 7554/eLife . 24149 . 018 Our structure highlights the critical role of SUR1-TMD0 in the association of the two subunits . In addition to contacts made by TM1 of Kir6 . 2 and the first TM helix of TMD0 which are consistent with previous structure-function studies ( Schwappach et al . , 2000 ) , there are also new interactions revealed by the structure in the extracellular domain of TMD0 and the turret/pore loop of Kir6 . 2 as well as the cytoplasmic domains of TMD0 and Kir6 . 2 . Indeed , TMD0 appears to harbor more mutations that disrupt channel biogenesis and trafficking than other regions of SUR1 ( Martin et al . , 2013 , 2016 ) . It is worth noting that many mutations in TMD0 which impair channel biogenesis and trafficking can be rescued by pharmacological chaperones , specifically sulfonylureas such as GBC ( Chen et al . , 2013a; Martin et al . , 2016; Yan et al . , 2004 , 2007 ) . As our structure is obtained in the presence of GBC , an important question to address in the future is whether GBC alters structural interactions between TMD0 and Kir6 . 2 to correct biogenesis/trafficking defects caused by TMD0 mutations . The interface between TMD0-L0 and Kir6 . 2 in the cytoplasmic domain near the proposed PIP2 binding site and where ATP density is observed suggests TMD0-L0 may directly enforce PIP2 or ATP binding to enhance Kir6 . 2 sensitivity to both , and also explains the effects of many disease mutations in this region . Although in our structure the Kir6 . 2 is bound to ATP with the pore in a closed conformation , a gating scheme whereby in the presence of PIP2 remodeling of the interfaces near the ATP and PIP2 sites leads to channel opening may be envisioned . Future studies comparing structures in the absence of ATP and with or without PIP2 are needed to understand in detail the structural changes involved in gating . The L0 region before the elbow helix leading to TMD1 in SUR1 was modeled de novo , with an amphipathic helix from L224-A240 and a helix from L260-D277 that are strongly supported by the density map . Part of L0 ( from a . a . 214 on ) is conserved in CFTR and the multidrug resistance-associated proteins MRPs ( Zhang and Chen , 2016 ) . Interestingly , in the recently reported CFTR structure , this loop which the authors named the ‘lasso motif’ also contains an amphipathic helix followed by another helix before the elbow helix ( Zhang and Chen , 2016 ) . Our structural model of L0 is in line with the CFTR model of the corresponding loop . In CFTR or MRP-1 , this loop has been shown to be involved in trafficking regulation by syntaxin 1A ( Naren et al . , 1998; Peters et al . , 2001 ) or association with the plasma membrane ( Bakos et al . , 2000 ) , respectively . It would be interesting to determine whether L0 of SUR1 has similar roles . A striking feature observed in our structure is the unexpected twisted inward-facing conformation of the SUR1-ABC core that is distinct from other ABC transporter apo-state structures ( Wilkens , 2015 ) . This observation suggests a possible mechanism in which GBC inhibits channel activity by preventing dimerization of NBDs in the presence of Mg-nucleotides ( Figure 8B ) . As GBC is known to inhibit the activity of other ABC transporter proteins including CFTR ( Schultz et al . , 1996 ) and the multidrug resistance protein MDR ( Golstein et al . , 1999 ) , the mechanism we propose could have broader implications . Intriguingly , close examination of the recently published zebrafish CFTR structure where the inhibitory R-domain is present ( Zhang and Chen , 2016 ) and the TAP transporter structure with an inhibitory viral peptide bound ( Oldham et al . , 2016 ) also indicates misalignment of the two NBDs albeit to lesser degrees , further suggesting that NBDs misalignment may be a common theme in ABC transporters bound to inhibitory ligands . In summary , the novel insight gained from our structure lays the foundation for future structural and functional studies . In particular , structures bound with various stimulatory and inhibitory ligands will further advance understanding of the detailed mechanisms of channel gating . Some regions known to be important for channel assembly and gating such as the distal N- and C-termini of Kir6 . 2 as well as several linker loops in SUR1 are not well resolved in the current map ( see Materials and methods for details ) . An equally important future goal is to stabilize these regions and obtain higher resolution structures to fully visualize the channel . Construction of the hamster SUR1 ( 94 . 5% protein sequence identity with human SUR1 ) with an N-terminal FLAG-tag ( f-SUR1 ) and rat Kir6 . 2 ( 96 . 15% protein sequence identity with human Kir6 . 2 ) recombinant adenoviruses was as described previously ( Lin et al . , 2005; Pratt et al . , 2009 ) . A FLAG tag ( DYKDDDDK ) was engineered at the N-terminus of SUR1 for affinity purification of the channel complex . In brief , the gene encoding the rat Kir6 . 2 was cloned into pShuttle , and recombined with the pAdEasy vector in the BJ5183 strain of Escherichia Coli . Positive recombinants were selected , and pAdEasy plasmids containing the correct insert were used to transfect HEK293 cells ( RRID:CVCL_0045 ) for virus production . The SUR1 recombinant adenovirus was constructed using a modified pShuttle plasmid ( AdEasy kit , Stratagene , San Diego , CA ) containing a tetracycline-inducible promoter . Recombinant viruses were amplified in HEK293 cells and purified according to the manufacturer's instructions . INS-1 cells clone 832/13 ( RRID:CVCL_7226 ) ( from Dr . Christopher Newgard ) ( Hohmeier et al . , 2000 ) were plated in 15 cm plates and cultured for 24 hr in RPMI 1640 with 11 . 1 mM D-glucose ( Invitrogen , Carsbad , CA ) supplemented with 10% fetal bovine serum , 100 units/ml penicillin , 100 μg/ml streptomycin , 10 mM HEPES , 2 mM glutamine , 1 mM sodium pyruvate , and 50 μM β-mercaptoethanol . For channel expression , cells were co-infected with three recombinant adenoviruses , one encoding Kir6 . 2 , one f-SUR1 , and one encoding tetracycline-inhibited transactivator ( tTA ) for the tTA-regulated f-SUR1 expression ( Pratt et al . , 2009 ) . Cells at ∼70% confluent density were washed once with phosphate-buffered saline ( PBS ) and then incubated for 3 hr at 37°C in OPTI-MEM without serum and a mixture of viruses with the multiplicity of infection ( M . O . I . ) of each virus determined empirically to optimize the maturation efficiency of the channel complex as judged by the abundance of the SUR1 and Kir6 . 2 bands as well as the ratio of the mature complex glycosylated versus the immature core-glycosylated SUR1 bands . Medium was then replaced with fresh growth medium plus 1 mM sodium butyrate and 1 µM glibenclamide ( GBC ) to enhance expression and maturation ( Yan et al . , 2004 ) , and the cells were further incubated at 37°C for 36–48 hr . Cells were harvested in PBS , pelleted , flash frozen in liquid nitrogen , and stored at −80°C until purification . For channel purification , cells were resuspended in hypotonic buffer ( 15 mM KCl , 10 mM HEPES , 1 . 5 mM MgCl2 ) and allowed to swell for 20 min on ice . Cells were then lysed with a tight-fitting Dounce homogenizer , then centrifuged at 20 , 000 xg for 60 min . Membranes were resuspended in buffer A ( 150 mM NaCl , 25 mM HEPES , 50 mM KCl , 1 mM ATP , 1 µM GBC , 4% Trehalose ) with protease inhibitors ( cocktail tablets from Roche ) and then solubilized with 0 . 5% Digitonin for 90 min . Solubilized membranes were separated from insoluble materials by centrifugation ( 100 , 000 xg for 30 min at 4°C ) and then incubated with anti-FLAG M2 affinity agarose gel for 4–5 hr . The protein-bound agarose gel was washed with five column volumes of buffer B ( 150 mM NaCl , 25 mM HEPES , 50 mM KCl , 1 mM ATP , 1 µM GBC , 0 . 05% Digitonin ) and bound proteins eluted in the same buffer with FLAG peptide . Eluted proteins were concentrated using a centricon filter ( 100 kD cutoff ) to a final concentration of ~0 . 7–1 mg/ml . Purified proteins were further fractionated by size exclusion chromatography using a Suprose six column and fractions analyzed by blue native gel electrophoresis and SDS-PAGE ( Figure 1A , B ) . Digitonin solubilized KATP complexes ( in the presence of 1 mM ATP and 1 µM GBC ) were first examined by negative-staining EM ( 1% w/v uranyl acetate , on continuous thin-carbon coated grids ) to confirm the integrity of the full complex ( Figure 1C ) . For cryo-EM imaging , due to low particle distribution with holey-carbon grids , we experimented with two types of grids: UltrAufoil gold grids and C-flat grids coated in-house with 5 nm of gold on each side , and used both in the final data collection . The grids were first glow-discharged by EasyGlow at 20 mA for 45 s , then 3 µl of purified KATP complex was loaded onto the grid , blotted ( 2–4 s blotting time , force −4 , and 100% humidity ) and cryo-plunged into liquid ethane cooled by liquid nitrogen using a Vitrobot Mark III ( FEI , Hillsboro , OR ) . Single-particle cryo-EM data weres collected on a Titan Krios 300 kV cryo-electron microscope ( FEI ) in the Multi-Scale Microscopy Core at Oregon Health and Science University , assisted by the automated acquisition program SerialEM . Images were recorded on the Gatan K2 Summit direct electron detector in the counting mode at the nominal magnification 81 , 000 x ( calibrated image pixel-size 1 . 720 Å ) , with varying defocus ranging between 1 . 2 and 3 . 5 µm across the dataset ( Figure 1D ) . To contain the beam radiation damage and reduce electron coincidence loss in the K2 counting-mode recording , the dose rate was kept around 2 . 0 e-/Å2/s , frame rate at 2 frames/s and 40 frames in each movie , which gave the total dose of approximately 40 e-/Å2 . In total , 4339 movies were recorded , from which ~35 , 000 particles were used in final reconstructions ( Figure 1—figure supplements 1 and 2 ) . The raw frame stacks were gain-normalized and then aligned and dose-compensated using Unblur ( Grant and Grigorieff , 2015 ) ( Table 1 ) . CTF was estimated from the aligned frame sums using CTFFIND4 ( Rohou and Grigorieff , 2015 ) . To reduce the possibility of bias and capture every possible particle view , an initial set of 350 , 000 potential particles ( referred to as ‘peaks’ in Figure 1—figure supplement 1 ) were picked using DoGPicker ( Voss et al . , 2009 ) with a broad threshold range for subsequent 2D classification using RELION ( Scheres , 2012 ) . 2D classification was able to remove the large number of false positives and aggregates , and resulted in ~35 , 000 particles with 2D classes in which secondary structure was already apparent ( Figure 1E ) . These class averages revealed that the side views also adopted a preferred orientation . Upon imposing C4 symmetry , the angular sampling space was filled in along three orthogonal axes ( Figure 1—figure supplement 2A ) , which greatly improved the quality of the 3D reconstruction . The final rounds of refinement with C4 symmetry revealed two 3D classes ( Figure 1—figure supplement 1 ) . The dominant class ( EMDB ID: EMD-8470 ) , derived from 20 , 707 particles had an overall resolution of ~6 . 7 Å , and application of a mask improved the resolution of the overall structure to 5 . 8 Å and the central Kir6 . 2 domain to 5 . 1 Å ( Figure 1—figure supplement 2B ) . The second class , derived from 14 , 115 particles , had an overall unmasked resolution of ~7 . 6 Å , and masking improved the resolution of the overall structure to 7 . 2 Å and for the central Kir6 . 2 domain to 6 . 9 Å ( Figure 1—figure supplement 2C ) . All resolutions were reported using the 0 . 143 criterion with gold-standard FSC and phase-randomization correction for the use of masks ( Chen et al . , 2013b ) . Resolution was further confirmed using local-resolution as measured using ResMap ( Kucukelbir et al . , 2014 ) , and by observing criterion such as helical pitch starting to become visible , and density bumps for some of the larger side chains ( see examples shown in Figure 3B ) . Maps were B-factor corrected during post-processing using the K2 MTF , and the fitting procedure described by Rosenthal and Henderson ( Rosenthal and Henderson , 2003 ) . The two 3D classes differ in the cytoplasmic domain of Kir6 . 2 where a rotation of ~14° relative to each other was observed ( Figure 1—figure supplement 2F ) . Local resolution measurements using ResMap and masked FSCs showed that some parts of the complex including Kir6 . 2 and TMDs of SUR1 had significantly better resolution , in the 5 Å range , than the overall resolution of 5 . 8 Å , while other parts such as the NBDs of SUR1 had worse resolution , estimated to be in the 8 Å range . Moreover , some parts of the channel complex , such as the TMD0 and L0 of SUR1 do not have existing homology models . Therefore , different strategies were used to model the channel complex , as detailed below . For Kir6 . 2 , a homology model was built from Kir3 . 2 ( PDB ID: 3SYA ) using MODELLER ( Webb and Sali , 2016 ) and served as the initial model . The model was docked into the density in UCSF Chimera ( Pettersen et al . , 2004 ) ; the fit was improved by rigid body refinement of domains in RSRef ( Chapman et al . , 2013 ) , followed by iterative rounds of real-space refinement in COOT ( Emsley et al . , 2010 ) and stereochemically restrained torsion angle refinement in CNS ( Brünger et al . , 1998 ) , substituting in the RSRef real-space target function ( Chapman et al . , 2013 ) , adding ( φ , ψ ) backbone torsion angle restraints , and imposing non-crystallographic symmetry ( NCS ) constraints . The final model contained residues 32–356 ( Figure 2—figure supplement 1 ) . The distal N- and C-termini of Kir6 . 2 , although interesting regions implicated in channel assembly and gating ( Devaraneni et al . , 2015; Enkvetchakul et al . , 2000; Zerangue et al . , 1999 ) , lacked strong density . Therefore , they were not included in the model . For the SUR1 core structure , the sequence was divided into three segments: TMD1 , NBD1 , and TMD2-NBD2 . A TMD1 homology model was built using PCAT-1 ( PDB ID: 4RY2 ) ( Figure 2—figure supplement 2 ) , NDB1 was modeled from the NDB1 of mouse P-glycoprotein ( PDB ID: 4 M1M ) ( Figure 2—figure supplement 3 ) , and TMD2 and NBD2 were modeled together from chain B of TM287/288 ( PDB ID: 4Q4HB ) ( Figure 2—figure supplement 4 ) ; all homology models were built with MODELLER . These models were docked into the density in Chimera . SUR1 had some disordered regions ( 744–770 , 928–1000 , 1319–1343 ) , particularly in the linkers between TMDs and NBDs , and in NBD1 , that were not seen in our map . These regions were removed from the homology models before proceeding with refinement . The TM helices were then manually adjusted in COOT , as a substantial adjustment was needed to move them into density . The domains were then refined in the same steps as outlined for Kir6 . 2 , except that before the final manual adjustments in COOT and final density gradient optimization , a batch of torsion angle simulated annealing optimization was inserted , again using RSRef/CNS and the same torsion angle restraints and NCS constraints . The final model for the ABC core structure contained residues 284–616 ( TMD1 ) , 675–739 and 762–930 ( NBD1 ) , 981–1044 and 1060–1321 ( TMD2 ) , and 1325–1577 ( NBD2 ) . TMD0 and L0 domains of SUR1 ( a . a . 1–283 ) are some of the most interesting and novel regions of the KATP complex for which there is no existing homology model . These domains were therefore modeled de novo . Even though embedded in a micelle , all the transmembrane helices in TMD0 are clearly visible in the density map . The visibility of helical pitch and some side chains allowed confident modeling and refinement of the TM helices . With the predominantly alpha-helical nature of this domain , continuous loop density between most of the TM helices , and the presence of residues with bulky side chains , we were able to build the ~200 residues of TMD0 with a good degree of confidence . Of less certainty was the L0 region of SUR1 that sits between TMD0 and TMD1 . While there was an easily identifiable region of the map corresponding to L0 , the scarcity of secondary structures in this region made it difficult to build with the same degree of confidence . This was further complicated by the high likelihood that some of the observed density may be attributable to the ligand GBC , a high affinity antagonist which has been shown to interact with this region ( Bryan et al . , 2004 ) . Nonetheless , we made a best effort to model the residues in L0 primarily to verify that ( 1 ) a plausible model could be built into this density , and ( 2 ) that the observed density was sufficient to account for all the amino acids in this loop . The L0 model we built fulfilled both criteria , and as such , allowed for a better interpretation and understanding of the electron density map . We did not , however , attempt to draw any definitive conclusions about specific residues or GBC density from our tentative modeling of L0 . Note we used two different software suites , RSRef and PHENIX ( Adams et al . , 2010 ) , to confirm the consistency of our individual models of Kir6 . 2 and the SUR1 ABC core structure upon refinement into our electron density . The full final models were refined with all the constraints available in PHENIX real-space refinement: torsion angles , bond lengths , Ramachandran , and secondary structure . This was done initially with side-chains in place to ensure that the refinement did not place residues in implausible configurations ( Figure 3B shows examples of residues that were particularly well-resolved and served as anchor points for building and refining the model ) . Evaluation of these refined models confirmed that the model could be refined to fit the density quite well while maintaining good stereochemical statistics ( Table 1 ) . However , as many of the side chains did not have much , if any , supporting density , a final pass was made throughout the entire model to remove these side-chains prior to PDB deposition ( PDB ID: 5TWV ) . The resulting model was very similar to the full-atom refinement , but had better statistics ( Table 1 ) primarily due to the reduced possibility of clashes .	The hormone insulin reduces blood sugar levels by encouraging fat , muscle and other body cells to take up sugar . When blood sugar levels rise following a meal , cells within the pancreas known as beta cells should release insulin . In people with diabetes , the beta cells fail to release insulin , meaning that the high blood sugar levels are not corrected . When blood sugar levels are high , beta cells generate more energy in the form of ATP molecules . The increased level of ATP causes channels called ATP-sensitive potassium ( KATP ) channels in the membrane of the cell to close . This triggers a cascade of events that leads to the release of insulin . Some treatments for diabetes alter how the KATP channels work . For example , a widely prescribed medication called glibenclamide ( also known as glyburide in the United States ) stimulates the release of insulin by preventing the flow of potassium through KATP channels . It remains unknown exactly how ATP and glibenclamide interact with the channel’s molecular structure to stop the flow of potassium ions . KATP channels are made up of two proteins called SUR1 and Kir6 . 2 . To investigate the structure of the KATP channel , Martin et al . purified channels made of the hamster form of the SUR1 protein and the mouse form of Kir6 . 2 , which each closely resemble their human counterparts . The channels were purified in the presence of ATP and glibenclamide and were then rapidly frozen to preserve their structure , which allowed them to be visualized individually using electron microscopy . By analyzing the images taken from many channels , Martin et al . constructed a highly detailed , three-dimensional map of the KATP channel . The structure revealed by this map shows how SUR1 and Kir6 . 2 work together and provides insight into how ATP and glibenclamide interact with the channel to block the flow of potassium , and hence stimulate the release of insulin . An important next step will be to improve the structure to more clearly identify where ATP and glibenclamide bind to the KATP channel . It will also be important to study the structures of channels that are bound to other regulatory molecules . This will help researchers to fully understand how KATP channels located throughout the body operate under healthy and diseased conditions . This knowledge will aid in the design of more effective drugs to treat several devastating diseases caused by defective KATP channels .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics" ]	2017	Cryo-EM structure of the ATP-sensitive potassium channel illuminates mechanisms of assembly and gating
Comparing the impact of the COVID-19 pandemic between countries or across time is difficult because the reported numbers of cases and deaths can be strongly affected by testing capacity and reporting policy . Excess mortality , defined as the increase in all-cause mortality relative to the expected mortality , is widely considered as a more objective indicator of the COVID-19 death toll . However , there has been no global , frequently updated repository of the all-cause mortality data across countries . To fill this gap , we have collected weekly , monthly , or quarterly all-cause mortality data from 103 countries and territories , openly available as the regularly updated World Mortality Dataset . We used this dataset to compute the excess mortality in each country during the COVID-19 pandemic . We found that in several worst-affected countries ( Peru , Ecuador , Bolivia , Mexico ) the excess mortality was above 50% of the expected annual mortality ( Peru , Ecuador , Bolivia , Mexico ) or above 400 excess deaths per 100 , 000 population ( Peru , Bulgaria , North Macedonia , Serbia ) . At the same time , in several other countries ( e . g . Australia and New Zealand ) mortality during the pandemic was below the usual level , presumably due to social distancing measures decreasing the non-COVID infectious mortality . Furthermore , we found that while many countries have been reporting the COVID-19 deaths very accurately , some countries have been substantially underreporting their COVID-19 deaths ( e . g . Nicaragua , Russia , Uzbekistan ) , by up to two orders of magnitude ( Tajikistan ) . Our results highlight the importance of open and rapid all-cause mortality reporting for pandemic monitoring . The impact of COVID-19 on a given country is usually assessed via the number of cases and the number of deaths , two statistics that have been reported daily by each country and put together into international dashboards such as the ones maintained by the World Health Organization ( https://covid19 . who . int ) or by the Johns Hopkins University ( https://coronavirus . jhu . edu ) ( Dong et al . , 2020 ) . However , both metrics can be heavily affected by limited testing availability and by different definitions of ‘COVID-19 death’ used by different countries ( Riffe et al . , 2021 ) : for example , some countries count only PCR-confirmed COVID-19 deaths , while others include suspected COVID-19 deaths as well . Excess mortality , defined as the increase of the all-cause mortality over the mortality expected based on historic trends , has long been used to estimate the death toll of pandemics and other extreme events—from the Great Plague of London in 1665 ( as described in Boka and Wainer , 2020 ) , to the influenza epidemic in London in 1875 ( Farr , 1885; Langmuir , 1976 ) , the XX–XXI century influenza pandemics of 1918 , 1957 , 1968 , 2009 ( Murray et al . , 2006; Viboud et al . , 2005; Viboud et al . , 2016; Simonsen et al . , 2013 ) as well as seasonal influenza epidemics ( Housworth and Langmuir , 1974 ) , and more recently for example Hurricane Maria in Puerto-Rico in 2016 ( Milken Institute , 2018 ) . Even though the excess mortality does not exactly equal the mortality from COVID-19 infections , the consensus is that for many countries it is the most objective possible indicator of the COVID-19 death toll ( Beaney et al . , 2020; Leon et al . , 2020 ) . Excess mortality has already been used to estimate the COVID-19 impact in different countries , both in academic literature ( e . g . Kontis et al . , 2020; Alicandro et al . , 2020; Ghafari et al . , 2021; Woolf et al . , 2020a; Woolf et al . , 2020b; Weinberger et al . , 2020; Blangiardo et al . , 2020; Kobak , 2021a; Modi et al . , 2021; Bradshaw et al . , 2021; Islam et al . , 2021 , among many others ) and by major media outlets . It has also been used to compare COVID-19 impact to the impact of major influenza pandemics ( Faust et al . , 2020; Petersen et al . , 2020 ) . Measuring and monitoring excess mortality across different countries requires , first and foremost , a comprehensive and regularly-updated dataset on all-cause mortality . However , there has been no single resource where such data would be collected from all over the world . The World Mortality Dataset presented here aims to fill this gap by combining publicly available information on country-level mortality , culled and harmonized from various sources . Several teams have already started to collect such data . In April 2020 , EuroStat ( http://ec . europa . eu/eurostat ) began collecting total weekly deaths across European countries , ‘in order to support the policy and research efforts related to COVID-19’ . At the time of writing , this dataset covers 36 European countries and also contains sub-national ( NUTS1–3 regions ) data as well as data disaggregated by age groups and by sex for some countries . In May 2020 , the Human Mortality Database ( http://mortality . org ) , a joint effort by the University of California , Berkeley , and Max Planck Institute for Demographic Research ( Barbieri et al . , 2015 ) , started compiling the Short Term Mortality Fluctuations ( STMF ) dataset ( STMF , 2021; Islam et al . , 2021; Németh et al . , 2021 ) . This dataset consists of weekly data , disaggregated by five age groups and by sex , and currently contains 35 countries with 2020 data . STMF only includes countries with complete high-quality vital registration data in all age groups . Both datasets are regularly updated and have considerable overlap , covering together 44 countries . In parallel , the EuroMOMO project ( https://www . euromomo . eu ) , existing since 2008 , has been displaying weekly excess mortality in 23 European countries , but without giving access to the underlying data . Another source of data is the UNDATA initiative ( http://data . un . org; search for ‘Deaths by month of death’ ) by the United Nations , collecting monthly mortality data across a large number of countries . However , information there is updated very slowly , with January–June 2020 data currently available for only four countries . Media outlets such as the Financial Times , The Economist , the New York Times , and the Wall Street Journal have been compiling and openly sharing their own datasets in order to report on the all-cause mortality in 2020 . However , these datasets are infrequently updated and their future is unclear . For example , the New York Times announced in early 2021 that they would stop tracking excess deaths due to staffing changes . Here , we present the World Mortality Dataset that aims to provide regularly-updated all-cause mortality numbers from all over the world . The dataset is openly available at https://github . com/akarlinsky/world_mortality and is updated almost daily . Our dataset builds upon the EuroStat and the STMF datasets , adding 59 additional countries — many more than any previous media or academic effort . At the time of writing , our dataset comprises 103 countries and territories . After the initial release of our manuscript , the dataset has been incorporated into the excess mortality trackers by Our World in Data ( Giattino et al . , 2020 ) , The Economist , and the Financial Times . While not all countries provide equally detailed and reliable data , we believe that information from all 103 countries is reliable enough to allow computation of excess mortality ( see Discussion ) . Our analysis ( updated almost daily at https://github . com/dkobak/excess-mortality ) showed statistically significant positive excess mortality in 69 out of 103 countries . Moreover , it suggests that the true COVID-19 death toll in several countries is over an order of magnitude larger than the official COVID-19 death count . We collected the all-cause mortality data from 103 countries and territories from 2015 onward into the openly available World Mortality Dataset . This includes 50 countries with weekly data , 51 countries with monthly data , and two countries with quarterly data ( Figure 1 ) . See Materials and methods for our data collection strategy . Briefly , we obtained the data from the websites of National Statistics Offices ( NSOs ) . If we were unable to locate the data ourselves , we contacted the NSO for guidance . The data from EuroStat and STMF were included as is , with few exceptions ( see Materials and methods ) . An important caveat is that recent ( 2020 and 2021 ) data are often preliminary and subject to backwards revisions , which we incorporate into our dataset . Other caveats and limitations are listed in the Materials and methods section . For each country , we predicted the ‘baseline’ mortality in 2020 based on the 2015–2019 data ( accounting for linear trend and seasonal variation; see Materials and methods ) . We then obtained excess mortality as the difference between the actual 2020–2021 all-cause mortality and our baseline ( Figure 2 , Figure 2—figure supplement 1 ) . For each country , we computed the total excess mortality from the beginning of the COVID-19 pandemic ( from March 2020 ) ( Table 1 ) . The total excess mortality was positive and significantly different from zero in 69 countries; negative and significantly different from zero in seven countries; not significantly different from zero ( z<2 ) in 25 countries . For South Africa and Argentina , there was no historic data available in order to assess the significance , but the increase in mortality was very large and clearly associated with COVID-19 ( Bradshaw et al . , 2021; Rearte et al . , 2021 ) . In terms of the absolute numbers , the largest excess mortality in our dataset was observed in the United States ( 640 , 000 by June 6 , 2021; all reported numbers here and below have been rounded to two significant digits ) , Brazil ( 500 , 000 by May 31 , 2021 ) , Russia ( 500 , 000 by April 30 , 2021 ) , and Mexico ( 470 , 000 by May 23 , 2021 ) ( Figure 3 ) . Note that these estimates correspond to different time points as the reporting lags differ between countries ( Table 1 ) . See Figure 3—figure supplement 1 for the same analysis using the 2020 data alone . Some countries showed statistically significant negative excess mortality , likely due to lockdown measures and social distancing decreasing the prevalence of influenza ( Kung et al . , 2021 ) , as we discuss further below . For example , Australia had −3 , 700 excess deaths , Uruguay had −2 , 200 deaths , and New Zealand had −1 , 900 deaths . In these three cases , the decrease in mortality happened during the southern hemisphere winter season ( Figure 2 ) . Similarly , Norway had −1 , 500 excess deaths , with most of this decrease happening during the 2020/21 winter season . Note that Uruguay had a large COVID outbreak in 2021 , but we only have 2020 data available at the time of writing . The statistically significant mortality decrease in Malaysia , Mongolia , and Seychelles may also be related to the lockdown and social distancing measures but does not show clear seasonality , so may possibly also be due to other factors . As the raw number of excess deaths can be strongly affected by the country’s population size , we normalized the excess mortality estimates by the population size ( Table 1 ) . The highest excess mortality per 100 , 000 inhabitants was observed in Peru ( 590 ) , followed by some Eastern European and then Latin American countries: Bulgaria ( 460 ) , North Macedonia ( 420 ) , Serbia ( 400 ) , Mexico ( 360 ) , Ecuador ( 350 ) , Lithuania ( 350 ) , Russia ( 340 ) , etc . ( Figure 3 ) . Note that many countries with severe outbreaks that received wide international media attention , such as Italy , Spain , and United Kingdom , had lower values ( Table 1 ) . The infection-fatality rate ( IFR ) of COVID-19 is strongly age-dependent ( Levin et al . , 2020; O’Driscoll et al . , 2021 ) . As the countries differ in their age structure , the expected overall IFR differs between countries . To account for the age structure , we also normalized the excess mortality estimates by the annual sum of the baseline mortality , that is the expected number of deaths per year without a pandemic event ( Table 1 ) . This relative increase , also known as a P-score ( Aron and Muellbauer , 2020 ) , was by far the highest in Latin America: Peru ( 153% ) , Ecuador ( 80% ) , Bolivia ( 68% ) , and Mexico ( 61% ) ( Figure 3 ) . These Latin American countries have much younger populations compared to the European and North American countries , which is why the excess mortality per 100 , 000 inhabitants there was lower than in several Eastern European countries , but the relative increase in mortality was higher , suggesting higher COVID-19 prevalence . That the highest relative mortality increase was observed in Peru , is in agreement with some parts of Peru showing the highest measured seroprevalence level in the world ( Álvarez-Antonio et al . , 2021 ) . For each country , we computed the ratio of the excess mortality to the officially reported COVID-19 death count by the same date . This ratio differed very strongly between countries ( Table 1 ) . Some countries had ratio below 1 , for example 0 . 7 in France and 0 . 6 in Belgium , where reporting of COVID deaths is known to be very accurate ( Sierra et al . , 2020 ) . The likely reason is that the non-COVID mortality has decreased , mostly due to the influenza suppression ( see below ) , leading to the excess mortality underestimating the true number of COVID deaths . Nevertheless , many countries had ratios above 1 , suggesting an undercount of COVID-19 deaths ( Beaney et al . , 2020 ) . At the same time , correlation between weekly reported COVID-19 deaths and weekly excess deaths was often very high ( Figure 4 ) : e . g . in Mexico ( undercount ratio 2 . 1 ) correlation was r=0 . 80 and in South Africa ( undercount ratio 2 . 7 ) it was r=0 . 93 . High correlations suggest that excess mortality during a COVID outbreak can be fully explained by COVID-19 mortality , even when the latter is strongly underreported . Peru deserves a special mention: until early June , the undercount ratio in Peru was 2 . 7 , with correlation r=0 . 87 . Peru changed the definition of reported ‘COVID deaths’ to be more inclusive and submitted backwards revisions to WHO ( Ministry of Health , 2021 ) ; as a result , the undercount ratio dropped to 1 . 0 and correlation increased to r=0 . 99 ( Figure 4 ) . This example clearly illustrates that undercount ratios above 1 . 0 primarily arise from undercounting deaths from COVID infections . See Discussion for additional considerations . Interestingly , in many countries , the undercount ratio was not constant across time . For example , the undercount ratio in Italy , Spain , Netherlands , and United Kingdom was ∼1 . 5 during the first wave ( Figure 4 ) , but decreased to ∼1 . 0 during the second wave . This decrease of the undercount ratio may be partially due to improved COVID death reporting , and partially due to the excess mortality underestimating the true COVID mortality in winter seasons due to influenza suppression . On the other hand , several countries showed very accurate reporting of the COVID-19 deaths with the undercount ratio being close to 1 . 0 ( Sierra et al . , 2020 ) from the beginning of the pandemic and up until the middle of the second wave ( e . g . Austria , Belgium , France , Germany , Slovenia , Sweden; Figure 4 ) . However , starting from December 2020 and up until March–April 2021 the excess deaths were underestimating the COVID-19 deaths in all these countries . The difference between the officially reported COVID-19 deaths and the excess deaths may correspond to the number of deaths typically caused by influenza and other infectious respiratory diseases in winter months . This difference ( computed starting from week 40 of 2020 and until week 15 of 2021 ) , as a fraction of baseline annual deaths , was in the 2 . 3–5 . 9% range ( Austria: 2 . 3% , Belgium: 5 . 9% , France: 4 . 3% , Germany: 3 . 9% , Slovenia: 3 . 5% , Sweden: 5 . 5% ) . This is in good agreement with the total negative excess deaths observed in Australia , New Zealand , Uruguay , and Norway ( −2 . 5% , −5 . 4% , −6 . 4% , −3 . 7% ) and coming mainly from the Southern and Northern hemisphere winter months respectively ( Kung et al . , 2021 ) . Per 100 , 000 inhabitants , the same difference was in the 20–60 range . The undercount ratio for most countries was below 3 . 0 ( Table 1 ) , but some countries showed much larger values . We found the highest undercount ratios in Tajikistan ( 100 ) , Nicaragua ( 51 ) , Uzbekistan ( 31 ) , Belarus ( 14 ) , and Egypt ( 13 ) ( Figure 3 ) . Such large undercount ratios strongly suggest purposeful misdiagnosing or underreporting of COVID-19 deaths , as argued by Kobak , 2021a for the case of Russia ( undercount ratio 4 . 5 ) . We presented the World Mortality Dataset — the largest international dataset of all-cause mortality , currently encompassing 103 countries . The dataset is openly available and regularly updated . We are committed to keep maintaining this dataset for the entire duration of the COVID-19 pandemic . The coverage and reliability of the data varies across countries , and some of the countries in our dataset may possibly report incomplete mortality numbers ( e . g . covering only part of the country ) , see caveats in the Data limitations and caveats section of Materials and methods . This would make the excess mortality estimate during the COVID-19 outbreak incomplete ( as an example , Lloyd-Sherlock et al . , 2021 estimate that the true excess mortality in Peru may be 30% higher than excess mortality computed here due to incomplete death registration in Peru ) . Importantly , the early pre-outbreak 2020 data for all countries in our dataset matched the baseline obtained from the historic 2015–2019 data , indicating that the data are self-consistent and the excess mortality estimates are not inflated . Another important caveat is that the most recent data points in many countries tend to be incomplete and can experience upwards revisions . Both factors mean that some of the excess mortality estimates reported here may be underestimations . Some of the countries in our dataset have excess death estimates available in the constantly evolving literature on excess deaths during the COVID-19 pandemic from academia , official institutions and professional associations . The largest efforts include the analysis of STMF data ( Kontis et al . , 2020; Islam et al . , 2021 ) and excess mortality trackers by The Economist and Financial Times . While the analysis is similar everywhere and the estimates broadly agree , there are many possible modeling choices ( the start date and the end date of the total excess computation; including or excluding historic influenza waves when computing the baseline; modeling trend over years or not , etc . ) making all the estimates slightly different . Conceptually , excess mortality during the COVID-19 pandemic can be represented as the sum of several distinct factors:Excessmortality=+ ( A ) DeathsdirectlycausedbyCOVIDinfection+ ( B ) DeathscausedbymedicalsystemcollapseduetoCOVIDpandemic+ ( C ) Excessdeathsfromothernaturalcauses+ ( D ) Excessdeathsfromunnaturalcauses+ ( E ) Excessdeathsfromextremeevents:wars , naturaldisasters , etc . We explicitly account for factor ( E ) and argue that for most countries , the contribution of factors ( B ) – ( D ) is small in comparison to factor ( A ) , in agreement with the view that excess mortality during an epidemic outbreak can be taken as a proxy for COVID-19 mortality ( Beaney et al . , 2020 ) . Below we discuss each of the listed factors . It is possible that when a country experiences a particularly strong COVID outbreak , deaths from non-COVID causes also increase due to the medical system being overloaded — factor ( B ) above . Our data show that this did not happen in Belgium ( undercount ratio during the first wave was close to 1 . 0 , i . e . all excess deaths were due to COVID-19 infections ) , despite a ∼100% weekly increase in all-cause mortality . Moreover , our data suggest that this did not happen in Peru , during one of the strongest registered COVID outbreaks in the world until now: despite ∼200% weekly increase in all-cause mortality , undercount ratio stayed close to 1 . 0 ( after Peru revised the number of reported COVID deaths , see above ) . Even if deaths due to other diseases did increase , then such collateral excess deaths can nevertheless be seen as indirect consequence of COVID-19 outbreaks . However , the available data suggest that this factor , if at all , plays only a minor role in the overall excess deaths . The data suggest that the contribution of factor ( C ) to the excess mortality is negative . Indeed , countries that implemented stringent lockdown and social distancing measures in the absence of COVID-19 community spread , such as Australia , New Zealand , and Uruguay , showed a clear winter-season decrease in all-cause mortality , likely due to reduced influenza transmission ( Kung et al . , 2021 ) . Here , using these three countries as well as some of the European data , we estimated that the influenza suppression alone can lead to a decrease of annual mortality by 3–6% . Other infectious diseases may also be suppressed by enforced social distancing . For example , in South Africa , lockdowns have noticeably decreased toddler mortality ( 0–4 years ) ( Bradshaw et al . , 2021 ) . This effect was not observed in the developed countries where toddler mortality is low ( Islam et al . , 2021 ) , but could be present in other developing countries as well . The effect of factor ( D ) appears to be country-specific . Traffic accident fatalities have decreased in the European Union and the Western Balkans ( European Commission , 2021; Transport Community , 2021 ) but have increased in the United States ( National Safety Council , 2021 ) . Homicides have increased in the United States ( Arthur and Asher , 2020; Faust et al . , 2021 ) and Germany ( Federal Criminal Police Office , 2021 ) but have decreased in Peru ( Calderon-Anyosa and Kaufman , 2021 ) , South Africa ( Bradshaw et al . , 2021 ) , and France ( Ministry of the Interior , 2021 ) . Suicides have first decreased and then increased in Japan , particularly in females ( Kurita et al . , 2021; Tanaka and Okamoto , 2021; Black and Kutcher , 2021 ) , have decreased in the United States ( Ahmad and Anderson , 2021; Faust et al . , 2021 ) , and in a study of 21 countries were found to have decreased in half of them and remained unchanged in the rest ( Pirkis et al . , 2021 ) . In the United States , deaths from drug overdoses and unintentional injuries have increased ( Faust et al . , 2021 ) . However , importantly , in all cited studies the combined effect of all changes in the frequency of unnatural changes did not exceed ∼1% of the baseline annual mortality , meaning that factor ( D ) plays only a minor role in excess mortality . Finally , factor ( E ) in 2020–2021 was mostly constrained to the Nagorno-Karabakh war between Armenia and Azerbaijan and the August 2020 heat wave in Europe , which we explicitly accounted for . We could have possibly missed some other similar events in other countries , but we believe they could only play a minor role compared to COVID-19 , thanks to the absence of other major wars or natural disasters in 2020–2021 in the countries included in our dataset . Together , the evidence suggests that the contribution of factor ( C ) is negative and contribution of factor ( D ) is small in comparison , so we believe that lockdown and social distancing measures on their own decrease — and not increase — the death rate , at least in short-term . The contribution of factor ( B ) , at least in developed countries , appears to be small , and so , in the absence of wars and natural disasters , one can expect excess mortality to provide a lower bound on the true number of COVID-19 deaths . In other words , we speculate that whenever COVID deaths are counted perfectly , they should exceed the excess mortality , leading to undercount ratio below 1 . This is indeed what we observed in several countries with strong COVID-19 outbreaks but accurate accounting of COVID deaths , for example Belgium , France , and Germany ( undercount ratios 0 . 6 , 0 . 7 , and 0 . 4 , respectively ) . The World Mortality Dataset is open for researchers and policy makers from all fields . Avenues for future research include the relation between various measures of excess mortality and economic development , population structure , lockdown and social distancing measures , border controls and travel restrictions ( Hale et al . , 2020 ) , properties of the health-care systems , vaccinations , institutional quality ( e . g . the Democracy Index ) , climate , geography , population density , and many more . Conversely , future research can use excess mortality estimates to study negative social or economic impact of high COVID-19 death toll . So far we were able to collect data from 103 nations out of ∼200 , with particularly sparse coverage in Africa , Asia , and the Middle East ( Figure 1 ) . During the pandemic , many countries have sped up collection and dissemination of preliminary all-cause mortality data , yet many other countries did not , and will report 2020 information with a substantial lag , in the coming months or even years . Once released , this information will be included in the World Mortality Dataset . Unfortunately , many countries do not keep reliable vital statistics and excess mortality may remain unknown for a long time . Summing up the excess mortality estimates across all countries in our dataset gives 4 . 0 million excess deaths . In contrast , summing up the official COVID-19 death counts gives 2 . 9 million deaths , corresponding to the global undercount ratio of 1 . 4 . However , there is ample evidence that among the countries for which the all-cause mortality data are not available the undercount ratio is much higher ( Watson et al . , 2020; Djaafara et al . , 2021; Watson et al . , 2021; Mwananyanda et al . , 2021; Koum Besson et al . , 2021; Leffler , 2021 ) . Using a statistical model to predict the excess mortality in the rest of the world based on the existing data from our dataset , The Economist in May 2021 estimated 7–13 million excess deaths worldwide ( The Economist , 2021 ) , which was 2–4 times higher than the world’s official COVID-19 death count at the time ( 3 . 5 million ) . In conclusion , the COVID-19 pandemic highlighted the great importance of reliable and up-to-date all-cause mortality data . Just as countries around the world collect and regularly report estimates of economic output such as the gross domestic product ( GDP ) , and just as they have been reporting COVID-19 mortality , they should be reporting all-cause mortality into a comprehensive multi-national repository ( Leon et al . , 2020 ) . For all countries that are not covered by EuroStat or STMF , we aimed to collect weekly , monthly , or quarterly all-cause mortality data from their National Statistics Offices ( NSOs ) , Population Registries , Ministries of Health , Ministries of Public Health , etc . , collectively referred to here as ‘NSOs’ . Our strategy was to search for mortality numbers for every country on their NSO’s website . The data may be present in the form of a spreadsheet , a table generator , a periodical bulletin , a press release , a figure that required digitizing , etc . If we were unable to locate such data , we contacted the NSO via email , a contact form on their website , or on social media , asking them if they have weekly , monthly , or quarterly data on all-cause mortality for 2020 . Responses from NSOs have varied substantially . Some have provided us with the requested information , some replied that no such data were available , some did not respond at all . For many countries , the email addresses did not work and returned an error message . Here are some representative examples of the declining responses: ‘‘We are sorry to inform you that we do not have the data you requested’’ ( China ) ; ‘‘These are not available’’ ( India ) ; ‘‘Unfortunately we don’t have this data . Currently , we only have the number of people dying from traffic accidents ( by month ) ’’ ( Vietnam ) ; ‘‘Unfortunately , we do not have a mechanism in place at the moment to capture routine mortality data in-country nation-wide […] As you may also be aware , death or mortality registration or reporting is yet a huge challenge in developing countries […]’’ ( Liberia ) . We included the data from 2015 onwards into our dataset if it satisfied the following inclusion criteria: ( 1 ) data were in weekly , monthly , or quarterly format ( we preferred weekly data whenever available ) ; ( 2 ) data existed at least until June 2020; ( 3 ) there were data for at least one entire year before 2020 ( or a forecast for 2020 , see below ) . At the time of writing , our dataset comprises 103 countries and territories ( Figure 1 ) . For the weekly data , we preferred ISO weeks whenever possible ( for Peru , Sweden , Ecuador , and Guatemala we converted daily data into ISO weeks ) . For Liechtenstein and Taiwan we preferred monthly format over weekly data from STMF/EuroStat , because weekly data were very noisy or less up-to-date . The data for Iran are available in quarterly format , where quarters start on December 21 , March 21 , June 21 , and September 21 ( Solar Hijri seasons ) . We treated the season starting on December 21 as the first data point for the following year . Unlike STMF ( Islam et al . , 2021 ) , we only collected country-level data , without age or gender stratification , since for most countries this information was not available . In some cases , we had to combine several data sources , for example taking the 2015–2018 monthly data from UNDATA and 2019–2020 monthly data from a country’s NSO . For two countries ( Gibraltar and Nicaragua ) some of the values were taken from media reports which in turn obtained them from the respective NSOs . Some data points for Cuba and Uruguay were taken from Castanheira et al . , 2021 and the data for Argentina from Rearte et al . , 2021 . A detailed description of all data sources for each country can be found at https://github . com/akarlinsky/world_mortality . In this manuscript , we treat Taiwan , Hong Kong , and Macao as separate countries . They release monthly all-cause mortality data , whereas China does not . We also treat Gibraltar , Greenland , and Transnistria as separate territories as the United Kingdom , Denmark , and Moldova do not report these deaths in their figures . Similarly , the French overseas departments of French Guiana , Guadeloupe , Martinique , Mayotte , and Réunion , the French overseas collectivity of French Polynesia , and Aruba , a constituent country of the Kingdom of the Netherlands , are included as separate territories as France and Netherlands do not report these deaths in their figures . The data in our dataset come with several important caveats . First , the 2020–2021 data are often preliminary and subject to backward revisions . The more recent the data point , the more incomplete it usually is . Some countries only publish complete data ( with a substantial delay ) while others release very early and incomplete data as well . We excluded the most recent data points whenever there was an indication that the data were substantially incomplete . For the United States , we used the ‘weighted’ mortality counts from the Centers for Disease Control and Prevention ( CDC ) that account for undercount in recent weeks , instead of the STMF data . Second , the completeness and reliability of all-cause mortality data varies by country . According to the United Nations Demographic Yearbook ( UNSD , 2019 ) , 83 of the countries in our dataset have a death registration coverage rate of 90% and above . In the remaining 20 countries , coverage is either estimated to be below 90% ( e . g . Peru , Ecuador , Bolivia ) or no estimate exists at all ( Kosovo , Taiwan , Transnistria ) . However , some of the available coverage estimates are outdated . For example , the estimate for Bolivia is from 2000 . The estimate for Peru is from 2015 , that is before the SINADEF reform in 2016 ( Vargas-Herrera et al . , 2018 ) which has likely substantially improved the coverage . For Taiwan , Human Mortality Database estimates that the data are over 99% complete . At the same time , note that the coverage estimates refer to the finalized data so preliminary 2020–21 data may be less complete , as explained above . It is also possible that COVID-19 pandemic could have affected the quality of the vital registration . Third , whereas we preferred data by date of death , the data from many countries are only available by the date of registration . Most weekly data are by the date of death ( one notable exception is United Kingdom ) , but monthly data are often by the date of registration . Whenever the data are organized by the date of registration , it will show spurious drops in weeks or months with public holidays ( e . g . last week of August and last week of December in the United Kingdom ) or during national lockdowns ( e . g . April 2020 in Kyrgyzstan , Kazakhstan , and Panama ) . Fourth , we aimed to collect information from all countries from 2015 onward , yet currently we only have later data for four countries: Chile ( 2016 ) , Germany ( 2016 ) , Transnistria ( 2016 ) , Peru ( 2017 ) . Two other countries , South Africa and Argentina , did not release any pre-2020 data at all , but instead published a forecast for 2020 based on the prior data ( Bradshaw et al . , 2021; Rearte et al . , 2021 ) . We included this published forecast into the dataset as year 0 . Fifth , for most countries , the data are provided as-is , but for three countries ( Brazil , Lebanon , and Sweden ) we performed some processing to assure consistency across years , resulting in non-integer values . In Brazil , there are two mortality monitoring systems: ‘Registro Civil’ ( RC ) and ‘Sistema de Informaçāo sobre Mortalidade’ ( SIM ) . RC is more up-to-date , whereas SIM is more complete . We used the SIM data up until October 2020 and RC data afterwards , multiplied by the ratio between total January–October 2020 deaths in SIM and in RC ( 1 . 08 ) . Lebanon has reported total deaths from 2015 to 2019 and hospital deaths from 2017 to 2021 . Total deaths in 2020–2021 were estimated by multiplying the hospital deaths by the ratio between total deaths in 2019 to hospital deaths in 2019 ( 1 . 34 ) . Sweden has a substantial number of deaths ( 2 . 9% of all deaths in 2019; 2 . 7% in 2020 ) reported with an ‘unknown’ week . However , ∼95% of these have a known month of death . In order to account for this , we redistributed deaths with known month but unknown week uniformly across weeks of the respective month , and similarly redistributed the remaining deaths with known year but unknown month . Despite the caveats and limitations listed above , all our data are self-consistent: the baseline mortality that we predict for 2020 agrees very well with the pre-COVID early 2020 mortality in all cases . Note that our projection for 2020 uses a linear trend ( see below ) and so can implicitly account for improvements in death registration over the recent years . We therefore believe that for countries with incomplete death registration coverage , our excess mortality estimates provide a lower bound to the true excess mortality . In order to estimate the excess mortality , we first estimated the expected , or baseline , mortality for 2020 using the historical data from 2015 to 2019 ( or as many years from this interval as were available; see above ) . We fitted the following regression model separately for each country: ( 1 ) Dt , Y=αt+β⋅Y+ϵ . Here , Dt , Y is the number of deaths observed on week ( or month , or quarter ) t in year Y , β is a linear slope across years , αt are separate intercepts ( fixed effects ) for each week ( month/quarter ) , and ϵ∼𝒩⁢ ( 0 , σ2 ) is Gaussian noise . This model can capture both seasonal variation in mortality and a yearly trend over recent years due to changing population structure or socio-economic factors . As an example , using monthly death data from Russia ( R2=0 . 72 , F=10 . 2 ) , we obtained β^=-2346±528 ( ± standard error ) , meaning that each year the number of monthly deaths decreases on average by ∼2300 , and so the predicted monthly deaths for 2020 are ∼7000 lower than the 2015–2019 average . In contrast , using weekly data from the United States ( R2=0 . 89 , F=31 . 7 ) , we obtained β^=773±57 , meaning that each year the number of weekly deaths increases on average by ∼800 . In these two cases , as well as in many others , the yearly trend was strong and statistically significant , and using the average 2015–2019 data as baseline , as is sometimes done , would therefore not be appropriate . We took the model prediction for 2020 as the baseline for excess mortality calculations: ( 2 ) B^t=α^t+β^⋅2020 . For the countries with weekly data , the model was fit using weeks 1–52 , as the week 53 only happens in rare years ( including 2020 ) . The baseline for week 53 was then taken as equal to the value obtained for week 52 . We took the same baseline for 2021 as for 2020 , to avoid further extrapolation . The excess mortality in each week ( or month , or quarter ) was defined as the difference between the actually observed death number and the baseline prediction . Note that the excess mortality can be negative , whenever the observed number of deaths is below the baseline . We summed the excess mortality estimates across all weeks starting from March 2020 ( week 10; for monthly data , we started summation from March 2020; for quarterly data , from the beginning of 2020 ) . This yields the final estimate of the excess mortality: ( 3 ) Δ=∑t≥t1 ( Dt , 2020-B^t ) +∑t ( Dt , 2021-B^t ) , where t1 denotes the beginning of summation in 2020 . We computed the variance Var[Δ] of our estimator Δ as follows . Let 𝐗 be the predictor matrix in the regression , 𝐲 be the response vector , 𝜷^= ( 𝐗⊤⁢𝐗 ) -1⁢𝐗⊤⁢𝐲 be the vector of estimated regression coefficients , and σ^2=∥𝐲-𝐗⁢𝜷^∥2/ ( n-p ) be the unbiased estimate of the noise variance , where n is the sample size and p is the number of predictors . Then Cov[β^]=σ^2 ( X⊤X ) −1 is the covariance matrix of 𝜷^ and S=Cov[B^t]=Cov[X2020β^]=σ^2X2020 ( X⊤X ) −1X2020⊤ is the covariance matrix of the predicted baseline values B^t where 𝐗2020 is the predictor matrix for the entire 2020 . We introduce vector 𝐰 with elements wt of length equal to the number of rows in 𝐗2020 , set all elements before t1 to zero , all elements starting from t1 to 1 , and increase by one all elements corresponding to the existing 2021 data . Then the ‘predictive’ variance of Δ is given by ( 4 ) Var[Δ]=Var[∑twtB^t]+∑twtσ^2=w⊤Sw+σ^2‖w‖1 , where the first term corresponds to the uncertainty of B^t and the second term corresponds to the additive Gaussian noise that 2020–2021 observations would have had on top of Bt without the pandemic event ( Abramovich and Ritov , 2013 ) . We took the square root of Var[Δ] as the standard error of Δ . Whenever the fraction z=\|Δ\|/Var[Δ] was below 2 , we considered the excess mortality for that country to be not significantly different from zero . Note that we could not estimate the uncertainty for Argentina and South Africa because raw historical data were not available ( see above ) . There exist more elaborate statistical approaches for estimating the baseline ( and thus the excess ) mortality , for example modeling the seasonal variation using periodic splines or Fourier harmonics , or controlling for the time-varying population size and age structure , or using a Poisson model ( Farrington et al . , 1996; Noufaily et al . , 2013 ) , etc . We believe that our method achieves the compromise between flexibility and simplicity: it is the simplest approach that captures both the seasonal variation and the yearly trend , and is far more transparent than more elaborate methods . Note that our uncertainty estimation assumes iid noise in Equation 1 . In reality , the noise may be temporally or spatially autocorrelated , which would affect the variance of B^t . Past ( 2015–2019 ) influenza outbreaks contributed to the estimation of the baseline B^t . As a consequence , our baseline captures the expected mortality without the COVID-19 pandemic , but in the presence of usual seasonal influenza . This differs from the approach taken by EuroMomo as well as by some studies of excess mortality due to influenza pandemics ( Viboud et al . , 2005; Simonsen et al . , 2013 ) , where the baseline is constructed in a way that weighs down previous influenza outbreaks so that each new outbreak would result in positive excess mortality . A parallel work on COVID-19 excess mortality based on the STMF dataset ( Islam et al . , 2021 ) also used that approach , which explains some of the differences between our estimates . We subtracted 4000 from the excess mortality estimates for Armenia and Azerbaijan to account for the 2020 Nagorno-Karabakh war . By official counts , it cost ∼3400 lives in Armenia and ∼2800 in Azerbaijan ( Welt and Bowen , 2021 ) , but we took 4000 deaths in each country to obtain a conservative estimate of COVID-related excess mortality . To the best of our knowledge , no other armed conflict in 2020–2021 resulted in more than 100 casualties in countries included in our dataset . Another correction was done for Belgium , Netherlands , France , Luxembourg , and Germany , where our data show a peak of excess deaths in August 2020 , not associated with COVID-19 ( see below and Figure 4 ) and likely corresponding to a heat wave ( Fouillet et al . , 2006; Fouillet et al . , 2008; Flynn et al . , 2005 ) . We excluded weeks 32–34 from the excess mortality calculation in these five countries . This decreased the excess mortality estimates for these countries by 1500 , 660 , 1600 , 35 , and 3700 , respectively . The EM-DAT database of natural disasters ( https://www . emdat . be ) lists only the following four natural disasters with over 200 fatalities in 2020–2021 in countries included in our dataset: the August heat wave in Belgium , France , and the Netherlands , and a sequence of heat waves in the United Kingdom in June–August 2020 ( 2500 casualties ) . We do not see clear peaks in our data associated with these heat waves , possibly because our United Kingdom data are by the date of registration and not by the date of death . We have therefore chosen not to adjust our excess mortality estimate for the United Kingdom . Note that other countries may also have experienced non-COVID-related events leading to excess mortality . However , as these events are not included in the EM-DAT database , we assume that their effect would be small in comparison to the effect of COVID-19 . For example , Russian data suggest ∼10 , 000 excess deaths from a heat wave in July 2020 in Ural and East Siberia ( Kobak , 2021a ) . We do not correct for it in this work as it is difficult to separate July 2020 excess deaths into those due to COVID and those due to the heat wave , based on the Russian country-level data alone , and we are not aware of any reliable published estimates . Importantly , 10 , 000 is only a small fraction of the total number of excess deaths in Russia . Another case is the February 2021 power crisis in Texas , USA , that has been estimated by BuzzFeed News to have yielded ∼700 excess deaths ( Aldhous et al . , 2021 ) . Again , this number is small compared to the total number of excess deaths in the United States . We took the officially reported COVID-19 death counts from the World Health Organization ( WHO ) dataset ( https://covid19 . who . int ) . To find the number of officially reported COVID-19 deaths at the time corresponding to our excess mortality estimate , we assumed that all weekly data conform to the ISO 8601 standard , and took the officially reported number on the last day of the last week available in our dataset . Some countries use non-ISO weeks ( e . g . starting from January 1st ) , but the difference is at most several days . ISO weeks are also assumed in the ‘Data until’ column in Table 1 . Officially reported numbers for Hong Kong , Macao , and Taiwan , absent in the WHO dataset , were taken from the Johns Hopkins University ( JHU ) dataset ( https://coronavirus . jhu . edu ) ( Dong et al . , 2020 ) as distributed by Our World in Data . We manually added officially reported numbers for Transnistria ( 1 , 195 by the end of May 2021: taken from the Telegram channel https://t . me/novostipmrcom ) . Note that for some countries there exist different sources of official data , e . g . Russia officially reports monthly numbers of confirmed and suspected COVID deaths that are substantially larger than the daily reported numbers ( Kobak , 2021a ) . However , it is the daily reported numbers that get into the WHO and JHU dashboards , so for consistency , here we always use the daily values . We defined undercount ratio as the number of excess deaths divided by the official number of COVID-19 deaths reported by the same date . If the number of excess deaths is negative , the undercount ratio is not defined . Additionally , we chose not to show undercount ratios for countries with positive number of excess deaths where there is no evidence that excess deaths were due to COVID ( Cuba , Hong Kong , Thailand ) . For these three countries there was no correlation between the monthly excess deaths and reported monthly numbers of COVID-19 deaths or cases , and no media reports of COVID outbreaks . To estimate excess deaths per 100 , 000 population , we obtained population size estimates for 2020 from the United Nations World Population Prospect ( WPP ) dataset ( https://population . un . org/wpp/ ) . The value for Russia in that dataset does not include Crimea due to its disputed status , but all Russian data of all-cause and COVID-19 mortality does include Crimea . For that reason , we used the population value of 146 , 748 , 590 , provided by the Russian Federal State Statistics Service , and similarly changed the value for Ukraine to 41 , 762 , 138 , provided by the Ukranian State Statistics Service . The number for Transnistria was absent in the World Population Prospect dataset , so we used the value obtained from its NSO ( 465 , 200 ) . Additionally , WPP reports population data for Serbia and Kosovo combined . We thus obtained population values for these countries from the World Bank Dataset ( https://data . worldbank . org/indicator/SP . POP . TOTL ) . Note that some of the population size estimates in the World Population Prospect dataset may be outdated or unreliable . Therefore , for some of the countries our excess death rates may be only approximate ( Spoorenberg , 2020 ) . The World Mortality Dataset is available at https://github . com/akarlinsky/world_mortality , ( copy archived at swh:1:rev:03534f5db091e7dbada157e4eb92d663b1d1287f , Karlinsky and Kobak , 2021 ) . The analysis code is available at https://github . com/dkobak/excess-mortality ( copy archived at swh:1:rev:f765cf8bb7d3246bed22f85c832a63b0cf58b904 , Kobak , 2021b ) . Our baseline estimates for all countries and all values shown in Table 1 are available there as CSV files . Frozen data , data sources and code for the paper are available at https://github . com/dkobak/excess-mortality/tree/main/elife2021 ( data update from July 3 , 2021 ) .	Countries around the world reported 4 . 2 million deaths from SARS-CoV-2 ( the virus that causes COVID-19 ) from the beginning of pandemic until the end of July 2021 , but the actual number of deaths is likely higher . While some countries may have imperfect systems for counting deaths , others may have intentionally underreported them . To get a better estimate of deaths from an event such as a pandemic , scientists often compare the total number of deaths in a country during the event to the expected number of deaths based on data from previous years . This tells them how many excess deaths occurred during the event . To provide a more accurate count of deaths caused by COVID-19 , Karlinsky and Kobak built a database called the World Mortality Dataset . It includes information on deaths from all causes from 103 countries . Karlinsky and Kobak used the database to compare the number of reported COVID-19 deaths reported to the excess deaths from all causes during the pandemic . Some of the hardest hit countries , including Peru , Ecuador , Bolivia , and Mexico , experienced over 50% more deaths than expected during the pandemic . Meanwhile , other countries like Australia and New Zealand , reported fewer deaths than normal . This is likely because social distancing measures reduced deaths from infections like influenza . Many countries reported their COVID-19 deaths accurately , but Karlinsky and Kobak argue that other countries , including Nicaragua , Russia , and Uzbekistan , underreported COVID-19 deaths . Using their database , Karlinsky and Kobak estimate that , in those countries , there have been at least 1 . 4 times more deaths due to COVID-19 than reported – adding over 1 million extra deaths in total . But they note that the actual number is likely much higher because data from more than 100 countries were not available to include in the database . The World Mortality Dataset provides a more accurate picture of the number of people who died because of the COVID-19 pandemic , and it is available online and updated daily . The database may help scientists develop better mitigation strategies for this pandemic or future ones .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "epidemiology", "and", "global", "health", "tools", "and", "resources" ]	2021	Tracking excess mortality across countries during the COVID-19 pandemic with the World Mortality Dataset
Enhancers play a central role in cell-type-specific gene expression and are marked by H3K4me1/2 . Active enhancers are further marked by H3K27ac . However , the methyltransferases responsible for H3K4me1/2 on enhancers remain elusive . Furthermore , how these enzymes function on enhancers to regulate cell-type-specific gene expression is unclear . In this study , we identify MLL4 ( KMT2D ) as a major mammalian H3K4 mono- and di-methyltransferase with partial functional redundancy with MLL3 ( KMT2C ) . Using adipogenesis and myogenesis as model systems , we show that MLL4 exhibits cell-type- and differentiation-stage-specific genomic binding and is predominantly localized on enhancers . MLL4 co-localizes with lineage-determining transcription factors ( TFs ) on active enhancers during differentiation . Deletion of Mll4 markedly decreases H3K4me1/2 , H3K27ac , Mediator and Polymerase II levels on enhancers and leads to severe defects in cell-type-specific gene expression and cell differentiation . Together , these findings identify MLL4 as a major mammalian H3K4 mono- and di-methyltransferase essential for enhancer activation during cell differentiation . Enhancers are gene regulatory elements critical for cell-type-specific gene expression in eukaryotes ( Bulger and Groudine , 2011 ) . Lineage-determining or signal-dependent transcription factors ( TFs ) bind to enhancers and recruit chromatin modifying and remodeling enzymes and the Mediator coactivator complex to initiate RNA polymerase II ( Pol II ) -mediated transcription at promoters ( Roeder , 2005 ) . Recent genome-wide analyses of histone modifications , binding of transcription coactivators p300 and MED1 and lineage-determining TFs , coupled with functional assays of gene regulatory elements , have enabled the identification of enhancer chromatin signatures ( Heintzman et al . , 2007; Lupien et al . , 2008; Wang et al . , 2008; Heintzman et al . , 2009; Visel et al . , 2009; Creyghton et al . , 2010; Ghisletti et al . , 2010; Heinz et al . , 2010; Ernst et al . , 2011; Rada-Iglesias et al . , 2011 ) . In contrast to active promoters that are marked by H3K4me3 , enhancers are marked by H3K4me1 and H3K4me2 ( H3K4me1/2 ) but little H3K4me3 , and are often bound by cell-type-specific TFs . Active enhancers are further marked by histone acetyltransferases CBP/p300-mediated H3K27ac ( Creyghton et al . , 2010; Rada-Iglesias et al . , 2011 ) . H3K4me1 on enhancers often precedes H3K27ac and activation of enhancers . A large number of recent publications have validated the predictive power of such enhancer chromatin signatures in identification of novel enhancers critical for cell-type-specific gene expression and cell differentiation ( reviewed in Calo and Wysocka , 2013 ) . In yeast , the Set1 complex is the sole H3K4 methyltransferase . Through the enzymatic subunit SET1 , it is responsible for all mono- , di- and tri-methylations on H3K4 ( Ruthenburg et al . , 2007 ) . In Drosophila , there are three Set1-like H3K4 methyltransferase complexes that use dSet1 , Trithorax ( Trx ) and Trithorax-related ( Trr ) as the respective enzymatic subunits ( Mohan et al . , 2011 ) . Mammals possess six Set1-like H3K4 methyltransferase complexes ( Figure 1—figure supplement 1A ) . Based on the sequence homology of enzymatic subunits and the subunit compositions ( Cho et al . , 2007; Ruthenburg et al . , 2007; Vermeulen and Timmers , 2010; Cho et al . , 2012 ) , they fall into three sub-groups: MLL1/MLL2 ( also known as KMT2A and KMT2B , respectively ) , MLL3/MLL4 ( MLL3 is also known as KMT2C; MLL4 is also known as KMT2D , ALR and sometimes MLL2 ) , and SET1A/SET1B ( also known as KMT2E and KMT2F , respectively ) , which correspond to Drosophila Trx , Trr and dSet1 complexes , respectively . dSet1 is responsible for the bulk of H3K4me3 in Drosophila ( Ardehali et al . , 2011; Mohan et al . , 2011 ) . Consistently , depletion of the unique CFP1 subunit of SET1A/SET1B complexes in mammalian cells markedly decreases global H3K4me3 level , suggesting that SET1A/SET1B are the major H3K4 tri-methyltransferases in mammals ( Clouaire et al . , 2012 ) . In contrast , knockdown of Trr , the Drosophila homolog of MLL3/MLL4 , decreases global H3K4me1 levels , indicating that Trr regulates H3K4me1 in Drosophila ( Ardehali et al . , 2011; Mohan et al . , 2011 ) . However , the histone methyltransferases ( HMTs ) responsible for H3K4me1/2 on mammalian enhancers remain elusive . Further , the functions of these H3K4 mono-/di-methyltransferases on enhancers and in regulating cell-type-specific gene induction and cell differentiation are unclear . Finally , how these HMTs are recruited to enhancers needs to be clarified ( Calo and Wysocka , 2013 ) . Adipogenesis and myogenesis are robust and synchronized models of cell differentiation . Differentiation of preadipocytes towards adipocytes , that is adipogenesis , is regulated by a network of sequentially expressed adipogenic TFs ( Rosen and MacDougald , 2006 ) . Peroxisome Proliferator-Activated Receptor-γ ( PPARγ ) is considered the master regulator of adipogenesis and controls adipocyte gene expression cooperatively with CCAAT/enhancer-binding protein-α ( C/EBPα ) ( Rosen et al . , 2002; Lefterova et al . , 2008 ) . The early adipogenic TF C/EBPβ marks a large number of TF ‘hotspots’ before induction of adipogenesis . C/EBPβ not only controls the induction of PPARγ and C/EBPα expression but also acts as a pioneer factor to facilitate the genomic binding of PPARγ , C/EBPα and other adipogenic TFs during adipogenesis ( Siersbaek et al . , 2011 ) . Adipogenesis in cell culture is synchronized , with the vast majority of cells in the confluent population differentiating into adipocytes within 6–8 days , thus providing a robust model system for studying transcriptional and epigenetic regulation of gene expression during cell differentiation ( Ge , 2012 ) . Myogenesis is another robust model system for cell differentiation . Ectopic expression of the myogenic TF MyoD in fibroblasts and preadipocytes is sufficient to induce muscle differentiation program characterized by expression of myogenesis markers such as Myogenin ( Myog ) and Myosin ( Tapscott et al . , 1988; Lassar et al . , 1991 ) . Using adipogenesis and myogenesis as model systems , here we show MLL4 is partially redundant with MLL3 and is required for cell differentiation and cell-type-specific gene expression . By ChIP-Seq analyses , we observe cell-type- and differentiation-stage-specific genomic binding of MLL4 . MLL4 is mainly localized on enhancers and co-localizes with lineage-determining TFs on active enhancers during differentiation . We demonstrate that MLL4 is partially redundant with MLL3 and is a major H3K4 mono- and di-methyltransferase in mouse and human cells . Furthermore , MLL4 is required for H3K4me1/2 , H3K27ac , Mediator and Pol II levels on active enhancers , indicating that MLL4 is required for enhancer activation . Finally , we provide evidence to suggest that lineage-determining TFs recruit and require MLL4 to establish cell-type-specific enhancers . Among the six SET1-like H3K4 methyltransferases found in mammals , we initially knocked out Mll3 and Mll4 individually in mice using gene trap approaches ( Figure 1—figure supplement 1A–E and data not shown ) . Mll3 knockout ( KO ) mice died around birth with no obvious morphological abnormalities in embryonic development . Mll4 KO mice showed early embryonic lethality around E9 . 5 . We then generated Mll4 conditional KO mice ( Mll4f/f ) ( Figure 1A–B ) , which were viable and allowed us to investigate MLL4 function in a tissue-specific manner . Conditional KO of Mll4 was also verified in cell culture . Deletion of Mll4 led to the disruption of MLL4 complex in cells ( Figure 1—figure supplement 2A-B ) . 10 . 7554/eLife . 01503 . 003Figure 1 . MLL4 is required for brown adipose tissue and muscle development . ( A and B ) Generation of Mll4 conditional KO mice ( Mll4f/f ) . ( A ) Schematic representation of mouse Mll4 wild-type ( WT ) allele , targeted allele , conditional KO ( flox ) allele and KO allele . In the targeted allele , a single loxP site was inserted in the intron before exon 16 . A neomycin ( neo ) selection cassette flanked by FRT sites and the second loxP site was inserted in the intron after exon 19 . The locations of PCR genotyping primers P1 , P2 , and P3 are indicated by arrows . ( B ) PCR genotyping of cell lines using mixtures of P2 + P3 or P1 + P3 primers . The genotypes are indicated at the top . ( C ) Genotype of E18 . 5 embryos isolated from crossing Mll4f/+;Myf5-Cre with Mll4f/f mice . The expected ratios of the four genotypes are 1:1:1:1 . Mll4f/f;Myf5-Cre mice died immediately after cesarean section because of breathing malfunction due to defects in muscles of the rib cage . ( D ) Representative pictures of E18 . 5 embryos of the indicated genotypes . ( E ) E18 . 5 embryos were sagittally sectioned along the midline . The sections of the cervical/thoracic area indicated in the schematic were stained with H&E ( upper panels ) or with antibodies recognizing BAT marker UCP1 ( green ) and skeletal muscle marker Myosin ( red ) ( lower panels ) . Scale bar = 800 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01503 . 00310 . 7554/eLife . 01503 . 004Figure 1—figure supplement 1 . Generation of Mll3 and Mll4 whole body KO mice . ( A ) Protein domains in yeast SET1 and the homologous mouse SET1-like H3K4 methyltransferases . ( B–E ) Generation of Mll3 and Mll4 whole body KO mice . Wild-type ( WT ) and KO alleles of Mll3 and Mll4 genomic loci are shown in ( B ) and ( D ) , respectively . The exons flanking the gene trap vectors are numbered . Mll3 KO mice are perinatal lethal ( C ) , while Mll4 KO mice are early embryonic lethal ( E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01503 . 00410 . 7554/eLife . 01503 . 005Figure 1—figure supplement 2 . Confirmation of Mll4 deletion . Immortalized Mll3−/−Mll4f/f brown preadipocytes were infected with adenoviral GFP or Cre . ( A ) Genome browser view of RNA-Seq analysis on Mll4 gene locus . The targeted exons 16–19 are highlighted in red box . ( B ) Deletion of Mll4 disrupts MLL4 complex in cells . Nuclear extracts were incubated with MLL4 , UTX , PTIP or PA1 antibody . The immunoprecipitates were analyzed by Western blotting using antibodies indicated on the right . DOI: http://dx . doi . org/10 . 7554/eLife . 01503 . 005 To understand the role of MLL4 in adipogenesis and myogenesis , we generated Mll4f/f;Myf5-Cre mice by crossing Mll4f/f with Myf5-Cre mice ( Tallquist et al . , 2000 ) . Myf5-Cre specifically deletes genes flanked by loxP sites in somitic precursor cells that give rise to both brown adipose tissue ( BAT ) and skeletal muscle in the back ( Seale et al . , 2008 ) . Mll4f/f;Myf5-Cre pups and E18 . 5 embryos were obtained at the expected Mendelian ratio but displayed marked reduction in back muscles and died immediately after birth due to breathing malfunction that is controlled by muscle groups in the rib cage ( Figure 1C–D and data not shown ) . Sagittal sections along the midline of E18 . 5 embryos were subjected to immunohistochemistry analysis using antibodies against BAT marker UCP1 and muscle marker Myosin ( Figure 1E ) . Compared with littermate controls and Mll3 KO , Mll4f/f;Myf5-Cre embryos showed marked decreases of BAT and muscle mass , suggesting that MLL4 is required for adipogenesis and myogenesis . To understand how MLL4 regulates BAT development , we induced adipogenesis in brown preadipocytes in culture . KO of Mll4 led to a moderate differentiation defect along with a transient up-regulation of Mll3 expression in the early phase of adipogenesis whereas KO of Mll3 had no effect on adipogenesis ( Figure 2—figure supplement 1A-F ) , suggesting a more prominent role of MLL4 in development and a partial compensation of MLL4 loss by MLL3 . To eliminate the compensatory effect , we isolated primary brown preadipocytes from E18 . 5 Mll3−/−Mll4f/f embryos . After SV40T immortalization ( Wang et al . , 2010 ) , cells were infected with adenoviral Cre ( Ad-Cre ) or GFP ( Ad-GFP ) to generate Mll3−/−Mll4−/− and Mll3−/− cells , respectively . Deletion of Mll4 by Cre from the immortalized Mll3−/−Mll4f/f brown preadipocytes had little effect on cell growth ( Figure 2—figure supplement 1G-H ) , but led to severe defects in adipogenesis and associated expression of adipogenesis markers and key regulators Pparg and Cebpa as well as brown adipocyte markers Prdm16 and Ucp1 ( Figure 2A–B and Figure 2—figure supplement 1I ) , indicating that MLL3 and MLL4 are functionally redundant and are essential for adipogenesis . Furthermore , ectopic expression of PPARγ or C/EBPβ failed to rescue adipogenesis in Mll3/Mll4 double KO preadipocytes ( Figure 2—figure supplement 1J and Figure 2C ) . Consistent with the results from brown preadipocytes , knockdown of Mll4 in 3T3-L1 white preadipocytes inhibited adipogenesis ( Figure 2—figure supplement 1K–M ) . Furthermore , MLL3 and MLL4 were also required for PPARγ-stimulated adipogenesis in mouse embryonic fibroblasts ( MEFs ) ( Figure 2—figure supplement 1N–P ) . 10 . 7554/eLife . 01503 . 006Figure 2 . MLL4 controls induction of cell-type-specific genes during differentiation . ( A–E ) Adipogenesis of Mll3−/−Mll4f/f brown preadipocytes . ( A and B ) MLL4 is required for adipogenesis . Immortalized Mll3−/−Mll4f/f brown preadipocytes were infected with adenoviral GFP or Cre , followed by adipogenesis assay . ( A ) 6 days after induction of differentiation , cells were stained with Oil Red O . Upper panels , stained dishes; lower panels , representative fields under microscope . ( B ) qRT-PCR of Mll4 , Pparg and Cebpa expression at indicated time points of adipogenesis . Quantitative PCR data in all figures are presented as means ± SD . D1 , day 1 . ( C ) MLL4 is required for C/EBPβ- and PPARγ-stimulated adipogenesis . Mll3−/−Mll4f/f brown preadipocytes were infected with retroviruses expressing vector ( vec ) , C/EBPβ or PPARγ . After hygromycin selection , cells were infected with adenoviral GFP or Cre , followed by adipogenesis assay . ( D–E ) MLL4 is required for induction of cell-type-specific genes during adipogenesis . Adipogenesis was done as in ( A ) . Cells were collected before ( day 0 ) and during ( day 2 ) adipogenesis for RNA-Seq . ( D ) Schematic of identification of MLL4-dependent and -independent up-regulated genes during adipogenesis . The threshold for up- or down-regulation is 2 . 5-fold . ( E ) Gene ontology ( GO ) analysis of gene groups defined in ( D ) . ( F–J ) MLL4 is required for MyoD-stimulated myogenesis . Immortalized Mll3−/−Mll4f/f brown preadipocytes were infected with retroviruses expressing Vec or MyoD . After hygromycin selection , cells were infected with adenoviral GFP or Cre , followed by myogenesis assay . ( F ) Western blot analysis of MyoD expression before differentiation . RbBP5 was used as a loading control . The asterisk indicates a non-specific band . ( G ) 5 days after induction of differentiation , cell morphologies were observed under microscope . ( H ) qRT-PCR analysis of myogenic gene expression after differentiation . ( I and J ) MLL4 is required for induction of cell-type-specific genes during myogenesis . Brown preadipocytes and myocytes were collected for RNA-Seq . ( I ) Schematic of identification of MLL4-dependent and -independent up-regulated genes during myogenesis . The threshold for up- or down-regulation is 2 . 5-fold . ( J ) GO analysis of gene groups defined in ( I ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01503 . 00610 . 7554/eLife . 01503 . 007Figure 2—figure supplement 1 . MLL4 is required for adipogenesis . ( A–C ) MLL3 is dispensable for adipogenesis of brown preadipocytes . Brown preadipocytes isolated from Mll3+/+ and Mll3−/− E18 . 5 embryos were immortalized by retroviruses expressing SV40T . Cells were induced for adipogenesis for 6–7 days . ( A ) qRT-PCR analysis of Mll3 expression before adipogenesis . ( B ) Oil Red O staining after adipogenesis . ( C ) qRT-PCR analysis of gene expression before ( D0 ) and after ( D6 ) adipogenesis . ( D–F ) Single KO of Mll4 leads to reduced adipogenesis . ( D ) qRT-PCR analysis of Cre-mediated Mll4 deletion before adipogenesis . ( E ) Oil Red O staining after adipogenesis . ( F ) qRT-PCR analysis of gene expression at indicated time points during adipogenesis . ( G and H ) Deletion of Mll4 does not affect growth of immortalized preadipocytes . ( G ) qRT-PCR analysis of Mll4 deletion in Cre-infected Mll3−/−Mll4f/f preadipocytes . ( H ) Cell growth curves . ( I ) qRT-PCR of BAT-specific Prdm16 and Ucp1 expression during adipogenesis . ( J ) Western blot analyses of retroviral C/EBPβ and PPARγ expression in adenoviral GFP- or Cre-infected Mll3−/−Mll4f/f brown preadipocytes before adipogenesis . RbBP5 was used as a loading control . ( K–M ) Knockdown of Mll4 inhibits adipogenesis of 3T3-L1 white preadipocytes . 3T3-L1 cells were infected with lentivirus shRNA targeting Mll4 or control ( Con ) virus , followed by adipogenesis assay . ( K ) qRT-PCR confirmation of Mll4 knockdown before adipogenesis . ( L ) Oil red O staining after adipogenesis . ( M ) qRT-PCR analysis of gene expression before ( D0 ) and after ( D6 ) adipogenesis . ( N–P ) MLL4 is required for PPARγ-stimulated adipogenesis of MEFs . 3T3-immortalized Mll3−/−Mll4f/f MEFs were infected with retroviruses MSCVhyg-PPARγ , followed by infection of MSCVpuro-Cre . Adipogenesis was induced with PPARγ ligand Rosiglitasone ( Rosi ) or vehicle DMSO alone . ( N ) Western blot analysis of retroviral PPARγ expression before adipogenesis . GAPDH was used as a loading control . ( O ) Oil red O staining and ( P ) qRT-PCR analysis of gene expression after adipogenesis . DOI: http://dx . doi . org/10 . 7554/eLife . 01503 . 00710 . 7554/eLife . 01503 . 008Figure 2—figure supplement 2 . Confirmation of RNA-Seq data by qRT-PCR . DOI: http://dx . doi . org/10 . 7554/eLife . 01503 . 008 To understand how MLL4 regulates myogenesis , we used retroviral vector to stably express ectopic MyoD in Mll3−/−Mll4f/f brown preadipocytes . Cells were then infected with adenoviral Cre or GFP , followed by induction of myogenesis ( Figure 2F–H ) . We found that deletion of Mll4 from Mll3−/−Mll4f/f brown preadipocytes led to severe defects in MyoD-stimulated myogenesis of preadipocytes and the associated expression of myogenesis markers such as Myog , Myh and Mck . Taken together , these data indicate that MLL4 is essential for adipogenesis and myogenesis . Next , we investigated how MLL4 regulates gene expression during differentiation . Adipogenesis was induced in Ad-GFP- or Ad-Cre-infected Mll3−/−Mll4f/f brown preadipocytes . Samples were collected before ( day 0 ) and during ( day 2 ) adipogenesis for RNA-Seq ( Figure 2D and Figure 2—figure supplement 2 ) . In total there were 14 , 902 genes expressed in Ad-GFP-infected cells at either time point . Among them , 1 , 175 ( 7 . 9% ) and 1 , 302 ( 8 . 7% ) showed over 2 . 5-fold down- and up-regulation , respectively , from day 0 to day 2 . Among the 1 , 302 up-regulated genes , a significant number ( 588 , p=1 . 4E-268 , hypergeometric test ) was induced in an MLL4-dependent manner . Strikingly , gene ontology ( GO ) analysis revealed that only the MLL4-dependent up-regulated gene group was associated with fat cell differentiation ( Figure 2E ) , suggesting that MLL4 selectively regulates the induction of adipogenesis genes . RNA-Seq analysis of myogenesis of MyoD-expressing brown preadipocytes revealed a similar trend . 1 , 108 ( 6 . 6% ) and 2 , 774 ( 16 . 5% ) genes showed over 2 . 5-fold down- and up-regulation , respectively , from brown preadipocytes to myocytes ( Figure 2I ) . Among the 2 , 774 up-regulated genes , a significant number ( 836 , p<1E-300 , hypergeometric test ) was induced in an MLL4-dependent manner . Furthermore , among those gene groups , only the MLL4-dependent up-regulated gene group was highly associated with muscle development ( Figure 2J ) . Together , these results indicate that MLL4 is required for induction of cell-type-specific genes during differentiation . To find out whether MLL4 directly regulates the induction of cell-type-specific genes during differentiation , we performed ChIP-Seq of MLL4 in adipogenesis and myogenesis using an anti-MLL4 antibody . To exclude false-positive MLL4 targets as a result of off-target antibody binding , ChIP-Seq was done in both Ad-GFP- and Ad-Cre-infected Mll3−/−Mll4f/f cells ( i . e . , Mll3 KO and Mll3/Mll4 double KO cells ) . High-confidence MLL4 binding signals were obtained by filtering out non-specific signals observed in MLL4-deficient cells ( Figure 3—figure supplement 1A-E ) . ChIP-Seq results were independently verified by quantitative ChIP assays at MLL4+ regions on multiple adipogenesis genes ( Figure 3—figure supplement 2 ) . ChIP-Seq of MLL4 in adipogenesis was done at three time points , day 0 ( preadipocytes ) , day 2 ( during adipogenesis ) and day 7 ( adipocytes ) . We identified 6 , 937 , 14 , 581 and 25 , 005 high-confidence MLL4 genomic binding regions at day 0 , day 2 , and day 7 , respectively ( Figure 3A ) . The average length of MLL4 binding regions was 350–400bp . Interestingly , MLL4 binding regions changed dramatically from day 0 to day 2 but were largely maintained from day 2 to day 7 , suggesting differentiation-stage-specific genomic binding of MLL4 during adipogenesis ( Figure 3A ) . ChIP-Seq of MLL4 in myocytes identified 26 , 581 high-confidence MLL4 genomic binding regions . The MLL4-binding regions in adipocytes and myocytes were largely non-overlapping ( Figure 3B ) , suggesting cell-type-specific genomic binding of MLL4 . 10 . 7554/eLife . 01503 . 009Figure 3 . Cell-type- and differentiation-stage-specific genomic binding of MLL4 . Adipogenesis and myogenesis were done as in Figure 2A , F , G , respectively . Cells were collected for ChIP-Seq analysis of MLL4 . ( A ) Venn diagram of MLL4 binding regions at day 0 ( preadipocytes ) , day 2 ( during adipogenesis ) and day 7 ( adipocytes ) of adipogenesis . ( B ) Venn diagram of MLL4 binding regions in adipocytes and myocytes . ( C ) ChIP-Seq profiles of MLL4 binding on gene loci encoding PPARγ and myogenin ( Myog ) at indicated time points and cell types . ( D ) GO analysis of genes associated with emergent MLL4 binding regions at indicated time points and cell types . DOI: http://dx . doi . org/10 . 7554/eLife . 01503 . 00910 . 7554/eLife . 01503 . 010Figure 3—figure supplement 1 . MLL4 binds to adipogenesis genes . ChIP-Seq of MLL4 was performed during adipogenesis as described in Figure 2 . ( A–E ) MLL4 binding on adipogenic genes at day 2 of adipogenesis . In each panel , the top 2 tracks show the raw data while the bottom track shows the filtered data . ( F–J ) MLL4 binding on Cebpa ( F ) , Klf15 ( G ) , Fabp4 ( H ) , Lpl ( I ) and Ucp1 ( J ) gene loci during adipogenesis . ( K ) MLL4 exhibits little binding to Cebpb locus during adipogenesis . DOI: http://dx . doi . org/10 . 7554/eLife . 01503 . 01010 . 7554/eLife . 01503 . 011Figure 3—figure supplement 2 . ChIP-qPCR confirmation of ChIP-Seq data . ( A ) Schematic of genomic locations of MLL4+ enhancers on Pparg , Cebpa , Fabp4 ( aP2 ) and Prdm16 loci . E , enhancer . ( B ) ChIP-qPCR confirmation of ChIP-Seq data on MLL4+ enhancers . DOI: http://dx . doi . org/10 . 7554/eLife . 01503 . 011 Cell-type- and differentiation-stage-specific genomic binding of MLL4 was confirmed on Pparg and Myog gene loci that encoded the adipogenic TF PPARγ and the myogenic TF Myog , respectively ( Figure 3C ) . MLL4 was also found to bind gene loci encoding other adipogenesis markers such as C/EBPα , KLF15 , aP2 ( Fabp4 ) , LPL and UCP1 at day 2 and day 7 of adipogenesis ( Figure 3—figure supplement 1F–K ) . Interestingly , MLL4 was largely absent from the Cebpb gene locus encoding the pioneer adipogenic TF C/EBPβ . To identify MLL4-associated genes , we used proximity to assign the top 2 , 000 emergent MLL4 binding regions at each time point to the nearest genes . GO analysis of MLL4-associated genes identified brown fat cell differentiation and fat cell differentiation as the top two functional categories at day 2 ( during adipogenesis ) and day 7 ( adipocytes ) but not at day 0 ( preadipocytes ) or in myocytes ( Figure 3D ) . GO analysis also identified muscle organ/tissue development and muscle cell differentiation as the top functional categories specifically in myocytes ( Figure 3D ) . These results are consistent with GO analysis of RNA-Seq data and suggest that cell-type- and differentiation-stage-specific genomic binding of MLL4 is directly involved in the regulation of genes critical for cell differentiation . Next , we performed motif analysis of the top 2 , 000 emergent MLL4 binding regions at each time point ( Figure 4A ) . At preadipocyte stage before adipogenesis ( day 0 ) , MLL4 binding regions were enriched with motifs of TFs functioning in various developmental lineages . However , at day 2 and day 7 of adipogenesis , MLL4 binding regions were highly enriched with motifs of major adipogenic TFs , including C/EBPα , C/EBPβ , PPARγ , EBF1 and GR ( Rosen and MacDougald , 2006; Ge , 2012 ) . In myocytes , MLL4 binding regions were highly enriched with motifs of myogenic TF MyoD and its binding partner TCF3 ( also known as E2A ) ( Lassar et al . , 1991 ) , as well as myogenic TFs Runx1 and TEAD4 ( MacQuarrie et al . , 2013 ) . 10 . 7554/eLife . 01503 . 012Figure 4 . Genomic co-localization of MLL4 with lineage-determining TFs during differentiation . ( A ) Adipogenic TF binding motifs are enriched at MLL4 binding regions during adipogenesis while myogenic TF binding motifs are enriched at MLL4 binding regions in myocytes . Top 2 , 000 emergent MLL4 binding regions at each time point were used for motif analysis . Only TFs that are expressed at the indicated cell type or differentiation stage are included . ( B and C ) Venn diagram ( B ) and heat maps ( C ) of genomic co-localization of MLL4 with C/EBPs ( C/EBPα or β ) and PPARγ at day 2 of adipogenesis . ( D and E ) Venn diagram ( D ) and heat maps ( E ) of genomic co-localization of MLL4 with MyoD in myocytes . ( F ) MLL4 physically interacts with C/EBPβ during adipogenesis . Nuclear extracts prepared at day 2 of adipogenesis were immunoprecipitated with MLL4 antibody . The immunoprecipitates were analyzed by Western blot using antibodies against MLL3/MLL4 complex components ( UTX , PTIP , RbBP5 and PA1 ) , Menin , or C/EBPβ . DOI: http://dx . doi . org/10 . 7554/eLife . 01503 . 01210 . 7554/eLife . 01503 . 013Figure 4—figure supplement 1 . ChIP-Seq and RNA-Seq data on Pparg gene during adipogenesis . D0 and D2 , day 0 and day 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 01503 . 01310 . 7554/eLife . 01503 . 014Figure 4—figure supplement 2 . ChIP-Seq and RNA-Seq data on Cebpa gene during adipogenesis . DOI: http://dx . doi . org/10 . 7554/eLife . 01503 . 01410 . 7554/eLife . 01503 . 015Figure 4—figure supplement 3 . PPARγ interacts with MLL3/MLL4-containing H3K4 methyltransferase complex in cells . ( A ) Anti-FLAG M2 agarose was used to immunoprecipitate from nuclear extracts of HeLaS cells stably expressing FLAG-tagged PPARγ ( F-PPARγ ) ( Ge et al . , 2008 ) . The immunoprecipitates were analyzed by Western blot using antibodies indicated on the right . MLL1C , MLL1 c-terminal fragment . ( B ) Anti-PPARγ antibody or IgG was used to immunoprecipitate from nuclear extracts of HeLaS cells stably expressing F-PPARγ . The immunoprecipitates were subjected to HMT assay on an histone H3 peptide as described previously ( Cho et al . , 2007 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01503 . 015 To experimentally validate the predicted motifs , we performed ChIP-Seq of C/EBPα , C/EBPβ and PPARγ at day 2 of adipogenesis and of MyoD in myocytes . By comparing the genomic localizations of MLL4 with those of C/EBPα , C/EBPβ and PPARγ , we found that consistent with the motif analysis , ∼64% of MLL4 binding regions overlapped with those of C/EBPα/β or PPARγ at day 2 of adipogenesis ( Figure 4B–C ) . In particular , genomic co-localization of MLL4 with C/EBPβ , C/EBPα , and PPARγ was observed on Pparg and Cebpa loci at day 2 but not day 0 of adipogenesis ( Figure 4—figure supplement 1 and 2 ) . In myocytes , 40% of MLL4 binding regions overlapped with those of MyoD ( Figure 4D–E ) . Consistent with these results , we observed a physical interaction of MLL4 with C/EBPβ during adipogenesis ( Figure 4F ) . We also found that PPARγ associated with MLL3/MLL4-containing H3K4 methyltransferase complex in cells ( Figure 4—figure supplement 3 ) , which is consistent with the reported direct interaction between PPARγ and MLL3/MLL4 complex ( Lee et al . , 2008 ) . Together , these results indicate significant genomic co-localization of MLL4 with lineage-determining TFs during adipogenesis and myogenesis . To characterize the genomic features of MLL4 binding regions during differentiation , we performed ChIP-Seq of H3K4me1/2/3 , H3K27ac and Pol II during adipogenesis . Then , we used histone marks to define four types of gene regulatory elements: active enhancer , silent enhancer , active promoter , and silent promoter ( Figure 5A ) ( Creyghton et al . , 2010; Rada-Iglesias et al . , 2011 ) . Average profile plots revealed that at day 2 of adipogenesis , MLL4 and adipogenic TFs C/EBPα , C/EBPβ and PPARγ were enriched on active enhancers ( Figure 5B ) . Genomic distribution analyses showed that among the 14 , 581 MLL4 binding regions at day 2 of adipogenesis , 9 , 642 ( 66 . 1% ) and 1 , 836 ( 12 . 6% ) were located on active and silent enhancers while only 451 ( 3 . 1% ) and 29 ( 0 . 2% ) were located on active and silent promoters , respectively , indicating that MLL4 mainly bound to enhancers and preferentially active enhancers ( p<1E-300 , binomial test ) ( Figure 5C ) . 10 . 7554/eLife . 01503 . 016Figure 5 . MLL4 co-localizes with lineage-determining TFs on active enhancers during differentiation . ChIP-Seq analyses of MLL4 , TFs , H3K4me1/2/3 and H3K27ac were done at day 2 of adipogenesis . ( A ) Table depicting histone modifications used to define gene regulatory elements . TSS , transcription start site . ( B and C ) MLL4 is mainly localized on active enhancers during adipogenesis . ( B ) Average binding profiles of MLL4 , adipogenic TFs C/EBPα , C/EBPβ and PPARγ , and RNA polymerase II ( Pol II ) around the center of each type of gene regulatory elements . ( C ) Pie charts depicting the genomic distributions of MLL4 , C/EBPα , C/EBPβ , PPARγ and Pol II binding regions . ( D ) MLL4 co-localizes with adipogenic TFs on active enhancers during adipogenesis . The binding profiles of C/EBPα , C/EBPβ , PPARγ and MLL4 on the three types of adipogenic enhancers ( C/EBP+PPARγ− , C/EBP−PPARγ+ and C/EBP+PPARγ+ ) are shown in heat maps . Adipogenic enhancers are defined as active enhancers bound with C/EBPα , C/EBPβ or PPARγ at day 2 of adipogenesis . DOI: http://dx . doi . org/10 . 7554/eLife . 01503 . 01610 . 7554/eLife . 01503 . 017Figure 5—figure supplement 1 . MLL4 co-localizes with lineage-determining TFs on active enhancers during myogenesis . MyoD-stimulated myogenesis of brown preadipocytes was done as in Figure 2 . ( A ) Average binding profiles of MLL4 and MyoD around the center of each type of gene regulatory elements in myocytes . ( B ) Pie charts depicting the genomic distributions of MLL4 and MyoD binding regions in myocytes . ( C ) MLL4 co-localizes with MyoD on active enhancers in myocytes . Heat maps illustrating the binding of MyoD and MLL4 on MyoD+ active enhancers are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 01503 . 017 Genomic distribution analyses also revealed preferential localization of C/EBPs and PPARγ on active enhancers during adipogenesis ( Figure 5C ) . Interestingly , 57 . 3% of the C/EBP-binding and 80 . 0% of the PPARγ-binding active enhancers were MLL4+ , significantly higher than the genome-wide level of co-localization ( 31 . 3% for C/EBPs and 56 . 3% for PPARγ , respectively ) , indicating that MLL4 preferentially co-localizes with these TFs on active enhancers than on other genomic locations ( p<1E-300 for either C/EBPs or PPARγ , hypergeometric test ) . At day 2 of adipogenesis , 11 , 280 active enhancers were bound with adipogenic TFs C/EBPα , C/EBPβ or PPARγ . These 11 , 280 active enhancers , which we termed adipogenic enhancers , could be clustered into 3 groups: C/EBP+PPARγ− ( C/EBPα- or β-positive but PPARγ-negative , 8 , 098 active enhancers ) , C/EBP−PPARγ+ ( 1 , 219 ) , and C/EBP+PPARγ+ ( 1 , 963 ) ( Figure 5D ) . MLL4 binding was found on 46 . 9% and 61 . 7% of C/EBP+PPARγ− and C/EBP−PPARγ+ adipogenic enhancers , respectively . Remarkably , MLL4 binding was found on 91 . 3% of C/EBP+PPARγ+ adipogenic enhancers , which were high-confidence adipogenic enhancers ( Lefterova et al . , 2008 ) ( Figure 5D ) . Similarly , MLL4 and the myogenic TF MyoD were enriched on active enhancers in myocytes ( Figure 5—figure supplement 1A-B ) . We identified 8 , 091 myogenic enhancers that were active enhancers bound with MyoD . Among the 8 , 091 myogenic enhancers , 5 , 228 ( 64 . 6% , p<1E-300 , hypergeometric test ) were MLL4+ ( Figure 5—figure supplement 1C ) . Together , these results indicate that MLL4 co-localizes with lineage-determining TFs on active enhancers during differentiation . The enrichment of H3K4 methyltransferase MLL4 on active enhancers , which are marked by H3K4me1/2 and H3K27ac , prompted us to examine the enzymatic properties of MLL4 . For this purpose , we affinity-purified endogenous SET1A/SET1B complex ( SET1A/B . com ) and MLL3/MLL4 complex ( MLL3/4 . com ) from 293 cells expressing FLAG-tagged CFP1 and PA1 , the unique subunit of SET1A/B . com and MLL3/4 . com , respectively ( Lee and Skalnik , 2005; Cho et al . , 2007 ) . In an in vitro HMT assay using core histones as substrate , we found that compared with SET1A/B . com , the major H3K4 tri-methyltransferase in mammalian cells , MLL3/4 . com carried much stronger mono- and di-methyltransferase activities but much weaker tri-methyltransferase activity on H3K4 ( Figure 6A ) . Time-course HMT assays confirmed H3K4me1/2 methyltransferase activity of MLL3/4 . com in vitro ( Figure 6B ) . 10 . 7554/eLife . 01503 . 018Figure 6 . MLL4 is a major H3K4 mono- and di-methyltransferase in cells . ( A and B ) In vitro histone methyltransferase ( HMT ) assay . ( A ) SET1A/SET1B complex ( SET1A/B . com ) and MLL3/MLL4 complex ( MLL3/4 . com ) that were affinity-purified from 293T nuclear extracts were incubated with core histones in an HMT assay for 3 hr , followed by Western blot analysis . ( B ) Time-course HMT assay with MLL3/4 . com . ( C ) Mll3−/−Mll4f/f brown preadipocytes were infected with adenoviral GFP or Cre as in Figure 2 . Cells were collected at day 0 and day 2 of adipogenesis for Western blot analysis of histone modifications . ( D ) Western blot analysis of histone modifications in MLL3−/−MLL4−/− HCT116 human colon cancer cells . ( E ) 87 . 7% of MLL4-binding regions at day 2 of adipogenesis were marked by H3K4me1/2 as revealed by ChIP-Seq analyses of MLL4 and H3K4me1/2 . DOI: http://dx . doi . org/10 . 7554/eLife . 01503 . 01810 . 7554/eLife . 01503 . 019Figure 6—figure supplement 1 . MLL3 and MLL4 are H3K4 mono- and di-methyltransferases in mammalian cells . ( A ) Western blot analyses of histone modifications in Mll3−/− brown preadipocytes . ( B ) Western blot analyses of histone modifications before and after MyoD-stimulated myogenesis . Pre-ad , preadipocytes . DOI: http://dx . doi . org/10 . 7554/eLife . 01503 . 019 In Mll3 single KO brown preadipocytes , we observed a moderate decrease of H3K4me1 ( Figure 6—figure supplement 1A ) . Deletion of Mll4 from Mll3−/−Mll4f/f cells led to global decreases of H3K4me1 and H3K4me2 but much less effect on H3K4me3 levels at day 0 and day 2 of adipogenesis . H3K27ac levels also decreased significantly ( Figure 6C ) . In Mll3/Mll4 double KO myocytes and human colon cancer cells ( Guo et al . , 2012 ) , we also observed significant decreases of H3K4me1 ( Figure 6—figure supplement 1B and Figure 6D ) . ChIP-Seq analyses of MLL4 and histone modifications also revealed that among the 14 , 581 MLL4 binding sites during adipogenesis , 12 , 783 ( 87 . 7% ) were marked by both H3K4me1 and H3K4me2 ( Figure 6E ) . These data indicate that endogenous MLL3 and MLL4 are major H3K4 mono- and di-methyltransferases in mammalian cells . The enrichment of H3K4 mono- and di-methyltransferase MLL4 on active enhancers during differentiation prompted us to investigate whether MLL4 is required for H3K4me1/2 on these enhancers . ChIP-Seq analyses revealed a dramatic increase of MLL4 levels on the 6 , 342 MLL4+ adipogenic enhancers from day 0 to day 2 of adipogenesis ( Figure 7A ) . Consistently , we observed marked increases of H3K4me1/2 on these enhancers . H3K27ac , Mediator ( represented by the MED1 subunit ) and Pol II levels also increased markedly on MLL4+ adipogenic enhancers from day 0 to day 2 . Deletion of Mll4 from Mll3−/−Mll4f/f cells prevented the marked increases of not only H3K4me1/2 but also H3K27ac , Mediator and Pol II on the 6 , 342 MLL4+ adipogenic enhancers from day 0 to day 2 ( Figure 7A ) . On Pparg and Cebpa gene loci , deletion of Mll4 also prevented the increases of H3K4me1/2 , H3K27ac , Mediator and Pol II on MLL4+ adipogenic enhancers during adipogenesis ( Figure 4—figure supplement 1 and 2 ) . On the 480 MLL4+ promoters identified during adipogenesis , MLL4 was required for H3K4me1/2 but not H3K4me3 levels . However , the effect of Mll4 deletion on promoter H3K4me1/2 is not as severe as on enhancers ( Figure 7—figure supplement 1 vs Figure 7A ) . 10 . 7554/eLife . 01503 . 020Figure 7 . MLL4 is required for enhancer activation during differentiation . Mll3−/−Mll4f/f brown preadipocytes were infected with adenoviral GFP or Cre as in Figure 2 . Cells were collected at day 0 and day 2 of adipogenesis for ChIP-Seq of H3K4me1/2/3 , H3K27ac , MED1 and Pol II , and RNA-Seq . ( A ) Deletion of Mll4 dramatically decreases H3K4me1/2 , H3K27ac , MED1 and Pol II levels on MLL4 positive ( MLL4+ ) adipogenic enhancers . Average profiles of histone modifications , MED1 and Pol II on MLL4+ adipogenic enhancers are shown . ( B ) MLL4 promotes induction of genes associated with MLL4+ adipogenic enhancer during adipogenesis . ( C ) Deletion of Mll4 reduces expression of MLL4+ adipogenic enhancer-associated genes . Gene expression fold changes were obtained by comparing Cre-infected with GFP-infected cells at day 2 of adipogenesis and are shown in the box plot . DOI: http://dx . doi . org/10 . 7554/eLife . 01503 . 02010 . 7554/eLife . 01503 . 021Figure 7—figure supplement 1 . MLL4 is required for H3K4me1/2 on MLL4+ promoters . The average profiles of H3K4me1/2/3 levels on the 480 ( 29 silent plus 451 active ) MLL4+ promoters identified during adipogenesis ( Figure 5C ) are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 01503 . 02110 . 7554/eLife . 01503 . 022Figure 7—figure supplement 2 . MLL4 is required for enhancer activation during myogenesis . MyoD-stimulated myogenesis of brown preadipocytes was done as in Figure 2 . ( A ) Deletion of MLL4 markedly decreases H3K4me1 and H3K27ac levels on MLL4+ MyoD+ active enhancers . Average profiles of histone modifications on MLL4+ MyoD+ active enhancers are shown . ( B ) ChIP-Seq of MyoD , MLL4 , H3K4me1 , H3K4me3 and H3K27ac on Myog gene locus is visualized on UCSC genome browser . Myo , myocytes . ( C ) Deletion of Mll4 reduces expression of MLL4+ MyoD+ active enhancer-associated genes . Gene expression fold changes were obtained by comparing adenoviral Cre-infected with GFP-infected cells in myocytes and are shown in the box plot . DOI: http://dx . doi . org/10 . 7554/eLife . 01503 . 022 During myogenesis , MLL4 and MyoD levels increased dramatically on the 5 , 228 MLL4+ myogenic enhancers . Deletion of Mll4 from Mll3−/−Mll4f/f cells also prevented the marked increases of H3K4me1 and H3K27ac on MLL4+ myogenic enhancers during myogenesis ( Figure 7—figure supplement 2A-B ) . Thus , MLL4 is the major H3K4 mono- and di-methyltransferase on adipogenic and myogenic enhancers . Because H3K27ac is a mark for active enhancers , these results indicate that MLL4 is required for the activation of cell-type-specific enhancers during adipogenesis and myogenesis . We next asked how MLL4 affects the induction and expression of genes regulated by cell-type-specific enhancers . To focus on the direct effect of MLL4 , we examined genes associated with MLL4+ adipogenic enhancers ( C/EBP+PPARγ− , C/EBP−PPARγ+ , or C/EBP+PPARγ+ ) . We first looked at the induction of genes from day 0 to day 2 of adipogenesis ( Figure 7B ) . Genes associated with MLL4+ adipogenic enhancers at day 2 were better induced than those associated with MLL4− adipogenic enhancers ( p=1 . 5E-22 , Wilcoxon test ) . The induction was even stronger if the associated MLL4+ adipogenic enhancers were C/EBP+PPARγ+ ( p=1 . 5E−76 , Wilcoxon test ) . We then examined how MLL4 affects expression of genes at day 2 of adipogenesis . As shown in Figure 7C , deletion of Mll4 significantly decreased expression of genes associated with MLL4+ adipogenic enhancers ( C/EBP+PPARγ− or C/EBP−PPARγ+ , p=2 . 2E−10 , Wilcoxon test ) . An even stronger effect of Mll4 deletion was observed if the associated MLL4+ adipogenic enhancers were C/EBP+PPARγ+ ( p=1 . 8E−31 , Wilcoxon test ) . In contrast , the deletion of Mll4 had little effect on the expression of genes associated with MLL4− adipogenic enhancers . In myocytes , deletion of Mll4 significantly decreased expression of genes associated with MLL4+ myogenic enhancers but not those associated with MLL4− myogenic enhancers ( Figure 7—figure supplement 2C ) . Together , these results indicate that MLL4 is required for activation of enhancers that are important for cell-type-specific gene expression during differentiation . The physical interaction and the genome-wide co-localization of C/EBPβ with MLL4 during adipogenesis suggest that the early adipogenic TF C/EBPβ may recruit MLL4 to establish at least a subset of adipogenic enhancers . To directly test this possibility , we ectopically expressed C/EBPβ in preadipocytes ( Figure 8A ) , followed by ChIP-Seq of C/EBPβ , MLL4 , H3K4me1 and H3K27ac without inducing differentiation . Of the 4 , 965 C/EBPβ+ MLL4+ active enhancers identified at day 2 of adipogenesis , 66 . 6% ( 3 , 309/4 , 965 ) were bound by ectopic C/EBPβ in undifferentiated preadipocytes . Ectopic C/EBPβ was able to recruit MLL4 to a subset of enhancers ( 666 out of 3 , 309 ) ( Figure 8B ) . Among them , 88 . 7% ( 591/666 ) showed the characteristics of active enhancers , as indicated by the presence of H3K4me1 and H3K27ac ( Figure 8C ) . Among the 591 recovered C/EBPβ+ MLL4+ active enhancers , 375 displayed ectopic C/EBPβ-induced de novo MLL4 binding , which led to significant increases of H3K4me1 and H3K27ac . The remaining 216 recovered C/EBPβ+ MLL4+ adipogenic enhancers , which were pre-marked by MLL4 in the control cells , showed ectopic C/EBPβ-enhanced MLL4 binding as well as H3K4me1 and H3K27ac ( Figure 8C ) . Importantly , deletion of Mll4 markedly decreased H3K4me1 and H3K27ac on over 90% of the 591 recovered C/EBPβ+ MLL4+ adipogenic enhancers ( Figure 8C–D ) . MLL4-dependent de novo H3K4me1 and H3K27ac were also observed on the C/EBPβ+ adipogenic enhancers located on Pparg gene locus ( Figure 8E ) . Together , these results suggest that lineage-determining TF C/EBPβ recruits and requires MLL4 to establish at least a subset of adipogenic enhancers . 10 . 7554/eLife . 01503 . 023Figure 8 . C/EBPβ recruits and requires MLL4 to establish a subset of adipogenic enhancers . Mll3−/−Mll4f/f brown preadipocytes were infected with retroviral C/EBPβ or Vec only , and then infected with adenoviral GFP or Cre . Cells were collected 2 days after confluence without induction of differentiation for Western blot ( A ) and for ChIP-Seq of C/EBPβ , MLL4 , H3K4me1 and H3K27ac ( B–E ) . ( A ) Western blot analyses of C/EBPβ expression and histone modifications . Nuclear protein RbBP5 was used as a loading control . ( B ) Ectopic expression of C/EBPβ in undifferentiated preadipocytes leads to MLL4 binding on a subset of C/EBPβ+ MLL4+ active enhancers identified at day 2 of adipogenesis . ( C ) Heat maps of the relative signals of C/EBPβ , MLL4 , H3K4me1 and H3K27ac on the recovered C/EBPβ+ MLL4+ active enhancers . ( D ) MLL4 is required for H3K4me1 on the recovered C/EBPβ+ MLL4+ active enhancers . ( E ) Genome browser view of ectopic C/EBPβ-induced MLL4 binding as well as H3K4me1 and H3K27ac on Pparg locus . ( F ) Model depicting the role of MLL4 in transcriptional regulation of adipogenesis . The early adipogenic TF C/EBPβ serves as a pioneer TF and recruits MLL4 to establish active enhancers on Pparg and Cebpa loci . After PPARγ and C/EBPα are induced , they recruit MLL4 and cooperate with other adipogenic TFs to shape the enhancer landscape important for adipocyte gene expression . ( G ) Model depicting the role of MLL4 in the step-wise enhancer activation . DOI: http://dx . doi . org/10 . 7554/eLife . 01503 . 02310 . 7554/eLife . 01503 . 024Figure 8—figure supplement 1 . Time-course ChIP-qPCR on C/EBPβ+MLL4+ enhancers on Pparg gene locus . ( A ) Schematic of C/EBPβ+MLL4+ adipogenic enhancers 1 , 2 , and 3 ( E1 , E2 and E3 ) on Pparg gene locus . ( B ) Time-course ChIP-qPCR on C/EBPβ+MLL4+ enhancers on Pparg gene locus at 0 , 2 , 4 , 8 , 12 , 24 , and 48 hr after induction of adipogenesis in wild-type brown preadipocytes . For clarity , only the upper half of error bars are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 01503 . 024 Consistently , by quantitative time-course ChIP assays performed during the first 48 hr of adipogenesis , we observed sequential enrichments of lineage-determining TF C/EBPβ , H3K4me1/2 methyltransferase MLL4 , H3K4me1/2 , and H3K27ac on C/EBPβ+MLL4+ adipogenic enhancers identified on Pparg gene locus ( Figure 8—figure supplement 1 ) . Combined with our earlier observations , these results suggest a step-wise model of enhancer activation during cell differentiation ( Figure 8G and ‘Discussion’ ) . Drosophila Trr and its mammalian homologs MLL3/MLL4 represent the major candidate methyltransferases for H3K4me1 on enhancers ( Calo and Wysocka , 2013 ) . Our in vitro HMT assays show that MLL3/4 . com carries predominantly H3K4 mono- and di-methyltransferase activity with little H3K4 tri-methyltransferase activity , in agreement with two recent studies ( Tang et al . , 2013; Wu et al . , 2013 ) . By generating Mll3/Mll4 double KO cells , we demonstrate that endogenous MLL3 and MLL4 are major H3K4me1/2 methyltransferases in mammalian cells and that MLL4 has a partial functional redundancy with MLL3 in cells . False-positive signals due to potential cross-reactivity of antibodies are a major concern in ChIP-Seq analyses ( Kidder et al . , 2011 ) . To exclude false-positive signals and identify bona fide MLL4 binding regions , we performed ChIP-Seq of MLL4 in both Mll3 KO and Mll3/Mll4 double KO cells . The results provide the direct evidence that MLL4 is a major H3K4me1/2 methyltransferase on enhancers during cell differentiation . More importantly , our data indicate that MLL4 is required for enhancer activation , which suggests that H3K4me1/2 may be important for enhancer activation . The partial functional redundancy between MLL3 and MLL4 in cells suggests that endogenous MLL3 is likely an H3K4me1/2 methyltransferase on enhancers and is likely involved in the activation of cell-type-specific enhancers . Significant levels of H3K4me1/2 remain in Mll3/Mll4 double KO cells ( Figure 6C ) , indicating that other H3K4 methyltransferases contribute to H3K4me1/2 in cells . Studies have shown possible roles of MLL1 and Set7 in catalyzing H3K4me1 on specific enhancers ( Blum et al . , 2012; Kaikkonen et al . , 2013 ) , but whether these enzymes are required for enhancer activation remains to be determined . It was shown recently that knockdown of several H3K4 methyltransferases including MLL1 , MLL3 , MLL4 in macrophages led to significant decreases of H3K4me1/2 on latent enhancers ( Kaikkonen et al . , 2013 ) . Our data indicate that MLL4 is required for the activation of cell-type-specific enhancers during differentiation . Whether MLL4 is also required for the activation of latent enhancers remains to be determined . During adipogenesis and myogenesis , MLL4 exhibits cell-type- and differentiation-stage-specific genomic binding and co-localizes with lineage-determining TFs and H3K4me1/2 on active enhancers . Thus , MLL4 appears to mark lineage-determining enhancers during differentiation . It will be interesting to test whether MLL4 binding can predict enhancers in other cell types . Interestingly , MLL4-dependent H3K4me1/2 distribute well beyond the binding sites of lineage-determining TFs and MLL4 . A similar observation can be made for p300 and H3K27ac ( Rada-Iglesias et al . , 2011 ) , where H3K27ac is distributed much broader than p300 . We speculate that lineage-determining TFs along with MLL4 may induce physical proximity of neighboring nucleosomes to MLL4 , which enables MLL4-mediated H3K4me1/2 over a much broader region . C/EBPβ is a lineage-determining TF for adipogenesis . It can bind to relatively closed chromatin and thus behaves as a pioneer factor in adipogenesis ( Siersbaek et al . , 2011 ) . By ectopic expression of C/EBPβ in preadipocytes without inducing differentiation , we show that C/EBPβ recruits and requires MLL4 to establish a subset of adipogenic enhancers . Our data are consistent with earlier reports that genomic binding of the lineage-determining TF PU . 1 leads to H3K4me1 in macrophages and B cells and that ectopic expression of PU . 1 in non-hematopoietic fibroblasts induces H3K4me1 on macrophage-specific enhancers ( Ghisletti et al . , 2010; Heinz et al . , 2010 ) . Although it remains to be determined whether MLL4 and/or MLL3 play a major role in establishing enhancers in macrophages , B cells and other cell types , the available results suggest that lineage-determining TFs recruit H3K4me1/2 methyltransferases to prime enhancer-like regions in a particular cell type . Ectopic expression of C/EBPβ in preadipocytes without differentiation only recovers a subset of endogenous C/EBPβ+MLL4+ enhancers , indicating that the recruitment of MLL4 to enhancers during adipogenesis involves additional mechanisms and factors . Because differentiation-dependent phosphorylation of C/EBPβ increases its DNA binding activity , the optimal genomic binding of C/EBPβ and the subsequent recruitment of MLL4 are likely dependent on induction of adipogenesis . Similarly , the binding of MyoD to myogenic enhancers on Myog locus is differentiation-dependent ( Figure 7—figure supplement 2B ) . The direct interaction between MLL3/4 . com and PPARγ and the genomic co-localization of MLL4 with PPARγ and C/EBPα suggest that PPARγ and C/EBPα also play critical roles in recruiting MLL4 to establish adipogenic enhancers . Motif analyses of MLL4 binding sites during adipogenesis and myogenesis have identified motifs of additional lineage-determining TFs , such as adipogenic TFs EBFs , GR and NFIC and myogenic TFs TCF3 , Runx1 and TEAD4 ( Figure 4A ) . Future ChIP-Seq analyses will tell whether these additional lineage-determining TFs can recruit MLL4 to regions that are not bound by C/EBPα/β , PPARγ or MyoD . On the other hand , MLL4 is absent from many genomic regions that are occupied by C/EBPα/β or PPARγ during adipogenesis or by MyoD in myocytes ( Figure 4B , D ) , suggesting that lineage-determining TFs alone are not always sufficient for recruiting MLL4 to establish enhancers and that additional mechanisms and factors may be involved . Ectopic expression of C/EBPβ in preadipocytes partially mimics the early phase of adipogenesis when PPARγ and C/EBPα have not been induced yet . Nevertheless , endogenous C/EBPβ co-localizes with MLL4 on active enhancers located on both Pparg and Cebpa gene loci ( Figure 4—figure supplements 1 and 2 ) . Because C/EBPβ not only transcriptionally activates Pparg and Cebpa gene expression but also facilitates PPARγ and C/EBPα genomic binding during adipogenesis ( Rosen et al . , 2002; Siersbaek et al . , 2011 ) , our data suggest the following model on how MLL4 regulates adipogenesis ( Figure 8F ) . The pioneer TF C/EBPβ recruits MLL4 to activate adipogenic enhancers on , and the subsequent induction of , Pparg and Cebpa gene expression . After the master adipogenic TFs PPARγ and C/EBPα are induced , they cooperate to recruit MLL4 to establish the downstream enhancer landscape critical for adipocyte gene expression . In cell culture , MLL3 and MLL4 are partially redundant in adipogenesis . In mice , deletion of Mll4 alone causes severe defects in BAT development . Deletion of Mll3 in mice has been shown to decrease white adipose tissue size ( Lee et al . , 2008 ) . These results suggest that while MLL4 is the dominant one during mouse embryonic development , MLL3 is partially redundant with MLL4 in mice . The critical roles of MLL3 and MLL4 in adipogenesis are consistent with our previous report that PTIP , a component of the MLL3/4 . com , is essential for Pparg and Cebpa expression and adipogenesis ( Cho et al . , 2009 ) . The presence of H3K4me1 on enhancers often precedes the active enhancer mark H3K27ac and gene expression , suggesting that H3K4me1 broadly defines a window of future active enhancers ( Calo and Wysocka , 2013 ) . Consistently , our time-course ChIP assays revealed sequential appearances of H3K4me1/2 and H3K27ac on C/EBPβ+MLL4+ active enhancers on Pparg gene locus in the early phase of adipogenesis ( Figure 8—figure supplement 1 ) . Further , we show that the H3K4me1/2 methyltransferase MLL4 is required for H3K27ac , Mediator and Pol II on enhancers , which indicates that MLL4 is required for enhancer activation and suggests that MLL4 exerts its function in enhancer commissioning through H3K4me1/2 . Collectively , our data suggest a stepwise model of enhancer activation during adipogenesis ( Figure 8G ) . Step 1 , pioneer TF binding . Pioneer TFs such as C/EBPβ bind enhancer-like regions . Step 2 , enhancer commissioning by MLL4 . Lineage-determining TFs , such as C/EBPβ , PPARγ and C/EBPα , cooperatively recruit MLL4 to perform H3K4me1/2 on enhancer-like regions . Step 3 , enhancer activation by H3K27 acetyltransferase p300 , followed by Pol II recruitment , establishment of enhancer–promoter interaction , and activation of cell-type-specific gene expression . Thus , the enhancer-associated H3K4me1/2 methyltransferase MLL4 appears to perform the opposite function of LSD1 , an H3K4me1/2 demethylase required for decommissioning of embryonic stem ( ES ) cell-specific enhancers during ES cell differentiation ( Whyte et al . , 2012 ) . Frequent loss-of-function mutations in MLL4 ( sometimes called MLL2 ) and its homolog MLL3 have been identified in developmental diseases such as Kabuki syndrome ( Ng et al . , 2010 ) , congenital heart disease ( Zaidi et al . , 2013 ) , and in cancers such as medulloblastoma ( Parsons et al . , 2011; Jones et al . , 2012; Pugh et al . , 2012 ) , non-Hodgkin lymphomas ( Morin et al . , 2011; Pasqualucci et al . , 2011 ) , breast cancer ( Ellis et al . , 2012 ) , and prostate cancer ( Grasso et al . , 2012 ) . Our findings suggest that loss-of-function mutations in MLL3 and MLL4 would impair H3K4me1/2 on enhancers , which lead to defects in enhancer activation , cell-type-specific gene expression and cell differentiation . Such a mechanism may contribute to the pathogenesis of these developmental diseases and cancers . The retroviral plasmids pMSCVhygro-PPARγ2 , pWZLhygro-C/EBPβ , and MSCVpuro-Cre have been described ( Ge et al . , 2008; Wang et al . , 2010 ) . MyoD cDNA was subcloned from MSCVpuro-MyoD into pWZLhygro . The shRNA sequence-targeting mouse Mll4 gene ( GCATGTTCTTCAAGGACAAGA ) was cloned into lentiviral vector pLKO . 1 . The following homemade antibodies have been described: anti-UTX ( Hong et al . , 2007 ) ; anti-PTIP , anti-PA1#2 , anti-MLL4#3 ( Cho et al . , 2009 ) . Anti-C/EBPα ( sc-61X ) , anti-C/EBPβ ( sc-150X ) , anti-PPARγ ( sc-7196X ) , and anti-MyoD ( sc-760 ) were from Santa Cruz Biotechnology ( Dallas , TX , USA ) . Anti-Menin ( A300-105A ) , anti-RbBP5 ( AA300-109A ) , and anti-MED1/TRAP220 ( A300-793A ) were from Bethyl Laboratories ( Montgomery , TX , USA ) . Anti-H3K4me1 ( ab8895 ) was from Abcam ( Cambridge , MA , USA ) . Anti-Pol II ( 17-672 ) and anti-H3K4me3 ( 07-473 ) were from Millipore ( Billerica , MA , USA ) . Other histone methylation and acetylation antibodies have been described ( Jin et al . , 2011 ) . Mll3+/− and Mll4+/− mice were generated from BayGenomics gene trap ES cell lines following standard procedures at NIDDK Mouse Knockout Core ( Stryke et al . , 2003 ) . Mll3+/− mice were derived from ES cell line XM083 , in which the gene trap vector pGT0lxf was inserted between exon 9 and 10 of one Mll3 allele . Mll4+/− mice were derived from ES cell line XT0709 , in which the gene trap vector pGT0lxf was inserted between exon 19 and 20 of one Mll4 allele . To generate Mll4 conditional KO mice , the loxP/FRT-flanked neomycin cassette was inserted at the 3’ of exon 19 and the single loxP site was inserted at the 5’ of exon 16 . The targeted region includes exons 16–19 ( Figure 1A ) . The Mll4floxneo/+ ES cell was injected into blastocysts to obtain male chimera mice that were crossed with wild-type C57BL/6J females to screen for germ line transmission . Mice bearing germline transmission ( Mll4floxneo/+ ) were crossed with FLP1 mice ( Jackson no . 003946 ) to generate Mll4f/+ ( i . e . , Mll4flox/+ ) mice . Mll4f/+ mice were crossed with Myf5-Cre ( Jackson no . 007893 ) . The resulting Mll4f/+;Myf5-Cre were then crossed with Mll4f/f to generate Mll4f/f;Myf5-Cre mice . Deletion of exons 16–19 by Cre causes open reading frame shift and creates a stop codon in exon 20 . The resulting Mll4 KO allele encodes a truncated MLL4 protein lacking the C-terminal ∼4200aa . For genotyping the Mll4 alleles , allelic PCR was developed as shown in Figure 1A–B . P1:5′-GTTCACTCAGTGGGGCTGTG-3′; P2: 5′-ATTGCATCAGGCAAATCAGC-3’; P3: 5’-GCAGAAGCCTGCTATGTCCA-3′ . Genotyping of Myf5-Cre was done by PCR using the following three primers: 5’-CGTAGACGCCTGAAGAAGGTCAACCA-3’ , 5’-CACATTAGAAAACCTGCCAACACC-3’ , and 5’-ACGAAGTTATTAGGTCCCTCGAC-3’ . PCR amplified wild-type ( 603 bp ) and Myf5-Cre allele ( 400 bp ) . All mouse work was approved by the Animal Care and Use Committee of NIDDK , NIH . E18 . 5 embryos were dissected out by Cesarean section and fixed in 4% paraformaldehyde overnight at 4°C . The embryos were further dehydrated and embedded in paraffin and sectioned at 7–10 µm with a microtome . H&E staining and immunohistochemistry ( IHC ) on paraffin sections were done as described ( Feng et al . , 2010 ) . The primary antibodies used for IHC were 1:20 dilution of anti-Myosin ( MF20; Developmental Studies Hybridoma Bank ) and 1:400 dilution of anti-UCP1 ( ab10983; Abcam ) . Fluorescent secondary antibodies used were Alexa Fluor 488 goat anti-mouse IgG2b and Alexa Fluor 555 goat anti-rabbit IgG ( Life Technologies , Carlsbad , CA , USA ) . Primary brown preadipocytes were isolated from interscapular BAT of E18 . 5 embryos and were immortalized with SV40T-expressing retroviruses pBabepuro-large T as described ( Wang et al . , 2010 ) . Adenoviral infection of preadipocytes was done at 50 moi . Adipogenesis of immortalized brown preadipocytes and 3T3-L1 white preadipocyte cell line was done as described ( Wang et al . , 2013 ) . PPARγ- or C/EBPβ-stimulated adipogenesis and MyoD-stimulated myogenesis were done as described ( Ge et al . , 2002 ) . Myogenesis was induced in near-confluent cells . Total RNA extraction and qRT-PCR were done as described ( Cho et al . , 2009 ) . Taqman probe for Mll3 ( assay ID Mm01156965_m1 ) was from Life Technologies . qRT-PCR of Mll4 was done using SYBR green primers: forward 5’-GCTATCACCCGTACTGTGTCAACA-3’ and reverse 5’-CACACACGATACACTCCACACAA-3’ . Taqman probes for Pparg1 and Pparg2 and other SYBR green primers for qRT-PCR have been described ( Wang et al . , 2013 ) . SYBR green primers for ChIP-qPCR are listed in Supplementary file 1B . ChIP was performed by following a protocol from Myers’ laboratory ( http://www . hudsonalpha . org/myers-lab/protocols ) with modifications . The cells were crosslinked with 1–2% formaldehyde for 10 min at room temperature . Crosslinking reaction was stopped by adding 125 mM glycine . The cells were washed with cold PBS twice . 2 × 107 cells were collected in 10 ml Farnham lysis buffer ( 5 mM PIPES pH 8 . 0/85 mM KCl/0 . 5% NP-40 , supplemented with protease inhibitors ) and centrifuged at 4 , 000 g for 5 min at 4°C . Cell pellet was washed with 10 ml Farnham lysis buffer , followed by centrifugation . Resulting nuclear pellet was resuspended in 1 ml TE buffer ( 10 mM Tris-Cl pH 7 . 7/1 mM EDTA , supplemented with protease inhibitors ) and sonicated for 17 min ( 30 s on/off cycle ) . Lysates were supplemented with detergents to make 1X RIPA buffer ( 10 mM Tris-Cl pH 7 . 7/1 mM EDTA/0 . 1% SDS/0 . 1% Na-DOC/1% triton X-100 ) and centrifuged to remove debris . For each ChIP , 6–8 µg antibodies were prebound to 50 µl Dynabeads Protein A ( 100 . 02D; Life Technologies ) overnight at 4°C . Next day , antibody-beads complex was added to chromatin from 2 × 107 cells and further incubated overnight at 4°C . The beads were washed twice with RIPA buffer , twice with RIPA + 0 . 3M NaCl , twice with LiCl buffer ( 50 mM Tris-Cl pH 7 . 5/250 mM LiCl/0 . 5% NP-40/0 . 5% Na-DOC ) and twice with PBS . DNA was eluted and reverse crosslinked in 200 µl elution buffer ( 1% SDS/0 . 1M NaHCO3 , supplemented with 20 µg proteinase K ) overnight at 65°C . DNA was purified by QIAquick PCR Purification Kit ( QIAGEN ) and quantified . DNA and libraries were constructed as described ( Wei et al . , 2012 ) . All ChIP-Seq and RNA-Seq samples were sequenced on Illumina HiSeq 2000 . All datasets described in this paper , including 8 RNA-Seq samples , 62 ChIP-Seq samples and 12 ChIP-Seq inputs , have been deposited in NCBI Gene Expression Omnibus under access # GSE50466 .	Almost every cell in a human body carries the same genes , but not every cell will express all of these genes as proteins . As different types of cells , such as brain , liver , fat or muscle cells , develop , they will express different genes; or they will express the same genes , but at different times and in different amounts . Enhancers are short stretches of DNA that boost the amount of protein that is produced when a gene is expressed , and they are particularly important for those genes that are expressed differently between cell types . Enhancers bolster expression of a gene by interacting with the DNA nearby . Even genes separated from enhancers by a long stretches of DNA can benefit because the way that DNA is tightly packed inside the nucleus means that two distant sequences can actually end up close together . Proteins called transcription factors will bind to enhancers and recruit the cell’s protein ‘machinery’ required to express nearby genes . Enhancers can be identified by specific chemical marks associated with their DNA , but little is known about the enzymes that leave these marks in mammals . Moreover , it is not clear which genes are influenced by these marks . Now , by examining fat cells and muscle cells as they mature , Lee et al . have found that an enzyme called MLL4 is responsible for adding chemical marks to enhancers in both humans and mice . Further , MLL4 is required both to allow cells to specialize into different cell types , and to boost the expression of genes that are specific to each type of mature cells . Since faulty MLL4 has been implicated in several cancers and developmental defects , the findings of Lee et al . may lead to a better understanding of these diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "cell", "biology" ]	2013	H3K4 mono- and di-methyltransferase MLL4 is required for enhancer activation during cell differentiation
Nuclear pore complexes ( NPCs ) form a selective filter that allows the rapid passage of transport factors ( TFs ) and their cargoes across the nuclear envelope , while blocking the passage of other macromolecules . Intrinsically disordered proteins ( IDPs ) containing phenylalanyl-glycyl ( FG ) -rich repeats line the pore and interact with TFs . However , the reason that transport can be both fast and specific remains undetermined , through lack of atomic-scale information on the behavior of FGs and their interaction with TFs . We used nuclear magnetic resonance spectroscopy to address these issues . We show that FG repeats are highly dynamic IDPs , stabilized by the cellular environment . Fast transport of TFs is supported because the rapid motion of FG motifs allows them to exchange on and off TFs extremely quickly through transient interactions . Because TFs uniquely carry multiple pockets for FG repeats , only they can form the many frequent interactions needed for specific passage between FG repeats to cross the NPC . Nuclear pore complexes ( NPCs ) are the sole mediators of bi-directional nucleocytoplasmic trafficking . Transport is rapid and reversible , with the entire process of transport factor ( TF ) docking , passage and release across the NPC taking only a few milliseconds ( Strawn et al . , 2004; Hulsmann et al . , 2012 ) . The NPC consists of a ∼30-nm diameter central channel filled with phenylalanyl-glycyl-repeat-rich nucleoporins ( FG Nups ) which provide the selective filter . Depletion or deletion of the FG Nups results in leaky , non-selective barriers ( Strawn et al . , 2004; Di Nunzio et al . , 2012; Hulsmann et al . , 2012; Funasaka et al . , 2013 ) . Moreover , the FG domains in isolation facilitate selective passage of TFs through nanopores ( Jovanovic-Talisman et al . , 2009; Kowalczyk et al . , 2011 ) or accumulation in hydrogels ( Frey et al . , 2006; Frey and Görlich , 2007; Schmidt and Gorlich , 2015 ) . It is generally accepted that TF interaction with FG repeats reduce the diffusional barrier to enable selective transport ( Vasu and Forbes , 2001; Rout et al . , 2003; Suntharalingam and Wente , 2003; Zeitler and Weis , 2004 ) , though the molecular mechanism of TF passage through the NPC remains largely unknown . TFs may alter the properties of the FG permeability barrier . Mesoscale observations of TF-FG repeat interactions in vitro have shown that TFs can change the height of FG brushes on planar surfaces ( Lim et al . , 2007; Eisele et al . , 2010; Kapinos et al . , 2014; Wagner et al . , 2015 ) , modulate the transport of inert cargo ( Lowe et al . , 2015 ) , assemble with FG Nups into large assemblies ( Lowe et al . , 2015 ) , and inhibit amyloid hydrogel formation observed for some FG Nups ( Milles et al . , 2013 ) . Whether these behaviors arise from changes in FG structure or as a result of the multivalent FG-TF interaction remains undetermined , as do their contributions to in vivo nuclear transport . Several mesoscale models have attempted to explain how FG Nups prevent the passage of most macromolecules while allowing selective transport of TFs alone and with their cargo . The FG Nups have been proposed to form a selective barrier due to their reversible inter-chain cohesion ( hydrogel [Frey et al . , 2006; Frey and Görlich , 2007; Ader et al . , 2010] and bundle models [Gamini et al . , 2014] ) , entropic exclusion ( virtual gating model ) ( Rout et al . , 2000 , 2003; Lim et al . , 2006 ) , collapse upon TF binding ( reduction in dimensionality model ) ( Peters , 2005 ) , or a combination thereof ( forest model ) ( Yamada et al . , 2010 ) . These models differ in their predictions of FG behavior on their own ( ranging from highly mobile to fully self-associated ) and upon binding of TFs . For example , the hydrogel , forest and reduction in dimensionality models invoke large changes in FG Nup behavior upon TF interaction ( Lim et al . , 2007; Eisele et al . , 2010; Kapinos et al . , 2014; Wagner et al . , 2015 ) . Crucially , none of these models or current observations fully explains how transport can be both selective and rapid , as is seen in vivo . These conflicting models of the mechanism of NPC selectivity remain ( Schmidt and Gorlich , 2015 ) because of a lack of atomic-scale experimental data describing FG Nup behaviors and interactions . Therefore , we used nuclear magnetic resonance ( NMR ) techniques to provide a rich readout of FG Nup behaviors at atomic detail ( Figure 1A ) . We defined a minimal system including FG attachment , FG repeat type , and TF interaction , the essential features necessary to recapitulate selective transport in vitro ( Jovanovic-Talisman et al . , 2009; Kowalczyk et al . , 2011 ) . We measured the physical state of FG repeats with and without TFs bound . Because environment strongly affects intrinsically disordered proteins ( IDPs ) , including FG Nups ( Uversky , 2009; Wang et al . , 2011; Tetenbaum-Novatt et al . , 2012; Phillip and Schreiber , 2013 ) , we mimicked the normal environment of NPCs using Xenopus egg extract , the best characterized environment for in vitro nuclear transport measurements ( Dabauvalle et al . , 1991 ) . In addition , we tested the following: cytoplasm of living Escherichia coli using in cell NMR ( Serber et al . , 2005 ) ; E . coli high speed lysate ( Tetenbaum-Novatt et al . , 2012 ) ; and buffer alone , the latter lacking crowding agents or competitors and being the milieu in which these proteins have been most studied previously ( Frey et al . , 2006; Lim et al . , 2006; Ader et al . , 2010; Yamada et al . , 2010 ) . 10 . 7554/eLife . 10027 . 003Figure 1 . FG Nups are normally in a fully disordered and highly dynamic fluid state . ( A ) Our experimental approach includes key features of the NPC; a mixture of FG flavors , attachment at one end , and both specific ( TF ) and non-specific interactions with the cellular milieu . For example , our largest construct ( FG-N-FSFG-K-tet ) contains two fragments from Nsp1 ( FG-N , turquoise; FSFG-K , green; full-length Nsp1 also shown with residue numbering ) , a separator ( white ) and the tetramerization domain of p53 ( yellow ) . NMR analysis is performed on this construct and its variants , in milieu of various types; changes in position or intensity of peaks ( bottom right ) indicate changes in structure or interactions of the FG motifs . ( B ) Deviations of chemical shift values in cell ( Escherichia coli ) from predicted ( colored bars ) showing that FG-N and FSFG-K fragments are fully disordered , with no propensity for secondary structure . Also shown are standard errors of the mean ( gray bars ) and positions of FG motifs in the sequence ( gray columns ) . Chemical shift values expected for an α-helix or β-sheet are approximately −4 and +4 ppm , respectively , as shown in the small insets . ( C ) The FG-N and FSFG-K constructs show significant interactions with the cellular milieu and exhibit very rapid motions . The upper panels show the ratio of R2/R1 indicative of overall motion and effects of multiple environments ( chemical exchange ) in cell ( E . coli ) , compared to buffer A . The heteronuclear nuclear overhauser effect ( nOe ) is shown in lower panels , indicative of backbone motions . Gray columns indicate the locations of the FG motifs . Full experimental details and interpretations are available in ‘Materials and methods’ and Figure 1—figure supplements 1–4 . DOI: http://dx . doi . org/10 . 7554/eLife . 10027 . 00310 . 7554/eLife . 10027 . 004Figure 1—figure supplement 1 . Lack of indicated secondary structure in FG Nup constructs ( panels A–H , text ) . The most generally accepted indicator of propensity for secondary structure is the differences in deviations of 13C shifts of Cα and Cβ's ( e . g . , Spera and Bax , 1991; Wishart and Sykes , 1994; Wishart and Case , 2001 , illustrated in e . g . , Fushman et al . , 1998; McDonnell et al . , 1999; Mittag and Forman-Kay , 2007 ) , with α-helical segments showing a value of ∼ +4 , and β sheets a value of ∼ −4 . However , the standard deviations of the underlying reference values have not been widely available until recently ( Tamiola et al . , 2010 ) . In the above panels , the observed values of the difference of deviations for each residues is shown in blue ( -● ) while the range of the calculated standard errors is shown in red ( -- ) . Values which are ambiguous or missing because of overlaps are indicated by an open circle ( O ) . Panels are ( A ) FSFG-K in cell; ( B ) FSFG-K in buffer A; ( C ) same as B but those ‘degenerate’ sequences otherwise unassigned , are duplicated at the additional positions; for example , in this case the sequence surrounding P397 was most similar to that of P282 ( ATSKPAFSF and ATSKPALEH ) and the P282 values are used at position P397 for the 13Cα and 13Cβ assignments; ( D ) assignments of B extended; ( E ) FG-N in cell; ( F ) sequence duplicated values of E; ( G ) FG-N in buffer D; ( H ) Sequence extended values of set G . The data presented indicate no significant runs of deviation of the Δδ shift value , and none exceeding the standard deviations of the measurements . All conclusions in the paper are based on specific assignment , or grouped assignments in the case of overlaps and extended assignment , and do not rely on specific assignment in ambiguous/overlapped cases . When in cell assignments were ambiguous , and in buffer assignments were consistent with in cell observation , the in buffer assignments for specific residues were used . DOI: http://dx . doi . org/10 . 7554/eLife . 10027 . 00410 . 7554/eLife . 10027 . 005Figure 1—figure supplement 2 . Details of NMR relaxation data and derived correlation times . Upper: 15N[1H] relaxation data for FG-N and FSFG-K in respectively E . coli cell , and in buffer A ( ‘in vitro’ ) . Measurements were at 900 MHz ( red ) , 800 MHz ( blue ) , and 500 MHz ( black ) . For those with dual field measurements , the ratio at the two fields is shown in purple . The nOe is used to analyze rapid backbone motion in the panel below , and data will be further examined in future work . Lower: derived rotational correlation times , τn , for the N-H bond vector from the heteronuclear nOe . Data indicate that the field dependence of the observed nOe for FG-N in cell ( red and black in low left above ) is almost solely from the expected field dependence of the expected correlation , since the derived τn's are very similar ( purple and cyan in left below ) . The in buffer average values , ∼0 . 9 ns , are similar for the two constructs , consistent with their similar chemical shift and dynamic light scattering properties ( Figure 1 , Figure 4G ) . The measured errors for the FG-N are greater than those of the FSFG-K set , predominantly because accurate estimates of peak heights are more difficult in the presence of overlaps in the spectra . Although there appears to be some significant sequence dependence in the data for FSFG-K in-buffer ( green , right ) , the S/N in the FG-N case obscures whether this is present there also . Sequence dependence of the τn may arise from local restriction by immediately adjacent side chains . Values of τn in cell are increased modestly ( to ∼1 . 17 ns ) for both constructs and are assumed to reflect partial restriction of fast internal motions ( Neuhaus and Williamson , 2000 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10027 . 00510 . 7554/eLife . 10027 . 006Figure 1—figure supplement 3 . HSQC spectra of FSFG-K construct in buffer A vary with pH . HSQC spectra of FSFG-K in buffer A adjusted to the indicated pH's . In buffer , fast solvent exchange of significant number of peaks is apparent as expected ( Croke et al . , 2008 , 2011; Burz et al . , 2012 ) . The His6 tag signal titrates with the pH change and is circled at the lower right of each panel . DOI: http://dx . doi . org/10 . 7554/eLife . 10027 . 00610 . 7554/eLife . 10027 . 007Figure 1—figure supplement 4 . Stability of FG Nup constructs by DLS . Stability of size for FG-N ( left ) and FSFG-K ( right ) in buffer A by dynamic light scattering . The raw decay times are used so no interpretive models are applied . FG-N shows a sharp transition to a gel-like state after about 7 hr . Protein concentrations 8 mg/ml . For detailed methods see ‘Dynamic Light Scattering’ . DOI: http://dx . doi . org/10 . 7554/eLife . 10027 . 00710 . 7554/eLife . 10027 . 008Figure 1—figure supplement 5 . Stability of the FG-N construct in cell ( left ) and in buffer ( right ) by NMR . Left panel: freshly made cells ( grown as ‘Materials and methods’ ) ; induced at OD 0 . 8 for 200 m were spun and resuspended in buffer E . Right panel: purified and concentrated FG-N was buffer exchanged directly from 8 M urea to buffer A . All spectrometer conditions remained constant between scans . Samples were contained in 5 mm Shigemi NMR tubes . The spectra are 1-D traces from the first block of HSQC spectra initiated at the listed times . DOI: http://dx . doi . org/10 . 7554/eLife . 10027 . 00810 . 7554/eLife . 10027 . 009Figure 1—figure supplement 6 . Transverse relaxation of FSFG-K in multiple environments . All R2 data were collected at 25°C with 140 μM 15N-labeled FSFG-K in buffer A with the following delays: 0 , 32 . 64 , 65 . 28 , 97 . 92 , 130 . 56 , 163 . 20 , 195 . 84 , 228 . 48 , 261 . 12 , and 326 . 40 ms . The concentrations of the different additives are as follows: E . coli lysate ( Wang et al . , 2011; Latham and Kay , 2014 ) 230 mg/ml; BSA 205 mg/ml Glycerol ( Li et al . , 2009; Wang et al . , 2011 ) 60% wt/wt; PVP-10 ( Li et al . , 2008 , 2009 ) 111 mg/ml; Trehalose ( Lins et al . , 2004; Chakrabortee et al . , 2010 ) 154 mg/ml . Sequence-dependent fluctuation is seen only with lysate or with BSA , indicating non-specific rapid interactions , in contrast to changes of viscosity ( Serber et al . , 2005; Li et al . , 2009; Wang et al . , 2011 ) , solvation ( Lim and Deng , 2009; Xue and Skrynnikov , 2011; Kim et al . , 2013 ) , or packing ( Li et al . , 2009; Uversky , 2009; Milles et al . , 2013 ) . The E . coli lysate concentration is comparable to the estimated in vivo total concentration ( Cayley et al . , 1991 ) . Interaction with BSA is consistent with previous observations ( Parks et al . , 1983; Higuchi et al . , 1994; Clarkson et al . , 1996; Fielding et al . , 2005; Simard et al . , 2006 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10027 . 009 We studied Nsp1 , the most tested and characterized FG Nup which has been shown in vitro to mimic transport faithfully ( Jovanovic-Talisman et al . , 2009; Hulsmann et al . , 2012 ) . We focused on two segments of Nsp1 for which there is a consensus that they prototypically represent the extreme flavors and behaviors of FG Nups ( Yamada et al . , 2010 ) ; ( i ) the N-terminal segment of low charge ( Asn-rich ) and irregularly spaced FG repeats observed to be highly cohesive and form amyloid hydrogels under certain conditions ( FG-N ) ( Ader et al . , 2010 ) and ( ii ) the central segment of significant charge ( Lys-rich ) with regular FSFG repeats typically observed to be highly soluble ( FSFG-K ) ( Patel et al . , 2007; Yamada et al . , 2010 ) . The isolated N-terminal domain forms selective hydrogels in vitro ( Ader et al . , 2010 ) and is able to replace Nup98-FG domain in reconstituted Xenopus nuclei to provide the primary barrier enabling selective transport ( Hulsmann et al . , 2012 ) . We studied these fragments individually and in combination in a wide range of constructs and conditions , including where the FG repeats are tethered to mimic their arrangement in the NPC ( Alber et al . , 2007 ) ( Table 1 , Figure 1—figure supplement 1 ) . 10 . 7554/eLife . 10027 . 021Table 1 . FG constructs preparedDOI: http://dx . doi . org/10 . 7554/eLife . 10027 . 021Abbreviation used in textSimple structureSequenceNumber of residues / monomerMWNsp 1 ref start numberData FigureFSFG-KM-Nsp1 ( 274–397 ) MDNKTTNTTPSFSFGAKSDENKAGATSKPAFSFGAKPEEKKDDNSSKPAFSFGAKSNEDKQDGTAKPAFSFGAKPAEKNNNETSKPAFSFGAKSDEKKDGDASKPFSFGAKPDENKASATSKPA12513 , 0702741 , 1S1 , 1S2FSFG-KM-Nsp1 ( 274–397 ) -LEHHHHHHMDNKTTNTTPSFSFGAKSDENKAGATSKPAFSFGAKPEEKKDDNSSKPAFSFGAKSNEDKQDGTAKPAFSFGAKPAEKNNNETSKPAFSFGAKSDEKKDGDASKPAFSFGAKPDENKASATSKPALEHHHHHH13314 , 1352741 , 2 , 1S1-6 , 2S1-2FG-NMGT-Nsp1 ( 48–172 ) -SHMHHHHHHMGTSAPNNTNNANSSITPAFGSNNTGNTAFGNSNPTSNVFGSNNSTTNTFGSNSAGTSLFGSSSAQQTKSNGTAGGNTFGSSSLFNNSTNSNTTKPAFGGLNFGGGNNTTPSSTGNANTSNNLFGATASHMHHHHHH13713 , 649481 , 1S1-2 , 1S4-5 , 2S1 ( FG-N ) - ( FSFG-K ) -Tet-6HisMGT-Nsp1 ( 48–172 ) -ASATSKPA-Nsp1 ( 284–397 ) -SHMGEYFTLQIRGRERFEMFRELNEALELKDAQAHMHHHHHHMGTSAPNNTNNANSSITPAFGSNNTGNTAFGNSNPTSNVFGSNNSTTNTFGSNSAGTSLFGSSSAQQTKSNGTAGGNTFGSSSLFNNSTNSNTTKPAFGGLNFGGGNNTTPSSTGNANTSNNLFGATAASATSKPAFSFGAKSDENKAGATSKPAFSFGAKPEEKKDDNSSKPAFSFGAKSNEDKQDGTAKPAFSFGAKPAEKNNNETSKPAFSFGAKSDEKKDGDASKPAFSFGAKPDENKASATSKPASHMGEYFTLQIRGRERFEMFRELNEALELKDAQAHMHHHHHH29230 , 23548 , 2843 , 4FSFG-KMGTSATSKPA-Nsp1 ( 284–397 ) -SHHHHHHMGTSATSKPAFSFGAKSDENKAGATSKPAFSFGAKPEEKKDDNSSKPAFSFGAKSNEDKQDGTAKPAFSFGAKPAEKNNNETSKPAFSFGAKSDEKKDGDASKPAFSFGAKPDENKASATSKPASHHHHHH13113 , 7212844 ( FSFG-K ) - ( FSFG-K ) -Tet-6HisMGTSATSKPA-Nsp1 ( 284–397 ) -ATSKPA-Nsp1 ( 284–397 ) -SHMGEYFTLQIRGRERFEMFRELNEALELKDAQAHMHHHHHHMGTSATSKPAFSFGAKSDENKAGATSKPAFSFGAKPEEKKDDNSSKPAFSFGAKSNEDKQDGTAKPAFSFGAKPAEKNNNETSKPAFSFGAKSDEKKDGDASKPAFSFGAKPDENKASATSKPASATSKPAFSFGAKSDENKAGATSKPAFSFGAKPEEKKDDNSSKPAFSFGAKSNEDKQDGTAKPAFSFGAKPAEKNNNETSKPAFSFGAKSDEKKDGDASKPAFSFGAKPDENKASATSKPASHMGEYFTLQIRGRERFEMFRELNEALELKDAQAHMHHHHHH28730 , 5042844 ( FG-N ) - ( FG-N ) -Tet-6HisMGCT-Nsp1 ( 48–172 ) -Nsp1 ( 48–172 ) -SHMGEYFTLQIRGRERFEMFRELNEALELKDAQAHMHHHHHHMGCTSAPNNTNNANSSITPAFGSNNTGNTAFGNSNPTSNVFGSNNSTTNTFGSNSAGTSLFGSSSAQQTKSNGTAGGNTFGSSSLFNNSTNSNTTKPAFGGLNFGGGNNTTPSSTGNANTSNNLFGATASAPNNTNNANSSITPAFGSNNTGNTAFGNSNPTSNVFGSNNSTTNTFGSNSAGTSLFGSSSAQQTKSNGTAGGNTFGSSSLFNNSTNSNTTKPAFGGLNFGGGNNTTPSSTGNANTSNNLFGATASHMGEYFTLQIRGRERFEMFRELNEALELKDAQAHMHHHHHH29629 , 927484 ( FG-N ) - ( FSFG-K ) -6HisMGT-Nsp1 ( 48–172 ) -ASATSKPA-Nsp1 ( 284–397 ) -SHHHHHHMGTSAPNNTNNANSSITPAFGSNNTGNTAFGNSNPTSNVFGSNNSTTNTFGSNSAGTSLFGSSSAQQTKSNGTAGGNTFGSSSLFNNSTNSNTTKPAFGGLNFGGGNNTTPSSTGNANTSNNLFGATAASATSKPAFSFGAKSDENKAGATSKPAFSFGAKPEEKKDDNSSKPAFSFGAKSNEDKQDGTAKPAFSFGAKPAEKNNNETSKPAFSFGAKSDEKKDGDASKPAFSFGAKPDENKASATSKPASHHHHHH25725 , 95548 , 2844 FG Nups were fully disordered and highly dynamic in all cellular milieu tested . In NMR measurements , the degree of secondary structure is correlated with the difference between the 13C chemical shift values of the α and β carbons relative to random coil values for each residue ( Schwarzinger et al . , 2001; Tamiola et al . , 2010 ) . The degree of secondary structure quantified in this way is near zero for all residues in FG-N and FSFG-K constructs in all conditions tested ( Figure 1B , Figure 1—figure supplement 1 ) ( Eliezer , 2009 ) . This behavior of intrinsic disorder , judged from chemical shifts , is strikingly robust , being seen under a wide variety of conditions and also ( for FSFG-K constructs ) in buffers of varying pH ( Figure 1—figure supplement 3 ) . Our observations in cellular milieux are in contrast to the behavior of the constructs containing the FG-N domain in buffer alone ( Hulsmann et al . , 2012 ) . Under these conditions , dynamic light scattering and NMR measurements indicate that FG-N forms a hydrogel-like material ( Figure 1—figure supplement 4 ) . The in buffer behavior of our construct is consistent with extensive previous observations of β-sheet formation and aggregation of the N-terminal domain of Nsp1 ( FG-N ) ( Frey et al . , 2006; Frey and Görlich , 2007; Ader et al . , 2010; Labokha et al . , 2013 ) . However , when FG-N constructs are observed by NMR inside a living cell , or in the presence of cell lysates or mimics , there is no appearance of high molecular weight components—NMR spectra of FG-N constructs in living E . coli did not change with time over more than 24 hr ( Figure 1—figure supplement 5 ) . These data indicate that the cellular milieu is a strong inhibitor of intermolecular FG repeat aggregation . The state of FG repeats in our system is therefore highly dynamic , as has been reported in vivo ( Mattheyses et al . , 2010 ) . To understand the differences between protein structure and dynamics in cellular mimics as compared to buffer , we measured spin relaxation parameters ( R1 , R2 , nOe ) which quantify the motion and interactions of the residues on timescales of μs to ms ( R1 , R2 , ) and ps to ns ( nOe ) . NMR relaxation properties of FG Nups are significantly different between buffer alone and a protein-rich environment ( Figure 1C ) . We measured large increases in R2/R1 , indicative of transient spectral changes on interaction between the constructs and the cellular milieu ( Figure 1C , Figure 1—figure supplement 2 ) . The increases in R2/R1 were not seen upon increasing viscosity or crowding by inert agents ( Figure 1—figure supplement 6 ) , indicating that these changes result from weak binding of the FG repeats to the proteinaceous milieu as suggested by previous studies ( Tetenbaum-Novatt et al . , 2012 ) . The phenylalanines and their adjacent residues show the primary interactions , with the greatest increases in R2/R1 , while the spacer sequences remained relatively unaffected . The nuclear Overhauser effect ( nOe ) data show that the FG-N are highly mobile and flexible ( Figure 1C , Figure 1—figure supplement 2 ) , indicating no sequence-specific compaction , folding , or molten globule formation ( Uversky , 2009; Theillet et al . , 2014 ) . The cellular milieu is thus in a state of constant , non-specific , dynamic interaction with the FG repeats . Our results suggest that interactions of FG repeats with the cellular milieu stabilize the unfolded state by engaging the hydrophobic phenylalanines in transient interactions that decrease their contact with the water , reducing the driving force for hydrophobic collapse and amyloid formation . This is consistent with previous work that demonstrated that mutation of the Fs to more hydrophilic residues inhibits aggregation ( Frey et al . , 2006 ) . Thus , exchange with the cellular milieu maintains the FG Nups in a highly dynamic , disordered state by inhibiting the intramolecular interactions that lead to aggregation in buffer . We used the atomic-level readout of our system to interrogate key aspects of FG Nup-TF interactions that have previously been inaccessible . It has long been established using crystallographic , computational , and NMR approaches that FG residues bind to hydrophobic pockets on TFs ( reviewed in Stewart , 2006 ) . However , the dynamics of this interaction and the behavior of the spacer regions between FG residues upon TF binding have remained uninterrogated , leading to proposals ranging from remaining disordered to significant structural rearrangements ( Rout et al . , 2003; Peters , 2005; Frey et al . , 2006; Lim and Deng , 2009 ) . Our assays provide a direct readout of the dynamics of the interaction and the state of the linker regions upon TF binding ( Figure 2 ) . 10 . 7554/eLife . 10027 . 010Figure 2 . The interaction of FG Nups with the transport factor Kap95 is specific to the FG sites and leaves the spacers highly mobile ( A ) Overlay of spectra with varying concentrations of Kap95 ( 0 , 6 . 125 , 12 . 5 , 25 , 50 μM ) in the presence of [U-15N] FSFG-K ( 25 μM ) in Xenopus egg extract , showing only the FG motifs strongly interact with the TF . Several peaks represent overlapping similar sequences with indistinguishable attenuation . ( B ) Superimposed values of attenuation ( ( 1 − I ) /Io ) on addition of Kap95 across the sequence; increased attenuation indicates a stronger interaction propensity for that residue ( positions of FG motifs in the sequence are indicated by gray columns ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10027 . 01010 . 7554/eLife . 10027 . 011Figure 2—figure supplement 1 . Reversibility of TF/FG Nup interactions . For each of FSFG-GK and FG-N's interactions with Kap95 , we demonstrated reversibility as follows . The spectra below can be cartooned as arising from the following , where the upper left panel represents the solution of an FG Nup . Addition of TFs ( red ) at high concentration forms a substantial amount of complex ( upper right ) . Addition of more FG Nup at the same concentration as upper left results in reversibility of a portion of the complex formed , and spectra identical to those formed by direct addition of TF in smaller amounts ( lower panels ) . If the complex was irreversible , then the dilution from the right upper complex , would lead to a summation of the upper two figures for the right lower , and would not show the intermediate chemical shifts seen . ( A ) HSQC spectra of complexes of FSFG-K and Kap95 , illustrating the formation of the same complex ( lower spectra ) by either admixture of the components ( lower left ) or dilution of the TF concentration from a higher value to the final by addition of FSFG-K ( lower right ) . At upper left , the region of the S , F residues of FSFG-K 50 μM is shown in the absence of Kap95 in buffer C . On makeup to 100 μM Kap95 , the resulting spectrum is at the upper right . Dilution with additional FSFG-K produces the spectra at the lower right . The lower spectra superimpose the starting spectrum ( upper left , red ) and the complex spectrum ( azure ) . ( B ) HSQC spectra of complexes of FG-N and Kap95 components ( lower left ) or dilution of the TF concentration from a higher value to the final by addition of FG-N ( lower right ) in buffer A . At upper left , the region of the N and F residues of FG-N 50 μM is shown in the absence of Kap95 . On addition of Kap95 to 0 . 5 molar equivalence ( 25 μM ) , the resulting spectrum is at the upper right . Dilution with additional FG-N produces the spectra at the lower right . The lower right spectra superimpose the starting spectrum ( upper left , blue ) and the complex spectrum ( green ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10027 . 01110 . 7554/eLife . 10027 . 012Figure 2—figure supplement 2 . 15N R2 titration of Kap95 and [15N]FSFG-K . Observed values of 15N R2 for FSFG-K at various concentrations , in the presence of 20 μM Kap95 . Experimental details in ‘NMR analysis’ in ‘Materials and methods’ . The 15N R2 values and their standard errors of the mean are plotted vs the log10 of the 15N FGFG-K concentration all in the presence of 20 μM Kap95 for the residue types of the FSFG motif . The values of 15N R2 are shown in green at the extreme right axis for free FSFG-K . The derived value fitted for the equation of Section S2 . 3 is shown as the dotted line with ‘X’ marks . The fitted values of Kd and Rb for the derived line are 36 . 1 μM and 25 . 5 s−1 . The apparent Kd is then equivalent to 216 μM for each FSFG site in FSFG-K . Lines connect the observed or calculated values . DOI: http://dx . doi . org/10 . 7554/eLife . 10027 . 012 Many previous in vitro observations of FG-TF interactions lacked demonstration of reversibility and showed strong affinities ( Frey and Görlich , 2009; Tetenbaum-Novatt et al . , 2012; Labokha et al . , 2013; Kapinos et al . , 2014 ) , incompatible with the rapid transport rates observed in vivo . By monitoring the recovery of the NMR spectral characteristics on dilution of Kap95 , the interacting TF , we showed that the FG repeat-TF interactions are fully reversible in our system ( Figure 2—figure supplement 1 ) . We used measured changes in amide resonances peak position and intensity as a function of Kap95 concentration using 2D Heteronuclear Single Quantum Coherence ( HSQC ) experiments ( Figure 2A ) . Each ( non-proline ) residue contributes a peak , with the amide proton resonance frequency on the x-axis and nitrogen resonance frequency on the y-axis . The signal intensity of each peak is sensitive to the mobility and environment of that residue; upon binding to a TF , the effects of slower motion and of differing environment result in the signal of the corresponding peak decreasing . Strikingly , although the phenylalanine repeats themselves are attenuated upon Kap95 interaction , the majority of peaks are minimally affected , showing that the corresponding residues remain disordered and dynamic ( Figure 2A ) . The spacer regions are thus highly mobile even within the interacting state . An extreme of this observation is seen for K332 , which shows no attenuation ( left subpanel in Figure 2A , lowest trace Figure 2C ) , and thus this residue remains flexible both in free and when bound to TF . When we examined attenuation as a function of residue number , we observed a striking periodic pattern of minimal attenuation for most residues and significant attenuation for FG repeats ( Figure 2B ) . The degree of signal attenuation falls off relatively uniformly at both sides of each FG repeat , consistent with the transient binding of individual FG residues , with the remaining residues remaining fully disordered and highly dynamic . The minimally attenuated residues include those that have previously been implicated in forming phase changes of the FG repeat regions ( Ader et al . , 2010 ) . We did not observe any major state change of spacers resulting from TF addition , such as caused by the formation or breakage of secondary structure , or gel/sol transitions ( Ader et al . , 2010; Labokha et al . , 2013 ) . We observe that the binding interface between FG Nups and TFs at each interacting site is small—just the 2–4 residues surrounding the FG repeat itself—and the residues that do not interact directly with the TF sites remain highly mobile and dynamic . The size of the binding interface is consistent with other known transient interactions ( Clackson and Wells , 1995; Morrison et al . , 2003; Tompa et al . , 2009; Dixon et al . , 2011; Ozbabacan et al . , 2011 ) . We estimated a minimum effective affinity of Kap95 for the FSFG-K from a titration observing 15N R2 ( Figure 2—figure supplement 2 ) ( Su et al . , 2007 ) . The true Kd is greater than 36 μM , fully consistent with rapid , reversible transport . Interaction or binding while remaining primarily disordered is highly unusual , as many functional IDPs are believed to adopt significant secondary structure formation upon interaction ( Wright and Dyson , 2009; Uversky and Dunker , 2013 ) , which we do not see here . This behavior is reminiscent of that proposed for random fuzzy complexes , in which one partner remains dynamically disordered and transient interactions predominate during selective recognition ( Sigalov , 2011; Fuxreiter and Tompa , 2012 ) . Many observations suggest that there are multiple transport routes across the NPC , with FG Nups arranged in a variety of geometries ( Akey and Radermacher , 1993; Goldberg and Allen , 1996; Gant et al . , 1998; Kiseleva et al . , 2004; Alber et al . , 2007; Burns and Wente , 2012; Laba et al . , 2014 ) . Thus , because the NPC contains a large number of FG Nups in different arrangements to each other , we investigated the common features of the effects of local packing and attachment on FG repeats ( Figure 3 , Table 1 ) . We designed chimeric proteins containing the FG domains in homotypic and heterotypic combinations . Some constructs included the tetramerization domain of p53 , allowing us to form complexes that mimic the attachment of the FG domains to the NPC . We found essentially no change in chemical shifts and only modest changes in linewidth upon attachment to the p53 tetramerization domain ( Figure 3 ) . Any ordered protein of this size ( ∼120 kDa ) would be invisible using our NMR experiments , demonstrating that FG Nups remain dynamic IDPs even in large complexes , as previously measured for other large IDPs ( Kosol et al . , 2013 ) . Notably , our largest constructs have a diameter measured by dynamic light scattering of 13 nm ( Figure 3B ) , a significant fraction of the 30-nm diameter of the NPC . Our results reveal that the behavior of a given FG repeat is not affected strongly by the number or type of other FG repeats surrounding it ( i . e . , no ‘emergent properties’ ) . Our results also strongly indicate that there are no significant FG–FG interactions , as interactions between different types of FGs should impact their average environment . However , we cannot completely exclude that extremely weak , dynamic interactions can occur between FG repeats , and indeed such interactions may modulate the long distance distribution of FG repeat regions in the NPC . The FG Nups thus remain a highly dynamic fluid even at sizes , arrangements , and packing densities commensurate with the NPC . We observed the same pattern of rapid exchange of the FG residues with TF binding sites while the spacer regions remain highly mobile in all environments studied and with all constructs tested , showing the robustness of our findings ( Figure 4 ) . 10 . 7554/eLife . 10027 . 013Figure 3 . Interaction of a high molecular-weight tetramerized hybrid construct indicates that the binding mechanism is robust to changes in FG type , packing , and environment . ( A ) 1H[15N] Heteronuclear Single Quantum Coherence ( HSQC ) spectra of the FG-N-FSFG-K-tet sequence ( Tables 1 and 2 ) ( MW 120 kDa ) in the unbound ( orange ) and Kap95-bound ( blue ) states with Xenopus egg extract as the milieu . Peaks that undergo signal attenuation in the Kap95-bound spectrum indicate binding of those residues to Kap95 . Typically , we observe that it is the FG residues ( pink boxes ) , but not intervening spacer residues ( no boxes ) , that attenuate upon binding to the transport factor Kap95; however , there are four FG repeats that are minimally attenuated ( blue boxes ) , and a small number of spacer residues that are attenuated ( purple boxes ) . ( B ) Schematic of the construct used , labeled as in ( A ) . Very small chemical shift changes are observable for a few sites , consistent with rapid exchange into multiple inhomogeneous sites . DOI: http://dx . doi . org/10 . 7554/eLife . 10027 . 01310 . 7554/eLife . 10027 . 014Figure 4 . FG constructs show no significant FG–FG interactions , and remain fully disordered even when tethered at one end and packed at high density . ( A–F ) We developed a wide range of constructs containing fragments from Nsp1 ( turquoise and green , numbering in Figure 1 ) and the small tetramerization domain from p53 ( yellow ) . Schematics illustrate the different constructs as studied within the cellular milieu ( blue circles ) which vary both in total molecular weight , crowding of the FG repeats via tetramerization , and composition of FG fragments ( homotypic or heterotypic ) . Below each schematic is the HSQC spectrum of the glycine region of the represented construct observed in cell ( E . coli ) . All constructs are tabulated in Table 1 . ‘M’ identifies apparent 15N-labeled metabolites which are variable in intensity from preparation to preparation . ( G ) Dynamic light scattering of three representative constructs in buffer , showing homogeneity and consistency with expected size of constructs . DOI: http://dx . doi . org/10 . 7554/eLife . 10027 . 014 While all overall behaviors are similar between FG-N and FSFG-K segments , in monomers or in tetramers , there are some intriguing differences . For example , in constructs where both FG types are present ( Figure 4 ) , not all FG-N phenylalanine residues are completely attenuated . This may be evidence for specificity of certain FG repeat types to binding particular TFs . The FG repeats fill the central channel of the NPC to form a barrier to non-specific macromolecular diffusion ( Patel et al . , 2007; Tetenbaum-Novatt and Rout , 2010 ) . We show that , even when densely packed and in mixed flavors as in the NPC , the FG repeat regions studied here remain fully disordered and highly dynamic . The two domains studied represent the extreme behaviors observed for FG Nups in vitro ( Patel et al . , 2007; Ader et al . , 2010; Yamada et al . , 2010 ) : FSFG-K is highly soluble and non-cohesive , while FG-N in buffer is highly cohesive , a prototypical hydrogel-forming FG Nup . It is clearly possible to make FG repeats aggregate ( Figure 1—figure supplements 4 , 5 ) ( Ader et al . , 2010 ) , as has been demonstrated for many proteins ( Goldschmidt et al . , 2010 ) . However , our data are inconsistent with any such hydrogel state in cellular milieu ( Frey et al . , 2006; Labokha et al . , 2013 ) . Re-arrangements of internal non-covalent crosslinks within hydrogels are on slow ( second to minute ) timescales , long enough to form a solid gel resistant to deformation , and too slow for solution NMR analysis ( Ader et al . , 2010 ) . Instead of such a highly internally interacting , slowly moving gel , our results demonstrate that FG repeat regions form a highly dynamic phase , consistent with the rapid rates of nuclear transport . A highly dynamic , fluid state for FG repeats in vivo is in agreement with measurements of the living NPC ( Atkinson et al . , 2013 ) . FG Nups do not appear to form a molten globule or collapsed state , as has also been proposed ( Yamada et al . , 2010 ) , instead behaving as a fully disordered IDP ( Rout et al . , 2003; Lim et al . , 2006; Lim and Deng , 2009; Atkinson et al . , 2013 ) . Our results also allow us to distinguish their IDP class ( Uversky , 2011; van der Lee et al . , 2014 ) . FG repeats do not appear to populate more folded states either inter- or intra-molecularly upon interaction with either TFs or other FG Nups ( Wright and Dyson , 2009; Yamada et al . , 2010; Han et al . , 2012; Kato et al . , 2012 ) . Instead , FG repeats lack any secondary structure , even when interacting with cognate binding partners , indicating that IDP mobility may be a key ingredient in nuclear transport . Our results highlight the importance of the cellular environment for determining the behavior of the FG Nups because multiple weak interactions with cellular components stabilize the FG Nups in a disordered state . We propose that the selectivity of the NPC is thus maintained by both non-specific and specific cellular interactions , consistent with recent work demonstrating that FG permeability barrier is modulated by TFs ( Lim et al . , 2007; Eisele et al . , 2010; Milles et al . , 2013; Kapinos et al . , 2014; Lowe et al . , 2015; Wagner et al . , 2015 ) . We show that the modulation of permeability of FG brush behavior by TFs is not a result of structural changes of the FG Nups , but must result instead from the rapid , multivalent interactions of TFs with FG repeats ( Tetenbaum-Novatt et al . , 2012; Schleicher et al . , 2014 ) . Our data , strongly supporting that FG repeats are highly dynamic tethered IDPs , imply that they must form entropic bristles ( i . e . , strongly sterically hindered regions ) around their tether site ( the central channel of the NPC ) ( Rout et al . , 2003; Strawn et al . , 2004; Lim et al . , 2006 , 2007; Lim and Deng , 2009; Tetenbaum-Novatt and Rout , 2010 ) . This hindrance increases with increasing size of the passing macromolecules; hence , small proteins can pass more easily than large ones ( Tetenbaum-Novatt and Rout , 2010 ) . These larger nonspecific macromolecules interact with the FG repeats with insufficient frequency to overcome the entropic barrier of the FG repeats' polymer bristle structure , thus being effectively excluded ( Tetenbaum-Novatt and Rout , 2010; Ma et al . , 2012 ) . TFs , however , have sufficient interaction frequency to do so—the original tenet of the ‘virtual gating’ idea , now delineated at the atomic scale by these results . In summary , the FG repeat regions , crowded around and within the central channel , may set up an entropic barrier that excludes macromolecules from their vicinity while permitting the approach of small molecules; however , macromolecules that interact with the FG repeats ( such as TFs ) can overcome this barrier ( Rout et al . , 2003; Lim et al . , 2007 ) by their multiple and frequent interactions with FG Nups . We are pursuing experiments which we hope in the near future will gage the precise nature and magnitude of the exclusion mechanism . Our results suggest that FG disorder is critical to the speed of selective nucleocytoplasmic transport . Selectivity is determined by the free energy gain upon TF-NPC binding , with the off rate ( and so speed of transport ) constrained by koff = Kd·kon . For a given selectivity ( Kd ) , the residence time within the NPC is constrained by the off rate . IDPs are able to engage far more rapidly than most ordered proteins ( Zhou , 2012 ) , allowing for high on and off rates while maintaining selectivity . For FGs in particular , several key lines of evidence indicate that the interaction of FG Nups with TFs is extraordinarily rapid , that is , each FG motif is in extremely fast exchange with TFs . We have directly shown that the binding interface is only 2–4 amino acids ( the FG repeat itself ) . The buried surface area of this interaction is <1000 Å2 , consistent with other known transient interactions ( Clackson and Wells , 1995; Tompa et al . , 2009; Dixon et al . , 2011 ) . The spacer regions remain highly mobile and the degree of signal attenuation falls off relatively uniformly at both sides of each FG repeat ( Figure 2B ) , fully consistent with the transient binding of individual FG residues . Within the NPC , a TF is able to rapidly diffuse , engaging transiently with multiple FG motifs across different FG Nups . The association rate of TFs with individual FG motifs is very fast because the FGs move very rapidly and the TF has multiple binding pockets available for interaction ( Bayliss et al . , 2000 , 2002; Bednenko et al . , 2003; Isgro and Schulten , 2005 , 2007; Liu and Stewart , 2005 ) . Taken together , these results indicate that the dynamics of FG-TF interaction are extraordinarily rapid . As a result , the interaction can be strong enough to be selective , and yet remain extremely fast , allowing rapid transit through the NPC . In essence , it is the quantity , rather than quality , of interactions possible per molecule with FG repeats that distinguishes a TF from other macromolecules ( Figure 5 ) . Transient non-specific interactions of the FG repeats with their milieu help maintain the disordered and dynamic nature of the barrier . Given the low sequence and structural complexity of the FG repeats , such non-specific interactions are to be expected but are not strong enough for these non-specific macromolecules to overcome the FG Nups' diffusion barrier ( Rout et al . , 2003 ) . In contrast , TFs bind specifically to the FG motifs through pockets tuned for this purpose ( Stewart , 2003 , 2006 ) while leaving the spacers between the motifs highly mobile . The fact that the spacers—and the FG repeats as a whole—remain fully mobile even when interacting with a TF also means that the barrier to non-specific passage formed by a dense fluid of FG repeats in the NPC remains fully intact , regardless of even large TF fluxes ( a significant issue with models that invoke state changes , which our results and model circumvent ) . 10 . 7554/eLife . 10027 . 015Figure 5 . Molecular description of the nuclear transport mechanism . Specific conclusions from our results , as discussed in the main text , are illustrated with ( A ) details from a docked molecular simulation of the TF Kap95 ( purple ) and the FG repeat region FSFG-K , rendered in Chimera ( see ‘Materials and methods’ ) , and ( B ) a diagrammatic view of the molecules involved . FG repeat ( green; an ensemble of disordered conformers is illustrated in A ) . Phe ( orange ) , Phe-binding pockets ( B , orange circles ) , TF ( pink ) , non-specific macromolecule ( B , blue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10027 . 015 While the FG repeats show little evidence of changed behavior upon oligomerization or TF interaction , it is likely that their structured attachment sites and diversity of flavors produce modulation of composition within the NPC . These variations are potentially important for the organization of alternate transport pathways ( Strawn et al . , 2004; Terry and Wente , 2007; Yamada et al . , 2010 ) . Our approach and its future extensions will thus be a powerful tool for the detailed characterization of these different nuclear transport pathways . The NPC is a biological example of the idiom ‘many hands make light work’; the combined effect of many weak , transient interactions provides a specific multifunctional transport system , which allows for the simultaneous passage of hundreds of different macromolecules with a wide range of sizes and for robustness to significant alterations in the cell ( Wente and Rout , 2010; Adams and Wente , 2013; Tran et al . , 2014 ) . The FG-N , and FSFG-K segments were derived as previously described ( Tetenbaum-Novatt et al . , 2012 ) with the following modifications . Expression plasmids ( pET24a , pRSFDuet ) containing FG Nup fragments were transformed into BL21DE3 Gold cells ( Agilent ) . Cells containing pET24a or pRSF constructs were grown to OD600 of 0 . 8 , induced with 1 mM IPTG and harvested after 2–4 hr . The cells' periplasm was removed by osmotic shock ( Magnusdottir et al . , 2009 ) , and lysed under denaturing conditions ( 8 M urea ) . Urea was maintained throughout the purification for FG-N , whereas FSFG-K had 8 M urea in the lysis , 3 M urea in the first wash , and no urea in the remaining washes and elution . Proteins were purified on Talon resin in 20 mM HEPES , 150 mM KCl , 2 mM MgCl2 , ( Buffer A , at pH 7 . 0 ) , with protease inhibitor cocktail ( PIC ) , PMSF , and pepstatin . Kap95 was prepared as previously described ( Tetenbaum-Novatt et al . , 2012 ) [U-15N] and [U-13C , 15N] materials were prepared using M9 media containing 15NH4Cl and [U-13C] glucose ( Cambridge Isotopes Limited , MA ) as needed . The tetramerization domain of p53 was used to provide intramolecular crowding by construction of fusions with the tetramerization segment ( Poon et al . , 2007 ) . As with all our constructs , the preparation was analyzed by size exclusion chromatography ( SEC ) to ensure that we were examining un-aggregated protein of the correct size and quantity ( data not shown ) and confirmed by dynamic light scattering . Experiments were conducted on Bruker spectrometers at 700 MHz and 298 K unless otherwise indicated . In cell NMR methods used the general procedures developed for STINT-NMR ( Burz et al . , 2006a , 2006b , 2012 ) , and other IDP studies . Assignment of resonances used the standard triple resonance approach as in Sattler et al . ( 1999 ) ; Muralidharan et al . ( 2006 ) ; Tait et al . ( 2010 ) ; Lemak et al . ( 2011 ) ; Kalinina et al . ( 2012 ) , including HNCO , HNCACO , HNCACM , CBCACINH , HNCA , HNCOCA , CCCONH , and HCCONH . Assignments were conducted in cell ( E . coli ) /buffer E and in buffer A , for the sets tabulated in Table 2 , and were confirmed in other buffers by examination of HSQCs . Relaxation methods used standard procedures ( Barbato et al . , 1992; Ferrage et al . , 2006 , 2009 , 2010 ) . No corrections were applied for possible changes of fast exchange with solvent water , as described for R2 ( CPMG ) ( Kim et al . , 2013 ) for the following reasons . The observed changes for the IDP α-synuclein are less than two fold , and sequence-independent for a change of pH from 7 . 4 to 6 . 2 ( Figure 1; Kim et al . , 2013 ) while for FG Nups we observe up to c . sixfold sequence dependent variation of R2 . Secondly , the same sequence dependent variations of R2 are seen in buffered E . coli extracts , with controlled pH ( Figure 1—figure supplement 6 ) . ‘Titration’ experiments were conducted at 800 or 700 MHz in buffers indicated , by preparation of separate samples for each change of concentration of Kap95 . 10 . 7554/eLife . 10027 . 022Table 2 . Sequence assignments filed with BRMB ( Ulrich et al . , 2008 ) DOI: http://dx . doi . org/10 . 7554/eLife . 10027 . 022FGFG-K in cell E . coli expression25 , 182FSFG-K in buffer25 , 183FG-N in cell E . coli expression25 , 184FG-N in buffer25 , 185 Titration experiments for 15N R2 were conducted at 500 MHz in buffer A , by preparing separate samples . Kap95 was fixed at 20 μM and 15N FSFG-K concentration was varied from 12 . 5 μM to 1200 μM . Each sample was prepared by addition of standard volumes of stocks to obtain the concentrations . No additional normalization was used . Each R2 data acquisition used 6 delay points . For FSFG:Kap95 molar ratio of 1:1 and 0 . 625:1 the following delay times were used 0 , 32 . 64 , 65 . 28 , 97 . 92 , 130 . 56 , 163 . 2 ms . For FSFG:Kap95 molar ratio of 4:1 the following delay times were used 0 , 81 . 6 , 163 . 2 , 244 . 8 , 326 . 4 , 408 ms . For FSFG:Kap95 molar ratio of 7:1 the following delay times were used 0 , 81 . 6 , 163 . 2 , 244 . 8 , 326 . 4 , 489 . 6 ms . For FSFG:Kap95 molar ratios of 60:1 , 28 . 75:1 and 11:1 the following delay times were used 0 , 81 . 6 , 163 . 2 , 326 . 4 , 489 . 6 , 652 . 8 ms . The same delays were used for an FSFG-N alone and the R2 values obtained provided the value of R2 free ( R2 ( 0 ) ) in the binding equation . The numbers of scans were optimized for S/N for each sample and total acquisition time per sample was ∼<13 hr . The chemical shifts and their standard deviations ( Tamiola et al . , 2010 ) were used in the standard equation ( Schwarzinger et al . , 2001 ) to derive deviations from predicted shift values for the sequences of the assigned proteins . The values for these for Δδ13Cα- Δδ 13Cβ are shown in Figure 1—figure supplement 1 . For NMR relaxation analysis , we derive approximate correlation times for ps/ns , associated with the different correlation times for hetero nuclear nOe by a direct calculation method that , in part , normalizes across different observation fields/frequencies , avoiding any complexities of spectral density analysis ( for example , Wirmer et al . , 2006 ) , of the unverified applicability of a ‘Model Free’ approach ( Buevich et al . , 2001 ) , and of analysis of simulated molecular dynamics which may not be generally applicable to all IDPs ( Kleckner and Foster , 2011; Xue and Skrynnikov , 2011 ) . The apparent correlation times for motion of backbone amides was derived from use of the heteronuclear 15N[1H] nOe using the standard dipolar treatment ( Neuhaus and Williamson , 2000 ) , ignoring chemical shift anisotropy contributions . For this pair , the field independent form from the standard formulae ( p 135 in Nicholas et al . , 2010 ) is given by ( 1 ) Onoe=0 . 782804− ( 0 . 010000683009+0 . 5132297763 ( fhτn ) 2 ) ( 0 . 002123332299+0 . 1102575629 ( fhτn ) 2+ ( fhτn ) 4 ) , where fH is the 1H frequency ( Hertz ) and τn is the apparent correlation time for the ns motion associated with direct dipolar relaxation ( Neuhaus and Williamson , 2000; Cavanagh et al . , 2007 ) . We use here τn to discriminate from the overall correlation time for folded protein motion usually denoted as τc . τn may be directly obtained as ( 2 ) τn=[ ( − ( 2 . 418830739 . 1030Onoe+2 . 02824096 . 1031 ( 0 . 005729037113Onoe2+0 . 07341631796Onoe+0 . 2024235645 ) 1/2+8 . 516066166 . 1030] ( 4 . 056481921 . 1030Onoe−3 . 175430273 . 1031 ( /fh . Dynamic Light Scattering ( DLS ) measurements were made on a Dynapro Plate Reader ( Wyatt Instruments , Santa Barbara , CA ) at 298 K , in a 384 well plate with typically triplicate samples . Curves in Figure 4G are summations from instrument measurements . The mass-weighted averaged radii ( and standard errors of the mean ) are 2 . 46 ( 0 . 15 ) nm; 2 . 63 ( 0 . 14 ) nm; 6 . 47 ( 0 . 30 ) nm for FG-N , FSFG-K and FG-N-FSFG-K-tet , respectively . Concentrations used were 8 , 4 , and 2 mg/ml . Concentration dependencies were small . For the distribution curves the Dynapro software DYNAMICS 7 . 1 . 0 . 25 was used to analyze the experimental correlations with multiple 1 s runs , standard filtering including exclusions of <1% mass peaks , and regularization using the coil setting ( Hanlon et al . , 2010 ) . The resulting binned regularized data was extracted and for each protein was summed across wells by scaled Gaussian summation at the peak position and width of the log-scale time base . Note that cumulant analysis ( Koppel , 1972 ) is not practical for multi-component systems with a large dynamic range . In each case the peak in the 1–10 nm regions represent >99% of the scattering mass . Averages and standard errors of the mean were calculated by pooling each mass-weighted average per well . Figure 2—figure supplement 4 shows raw decay time data from DLS for FG–N and FSFG-K constructs in standard buffer A , 8 mg/ml . The vertical axis represents the % of mass having a specific decay time , with period in hours after start shown in the NE axis . The presentation of decay time avoids any issue of model for specific molecular shape or density or validity of the Raleigh-Debye approximation ( Berne and Pecora , 2000 ) . Data are pooled from a triplicate reading using the setting indicated above . A: 20 mM HEPES , 150 mM KCl , 2 mM MgCl2 , pH 6 . 5 unless otherwise noted;B: Xenopus egg extract: protein egg extract prepared according to ( Shechter et al . , 2004 ) containing 31 mg/ml protein and was diluted with buffer A ( 3 + 10 v + v ) with a final pH of 6 . 5;C: E . coli lysate: lyophilized protein from high-speed spin dissolved in buffer A , at a final concentration of 30 mg/ml;D: Buffer A plus 100 mg/ml BSA;E: 20 mM Tris pH 7 . 5 , 150 mM NaCl ( Burz et al . , 2006 ) . HADDOCK , a rigid body docking algorithm ( Dominguez et al . , 2003 ) , was used to combine Kap95 and FSFG-K into possible orientations to provide a molecular representation of our findings . Specifically , the crystal structure of Kap95 ( 3ND2 ) and a random conformations of FSFG-K was used as inputs and the FSFG-K repeat residues and residues on Kap95 identified by Isgro et al . ( Isgro and Schulten , 2005 ) were chosen as active residues . Models were subsequently altered and images were rendered using Chimera ( Yang et al . , 2012 ) .	Eukaryotic cells have a nucleus that contains most of the organism's genetic material . Two layers of membrane form an envelope around the nucleus and protect its contents from the rest of the cell's interior . However , this protective barrier must also allow certain proteins and nucleic acids ( collectively called ‘cargo’ ) to move in and out of the nucleus . Cargo molecules can pass through channel-like structures called nuclear pore complexes , which are embedded in the nuclear envelope . However , transport across this barrier is highly selective . While small molecules can pass freely through nuclear pore complexes , larger cargo can only be transported when they are bound to so-called transport factors . The nuclear pore complex is a large structure made up of more than 30 different proteins called nucleoporins . Like all proteins , nucleoporins are built from amino acids . Many nucleoporins contain repeating units of two amino acids , namely phenylalanine ( which is often referred to as ‘F’ ) and glycine ( or ‘G’ ) . These ‘FG nucleoporins’ are found on the inside of the nuclear pore complex and interact with transport factors to allow them to transit across the nuclear envelope . Several models have been put forward to explain how FG nucleoporins block the passage of most molecules . But it was unclear from these models how these nucleoporins could do this while simultaneously allowing the selective and fast transport of nuclear transport receptors . There was also a major lack of experimental data that probed the behavior of FG nucleoporins in detail . Hough , Dutta et al . have now used a technique called nuclear magnetic resonance spectroscopy ( or NMR for short ) to address this issue . NMR can be used to analyze the structure of proteins and how they interact with other molecules . This analysis revealed that FG nucleoporins never adopt an ordered three-dimensional shape , even briefly; instead they remain unfolded or disordered , moving constantly . Nevertheless , and unlike many other unfolded proteins , FG nucleoporins do not aggregate into clumps . This is because they are constantly changing and continuously interacting with other molecules present inside the cell , which prevents them from aggregating . Hough , Dutta et al . also observed that the repeating units in the FG nucleoporins engaged briefly with a large number of sites or pockets present on the transport factors . These FG repeats can bind and then release the transport factors at unusually high speeds , which enables the transport factors to move quickly through the nuclear pore complex . This transit is specific because only transport factors have a high capacity for interacting with the FG repeats . These findings provide an explanation for how the nuclear pore complex achieves fast and selective transport . Further work is needed to see whether certain FG nucleoporins specifically interact with a particular type of transport factor , to provide preferred transport routes through the nuclear pore complex .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2015	The molecular mechanism of nuclear transport revealed by atomic-scale measurements
Human perception is invariably accompanied by a graded feeling of confidence that guides metacognitive awareness and decision-making . It is often assumed that this arises solely from the feed-forward encoding of the strength or precision of sensory inputs . In contrast , interoceptive inference models suggest that confidence reflects a weighted integration of sensory precision and expectations about internal states , such as arousal . Here we test this hypothesis using a novel psychophysical paradigm , in which unseen disgust-cues induced unexpected , unconscious arousal just before participants discriminated motion signals of variable precision . Across measures of perceptual bias , uncertainty , and physiological arousal we found that arousing disgust cues modulated the encoding of sensory noise . Furthermore , the degree to which trial-by-trial pupil fluctuations encoded this nonlinear interaction correlated with trial level confidence . Our results suggest that unexpected arousal regulates perceptual precision , such that subjective confidence reflects the integration of both external sensory and internal , embodied states . Our subjective feeling of confidence enables us to monitor experiences , identify mistakes , and adjust our decisions accordingly . It is therefore crucial to understand what underlies this feeling; for example , does only the quality of available sensory signals matter , or do our confidence reports also reflect internal bodily states , such as arousal ? Although confidence is thought to depend upon the quality or strength of sensory evidence , convergent computational theory and experimental data highlight the role of interoceptive inferences in guiding exteroceptive awareness . In this sense , confidence may be a metacognitive integration of both internal and external sources of uncertainty . Here , we address this possibility using a novel psychophysical design , in conjunction with signal theoretic modelling of confidence , to assess the degree to which sensory uncertainty depends upon unexpected arousal . Computationally , confidence is typically described as the output of a feed-forward ideal statistical observer monitoring sensory ( or decision ) evidence . Confidence is thus determined solely by the quality or strength of sensory inputs relative to a late-stage criterion or threshold . For example , in signal detection theory , sensory samples whose average intensity fall beyond a confidence criterion are ascribed a higher certainty ( Galvin et al . , 2003; Lau and Rosenthal , 2011; Maniscalco and Lau , 2012 ) . Similarly , ballistic accumulation models suggest that confidence relates to the speed of evidence accumulation relative to decision threshold ( Kiani and Shadlen , 2009; Kiani et al . , 2014 ) . In both cases , confidence is generated by the bottom-up read-out of sensory information relative to a decision variable , and is assumed to depend on the same information underlying the accuracy of the perceptual decision itself . However , emerging evidence suggests that confidence can be influenced independently of choice accuracy; for example magnetic stimulation of the motor cortex specifically disrupts confidence but not accuracy for perceptual choice ( Fleming et al . , 2015 ) . Similarly , increased sensory noise reduces confidence even when difficulty is equated ( Spence et al . , 2016 ) . A potential physiological mediator of these effects is bodily arousal , which regulates affective salience and perceptual variability ( Critchley et al . , 2001; Murphy et al . , 2014b ) . Sudden increases in arousal trigger a reciprocal cascade of central , autonomic , and peripheral responses in the brain , heart , and pupil . Centrally , arousal is mediated by a reciprocal noradrenergic network with projections throughout the prefrontal , sensory , and limbic cortices ( Aston-Jones and Cohen , 2005; Murphy et al . , 2014a ) . This network of areas is also important for integrating perceptual and interoceptive signals ( Singer et al . , 2009; Critchley and Harrison , 2013; Salomon et al . , 2016 ) , error-awareness ( Fiehler et al . , 2004; Klein et al . , 2013 ) , and expected confidence or volatility ( Iglesias et al . , 2013; Schwartenbeck et al . , 2015 ) . While substantial evidence supports the integration of arousal and sensory information , these observations are difficult to reconcile with ideal observer models . In contrast , predictive coding emphasizes interoceptive inference , in which confidence reflects the precision ( or inverse variance ) of a higher-order belief about both internal states and external sensations ( Friston and Kiebel , 2009; Clark , 2015 ) . Neurobiologically , precision is encoded by the gain of local pyramidal cells ( Bastos et al . , 2012 ) , which is regulated across the cortical hierarchy by neuromodulators such as dopamine and noradrenaline ( Feldman and Friston , 2010; Friston et al . , 2012; Moran et al . , 2013; Kanai et al . , 2015 ) . The global regulation of precision by neuromodulatory gain control entails that unexpected changes in internal states should influence the estimation of confidence for other , exteroceptive channels . Predictive coding thus hypothesizes that the weight given to sensory noise depends upon expected interoceptive states , such as arousal and cardiac acceleration ( Gu et al . , 2013; Seth , 2013; Barrett and Simmons , 2015 ) . On this basis , we reasoned that an unexpected increase in arousal should reduce the influence of sensory noise on confidence . To test this hypothesis , we presented effectively salient , masked disgust cues in advance of a visual stimulus of variable sensory precision . Crucially , performance was equated across conditions such that changes in subjective uncertainty could be attributed to a precision-weighting mechanism , independently from any effect on choice accuracy . We further modelled evoked physiological responses ( heart rate and pupil dilation ) , to determine whether the encoding of sensory noise in these measures also depended upon cue-induced ‘arousal prediction error’ ( APE ) , and if this encoding was reflected in the trial-by-trial fluctuations of subjective confidence . To test these hypotheses , 29 participants performed the motion discrimination task illustrated in Figure 1 . On every trial a global motion stimulus was preceded by a masked disgust or neutral cue . Participants then discriminated the average direction of a cloud of moving dots and rated their confidence in this decision . We used disgusted faces as arousal cues as they signal salient interoceptive and affective challenge ( Chapman and Anderson , 2012 ) , and elicit increased arousal and physiological responses , including heart rate acceleration and facial mimicry , even when presented without awareness ( Vrana , 1993; Phillips et al . , 1997; Dimberg et al . , 2000; Chapman and Anderson , 2012 ) . Furthermore , all faces were masked from awareness , allowing us to discount any role of conscious demand characteristic in our cue-related effects . 10 . 7554/eLife . 18103 . 003Figure 1 . Arousal-Cued global motion task . Trial schematic illustrating our arousal-cued global motion task , in which an unexpected , masked disgusted face increased arousal just prior to a motion judgement and confidence rating . On each trial motion stimulus of variable precision ( 15 or 25 degrees standard deviation , σ ) were preceded by either a masked disgust or neutral face , followed by the perceived neutral mask . Participants then made a forced-choice motion discrimination and subjective confidence rating . Histogram and average luminance-matching was applied between conditions and frames to eliminate pupillary artefacts , see Materials and methods for more details . DOI: http://dx . doi . org/10 . 7554/eLife . 18103 . 003© 1998 , Karolinska Institutet , Psychology section , All Rights Reserved1998Lundqvist D , Flykt A , Öhman A . The Karolinska directed emotional faces ( KDEF ) [CD-ROM] . Stockholm . Department of Clinical Neuroscience Psychology Karolinska InstitutetFace stimuli images taken from the Karolinska Directed Emotional Faces database and adapted with permission ( ID AM25DIS ) . To assess the independent influence of sensory variance ( or precision ) , the average mean and variance of motion signals were manipulated orthogonally ( see Figure 2A ) using a global-motion stimulus ( see Spence et al . , 2016 for a similar technique ) . Crucially , to preclude an impact of task difficulty on confidence , discrimination performance was held constant ( 71% for low-variance trials ) by adaptively adjusting the mean motion signal across trials ( Figure 2B ) . Finally , to quantify the impact of sensory noise and disgust cues on perceptual choice and uncertainty , we applied a signal-theoretic approach to modelling confidence reports ( Galvin et al . , 2003; Maniscalco and Lau , 2012 ) . 10 . 7554/eLife . 18103 . 004Figure 2 . Overview of behavioral results . ( A ) Manipulation of sensory precision; stimulus probability density functions show our low ( 15 σ ) and high ( 25 σ ) variance conditions; stimulus mean and variance were orthogonally manipulated using a global-motion stimulus . ( B ) The performance was held constant using adaptive thresholding separately for disgust vs . neutral trials; conditions labels are neutral low variance ( NL ) , neutral high variance ( NH ) , disgust low variance ( DL ) , disgust high variance ( DH ) . ( C ) Degraded sensory precision reduces perceptual sensitivity; cues had no impact on either motion detection ( i ) or threshold ( ii ) . Instead , disgust cues selectively increased rightward bias for low-variance stimuli ( iii ) , suggesting arousal altered stimulus expectations . As predicted by interoceptive inference , arousing cues significantly decrease the impact of noise on uncertainty ( M-bias ) ( iv ) . Curly brackets indicate F-test of 2-way interaction , square brackets indicate post-hoc t-tests ( * p<0 . 001 , p<0 . 01 , * p<0 . 05 ) . All error bars +/- SEM . ( D ) Although performance was held constant ( dark triangles , % correct ) , participants show considerable variability in metacognitive sensitivity ( light diamonds , M-Ratio ) , reproducing previous results using the signal-theoretic confidence model . ( E ) Participants had no awareness of cue valence in a post-task forced choice test using identical trial parameters; 95% confidence intervals for d-prime on all four face pairs overlap zero ( see Materials and methods , Valence Detection Task ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18103 . 00410 . 7554/eLife . 18103 . 005Figure 2—source data 2 . Table with variable codes used in Figure 2—source data 1 . Please see Materials and methods for full descriptions of all variables . DOI: http://dx . doi . org/10 . 7554/eLife . 18103 . 00510 . 7554/eLife . 18103 . 006Figure 2—source data 1 . This csv table contains the data for Figure 2 . All data are split by condition , NL = “neutral cue low variance” , NH = “Neutral cue high variance” , DL = “disgust cue low variance , DH = “disgust cue high variance” . DOI: http://dx . doi . org/10 . 7554/eLife . 18103 . 006 In a series of control analyses , we confirmed that ( 1 ) staircases were stable across trials and between conditions , ( 2 ) staircases successfully controlled for potential cue-induced differences in detection difficulty , and ( 3 ) the masking procedure successfully prevented the detection of cue valence . Analysis of detection accuracy across blocks showed that our adaptive staircases successfully held performance stable across blocks; ( F ( 1 , 24 ) , all conditions Ps >0 . 12 , Figure 2B ) . Further , cues exerted no influence on motion discrimination sensitivity ( d-prime , d’ ) , reaction times , or motion thresholds ( Figure 1B , i-ii , all ps > 0 . 1 ) , demonstrating that cues did not distract participants from the upcoming motion signal or otherwise alter stimulus sensitivity or detection performance . Analysis of d-prime for our forced-choice valence detection task at the end of the experiment showed that cues were not seen by participants ( all 95% CIs span zero , Figure 1D ) . Replicating previous results ( de Gardelle and Summerfield , 2011; Spence et al . , 2016 ) , increased sensory noise ( motion variance ) rendered motion discrimination more difficult , slowing reaction times ( median RT , main effect Variance , F ( 1 , 24 ) = 4 . 76 , p=0 . 039 , partial η2 = 0 . 17 , ) and decreasing sensitivity ( d’ , main effect Variance , F ( 1 , 24 ) = 185 . 15 , p<0 . 001 , partial η2 = 0 . 89 , see Figure 2B , i ) . We next assessed whether cues altered perceptual biases for motion , i . e . whether cues increased the influence of prior beliefs on stimulus classification . Although participants were generally unbiased in their tendency to respond left or right across conditions ( choice criterion ( c ) , grand mean F ( 1 , 24 ) = 1 . 45 , p=0 . 24 ) , a variance × cue interaction was found such that c was increased on low variance disgust-cued trials , but reduced on high variance disgust-cued trials ( V × P interaction , F ( 1 , 24 ) = 10 . 46 , p=0 . 004 , partial η2 = 0 . 30 ) . Follow-up paired-samples t-tests on this effect revealed that on trials following neutral cues , c did not differ between noise levels ( CB NH – NL; t ( 24 ) = 0 . 26 , p=0 . 80 ) , whereas disgust cues increased rightward bias for low variance trials ( CB DH – DL t ( 24 ) = 3 . 76 , p<0 . 001 ) . These results demonstrate that unseen , arousing cues selectively increased biases for low noise ( high precision ) stimuli , in the absence of any differences in the speed or accuracy of motion discrimination . To quantify the impact of sensory noise and arousing cues on choice uncertainty , we applied a signal-detection theoretic ( SDT ) approach to modelling confidence reports ( Galvin et al . , 2003; Maniscalco and Lau , 2012 ) . This model yielded M-Ratio and M-Bias parameters , which quantify the objective sensitivity and bias of confidence reports , respectively ( Maniscalco and Lau , 2012 ) . According to SDT , an M-Ratio ( m’/d’ ) of one indicates optimal metacognitive sensitivity ( i . e . , confidence ratings exhaust sensory information ) , with lower ratios indicating poorer metacognition . Alternatively , the M-Bias parameter describes the amount of sensory evidence needed to report a particular level of confidence , with higher values indicating a higher overall subjective uncertainty ( i . e . , a more conservative confidence bias ) . Consistent with interoceptive inference , we found that arousing disgust cues counter-acted the conservative bias induced by high sensory noise ( F ( 1 , 24 ) = 6 . 19 , p=0 . 020 , partial η2 = 0 . 21 ) , see Figure 2C , iv . Following neutral cues , confidence reports were significantly more conservative for noisy stimuli ( MB NH – NL; t ( 24 ) = 2 . 25 , p=0 . 034 ) , reproducing the previously reported impact of stimulus noise on uncertainty ( Spence et al . , 2016 ) . In contrast , disgust cues reduced this effect , decreasing uncertainty for high variance trials and increasing it for low-variance trials ( MB DH – DL; t ( 24 ) = −0 . 197 , p=0 . 85 ) . We also assessed whether these effects were independent of metacognitive sensitivity ( i . e . , that shifts in uncertainty related to an overall reduction of metacognitive sensitivity ) , repeating our factorial analysis for M-Ratio . Indeed , cues did not disrupt or alter metacognitive sensitivity; no significant effects were found for M-Ratio ( all p>0 . 6 ) . Additionally , overall M-Ratio and M-Bias did not correlate significantly with one another ( r = 0 . 37 , p=0 . 07 ) . These results demonstrate that perceptual and metacognitive biases for noisy stimuli are selectively altered by arousing disgust cues , even in the absence of performance differences in perception or metacognition . We next determined whether trial-by-trial fluctuations in confidence were related to cardiac or pupillary responses , and if cues successfully altered arousal to modulate these relationships . To do so , we applied a hierarchical general linear modelling approach to estimate the time course of pupillary and cardiac responses , and the encoding of our explanatory variables ( e . g . , cue valence , sensory noise , confidence and interactions thereof ) in these measures . We further performed post-hoc contrasts , or example on the main effect of cue valence or confidence , to delineate the shape of significant interactions . We used a non-parametric , cluster-based permutation t-test ( Hunt et al . , 2013; Hauser et al . , 2015 ) to determine when , with respect to trial time , our experimental variables were significantly encoded in evoked physiological responses . This procedure controlled for the family-wise error rate , while simultaneously accounting for variability in trial difficulty , as measured by RT and signal mean ( see Materials and methods for more details ) . Inspection of the grand mean response for each measure revealed a canonical orientation response locked to trial onset ( i . e . , the forward mask ) , as characterized by pupillary dilation ( grand mean peak at 2110 ms post-baseline , Figure 3A ) and heart rate deceleration ( grand mean trough at 1900 ms post-baseline , Figure 4A ) ( Sokolov , 1963; Graham and Clifton , 1966 ) . Consistent with its impact on discrimination difficulty , sensory noise increased pupil dilation ( peak effect = 2121 ms post-baseline , duration 554–2377 ms , max β = 24 . 74 , cluster p=0 . 014 ) ( Kahneman and Beatty , 1966; Murphy et al . , 2014b ) . Confidence showed a biphasic relationship with dilation depending on trial time , marked by greater dilation during stimulus presentation ( peak dilation effect = 712 ms post-baseline ( pb ) , max β = 11 . 35 , Minimum Cluster p=0 . 038 , duration 676–1560 ms post-baseline ) , but increased constriction during confidence rating ( peak constriction effect = 2273 ms post-baseline , max β = −18 . 53 , Minimum Cluster p=0 . 038 , duration 676–1560 ms post-baseline ) , see Figure 3—figure supplement 1A . This effect may reflect distinct neurophysiological contributions from stimulus processing vs post-stimulus evidence accumulation mechanisms ( Pleskac and Busemeyer , 2010; Lebreton et al . , 2015; Lempert et al . , 2015 ) . Confirming that our manipulation successfully modulated arousal , unseen disgust cues significantly increased both pupil dilation and cardiac acceleration ( Figure 3—figure supplement 1B , and Figure 4c ) , with increased pupil dilation during motion choice ( peak at 1596 ms post stimulus , duration 1686–2403 ms , max β = 21 . 85 , cluster p=0 . 032 ) and greater cardiac acceleration during confidence ratings ( peak effect 3200 ms , duration = 2900–3700 ms , max β = 0 . 31 , cluster p=0 . 044 ) . Confidence was also related to heart-rate acceleration throughout the trial , with greater confidence linked to a faster heart rate in the interval lasting from stimulus presentation to ratings ( peak effects at 500 ms and 3900 ms , durations 100–1000 ms and 1600–4100 ms , max β = 0 . 31 , cluster Ps = 0 . 046 and 0 . 002 ) , see Figure 4A . 10 . 7554/eLife . 18103 . 007Figure 3 . Pupillometry results . ( A ) Results of general linear modelling ( GLM ) of pupil responses; the pupil grand mean response function shows a canonical orientation response , peaking during confidence rating before returning to baseline in the 2–3 s jittered inter-trial interval . ( B ) As predicted , pupillary fluctuations encode the interaction of exteroceptive noise and unexpected internal arousal , time locked to the response interval and onset of confidence rating . ( C ) For illustration , mean response for each condition , extracted from significant time-window controlling for all explanatory and nuisance variables . GLMs were fit across all trials to each time point of the pupil series . Explanatory variables encoded main effects of stimulus noise , variance , confidence , and interactions thereof , revealing the amplitude and timing of each effect . The effects are independent from task-difficulty; trial-wise mean signal and RT were controlled in all analyses . Significance assessed using a cluster-based permutation t-test , cluster alpha = 0 . 05; cluster shown by shaded grey patch . See Materials and methods for more details . DOI: http://dx . doi . org/10 . 7554/eLife . 18103 . 00710 . 7554/eLife . 18103 . 008Figure 3—figure supplement 1 . Additional pupil effects of interest . ( A ) Confidence is biphasically encoded in pupil responses , with a stimulus-locked dilatory effect and a rating-locked constriction effect . Cues ( B , disgust > neutral ) increased dilation from response to rating . ( C ) Three-way interaction of cue valence , variance , and confidence showing that magnitude of cue-related pupil reversal correlates with trialwise confidence . Results of general linear modelling of the pupil , with explanatory variables encoding the main effects of stimulus noise , variance , confidence , and interactions thereof , revealing the amplitude and timing of each effect . The effects are independent from task-difficulty; trialwise mean signal and RT were controlled in all analyses . Significance assessed using a cluster-based permutation t-test , cluster alpha = 0 . 05; cluster shown by shaded grey patch . See Materials and methods for more details . DOI: http://dx . doi . org/10 . 7554/eLife . 18103 . 00810 . 7554/eLife . 18103 . 009Figure 4 . Cardiac results . ( A ) Grand mean cardiac response function showing canonical heart rate deceleration orientation response , and trial timings . ( B ) Subjective confidence ratings encoded by greater heart rate acceleration , beginning with stimulus onset and peaking during ratings . ( C ) Unseen disgust cues increase heart rate during confidence rating . ( D ) This effect interacts with confidence , effectively reversing the mapping of cardiac acceleration and subjective uncertainty . ( E ) To illustrate this effect , trials were median split into high and low confidence for each disgust condition ( e . g . , neutral low confidence , NLC ) , and mean response was extracted from within the significant cue by confidence window . Results of general linear modelling of instantaneous heart rate , with explanatory variables encoding the main effects of stimulus noise , variance , confidence , and interactions thereof , revealing the amplitude and timing of each effect . Effects are independent from task-difficulty; trial-wise mean signal and RT were controlled in all analyses . Significance assessed using a cluster-based permutation t-test , cluster alpha = 0 . 05; cluster shown by shaded grey patch . See Materials and methods for more details . DOI: http://dx . doi . org/10 . 7554/eLife . 18103 . 009 Pupil responses also encoded the interaction of cue and motion variance in the same time interval as the overall cue main effect , with cues reversing the dilatory effect of sensory noise ( peak effect 1467 ms post-baseline , duration 1492–2472 ms , min β = −21 . 67 , cluster p=0 . 034 , Figure 3C ) . Crucially , this effect was related to confidence in a positive three-way interaction ( peak effect 1512 ms post-baseline , duration 683–2099 ms , max β = 21 . 22 , cluster p=0 . 008 , Figure 3—figure supplement 1C ) , demonstrating that trial-by-trial fluctuations in subjective confidence tracked the cue-induced reversal of pupillary noise encoding . This finding mirrors our primary behavioural effect , indicating that the impact of disgust cue on confidence biases relates to a shift in the mapping between noise-induced uncertainty and physiological responses . In contrast , cardiac signals were insensitive to sensory noise or noise by cue interactions . Instead , the magnitude of the cue-related cardiac main effect negatively interacted with confidence ( peak effect 2800 ms post-baseline , duration 2800–3200 ms , min β = −0 . 30 , cluster p=0 . 044 ) , supporting a reversal in the mapping between heart rate acceleration and subjective uncertainty ( Figure 4C , D ) . This latter effect demonstrates that experimentally induced increases in arousal disrupt the typical relationship of heart-rate acceleration and confidence . In the present study , we demonstrate a close linkage of perceptual confidence , unexpected arousal , and related interoceptive signals . Across perceptual , physiological , and subjective measures we demonstrate that the encoding of sensory noise is weighted by interoceptive arousal . These results may have important implications for understanding medical and psychiatric disorders , in which patients exhibit chronic alterations in arousal or interoception . Substance abuse , psychosis , anxiety , and depression for example have been linked to altered heart-rate variability , physiological responses , and interoceptive sensitivity ( Dawson et al . , 1977; Hoehn-Saric and McLeod , 2000 ) . Our results suggest that the altered psychophysiology of these patients may cause them to perceive an unrealistically ( un ) -certain world . 29 participants took part in the experiment at University College London ( UCL ) . Previous studies examining the impact of sensory noise on confidence ( Zylberberg et al . , 2014; Spence et al . , 2016 ) and pupillometric responses during decision making ( Murphy et al . , 2014b ) have reported samples of 7–20 participants . To ensure a robust estimate of our behavioural and physiological effect while accounting for potential missing data ( due to e . g . , trials rejected due to blinks ) , we recruited a larger sample of 29 participants ( 17 F ) aged 18–39 ( M = 25 . 4 , SD = 5 . 0 ) in total . All participants had normal or corrected-to-normal vision with no history of neurological or psychiatric disorders . Participants received monetary compensation ( £15 ) for completing the experiment . Informed consent was obtained from all participants , and all procedures were conducted in accordance with the Declaration of Helsinki and with approval from the UCL Research Ethics Committee .	As you read the words on this page , you might also notice a growing feeling of confidence that you understand their meaning . Every day we make decisions based on ambiguous information and in response to factors over which we have little or no control . Yet rather than being constantly paralysed by doubt , we generally feel reasonably confident about our choices . So where does this feeling of confidence come from ? Computational models of human decision-making assume that our confidence depends on the quality of the information available to us: the less ambiguous this information , the more confident we should feel . According to this idea , the information on which we base our decisions is also the information that determines how confident we are that those decisions are correct . However , recent experiments suggest that this is not the whole story . Instead , our internal states – specifically how our heart is beating and how alert we are – may influence our confidence in our decisions without affecting the decisions themselves . To test this possibility , Allen et al . asked volunteers to decide whether dots on a screen were moving to the left or to the right , and to indicate how confident they were in their choice . As the task became objectively more difficult , the volunteers became less confident about their decisions . However , increasing the volunteers’ alertness or “arousal” levels immediately before a trial countered this effect , showing that task difficulty is not the only factor that determines confidence . Measures of arousal – specifically heart rate and pupil dilation – were also related to how confident the volunteers felt on each trial . These results suggest that unconscious processes might exert a subtle influence on our conscious , reflective decisions , independently of the accuracy of the decisions themselves . The next step will be to develop more refined mathematical models of perception and decision-making to quantify the exact impact of arousal and other bodily sensations on confidence . The results may also be relevant to understanding clinical disorders , such as anxiety and depression , where changes in arousal might lock sufferers into an unrealistically certain or uncertain world .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Unexpected arousal modulates the influence of sensory noise on confidence
The brain has evolved an internal model of gravity to cope with life in the Earth's gravitational environment . How this internal model benefits the implementation of skilled movement has remained unsolved . One prevailing theory has assumed that this internal model is used to compensate for gravity's mechanical effects on the body , such as to maintain invariant motor trajectories . Alternatively , gravity force could be used purposely and efficiently for the planning and execution of voluntary movements , thereby resulting in direction-depending kinematics . Here we experimentally interrogate these two hypotheses by measuring arm kinematics while varying movement direction in normal and zero-G gravity conditions . By comparing experimental results with model predictions , we show that the brain uses the internal model to implement control policies that take advantage of gravity to minimize movement effort . It is always fascinating to witness the ability of acrobats and dancers to accomplish complex and elegant movements , graciously interacting with gravito-inertial forces . Computational theory postulates that this captivating performance is due to the ability of the brain to learn and store internal representations of environmental dynamics ( Wolpert and Ghahramani , 2000 ) . On earth , gravity is the most ubiquitous and constant environmental feature . As such , a neural representation of gravity is created and stored through an internal model ( Papaxanthis et al . , 1998a; Angelaki et al . , 1999; Merfeld et al . , 1999; McIntyre et al . , 2001; Angelaki et al . , 2004; Indovina et al . , 2005; Miller et al . , 2008; Crevecoeur et al . , 2009; Gaveau and Papaxanthis , 2011; Laurens et al . , 2013a , 2013b ) . The need for an internal model of gravity arises because Einstein’s equivalence principle prevents any single sensory receptor from encoding gravity without simultaneously also encoding inertial accelerations ( Einstein , 1908 ) . The neural representation of gravity is thought to solve this ambiguity by multisensory statistical inference ( Angelaki et al . , 1999; Merfeld et al . , 1999; Angelaki et al . , 2004; Laurens et al . , 2013b ) . An internal model of gravity has been shown to benefit the anticipation of a free falling object motion ( Zago and Lacquaniti , 2005; Zago et al . , 2008; Lacquaniti et al . , 2013 ) , as well as the visual perception of allocentric vertical ( Van Pelt et al . , 2005; De Vrijer et al . , 2008; Elmore et al . , 2014 ) . However , whether and how an internal model of gravity benefits the planning and execution of skilled movement remains unknown . One influential viewpoint assumes a need for compensation; i . e . , the internal model of gravity is used to predict and compensate for its mechanical effects on the body . This hypothesis has been motivated by experimental findings on arm movements in the presence of externally applied Coriolis , viscous force fields and interaction torques ( Shadmehr and Mussa-Ivaldi , 1994; Gribble and Ostry , 1999; Pigeon et al . , 2003 ) . The main benefit of a compensation strategy would be the simplification of motor planning by allowing for invariant trajectories ( Hollerbach and Flash , 1982; Atkeson and Hollerbach , 1985 ) . Current research in fields as diverse as neurorehabilitation , movement perception , or motor control modularity , assumes such a compensation principle ( Prange et al . , 2009 , 2012; Cook et al . , 2013; Russo et al . , 2014 ) . An alternative perspective is based on a need for effort optimization , i . e . , the internal model of gravity could be used to predict and take advantage of its mechanical effects on the body . It has been proposed that adaptation to the Earth’s gravity field has allowed control policies to evolve that take advantage of environmental dynamics – use gravity as an assistive force to accelerate downward movements and as a resistive force to decelerate upward movements . This strategy , which has been formalized into a Minimum Smooth-Effort model , would result in movement kinematics that varies with direction ( Berret et al . , 2008a; Gaveau et al . , 2014 ) . Indeed , upward movements were shown to have shorter time to peak velocity and larger curvature than downward movements , but such comparisons were until now largely qualitative ( Papaxanthis et al . , 1998b; Gentili et al . , 2007 ) . Furthermore , upward/downward direction-dependent kinematics could also arise from the complex dynamics of the peripheral neuromuscular system . For example , the firing properties of extensor motoneurons ( pulling the arm downwards when upright ) are known to obey different rules from those of flexor motoneurons ( pulling upwards when uptight; Cotel et al . , 2009; Wilson et al . , 2015 ) . In addition , muscle force production generated by eccentric contraction ( elongation , e . g . downward movements for flexors ) is also known to obey differing rules from concentric contraction ( e . g . upward movement for flexors; for a review see Enoka , 1996 ) . Because of these asymmetrical neuromuscular peripheral properties , upward/downward kinematic asymmetries cannot necessarily be attributed to gravity effort optimality ( i . e . , minimization of muscular force to elevate and lower the arm ) . Thus , a solid test of the effort optimization hypothesis has been lacking . Here , we explicitly distinguish between the compensation and effort optimization hypotheses with two critical experiments . First , we contrast predictions of optimal control models that either compensate or take advantage of gravity force effects . Then , we quantitatively compare these predictions with actual kinematic features of arm movements in multiple directions . Second , in order to discard a possible influence of peripheral neuromuscular mechanisms , we measure how differences in upward versus downward arm movement kinematics are influenced by the lack of gravity during the zero-G phase of parabolic flight . If directional asymmetries originate from asymmetric firing of flexor/extensor motoneurons , they should persist in the absence of gravity because motoneurons properties are not expected to change during the short zero-G phase of a parabolic flight ( Ishihara et al . , 1996 , 2002; for a review see Nagatomo et al . , 2014 ) . On the other hand , if they originate from eccentric versus concentric force production differences , directional asymmetries should cease to exist instantly in 0g because movement dynamics no longer differ for upward and downward movements ( Enoka , 1996 ) . Alternatively , however , if directional asymmetries originate from neural planning processes that take advantage of the internal model of gravity , following a gradual recalibration of the gravity internal model , directional asymmetries should progressively decrease towards new optimal zero-G values ( McIntyre et al . , 2001; Izawa et al . , 2008; Snaterse et al . , 2011 ) . In support of the effort optimization hypothesis , we show that directional asymmetries are gradually eliminated during repeated exposure to the zero-G phase of parabolic flight . We asked fifteen humans to perform arm movements around the shoulder joint in different directions ( Figure 1A ) , whereby the work of gravity torque is systematically varied ( Figure 1B ) , while other dynamic variables ( interaction torques , Coriolis and centripetal forces ) are constant ( see Figure 1—figure supplement 1 for a geometrical illustration of the mechanical system and details on gravity torque computations ) . In single-degree of freedom movements , although the spatial shape of the endpoint trajectory is circular and constant across directions , the temporal shape of the endpoint trajectory ( the form of the velocity profile ) could change . As a measure of the shape of the velocity profile , we define a symmetry ratio ( SR: acceleration time divided by the total movement time ) . SR corresponds to the relative timing of peak velocity ( as in Figure 1C ) and thus allows quantifying whether arm kinematics remains invariant ( compensation hypothesis prediction ) or changes ( effort optimization hypothesis prediction ) with movement direction . Here we use the optimal control framework ( Bryson and Ho , 1975; Todorov , 2004 ) to simulate arm movement planning in different directions . Specifically , we change the cost function being minimized to understand how the neural model of gravity serves motor planning; i . e . , we compare simulations of optimal control models that use the internal model of gravity to either compensate or take advantage of its mechanical effects on the body . 10 . 7554/eLife . 16394 . 003Figure 1 . Task design and theoretical prediction . ( A ) Participants’ initial position and projection of the 17 targets onto the frontal plane . The angle γ , representing the target inclination with respect to horizontal , was used to calculate the gravity torque projecting onto the plane of motion ( see also Figure 1—figure supplement 1 ) . ( B ) The Work of Gravity Torque ( WGT ) , projected onto the plane of motion and integrated over the whole movement , is plotted as a function of movement direction ( same color code as in A ) . This non-linear ( cosine-tuned ) dependence is well fitted by a sigmoidal function ( black curve , average RMSE = 0 . 12; [min: 0 . 09 , max: 0 . 15] , see Materials and methods ) . Positive/negative values indicate that WGT has the same/opposite direction as the arm movement . ( C ) Mean velocity profiles ( for the average subject ) , normalized in both amplitude and duration , illustrate the predictions of the compensation hypothesis ( Jerk ) and the effort optimization hypothesis ( Smooth-Effort ) in the vertical plane ( upwards , −90; downwards , 90°; see arrows and color-coded direction definition in panel A ) . SR up ( red ) and SR down ( blue ) illustrate the calculation of a symmetry ratio ( acceleration time / movement time ) allowing quantification of kinematic differences/similarities . ( D ) Simulated symmetry ratio predicted by the compensation hypothesis: Minimum Jerk ( triangles , Flash and Hogan , 1985 ) ; and the effort optimization hypothesis: Minimum Smooth-Effort model ( dots , Gaveau et al . , 2014 ) ; as a function of movement direction ( −90° , upwards and 90° , downwards ) . Similarly to WGT ( see panels B ) , simulated symmetry ratio obtained from the Smooth-Effort model is well fitted by a sigmoidal function ( black curve ) . ( E ) Simulated symmetry ratio as a function of WGT ( grey triangles , Jerk and black dots , Smooth-Effort ) . Each data point represents the prediction for one subject moving in one direction ( n = 255 in each plot ) . It is noticeable that according to the effort optimization hypothesis , arm kinematics ( symmetry ratio ) should not be invariant but instead linearly correlate with WGT . ( F ) Simulated symmetry ratio predicted by two other well-known models minimizing dynamic cost functions: the Minimum Variance ( Harris and Wolpert , 1998 ) and the Minimum Torque Change ( Uno et al . , 1989 ) . It can be observed that the modulation of kinematics with movement direction is a specific feature of the effort-related optimization only . DOI: http://dx . doi . org/10 . 7554/eLife . 16394 . 00310 . 7554/eLife . 16394 . 004Figure 1—figure supplement 1 . Gravity torque projection in Experiment 1 . ( A ) Geometrical representation of the task and equations for computing gravity torque , projecting onto the plane of motion , from endpoint kinematics . Subjects were asked to perform single degree of freedom reaching movements , from a horizontal initial position , to a target requiring a θ = 45° shoulder rotation with an orientation relative to horizontal defined by an angle γ . l is the lever arm length . ( B ) Gravity torque projected onto the plane of motion for 7 exemplary plane orientations ( γ angle ) throughout the θ = 45° required movement amplitude . The gravity torque projecting onto the plane of motion was calculated using Equation 1 for all γ values presented panel C . The Work of Gravity Torque was then calculated using Equation 2 . ( C ) The convention on plane of motion angle values ( γ ) used to compute the Work of Gravity Torque in Experiment 1 is provided along with the exact positioning of each target defined as shoulder angular coordinates . DOI: http://dx . doi . org/10 . 7554/eLife . 16394 . 00410 . 7554/eLife . 16394 . 005Figure 1—figure supplement 2 . Clarification on the hybrid cost used in the Smooth-Effort model . Because the Smooth-Effort model minimizes a combination of the Absolute Work of force and Jerk , one may wonder whether the predicted linear correlation of symmetry ratio with the Work of Gravity Torque ( WGT , see Figure 1E ) emerges from the effort-related cost ( absolute work of force ) or from the kinematic cost ( Jerk ) . This figure presents the results of 1530 simulations performed for the 15 subjects * 17 targets * 6 different values of α ( the weighting factor for the Jerk cost in the Smooth-Effort model , see Methods ) . For larger values of α , the symmetry ratio saturates at SR = 0 . 5 . As illustrated , the higher the weight of the jerk , the flatter the correlation between symmetry ratio ( SR ) and WGT . This undeniably confirms the effort-related origin of the experimental findings presented in Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 16394 . 00510 . 7554/eLife . 16394 . 006Figure 1—figure supplement 3 . Supplemental minimum Variance simulations including muscle dynamics . Similarly to results presented in the main text ( Figure 1F ) , including muscle dynamics into the minimization of end-point variance predicts constant symmetry ratios for all movement directions . DOI: http://dx . doi . org/10 . 7554/eLife . 16394 . 006 Minimizing a kinematic cost only , without taking joint torque into account , such as the Jerk model ( Flash and Hogan , 1985 ) , is a perfect example of the compensation strategy . This is because the brain must compensate all perturbing dynamics to produce a consistent and invariant kinematically-defined motor plan ( Hollerbach and Flash , 1982; Atkeson and Hollerbach , 1985 ) , thus predicting constant SR values for all movement directions ( Jerk prediction in Figure 1D ) . In contrast , the Smooth-Effort model minimizes a hybrid cost that , in addition to the Jerk , takes the external dynamics into account to minimize the muscular force needed to move the arm ( absolute work of muscular torque ) . Thus , by design , the Smooth-Effort model implements the effort optimization strategy , whereby the brain uses the gravity internal model to predict and take advantage of gravity torques in accelerating and decelerating downward and upward movements , respectively ( Berret et al . , 2008a , 2008b; Gaveau et al . , 2011 , 2014 ) . The effort optimization hypothesis predicts a sigmoidal dependence of SR on movement direction ( Smooth-Effort prediction in Figure 1D ) . Critically , increasing the weight of movement smoothness ( Jerk ) in the Smooth-Effort hybrid cost , leads to a progressive disappearance of the gravity-torque related tuning of arm kinematics ( Figure 1—figure supplement 2 ) . Accordingly , the compensation hypothesis predicts that movement kinematics remains unchanged for various gravity torque conditions ( Jerk prediction in Figure 1E ) , whilst the effort optimization hypothesis predicts that movement kinematics strongly correlates ( average R = 0 . 99 [min: 0 . 986 , max: 0 . 994] , p<1e-07 ) with the amount of gravity torque engaged in the motion ( Smooth-Effort prediction in Figure 1E ) . Importantly , the SR dependence on movement direction ( i . e . , on gravity torque ) is a unique feature of effort-related optimization as minimizing other cost functions that take joint torques into account , but are not directly related to effort , such as the Variance ( Harris and Wolpert , 1998 ) or Torque Change ( Uno et al . , 1989 ) , predict constant SR as a function of arm movement direction ( Figure 1F; see also Figure 1—figure supplement 3 for additional simulations testing the effect of including muscle dynamics into the minimization of end-point Variance ) . In fact , minimization of an effort-related cost is a necessary and sufficient condition to predict directional asymmetries in the vertical plane ( Berret et al . , 2008a , 2008b ) . Participants accomplished rapid arm movements with single-peaked velocity profiles . Average duration did not vary with movement direction ( 0 . 40s ± 0 . 01 , SD; F16 , 224=0 . 67 , p=0 . 82 ) . Because arm movements were visually guided , systematic and variable errors were small and independent of movement direction ( -2 . 5°< SE < 2°; VE < 2 . 5°; F16 , 224=1 . 13 , p=0 . 32 and F16 , 224=1 . 02 , p=0 . 44 , respectively ) . As illustrated in Figure 2A , SR shows a sigmoidal modulation as a function of movement direction ( F16 , 224=28 . 36 , p<0 . 001 ) . The robustness of this result is illustrated in Figure 2B , which shows sigmoidal fits for individual subjects ( average RMSE = 9 . 77e-03 [min: 4 . 5e-03 , max: 1 . 53e-02] ) . When SR was regressed against gravity torque , correlation coefficients were high , averaging R = 0 . 84 [min: 0 . 67 , max: 0 . 92] , p=2 . 5e-05 ( Figure 2C ) . Furthermore , SR was independent of movement duration and movement amplitude ( average correlation coefficient for duration R = 0 . 22 [min: 0 . 04 , max: 0 . 57] , p=0 . 41 , see Figure 2—figure supplement 1A; for amplitude R = 0 . 29 [min: 0 . 02 , max: 0 . 69] , p=0 . 26 , see Figure 2—figure supplement 1B ) . Thus , arm kinematics is selectively modulated according to the gravity torque requirements of the movement , as predicted by the Smooth-Effort model and quantified by high correlation coefficients between predicted and experimental SR values ( average R = 0 . 82 [min: 0 . 61 , max: 0 . 91] , p=5 . 6e-05; compare Figure 1D and E to Figure 2A and C , respectively ) . These findings support the effort optimization hypothesis , whereby the brain implements control policies that exploit gravity effects to minimize muscular efforts . 10 . 7554/eLife . 16394 . 007Figure 2 . Experimental findings . ( A ) Experimentally recorded symmetry ratio , averaged across all subjects , is plotted as a function of movement direction . Similarly to the effort optimization hypothesis prediction ( see panels D in Figure 1 ) , experimental symmetry ratio is well fitted by a sigmoidal function ( black curve ) . Error bars illustrate SD . ( B ) Fits of a sigmoid function to symmetry ratio as a function of movement direction for data from individual subjects . ( C ) Symmetry ratio as a function of Work of Gravity Torque ( WGT ) . Each data point represents the mean of 12 trials for one subject moving in one direction ( n = 255; 3060 trials total ) . It is noticeable that , similarly to the effort optimization hypothesis prediction ( Figure 1E ) , arm kinematics ( symmetry ratio ) linearly correlates with gravity torque . DOI: http://dx . doi . org/10 . 7554/eLife . 16394 . 00710 . 7554/eLife . 16394 . 008Figure 2—figure supplement 1 . Supplemental analyses testing the effect of movement duration and amplitude on experimental findings . Each data point represents the mean of 12 trials for one subject in one direction ( n = 255 in each plot; 3060 trials total ) . ( A ) Symmetry ratio as a function of movement duration . ( B ) Symmetry ratio as a function of movement amplitude . These two plots reveal that the symmetry ratio was independent of movement duration ( individual correlation coefficients averaged R = 0 . 22 [min: 0 . 04 , max: 0 . 57] , p=0 . 41 ) and amplitude ( on average R = 0 . 29 [min: 0 . 02 , max: 0 . 69] , p=0 . 26 ) ; therefore ensuring that the modulation of arm kinematics observed in Figure 2 truly reflects gravity torque effects . DOI: http://dx . doi . org/10 . 7554/eLife . 16394 . 008 Although undeniably supportive , interpretation of these results is complicated by direction-dependent properties of the peripheral neuromuscular system ( Enoka , 1996; Cotel et al . , 2009; Wilson et al . , 2015 ) . As a further test , exploitation of microgravity environments ( e . g . , during the zero-G phase of parabolic flight ) offers a powerful tool to interrogate neural vs . peripheral origins of the directional asymmetries . This is because during the repeated transitions to zero-G , gravity torque is temporarily eliminated . According to the effort optimization hypothesis , kinematic asymmetries should gradually but systematically decrease to zero , because a progressive re-optimization neural process should take place gradually over multiple zero-G transitions ( McIntyre et al . , 2001; Izawa et al . , 2008; Snaterse et al . , 2011 ) . In contrast , a peripheral origin of kinematic asymmetries leads to different predictions . Specifically , if directional asymmetries originate from flexor/extensor motoneuron properties , SR should not change during the zero-G phase of parabolic flight ( Ishihara et al . , 1996 , 2002; Cotel et al . , 2009; for a review see Nagatomo et al . , 2014; Wilson et al . , 2015 ) . Else , if directional asymmetries originate from force production properties , SR asymmetries should be eliminated instantly , not gradually , in 0g ( Enoka , 1996 ) . A second experiment was designed to test these predictions . Eleven participants performed fast and visually guided single-degree of freedom arm reaching movements in two directions ( toward the head and toward the feet , Figure 3A ) , in zero-G conditions during 5 parabolas ( P1-P5 ) of a flight where centrifugal manoeuvers allow cancellation of gravity effects in the plane’s frame of reference . Arm movements were planar with comparable systematic and variable errors for different gravity and direction conditions ( shoulder abduction-adduction and internal/external rotation < 3 . 1°; -3°< SE < 3 . 3°; VE < 3 . 4°; gravity effect on SE , p=0 . 12 and VE , p=0 . 27; direction effect on SE , p=0 . 41 and VE , p=0 . 35 ) . Velocity profiles were single-peaked in both one-G and zero-G conditions and average movement durations ranged between 0 . 40s and 0 . 54s ( on average , 0 . 45±0 . 13s ) , without any statistical difference between gravity conditions ( F5 , 50=1 . 274 , p=0 . 29 ) and movement direction ( F1 , 10=0 . 549 , p=0 . 48 ) . 10 . 7554/eLife . 16394 . 009Figure 3 . Adaptation to microgravity . ( A ) Participants’ initial position and positioning of the 3 targets in the sagittal plane . Eleven participants performed fast and visually guided mono-articular arm movements ( shoulder rotations ) in the sagittal plane under normal gravity ( one-G ) and micro-gravity conditions ( zero-G ) during a parabolic flight ( parabola 1 , P1 to parabola 5 , P5 ) . ( B ) Symmetry ratios ( acceleration time / movement time ) predicted by the Minimum Smooth-Effort model in one-G and in zero-G conditions . ( C ) Symmetry ratios experimentally recorded before ( 1g ) and during adaptation to zero-G ( P1 to P5 ) . ( D ) Mean velocity profiles , normalized in amplitude and duration . Qualitative comparisons between upward and downward arm movements illustrate the progressive decrease of directional asymmetries when subjects adapted to the new microgravity environment . ( E ) Vertical bars represent the average symmetry ratio ( black vertical axis , left ) for the recorded ( black filled bars ) and simulated ( black open bars ) data in one-G and during zero-G ( P1 to P5 ) environments . For simulated data , the g value was fitted ( −2G<g<2G ) in order to best predict the measured symmetry ratios . These fitted g values , represented by the green dots ( green vertical axis , right ) , reveal a progressive decrease of the g internal model value during zero-G exposure . Error bars illustrate SD and color-coded arrows denote movement direction ( red = up; blue = down ) . See also Figure 3—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 16394 . 00910 . 7554/eLife . 16394 . 010Figure 3—figure supplement 1 . Supplemental analysis on the effect of subject order during the microgravity experiment . Because two subjects were successively tested during each parabolic flight , the second subject to be tested had therefore already been submitted to microgravity during the test period of the first subject . To reduce sensorimotor adaptation before the experiment was performed for the second subject , we restrained this subject on the ground and prevented him from moving his body limbs . Our rationale was that preventing motor interaction with the various unwanted gravito-inertial force fields would reduce sensorimotor adaptation . Vertical bars represent the directional difference ( Down-Up ) between average symmetry ratios for subjects who performed the experiment first ( black filled bars ) and those who performed the experiment second ( black open bars ) in one-G and during zero-G ( P1 to P5 ) environments . Subjects in both groups exhibited similar behavior ( Kruskal-Wallis ANOVA: H1 , 11<2 . 13 and p>0 . 14 in all cases ) ; i . e . directional asymmetry was present during the first parabola and then progressively disappeared . DOI: http://dx . doi . org/10 . 7554/eLife . 16394 . 010 Figure 3B shows the SR values predicted from the Smooth-Effort model for upward ( red ) and downward ( blue ) arm movements , both in one-G and zero-G environments . Optimization to the zero-G environment no longer predicts direction-dependent differences in velocity profiles , as was the case in one-G . In line with the effort optimization hypothesis , which predicts gradual adaptation to the zero-G environment , SR slowly converged towards the new direction-independent optimal value in zero-G ( Figure 3C; gravity condition and movement direction interaction effect , F5 , 50=10 . 54 , p<0 . 001 ) . Importantly , directional asymmetries persisted early in zero-G ( p=4 . 8e-05 for P1 and p=1 . 09e-03 for P2 ) . This indicates that , during initial exposure to zero-G , the brain still uses the internal model of the one-G environment to plan arm movements . However , directional asymmetries progressively disappeared in the following parabolas , suggesting a gradual re-optimization of motor commands through sensorimotor adaptation to the zero-G environment ( post-hoc , p=0 . 34 for P3 , p=0 . 92 for P4 , and p=0 . 99 for P5 ) . The corresponding average velocity profiles qualitatively illustrate the increase in acceleration/deceleration symmetry ( Figure 3D ) . For the sake of clarity , these experimental findings are replotted as SR Down - SR Up in Figure 3E ( black bars ) . To quantitatively validate the hypothesis of a gradual re-optimization of motor commands , we fitted this directional difference with the Smooth-Effort model ( white bars in Figure 3E ) by letting the internal model of gravity , g , being a free parameter ( -2G<g<2G ) . The progressive decrease in g ( internal model of gravity; green dots in Figure 3E ) shows that the progressive change in arm kinematic asymmetries can be well explained as a recalibration of the gravity internal model used for optimal motor planning based on the effort optimization ( minimization ) model ( McIntyre et al . , 2001; Izawa et al . , 2008; Snaterse et al . , 2011 ) . It is broadly believed that the brain develops and uses internal models of the sensor and effector dynamics , as well as physical laws of motion , to optimally interact with the external environment ( Shadmehr and Mussa-Ivaldi , 1994; Conditt et al . , 1997; Gribble and Ostry , 1999; Wolpert and Ghahramani , 2000; Pigeon et al . , 2003; Todorov , 2004; Ahmed et al . , 2008; Scott , 2012 ) . Having evolved in the Earth’s gravitational environment , our brains have thus acquired an internal model of gravity ( Angelaki et al . , 1999 , 2004; Indovina et al . , 2005; Miller 2008 ) . Here , we have conducted two critical experiments and showed that the brain takes advantage of this internal model to implement control policies that minimize movement effort under gravito-inertial constraints . First , we have shown that humans use versatile temporal trajectories that are linearly tuned to the gravity torque requirements of the task . Simulations of a Smooth-Effort model , which minimizes a hybrid cost composed of the absolute work of muscular forces ( mechanical energy expenditure ) and jerk ( inverse smoothness of the trajectory ) can predict not only differences in upward/downward arm movements , but also the linear modulation of endpoint kinematics according to the gravity torque requirements of the movement . Because the smoothness term of the Smooth-Effort model corresponds to the Jerk , the Smooth-Effort model can be considered an expansion of the minimum Jerk requirement to include how task dynamics shapes movement kinematics . Yet , inclusion of an effort-related cost is a necessary and sufficient condition to predict directional asymmetries in the vertical plane ( Berret et al . , 2008a , 2008b ) . Furthermore , minimizing other torque-related cost functions failed to predict direction-dependent kinematics ( Figure 1F ) . The present results therefore strongly support effort minimization in humans , further extending the growing idea that perceived effort plays an important role in the tailoring of human motor as well as non-motor behaviors ( Bramble and Lieberman , 2004; Walton et al . , 2006; Mazzoni et al . , 2007; Carrier et al . , 2011; Kurzban et al . , 2013; Selinger et al . , 2015; Farshchiansadegh et al . , 2016; Shadmehr et al . , 2016 ) . Second , we have also shown that the direction-dependent kinematics observed in normal gravity progressively vanishes during repeated exposure to a microgravity environment . Remarkably , Smooth-Effort model simulations nicely predict this adaptation to zero-G . Results of this second experiment are of major importance to disentangle peripheral and neural mechanisms for direction-dependent kinematics . This is because , if the observed direction-dependent kinematics were due to properties of the neuromuscular system , either an abrupt change or no change at all would be expected in the new gravity environment ( Enoka , 1996; Ishihara et al . , 1996 , 2002; Cotel et al . , 2009; Nagatomo et al . , 2014; Wilson et al . , 2015 ) . The fact that neither happens reveals a central mechanism . Furthermore , the fact that the observed progressive changes in kinematic asymmetries lead to new optimal values clearly supports the hypothesis of a progressive re-optimization procedure originating from planning processes ( Izawa et al . , 2008; Snaterse et al . , 2011; Selinger et al . , 2015 ) . The present findings on arm movements extend and supplement the recent results of Selinger and collaborators on human walking ( Selinger et al . , 2015 ) , suggesting that i ) the brain can easily and progressively adapt motor patterns to reduce energy expenditure and ii ) energy-related criteria ( such as effort ) are not only the result of , but actually tailor , motor patterns . The present results and conclusions stand in contrast to a broad view that the brain uses internal models of perturbing forces for their compensation such that stereotypic trajectories can be maintained ( Hollerbach and Flash , 1982; Atkeson and Hollerbach , 1985; Shadmehr and Mussa-Ivaldi , 1994 ) . Although very influential , such a compensation hypothesis has been challenged by results of studies that quantified velocity profiles and revealed that the temporal organization of arm kinematics shows a small , yet consistent , dependence on movement direction , speed and load ( Papaxanthis et al . , 1998b; Gaveau et al . , 2011 , 2014 ) . Furthermore , findings inconsistent with the compensation hypothesis have been largely ignored . For example , Virji-Babul et al . ( 1994 ) reported regression slopes of SR over movement amplitude that significantly differed between upward and downward movements ( see Figure 2C in Virji-Babul et al . , 1994 ) . Also , the consistent observation of negative periods on the phasic activation of arm muscles , resulting from the subtraction of the hypothesized gravity-compensatory activity from the full muscle activation ( Flanders et al . , 1996 ) , suggest that muscular activity does not compensate gravity torque ( Gaveau et al . , 2013 ) . Luckily , the application of the optimal control theory to the study of biological movement has given better insights into old phenomena . For example , recent studies , which framed motor adaptation as a process of re-optimization ( Izawa et al . , 2008; Crevecoeur et al . , 2009; Gaveau et al . , 2011; Cluff and Scott , 2015 ) , have reported subtly altered trajectories – by contrast to the traditional compensation view that assumes invariant trajectories . Thus , newly constructed/calibrated internal models may serve trajectory optimization rather than external force compensation . The present results provide further support for this notion and demonstrate the propensity of the motor system for multiple control policies ( i . e . , trajectories ) whose temporal organization shows small , but systematic , differences , such as to allow minimization of motor effort in our daily living ubiquitous gravity environment . Although we have only used simple , mono-articular arm movements in the present experiments , our conclusions are generalizable . Specifically , directional asymmetries in the vertical plane ( upwards versus downwards ) have also been observed in multi-articular arm reaching , reaching to grasp , grasping , hand drawing , and whole-body sit-to-stand / stand-to-sit movements ( Papaxanthis et al . , 1998c , 2003 , 2005; Yamamoto and Kushiro , 2014 ) . Thus , we propose that the directional tuning of movement kinematics is a general feature of motor control that may reflect an evolutionary and/or developmental advantage for effort optimization in the Earth’s gravity field ( Bramble and Lieberman , 2004; Carrier et al . , 2011; Selinger et al . , 2015 ) . Finally , it is important to speculate that the Smooth-Effort model pioneered here should not be considered solely an extension of the minimum jerk optimization model , which was proposed for planar horizontal movements that are unaffected by gravity ( Flash and Hogan , 1985 ) , to now include optimization of work against gravity for vertical movements . Even for movements in the horizontal plane , because the effort component of the cost function in the Smooth-Effort model is torque-dependent , the predictions of the Smooth-Effort model will change with the torque requirements of the movement . The minimum Jerk model predictions , however , will remain constant . Therefore , if interaction torques ( for multi-degree of freedom arm movements ) or additional external torques ( produced by a robotic manipulandum for example ) are experienced in the horizontal plane , the Smooth-Effort and Jerk model predictions would be different . Future experiments should test whether the need for optimization of work is limited to vertical movements pro or against gravity or , as we propose , represent a more general principle of motor control . Participants comfortably sat on a chair with their trunk in the vertical position ( Figure 1A ) . All trials started from a fixed initial position: shoulder elevation 90° , shoulder abduction 0° , elbow joint fully-extended , hand semi-pronated and aligned with the upper arm and the forearm . From that initial position , participants carried out rapid , visually guided , single degree of freedom arm movements ( rotation around the shoulder , 45° amplitude ) towards 17 targets ( plastic markers , diameter 1 cm ) placed in the right sagittal-frontal space . Note that results from previous experimental and theoretical studies have demonstrated that directional asymmetries in the vertical plane do not originate from the existence of inertial interaction torques at the elbow and wrist joints ( Le Seac'h and McIntyre , 2007; Gaveau et al . , 2014 ) . Figure 1A depicts the projection of targets' position onto the frontal plane . The inter-target angles are described in Figure 1—figure supplement 1C ( Plane angles ) . All targets were centered on the participants’ right shoulder at a distance equal to the length of their fully-extended arm . Reaching movements required a combination of shoulder abduction and shoulder flexion or extension . Participants performed 204 trials in a random order ( 12 trials for each movement direction , total trials in the experiment = 3060 ) . This experiment took place in an aircraft during parabolic flight . Participants comfortably sat on the aircraft’s floor with their legs strapped and their trunk in the vertical position ( Figure 3A ) . The general organization of the task was exactly as described in previous studies ( Gaveau et al . , 2011 , 2014 ) . Briefly , 3 targets were centered on the participants’ right shoulder ( parasagittal plane ) at a distance equal to the length of their fully extended arm . Participants were requested to perform fast visually guided upward and downward arm reaching movements ( 45° shoulder rotation ) . Participants first performed arm movements in normal gravity during the flight ( before parabolic maneuvers started , 40 trials for each movement direction ) and then in microgravity during 5 parabolas ( ≈75 trials , ≈15 trials per parabola ) . During the parabolic flight and based on pilots’ instructions about the gravity force level , the experimenter verbally instructed participants when to start a block of reaching movements and when to stop . This was important to ensure that movements started and finished within zero-G conditions; i . e . , participants made no movement in one-G or two-G conditions during or between parabolic manoeuvers . At the beginning of a block of movements within a parabola , participants repetitively reached between the middle , the upward , and the downward targets as follows: middle-upward ( stopped for roughly 1 s ) , upward-middle ( pause 1 s ) , middle-downward ( pause 1 s ) , downward-middle ( pause 1 s ) and so on for approximately 15 trials . As initial position did not influence adaptation results across parabolas , movements with different initial positions were pooled together within each direction . The experiments were carried out during 4 different flights . Each flight was composed of thirty parabolas , each parabola consisting of three successive phases: ( i ) hypergravity ~ 1 . 8 g , ( ii ) microgravity ~ 0 g , and ( iii ) hypergravity again ~ 1 . 8 g . Each of those three phases lasted ~30 s and the parabolas were separated by a time interval of ~ 2 min . The whole flight lasted ~ 2 hr . Here we present the results of an experiment performed during 5 parabolas for each participant . After this experiment , participants carried out a different experiment from which the results are not presented here . Two participants were tested on each flight . Therefore , to reduce motor adaptation to zero-G before the experiment took place , participants who did not perform the experiment during the first five parabolas were restrained on the aircraft floor so as to prevent any motion . Similar results were observed for participants who did the experiment at the beginning or at the end of the flight ( see Figure 3—figure supplement 1 ) . All analyses were performed using custom programs in Matlab ( Matworks , Natick , MA ) and have been described in details in previous studies ( Gaveau et al . , 2011 , 2014 ) . Arm movements in both Experiments were recorded using an optoelectronic system of motion analysis ( Smart , B . T . S . , Italy ) with 4 TV-cameras ( 120 Hz ) . Five reflective markers ( diameter: 4 mm ) were placed on the shoulder ( acromion ) , elbow ( lateral epicondyle ) , wrist ( middle of the wrist ) , hand ( first metacarpo-phalangeal joint ) , and the nail of the index . Kinematics was recorded in three dimensions ( X , Y and Z ) and low-pass filtered ( 10 Hz ) using a digital fifth-order Butterworth filter . The start and end of each movement was defined as the time at which finger tangential velocity went above or fell below 5% of maximum velocity . An automatic inspection of all trials revealed that shoulder angular velocity profiles were single-peaked and presented no motion ( <3° ) from any other joint . Following this analysis , we calculated the subsequent kinematic parameters: ( i ) movement duration ( MD ) , ( ii ) constant and variable angular final error and ( iii ) symmetry ratio ( SR ) of the finger velocity profile , defined as the ratio of acceleration time to total movement time ( a ratio equal to 0 . 5 indicates temporally symmetric velocity profiles ) . In the present study , we used SR to quantify kinematic variations with movement direction . SR is a standard parameter that has been routinely used in numerous studies , therefore ensuring an easy comparison of the present result with past ones . We calculated the gravity torque ( GT ) normal to the plane of motion with the following formula: ( 1 ) GT=mglcosθsinγ where m is the mass of the arm ( estimated for each subject from anthropometric tables , Winter , 1990 ) , g = 9 . 81 m . s-2 , l is the lever arm length ( Winter , 1990 ) , θ is the movement amplitude ( 0° to 45° ) and γ is the plane of motion inclination with respect to horizontal . The geometrical illustration of the mechanical system along with details on how to derive Equation 1 are displayed in Figure 1—figure supplement 1 . In Experiment 1 we report the Work of Gravity Torque calculated as follows: ( 2 ) WGT= ∫045 ( GT ) dθ Positive values indicate that WGT has the same direction as arm motion , whilst negative values indicate that WGT direction is opposite to arm motion direction ( see Figure 1—figure supplement 1 ) . In Experiment 1 , we also performed iterative least square minimizations to fit a sigmoid function on simulated as well as experimentally recorded SR . This was performed using the “nlinfit” Matlab function ( Mathworks ) and Equation 3: ( 3 ) F=p1+ p2/ ( 1+exp ( −x+p3p4 ) ) where p1 to p4 are free parameters and x is the angular scale . The goodness of fit was assessed by the Root-Mean-Square Error ( RMSE ) , which is essentially equivalent to Standard Deviation and has the same units as the variable being fitted . We predicted SR using an optimal control framework based on the Minimum Smooth-Effort model ( Gaveau et al . , 2014 ) . More sophisticated versions of the model focusing on multi-degree of freedom arm movements have been described in previous publications ( Berret et al . , 2008a , 2008b; Gaveau et al . , 2011 ) . Equation 4 describes the equation of motion for a single degree of freedom limb movement , with amplitude angle ( θ ) , plane of motion inclination with respect to horizontal ( γ ) , moment of inertia ( I ) , viscous friction coefficient B=0 . 87 ( as in Nakano et al . , 1999 ) and gravitational torque GT ( θ , γ ) . The net muscle torque acting at the shoulder is obtained as follows: ( 4 ) τ=Iθ¨+Bθ˙+GT ( θ , γ ) The Minimum Smooth-Effort model minimizes a combination of Effort and Smoothness . The mechanical effort related to a movement , i . e . , the amount of muscular force spent to move the arm , can be computed as the absolute work of the muscle torque: ( 5 ) Ceffort=∫0T\|τθ˙\| dt The rationale of this effort term is that the desired trajectory must take advantage of non-muscular torques – gravity torque in the present experiment – in order to minimize the amount of muscular torque required to move the arm up to the final posture . It has been shown that considering effort expenditure alone usually fails to account for several features of human trajectories , such as motion smoothness ( Berret et al . , 2011a , 2011b ) . Consequently , we considered that a complementary objective of motor planning was to maximize motion smoothness . This was achieved by penalizing large angle jerks . Thus an additional term that enters into the minimization is: ( 6 ) Csmooth=∫0T ( dθ¨/dt ) 2dt The Smooth-Effort model then relies on the following composite cost function: ( 7 ) C=Ceffort+αCsmooth where α is a weighting factor normalizing the relative magnitude of the jerk term in the total cost function . For all simulations , but Figure 1—figure supplement 2 , we set α=7e−5 . Figure 1—figure supplement 2 presents simulations where α was systematically varied to test the relative roles of both the effort and smoothness parts of the cost function in predicting direction dependent kinematics ( i . e . , the linear correlation of SR with the Work of Gravity Torque; WGT , see Figure 1E ) . Note that minimizing some function f ( x ) =bg ( x ) +ch ( x ) will provide the same solution x as minimizing f ( x ) /d=b/dg ( x ) +c/dh ( x ) for any d>0 . Hence , we normalized our cost function by setting b=d , thereby assuming a unit coefficient in front of one component of the cost . For a given α , the optimal control problem consists of finding a vector ( here the time derivative of the muscle torque ) driving the system from an initial ( θ0 ) to a final static posture ( θf ) , in time T ( adjusted for each subject based on experimental data ) , and yielding a minimum cost value C . We solved the minimization numerically using a Gauss pseudospectral method and the software GPOPS ( Benson et al . , 2006; Garg et al . , 2009; Rao et al . , 2010 ) . We verified that the control variable was smooth , the boundary values were not reached and the Pontryagin’s maximum principle necessary conditions ( such as the constancy of the Hamiltonian ) met . Predicted SR values were determined based on the simulated velocity profiles , as described above for experimental data . We also derived the solution of the Minimum Smooth-Effort model for a constant gravitational torque and obtained very similar results to those presented in the main text . This means that asymmetries are not totally due to GT variations along the movement amplitude but to the presence of non-zero GT . The linearized case for similar models has been analyzed in depth in Berret et al . ( 2008a ) ; where the solution of the Minimum Smooth-Effort model was derived explicitly and the origin of asymmetries in such a model was mathematically demonstrated – minimizing an effort-related cost is a necessary and sufficient condition for the production of a brief transient muscle inactivation near the peak of velocity which in turn induces the observed kinematic asymmetries with respect to movement direction . In Figure 1 we also present the results of simulations performed with three influential alternative models: the Minimum Jerk , the Minimum Torque Change and the Minimum Variance ( Flash and Hogan , 1985; Uno et al . , 1989; Harris and Wolpert , 1998 ) . Predicted SR values corresponding to the minimization of each subjective cost were obtained using the following equations and the same optimal control framework as described for the Minimum Smooth-Effort model . The cost to minimize for the Jerk model was: ( 8 ) Cj=∫0T ( dθ¨/dt ) 2dt As we deal with a 1-dof arm , this model accounts for both the angular and Cartesian versions of the Minimum Jerk ( which indeed provide equivalent predictions in the current case ) . Note that the smoothness term of the Smooth-Effort model corresponds to the minimum Jerk . The Smooth-Effort model thus represents an important and non-trivial extension of the minimum Jerk model , which also accounts for how task dynamics shapes movement kinematics . Because the effort part of our model is torque dependent , the prediction of the Smooth-Effort model will change with the torque requirement of the movement whilst the minimum Jerk prediction will not . Therefore , if interaction torques ( for multi-degree of freedom arm movements ) or additional external torques ( produced by a robotic manipulandum for example ) are experienced in the horizontal plane , the minimum Smooth-Effort and the minimum Jerk would predict very different solutions . The cost to minimize for the Torque Change model ( Uno et al . , 1989 ) was: ( 9 ) Cτ= ∫0T ( dτdt ) 2dt Simulations of the Minimum Torque change model for horizontal single degree of freedom arm movements performed under various dynamics by Tanaka et al . ( 2004 ) reported that the Minimum Torque Change model predicts velocity profiles that are always symmetrical , in line with Figure 1F . The Minimum Torque Change model in the gravity field has also been investigated in depth in ( Berret , 2009 ) . We also derived the solutions of the Minimum Variance model in the gravity field ( Harris and Wolpert , 1998 ) . We first linearized the arm’s dynamics ( Equation 4 ) in order to compute the optimal solution according to the method presented in Harris and Wolpert ( 1998 ) and Tanaka et al . ( 2004 ) . Here: ( 10 ) mglcosθ≈mglcosθ0=k Equation 10 formalizes that a constant gravitational torque is pushing the moving segment downwards . Note that such linearized case for similar models has been extensively used by previous studies ( Harris and Wolpert , 1998; Tanaka et al . , 2004; Berret et al . , 2008a ) . We considered the discrete time version of the Minimum Variance optimal control problem . Denoting by x= ( θ , θ˙ , τ ) T the state vector and by u=τ˙ the control variable , the linear state-space dynamics for the Minimum Variance model was expressed as follows: ( 11 ) xt+1=Axt+B ( ut+wt ) +C where wt~N ( 0 , σut2 ) is the signal dependent ( multiplicative ) noise at time t ( σ=0 . 2 in all simulations ) and C= ( 0 , −k/I , 0 ) TΔt with Δt the time step size after discretization ( Δt=10ms in our simulations ) . We obtained the solution to the optimal control problem by iteratively computing the distribution of the state vector at time t: ( 12 ) E[xt]=Atx0+∑i=0t−1At−1−i ( Bui+C ) ( 13 ) cov[ xt ]= σ∑i=0t−1 ( At−1−iB ) ( At−1−iB ) Tui2 The positional variance of the endpoint at time t ( denoted by Vt ) is thus given by the element ( 1 , 1 ) of the matrix cov[ xt ] ( Equation 13 ) . We can then define a quadratic programming problem by defining a cost related to the endpoint positional variance as follows: ( 14 ) Cvar ( u0 , u1 , … , uT+R ) =∑t=T+1T+RVt where R>1 is an integer defining the post-movement stabilization time , T is the a priori chosen movement time ( counting only the transient phase ) . In our simulations , we considered R=T . Because noise accumulates through time , Cvar is indeed a function of the control variable during the transient period of the motion . This minimization problem was solved using Matlab's fmincon function ( sqp algorithm ) . For the sake of completeness , we also derived the minimum variance solution when modeling some basic muscle dynamics , as performed by previous studies ( Harris and Wolpert , 1998; Tanaka et al . , 2004 ) . In agreement with previous results from Tanaka et al . ( 2004 ) ; we obtained similar results to those presented in the main text ( Figure 1F ) , as illustrated in Figure 1—figure supplement 3 . In that case , the state vector becomes x= ( θ , θ˙ , aag , aant ) T , the control variable becomes u= ( uag , uant ) T and the muscles are modeled as first order low-pass filters . The equations for the associated dynamics are as follows: ( 15 ) τag−τant=Iθ¨+Bθ˙+GT ( θ , γ ) ( 16 ) τag−τant=ρ ( aag−aant ) ( 17 ) σa˙ag=uag−aag ( 18 ) σa˙ant=uant−aant Equation 15 describes the equation of motion for a single degree of freedom limb movement , with amplitude angle ( θ ) , plane of motion inclination with respect to horizontal ( γ ) , moment of inertia ( I ) , viscous friction coefficient B=0 . 87 ( Nakano et al . , 1999 ) and gravitational torque GT ( θ , γ ) . In Equation 16 , the constant ρ is a gain factor relating agonist and antagonist muscle activations to joint torques . Equations 17 and 18 describe the muscle dynamics as a first order low-pass filter . The control variable is the motoneurons inputs uag and uant with the constraint: ( uag , aant ) ϵ [ 0 , 1 ]2 . This implies the positivity of muscle activations and , therefore , muscle torque . The net torque is simply obtained by subtracting the agonist and antagonist torques . For each participant , we calculated mean values for all recorded variables and checked for normal distribution ( Shapiro-Wilk tests ) and sphericity ( Mauchly tests ) . In Experiment 1 , statistical effects were accessed by within-subjects one-way repeated-measures ANOVA ( factor: 17 angles ) . In Experiment 2 , statistical comparisons were carried out by within-subjects two-way repeated-measures ANOVA . The factors were direction ( 2 levels ) and gravity-conditions ( six levels: 1g and 5 parabolas in 0g ) . For all statistical analyses , post-hoc differences were assessed with Scheffé's tests and significance was accepted at p<0 . 05 .	Many of the activities of humans and other animals require the limbs to be moved in a coordinated manner . For a movement to be successful , the brain must generate muscle contractions that take into account factors in the environment that might affect the movement . One such prominent environmental feature is gravity , and it is broadly believed that the brain develops and uses an internal representation of gravity to anticipate its effects on the limbs . How an internal representation of gravity helps limb movements to be made successfully is not known . Theorists have proposed that the brain could use the internal model of gravity to predict how to compensate for its mechanical effects – or , on the contrary , take advantage of them . Flying a plane in a “parabolic” arc creates a microgravity environment inside it that produces a feeling of weightlessness . Gaveau et al . asked volunteers to perform arm movements in normal earth gravity and in microgravity conditions . Under normal gravity , the volunteers made arm movements with speed profiles that differed according to movement direction . When they first performed these movements in microgravity , the speeds still differed according to direction . However , as the participants gained more experience of making the movements in microgravity , the speed at which upward and downward arm movements were made became more similar . Eventually movements were performed at the same speed in either direction . Comparing these results to numerical simulations revealed a sophisticated behavior where movements are organized to take advantage of the effects of gravity to minimize the effort that the muscles need to make . Further research into the neural mechanisms behind this optimization process could benefit the development of various rehabilitative and assistive technologies , such as brain-machine interfaces and robotic devices to guide and support limbs .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "neuroscience" ]	2016	Direction-dependent arm kinematics reveal optimal integration of gravity cues
Reaction times ( RTs ) are assumed to reflect the underlying computations required for making decisions and preparing actions . Recent work , however , has shown that movements can be initiated earlier than typically expressed without affecting performance; hence , the RT may be modulated by factors other than computation time . Consistent with that view , we demonstrated that RTs are influenced by prior experience: when a previously performed task required a specific RT to support task success , this biased the RTs in future tasks . This effect is similar to the use-dependent biases observed for other movement parameters such as speed or direction . Moreover , kinematic analyses revealed that these RT biases could occur without changing the underlying computations used to perform the action . Thus the RT is not solely determined by computational requirements but is an independent parameter that can be habitually set by prior experience . The reaction time ( RT ) is arguably the most widely used measure in neuroscience and psychology for noninvasively assessing processing in the brain: it is assumed to reflect the time needed to complete the perceptual and motor-planning computations required to prepare a response ( Donders , 1969; Sternberg , 1969; Friston et al . , 1996; Spivey , 2007; Sanders , 1998 ) . This assumption appears justified by evidence that RTs are modulated by factors such as stimulus complexity ( Kaswan and Young , 1965; Hick , 1952; Ratcliff , 2002 ) , stimulus-response compatibility ( Fitts and Seeger , 1953; Simon and Rudell , 1967; Simon and Wolf , 1963 ) , number of potential responses ( Henry and Rogers , 1960; Fischman , 1984; Christina et al . , 1982 ) , or required response accuracy ( Fitts , 1954; Fitts , 1966; Reddi and Carpenter , 2000 ) . According to this assumption , any reduction in the RT should negatively affect the quality of the resulting response . Indeed , instructing participants to respond as rapidly as possible can lead to a decrease in accuracy ( Fitts , 1966 ) , consistent with changes in accuracy that occur through more direct manipulation of allowed preparation time ( Schouten and Bekker , 1967; Ghez et al . , 1997; Stanford et al . , 2010; Haith et al . , 2016 ) . However , recent evidence showed that it was possible to reduce the RT considerably before any decline in movement accuracy was observed . For example , when individuals were placed under strict time constraints , movement accuracy and task success decreased only after the RT was shortened by ~80 ms ( Haith et al . , 2016 ) . Similarly , startle by a loud acoustic stimulus could evoke initiation of a fully prepared action ~70 ms earlier than typically observed ( Valls-SoleSolé et al . , 1999; Carlsen et al . , 2004 ) . These findings suggest that the RT includes some additional time that is not required for computing the upcoming response , raising the possibility that other factors might also influence the RT . An additional clue that the RT does not strictly represent computation time comes from evidence that the RT is influenced by context . This was illustrated in a recent study examining the planning of intentionally curved reaches ( Wong et al . , 2016 ) . In this study , the RTs of simple point-to-point reaches were observed to be shorter than those of more complex curved movements , consistent with the idea that RT reflects computation time . However , when point-to-point movements were interleaved among curved reaches , the RTs of those point-to-point reaches were surprisingly prolonged by ~90 ms , matching the RTs of the curved reaches . A similar contextual effect has been observed during mental rotation: during a letter discrimination task , identifying letters that were presented upright occurred at much shorter RTs when all the letters were upright compared to when other letters in the same block of trials were rotated ( Ilan and Miller , 1994 ) . These data suggest that RTs might be subject to experience-dependent biases – akin to other movement parameters such as speed or direction ( Diedrichsen et al . , 2010; Verstynen and Sabes , 2011; Hammerbeck et al . , 2014; Huang et al . , 2011 ) – rather than arising strictly from the outcome of computational requirements . That is , RTs may be subject to habit in the sense that an RT of a given magnitude may become more likely to be generated in the future simply because it has been generated in the past , regardless of current task requirements . Here we present results from two experiments which demonstrate that the RT is subject to experience-dependent effects . In the first experiment – a target-interception task – we showed that experience with initiating reaches at a short or long RT exerts a corresponding bias on the RTs of subsequently performed point-to-point reaches . In a second experiment , we demonstrated that these RT biases can also occur for more complex curved reaches around barriers . This barrier task can be performed with or without the presence of a direct cue illustrating the appropriate trajectory , corresponding to greater or lesser computational requirements respectively ( Wong et al . , 2016 ) . We found that participants’ RTs in the cued condition depended strongly on whether or not they had previously experienced the uncued condition . Analysis of the kinematics of these curved reaches allowed us to determine that the observed experience-dependent changes in RT were not due to habitually ignoring the cue and planning the movements in a different manner , but instead to habitually adopting a longer RT despite using a briefer computation ( i . e . , taking advantage of the cue ) to solve the task . Together , these two experiments reveal that the RT does not strictly reflect the time needed to complete the computations required for preparing responses , but may instead be selected habitually according to prior experience . In Experiment 1 , we tested whether performance of a task that encouraged movements to be generated with particular RTs could affect the RTs of subsequent movements performed in a different context . We asked participants to perform an interception task to hit a target that moved in a straight line toward or away from the participant ( Figure 1A ) . For one group of 10 participants , the target always moved outward , encouraging participants to initiate reaches at shorter RTs to intercept the target before it moved beyond an invisible boundary and disappeared ( see Materials and methods ) . For a second group of 10 participants , the target always moved inward . This encouraged participants to increase their RTs , since they could reduce the effort required to complete this task by waiting for the target to move closer before reaching out to hit it . In both cases , participants were given no specific RT instruction . Additionally , because participants were not required to stop inside the target on interception trials , participants typically generated shooting movements through the target . Immediately before and after participants performed this interception task , we measured their RTs for simple point-to-point reaches ( wherein the hand had to stop inside the target and move at a constrained speed ) to assess any experience-related changes in RT . At baseline , participants in both groups generated point-to-point reaches with comparable RTs ( Table 1; no significant difference between groups: t = −0 . 22 , p=0 . 82 ) , and were able to satisfy the speed requirements of the task ( see Table 1 and Materials and methods ) . Movement of the target , either toward or away from the participant’s starting position , exerted a reliable influence on behavior during performance of the interception task . When the target moved outward ( Figure 1B; Figure 1—source data 1 – Outward; Table 1 ) , participants increased their movement speed ( paired t-test comparing pre-training to the last training block , t ( 9 ) = 6 . 30 , p<0 . 001 ) and decreased their RT ( paired t-test comparing pre-training to the last training block , t ( 9 ) = 2 . 99 , p=0 . 02 ) relative to their behavior on baseline point-to-point reaches . These reductions in RT persisted in the final phase of the experiment when the target was again stationary: RTs during the post-training point-to-point reaching block were shorter than in the analogous pre-training block by 21 . 68 ± 6 . 32 ms ( Figure 1C; paired t-test comparing post to pre , t ( 9 ) = −3 . 43 , p=0 . 02 ) . This change in RT from pre-training to post-training was on average 76 . 6% of the total reduction in RT associated with the outward-interception task ( i . e . , pre-training compared to the last outward-interception training block ) . Importantly , these changes in RT for point-to-point movements occurred with no significant change in either peak velocity ( paired t-test comparing post to pre , t ( 9 ) = 0 . 98 , p=0 . 70 ) or endpoint error ( paired t-test comparing post to pre , t ( 9 ) = 0 . 95 , p=0 . 70 ) . Thus , repeatedly generating movements at low RTs led to a reduction of the RT on subsequent point-to-point reaches with no detectable decrement in performance . Interception of an inward-moving target had the opposite effect on behavior ( Figure 1D , E; Figure 1—source data 2 – Inward; Table 1 ) . Relative to baseline point-to-point movements , training on the inward-interception task led to a decrease of movement speed ( paired t-test comparing pre-training to the last training block , t ( 9 ) = −6 . 82 , p<0 . 001 ) and an increase in RT ( paired t-test comparing pre-training to the last training block , t ( 9 ) = 2 . 45 , p=0 . 04 ) . RTs during a subsequent block of point-to-point reaches increased on average by 12 . 87 ± 4 . 94 ms relative to baseline , although this change was not significant ( paired t-test comparing post to pre , t ( 9 ) = 2 . 60 , p=0 . 09 ) and corresponded to only 9 . 69% of the increase in RT observed during the interception task . There was also a trend for the velocity of these reaches to increase although this also was not significant ( paired t-test comparing post to pre , t ( 9 ) = 2 . 07 , p=0 . 14 ) and there was no clear change in endpoint error ( paired t-test comparing post to pre , t ( 9 ) = −0 . 56 , p=0 . 59 ) . A mixed-effects model comparing the change in point-to-point RTs across the two groups revealed a significant interaction ( likelihood ratio test; χ2 ( 1 ) =77 . 55 , p<0 . 001 ) between the direction of target motion during training ( outward or inward ) and the change in RT from pre-training to post-training blocks ( Figure 1F ) . This interaction was driven by a difference in RT only during the post-training block ( post-hoc test , p=0 . 03 ) , since RTs during the pre-training block were comparable . Together , these data reveal that the RT can be biased in an experience-dependent manner , suggesting that the RT does not strictly reflect obligate computation time . To further explore the experience-dependent bias in RT , we examined this effect in a more complex task that required participants to plan curved reaches around barriers . This task can be performed in one of two ways: either in the absence or presence of cues that illustrate a specific path around the barriers ( Figure 2A ) . Although the movements required to complete the task are identical in both cases , the presence of a path cue has been observed to provide a significant RT advantage ( Wong et al . , 2016 ) . This advantage is thought to reflect a difference in computational requirements between the cued and uncued conditions , with the latter condition requiring an additional trajectory-planning stage to represent the path shape to be executed . Hence , Experiment 2 provides an assay of habitual effects on RT between two sets of movements that have different planning requirements but the same execution requirements . Experiment 1 provides clear evidence that the RT can be modulated in an experience-dependent manner as if by a habit , and that such biases persist over a large number of trials . However , this paradigm was unable to ascertain whether these RT biases arose because of changes in the rate of computational processing , or because of effects on the non-computational portion of the RT . In the former case , changing the duration of computational processes could increase task success , particularly when shortening the RT in the outward-interception task . Thus there was a motivational incentive that could drive changes in the speed-accuracy trade-off in favor of shortened RTs while maintaining consistent accuracy ( e . g . , motivation could have driven an increase in the rate at which evidence accumulated in a drift-diffusion model of RT [Ratcliff , 1978] ) , much in the same way that reward can reduce RT without affecting accuracy ( Takikawa et al . , 2002; Hübner and Schlösser , 2010; Salinas et al . , 2014; Manohar et al . , 2015 ) . This modified speed-accuracy trade-off may then have been retained when planning subsequent point-to-point movements . Alternatively , if RT biases simply reflect modulation of the non-computational portion of the RT , the manner in which movements are planned should remain unchanged . Two pieces of evidence from Experiment 1 speak against the hypothesis that the observed changes in RT occurred through a motivational effect . First , the RT bias persisted across a large number of trials ( e . g . , the entire next block ) with no obvious decay toward the initially expressed RT . This persistence stands in contrast to the short-lived effects of motivation , which appear to decay after only a few subsequent trials ( Takikawa et al . , 2002; Xu-Wilson et al . , 2009; Wong et al . , 2015 ) . Second , the effect of motivation typically influences not just the RT , but also movement speed and accuracy; in general , these three parameters have often been observed to modulate together in response to motivation ( Takikawa et al . , 2002; Wong et al . , 2015 ) . However , we observed no obvious changes in other movement kinematics such as movement speed or accuracy that accompanied the RT biases . On the other hand , speed and accuracy were constrained by task requirements during the point-to-point movement blocks . Nevertheless , these data suggest that motivation alone cannot account for the long-lasting RT biases observed in Experiment 1 . The kinematically more complex trajectories required in Experiment 2 provided us with greater sensitivity to examine whether RT biases arose from effects on the computational or the non-computational portion of the RT . That is , we were better able to detect any execution-related differences that may have arisen from subtle changes in movement planning that were associated with changes in RT . During Experiment 2 , participants in the cued condition could have initiated their movements at short RTs; however , they exhibited persistently prolonged RTs when previously exposed to the more computationally demanding uncued condition . Such a bias in RT could have occurred for one of two reasons . Participants may have persisted in planning trajectories in the same manner as they had done without path cues , despite the availability of the path cue that would eliminate the need for this planning stage ( habitual planning ) . Alternatively , participants may have used the path cue to reduce the computational load for planning but simply did not express their prepared actions earlier; that is , perhaps the RT does not strictly reflect computation time , but instead contains a manipulable , non-computational component that may be subject to habit ( habitual initiation ) . The analysis of movement kinematics allowed us to distinguish between these two alternatives . In particular , we demonstrated that movement kinematics changed in a consistent manner in the presence or absence of the path cues , arguing that differences in movement kinematics do reflect changes in how these actions were being planned . However , we observed no effect of condition order on trajectory kinematics ( i . e . , reach kinematics were always influenced similarly by the path cues when they were available , regardless of whether participants had prior experience with uncued trials ) , suggesting that participants did not ever exhibit habitual cue-free preparation of the movement trajectory when generating reaches in the presence of a cue . This is particularly evident in Experiment 2B , when movement kinematics indicate a particularly strong reliance on the cue to plan the action in the second cued block even though the RT did not reflect such changes in movement planning . Moreover , reaches in the uncued and cued conditions were confined to different portions of the workspace , making it unlikely that participants habitually applied the identical motor plans from previously-performed uncued reaches . These findings stand in contrast to the RT , which did not consistently modulate according to the presence or absence of the path cue , but was instead experience-dependent . Hence these data suggest that unlike motor planning – which is determined by the current task at hand – movement onset may be influenced in a use-dependent manner consistent with the idea of habitual initiation , and therefore may not represent the actual time required to prepare a movement . If the RT can be habitual , this implies that the time of movement initiation is not simply the end result of computational processing but may actually be represented as a separate movement parameter during the pre-movement period , analogous to movement speed . Indeed , previous work ( Haith et al . , 2016; Brown and Robbins , 1991 ) has suggested that movement initiation may be independent of movement planning . Such a distinction between planning and initiation is consistent with recent data suggesting that an initiation signature can be observed in motor cortex preceding movement onset in a manner that is independent of the specific action being prepared ( Kaufman et al . , 2016 ) . Our finding that movement initiation can be determined by prior experience , rather than the time when planning has completed , further supports this view . Selection of RT based on prior experience rather than computation time implies that it should be possible to choose a RT that is too short to allow for all planning processes to be completed prior to movement initiation . Consistent with this idea , there indeed appear to be situations in which the RT is spontaneously chosen to be improperly short ( Orban de Xivry et al . , 2017; Haith et al . , 2016 ) . In such cases , however , online corrections and refinements that occur after movement initiation help ensure a successful movement outcome ( Orban de Xivry et al . , 2017; Kohen et al . , 2017; Wong and Haith , 2017 ) . Hence , in the event that the RT is biased to be shorter than the time required to complete all necessary decision-making and planning-related computations , the motor system has online correction mechanisms in place to maintain overall task success . Experience-dependent biases have been previously demonstrated for movement parameters such as speed or direction . For example , repetition of movements at a particular speed or toward a particular direction of the workspace strongly influences the kinematics of subsequently performed reaches ( Diedrichsen et al . , 2010; Verstynen and Sabes , 2011; Hammerbeck et al . , 2014; Huang et al . , 2011 ) . Repetition has also been shown to influence the direction of movement invoked by transcranial magnetic stimulation ( Classen et al . , 1998 ) . In these cases , however , the influence of prior experience on movement biases appears for the most part to be short-lived , unlike the long-lasting effects seen here in the RT . Why experience-dependent biases occur is still unclear . It has previously been proposed that such biases may simply be a result of Hebbian learning ( Bütefisch et al . , 2000; Bütefisch et al . , 2004 ) : repetition of the same action presumably strengthens the neural circuits that give rise to that movement , making it more likely to be invoked in the future . From a theoretical standpoint , experience-dependent biases have been framed in the context of a normative Bayesian model in which repetition leads to the construction of a strong prior that influences the preparation of future responses ( Verstynen and Sabes , 2011 ) . Regardless of whether one uses a mechanistic or computational framework , however , explanations for experience-dependent biases rely on the assumption that the parameter in question – e . g . , speed or direction – is represented as a movement parameter that can be specified prior to movement . Therefore , experience-dependent biases on the RT imply that the RT is not simply the passive consequence of the time required to complete computational processing , but may instead be represented as a separate movement parameter that can be subject to habit . In summary , these data support the hypothesis that the RT should be considered a distinct parameter that is selected during the pre-movement period; that is , the RT may in part reflect , but is not strictly dependent upon , underlying computational processes . Thus , caution must be taken when interpreting differences in RT as indicative of the existence of computational stages with smaller or larger processing demands . Such RT differences may simply reflect carryover from RTs recently exhibited in the past ( i . e . , habit ) , regardless of computational demands . Data were analyzed offline using programs written in MATLAB ( The MathWorks , Natick , MA ) and in R ( R Development Core Team , 2016 ) . Analysis code and data are available on GitHub at https://github . com/BLAM-Lab-Projects/RT_habit ( Wong , 2017 ) . A copy is archived at https://github . com/elifesciences-publications/RT_habit . Reaches were selected according to a velocity criterion ( tangential velocity greater than 0 . 05 m/s ) and verified by visual inspection . For each movement , RT was computed as the time between target onset and movement initiation . Inherent delays in the system were estimated to be 105 ms on average; all RTs have been corrected to compensate for this delay . Velocity was calculated by taking the numerical derivative of the hand position after smoothing using a second-order Savitzky-Golay filter with a frame size of 19 samples . Endpoint error was calculated as the absolute radial distance between the final position of the hand and the center of the target . All values are reported along with S . E . M . In Experiment 1 , RT , peak velocity , and interception amplitude were measured during interception training blocks . The amplitude of the target at the time of interception was calculated as the distance of the target away from the central starting position at the first time the hand entered the target; if the participant was not successful at intercepting the target , no target amplitude was recorded for that trial . During pre- and post-interception blocks , RT , peak velocity , and endpoint error were measured for point-to-point reaches . These three metrics were compared within groups using paired t-tests , with p-values adjusted for multiple comparisons using Bonferroni-Holm corrections . RT was also compared across groups using a mixed-effects model in R using the lme4 package ( Bates et al . , 2015 ) , with post-hoc pairwise tests performed using the generalized linear hypothesis testing function in the multcomp package ( Hothorn et al . , 2008 ) and adjusted for multiple comparisons using the Bonferroni-Holm correction . In Experiment 2 , trials were excluded if participants did not complete their reach within 1200 ms . Additionally , since most barrier configurations had more than one possible solution ( e . g . above or below a barrier ) , we pre-selected one possible path as ‘canonical’ and presented that solution on path-cued trials; any ‘non-canonical’ reaches were excluded to allow for a fair comparison of RTs for movements of comparable kinematics . Reaches were not excluded if participants simply hit one of the barriers or did not reach close enough to the target to be considered successful on that trial but otherwise satisfied the inclusion criteria noted above . On average across participants , about 7 . 5% of reaches were excluded from uncued blocks and 3 . 0% of reaches were excluded from cued blocks . RTs were compared in R with mixed-effects models using the lme4 package ( Bates et al . , 2015 ) . For Experiment 2A , this model treated order ( uncued reaches first or cued reaches first ) and condition ( cued or uncued ) as fixed effects and barrier configuration and participant as random effects . Significant effects were determined using a likelihood ratio test to compare pairs of models ( with and without the factor of interest ) . In Experiment 2B , there was only a main effect of block ( 1 , 2 , or 3 ) , with post hoc tests performed in R using the generalized linear hypothesis testing function in the multcomp package ( Hothorn et al . , 2008 ) . All p-values obtained from post hoc tests were adjusted for multiple comparisons using Bonferroni-Holm corrections . The time course of the change in RT during path-cued blocks was compared across groups by fitting generalized linear models in R with factors of group and trial using the nlme package ( Pinheiro et al . , 2014 ) to account for the autocorrelated covariance structure across time within participants . These models remove any autocorrelation structure across trials for each participant individually prior to examining main effects; the form of the autocorrelation structure was selected by fitting Autoregressive-Moving Average ( ARMA ) models to the data and selecting the model fit that yielded the lowest Aikake Information Criterion on average across participants; this led to a choice of an ARMA ( 1 , 1 ) process . Movement kinematics were examined using two methods . First , movements were compared using tools of functional data analysis ( Goldsmith and Kitago , 2016 ) . Briefly , trajectories were time-normalized and evenly resampled . For each trajectory , mean pairwise differences were examined using a function-on-scalar regression model fit using a Bayesian method that allows for correlations in the errors . A 95% simultaneous posterior credible interval was used to identify significant differences between conditions , accounting for the multiple comparisons made across time points within a single trajectory but not for multiple comparisons across trajectories . Second , movement trajectories were also examined by comparing the shape similarity of each movement to the path cue using a Procrustes distance metric ( Goodall , 1991 ) . The Procrustes distance finds the best combination of translation , rotation , and scaling to match a shape to its template , then estimates the remaining dissimilarity normalized between 0 and 1 , where 0 implies the two shapes are perfectly matched . For each movement , the Procrustes distance was estimated; then the overall distributions of Procrustes distances for the cued and uncued conditions were compared using a generalized linear mixed model with a log-normal link function using brms , an R interface to the Stan language ( Buerkner , 2016; Carpenter et al . , 2016; Hoffman and Gelman , 2014 ) . This generalized linear mixed model had main effects of order ( uncued reaches first or cued reaches first ) and condition ( cued or uncued ) ; significant effects were estimated by calculating Bayes factors to test the null hypothesis that the effect coefficient is equal to zero ( according to whether the confidence intervals of the effect included zero ) .	Often , we need to make split-second decisions , be it to avoid an accident , outwit someone in an argument or to win a game . The time that it takes to respond to a signal , i . e . , the reaction time , might be the crucial factor to help us succeed or even survive . Many people assume that the reaction time represents the time needed to prepare an action , and to respond sooner one must ‘think faster’ . However , what really happens in the brain is not well understood . While more complex tasks seem to require longer reaction times , recent evidence suggests that determining when an action begins may not depend on how long it takes to decide which specific action should be taken . Indeed , reaction times may be shortened without changing the accuracy of the planned movement . Using different performance tests , Wong et al . now demonstrate that the reaction time can be influenced by prior experience . In the first task , participants had to respond quickly to catch a moving target . When they later had to move toward a static target , their reaction times were reduced . In the second experiment , the participants practiced a task that required them to plan movements around obstacles . Participants were then given a hint that made it easier to plan their movements , but reaction times did not decrease as expected . Wong et al . then analyzed their movements and demonstrated that although reaction times remained the same , the hint did ease movement planning . This suggests that the reaction time did not always reflect how long it took to prepare a response , but was influenced by prior experience . A next step will be to test what other factors may influence the reaction time . A deeper knowledge of these factors will help to avoid misinterpretation of neural data .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2017	Reaction times can reflect habits rather than computations
β- and γ-cytoplasmic actins are ubiquitously expressed in every cell type and are nearly identical at the amino acid level but play vastly different roles in vivo . Their essential roles in embryogenesis and mesenchymal cell migration critically depend on the nucleotide sequences of their genes , rather than their amino acid sequences; however , it is unclear which gene elements underlie this effect . Here we address the specific role of the coding sequence in β- and γ-cytoplasmic actins’ intracellular functions , using stable polyclonal populations of immortalized mouse embryonic fibroblasts with exogenously expressed actin isoforms and their ‘codon-switched’ variants . When targeted to the cell periphery using β-actin 3′UTR; β-actin and γ-actin have differential effects on cell migration . These effects directly depend on the coding sequence . Single-molecule measurements of actin isoform translation , combined with fluorescence recovery after photobleaching , demonstrate a pronounced difference in β- and γ-actins’ translation elongation rates in cells , leading to changes in their dynamics at focal adhesions , impairments in actin bundle formation , and reduced cell anchoring to the substrate during migration . Our results demonstrate that coding sequence-mediated differences in actin translation play a key role in cell migration . Actin is one of the most essential and abundant eukaryotic proteins , highly conserved across the tree of life . Among the six mammalian actins , β- and γ-cytoplasmic actins are the only two that are ubiquitously expressed in every mammalian cell type and share the highest identity at the amino acid level , with only four conservative substitutions within their N-termini ( Vandekerckhove and Weber , 1978 ) . Despite this near-identity , the coding sequences for these two actin isoforms are different by approximately 13% due to synonymous substitutions ( Erba et al . , 1986 ) . Our work has previously shown that this coding sequence difference can lead to differential arginylation of these two actins . Arginylated β-actin accumulates in vivo , while arginylated γ-actin is degraded ( Zhang et al . , 2010 ) . In mice , β- and γ-cytoplasmic actins play vastly different physiological roles . β-actin knockout leads to defects in embryogenesis and early embryonic lethality ( Bunnell et al . , 2011; Shawlot et al . , 1998; Shmerling et al . , 2005; Strathdee et al . , 2008; Tondeleir et al . , 2013; Tondeleir et al . , 2014 ) , while γ-cytoplasmic actin knockout mice survive until birth and have much milder overall phenotypic defects ( Belyantseva et al . , 2009; Bunnell and Ervasti , 2010 ) . Our prior work has shown that β-actin’s nucleotide sequence , rather than its amino acid sequence , underlies its essential role in embryogenesis ( Vedula et al . , 2017 ) . Using CRISPR/Cas9 , we edited the five nucleotides at the beginning of the β-actin coding sequence within the β-actin gene ( Actb ) , causing it to encode γ-actin protein ( Actbc-g/c-g , β-coded γ-actin ) . Such Actbc-g/c-g mice developed normally and showed no gross phenotypic defects , thus demonstrating that the intact β-actin gene , rather than protein , defines its essential role in embryogenesis ( Patrinostro et al . , 2018; Vedula et al . , 2017 ) . Thus , nucleotide sequence constitutes a major , previously unknown determinant of actin function , even though it is unclear which specific nucleotide-based elements of the actin gene play a role in this effect . Here we tested whether the coding sequence alone , in the context of invariant non-coding elements and independent of the positional effects of the actin gene , plays a role in cytoplasmic actins’ intracellular function . To do this , we incorporated the coding sequences of β- and γ-actins and their ‘codon-switched’ variants ( β-coded γ-actin and γ-coded β-actin ) into otherwise identical constructs containing the human β-actin promoter , an N-terminal enhanced Greef Fluorescent Protein ( eGFP ) fusion , and the β-actin 3′UTR . Stable expression of these constructs in mouse embryonic fibroblasts ( MEFs ) resulted in dramatically different effects on directional cell migration . While cells expressing β-actin migrated at rates similar to wild-type untransfected cells in wound-healing assays , cells expressing γ-actin migrated nearly twofold faster . This difference depended directly on the coding sequence , as evident by the use of the ‘codon-switched’ actin variants . Expression of γ- or γ-coded β-actin led to changes in cell morphology and distribution of focal adhesions , which were larger in size and localized mostly at the cell periphery rather than under the entire cell , similarly to previously reported cellular phenotypes linked to poorer cell attachment and faster migration ( Kim and Wirtz , 2013 ) . Focal adhesions in γ-actin- or γ-coded-β-actin-expressing cells appeared to be poorly anchored , though often not visibly associated with long actin bundles . In contrast , long actin cables could be clearly seen anchoring focal adhesions in β-actin- or β-coded γ-actin-expressing cells . Single-molecule measurements of actin translation using the SunTag system at the focal adhesion sites showed an approximately twofold faster translation elongation of β-actin compared to γ-actin . Fluorescence recovery after photobleaching ( FRAP ) demonstrated that γ-actin accumulation in cells was slower than β-actin , further confirming global differences in actin isoform translation . Molecular simulations of actin assembly at the focal adhesions showed that differences in translation rates can directly impact actin bundle formation , leading to shorter actin bundles in the case of slower translating γ-actin , in agreement with our experimental data . Our results demonstrate that nucleotide coding sequence-dependent translation rates , coupled to zipcode-targeted actin mRNA localization , play an essential role in differentiating actin isoforms’ function in cell migration . To test the specific effect of coding sequences on intracellular functions of actin isoforms , we generated immortalized MEF cell cultures stably expressing β- and γ-actin coding sequences , as well as their codon-switched variants ( β-coded γ-actin and γ-coded β-actin ) , cloned into identical expression constructs under the human β-actin promoter , containing an N-terminal eGFP fusion and the β-actin 3′UTR ( Figure 1 , top left , and 'Materials and methods' ) . This construct design enabled us to confine our experiments to the effects of the coding sequence and exclude any potential contribution from other elements known to mediate differences between β- and γ-actins , including promoter-mediated transcription ( Tunnacliffe et al . , 2018 ) , differential 3′UTR-mediated mRNA targeting ( Hill and Gunning , 1993; Katz et al . , 2012; Kislauskis et al . , 1993; Kislauskis et al . , 1994 ) , and differential N-terminal processing ( Zhang et al . , 2010 ) . Cell populations stably expressing eGFP constructs were checked to ensure similar levels of eGFP mRNA , as well as to confirm that the expression of the exogenous eGFP-actin did not have any significant effect on the endogenous β-actin and γ-actin mRNA levels ( Figure 1—figure supplement 1 ) . We also confirmed that β-actin 3′UTR targeted the eGFP-actin mRNA to the cell periphery , using fluorescence in situ hybridization ( FISH ) ( Figure 1—figure supplement 2 ) . Finally , we confirmed that the level and distribution of F-actin in each of the cell cultures transfected with different actin isoforms was largely similar to each other ( Figure 1—figure supplement 3 ) . Thus , in these cell populations , the effects of exogenously expressed actin could be tested without perturbation of other actin-related processes that are essential for cell viability . β- and γ-cytoplasmic actins make up more than 50% of the total actin in these cells and have been previously shown to play major non-overlapping roles in directional cell migration ( Patrinostro et al . , 2017 ) . We , therefore , tested whether cells expressing eGFP-β-actin or eGFP-γ-actin showed any differences in cell migration using a wound-healing assay . Strikingly , while cells expressing eGFP-β-actin migrated at rates similar to wild-type untransfected cells , cells expressing eGFP-γ-actin migrated nearly twofold faster ( Figure 1 , top right and bottom; Figure 1—figure supplement 4; Videos 1 and 2 ) . This difference in cell migration rates was coding sequence dependent , as seen in cells expressing the codon-switched actin variants , γ-coded β-actin ( which migrated faster , similarly to those expressing γ-actin ) , and β-coded γ-actin ( which migrated slower , like β-actin-expressing cells ) ( Figure 1 , top left; Videos 3 and 4 ) . Thus , the effect of actin isoform expression on directional cell migration is mediated by their nucleotide coding sequence and does not appear to be influenced by their amino acid sequence . In normal cells , the β-actin 3′UTR contains a zipcode sequence that is required for its mRNA localization to the cell periphery ( Kislauskis et al . , 1993 ) and has been shown to be important for directional cell migration ( Condeelis and Singer , 2005; Katz et al . , 2012; Kislauskis et al . , 1994; Kislauskis et al . , 1997 ) . γ-actin mRNA has no such sequence and does not undergo targeting to the cell periphery ( Hill and Gunning , 1993 ) . All our constructs described above contained the β-actin 3′UTR with the zipcode sequence as one of the constant elements . To test whether 3′UTR-mediated targeting of actin mRNA affects the cell migration phenotypes observed in our stably transfected cell cultures , we performed the same experiment using cell cultures stably expressing similar actin constructs , but without the β-actin 3′UTR ( Figure 1—figure supplement 2 ) . These cells did not exhibit significant differences in cell migration rates ( Figure 1—figure supplement 5 ) . Thus , differences in the effects of cytoplasmic actin coding sequences on cell migration require mRNA targeting to the cell periphery . Changes in cell migration rates are normally associated with changes in actin dynamics at the leading edge , rate and persistence of leading edge protrusions and retractions , as well as focal adhesion formation and dynamics , which affect cell spreading , polarization , and attachment to the substrate . Focal adhesions’ strength and persistence are closely regulated by their association with actin filaments , which grow at the focal adhesion sites to form a dynamic actin bundle that participates in anchoring the cells to the substrate . Thus , focal adhesions critically depend on actin dynamics in the vicinity of the adhesion site . In turn , focal adhesions can regulate cell spreading and polarization , in addition to cell migration rates . To test these processes in eGFP-actin isoform-transfected cells , we first looked at the rate and persistence of leading edge protrusions and retractions , but found no consistent differences between the cell populations that correlated with cell migration rates ( Figure 2—figure supplement 1 ) . We next assessed focal adhesion dynamics in these cells using total internal reflection fluorescence microscopy ( TIRF-M ) of eGFP-β-actin and eGFP-γ-actin . Since the imaging volume in TIRF-M is limited to the basal 300 nm or less , we reasoned that most of the actin signals visible in this volume should be associated with focal adhesion patches . Imaging the long-term ( hours ) behavior of actin at focal adhesion patches during wound healing using TIRF-M revealed that in migrating cells , eGFP-β-actin patches appeared more prominent and persisted considerably longer than eGFP-γ-actin patches ( Figure 2A–C ) , suggesting that focal adhesions in eGFP-β-actin-expressing cells persist for longer periods of time . At the same time , testing short-term ( 5 min ) actin dynamics at the focal adhesions using Fluorescence Recovery After Photobleaching ( FRAP ) showed no notable differences in focal adhesion recovery rates that correlated with either coding or amino acid sequence ( Figure 2—figure supplement 2 ) . Thus , different actin isoforms affect long-term focal adhesion persistence without strongly affecting short-term focal adhesion or protrusion dynamics during persistent directional migration at the cell leading edge . To get deeper insights into the focal adhesion morphology and distribution in cells transfected with different actin isoforms , we grew cells at a low density , to enable visualization of the morphology and cytoskeleton-dependent structures in individual cells isolated on coverslips , without contacting their neighbors . Notably , cells in such scarce cultures are under no stimuli to migrate . Many of them remain stationary or move randomly around the same area , resulting in much slower rates of persistent migration and overall displacement over time . Consequently , such sparsely grown cells transfected with different actin isoforms do not prominently differ from each other in their migration ( Figure 3—figure supplement 1 ) , even though they are expected to undergo similar actin isoform-related changes at the subcellular level . To analyze focal adhesions and spreading in these cells , we first used TIRF-M to image single cells stained with antibodies to the focal adhesion protein paxillin . These assays revealed prominent differences in focal adhesion morphology and distribution between the cell populations transfected with different actin isoforms ( Figure 3 , top row of images; see also Figure 3—figure supplements 2–5 ) . In eGFP-β-actin-expressing cells , focal adhesions had a normal elongated morphology and were distributed throughout the entire cell footprint . In contrast , eGFP-γ-actin-expressing cells formed focal adhesions that localized mostly at the cell periphery ( Figure 3 , top row of images; Figure 3—figure supplement 6A , B ) . This trend depended on the actin coding sequence: focal adhesions in eGFP-β-coded-γ-actin-expressing cells resembled those in eGFP-β-actin , while focal adhesions in eGFP-γ-coded-β-actin-expressing cells were like those in eGFP-γ-actin-expressing cells ( Figure 3—figure supplement 6A , B ) . Imaging eGFP-actin in widefield showed that most focal adhesions in cells expressing eGFP-β-actin and eGFP-β-coded γ-actin were associated with long thick bundles of actin emanating from the focal adhesion point ( Figure 3 , top , and Figure 3—figure supplements 2–5 ) . In comparison , the dorsal bundles connecting to the focal adhesions were much less prominent in γ-actin- and γ-coded β-actin-expressing cells . Focal adhesion size uniquely predicts cell migration rate ( Kim and Wirtz , 2013 ) , with the larger focal adhesions correlating with faster migration speeds . We measured the focal adhesion area in all the four cell cultures transfected with different actin isoforms and found that faster migrating cells expressing γ-actin , and γ-coded β-actin indeed , had significantly larger focal adhesions than slower migrating cells expressing β-actin and β-coded γ-actin ( Figure 3 , bottom ) . Morphologically , focal adhesions in γ-actin- and γ-coded β-actin-expressing cells appeared wider and less elongated than in β-actin- and β-coded γ-actin-expressing cells; however , global measurements of their aspect ratios did not reveal any consistent statistically significant differences ( Figure 3—figure supplement 6C ) . This could be due to the fact that the majority of focal adhesions in all of these cells were small and dot-like , and only a few larger ones tended to exhibit potential differences in morphology . Thus , β- and γ-actin coding sequences determine the size and distribution of focal adhesions in migrating cells in a manner that correlates with changes in their migration speed . We also measured focal adhesion recovery rates in single-cell cultures using FRAP . Focal adhesions in cells transfected with β-actin and β-coded γ-actin recovered slightly faster than cells transfected with γ-actin and γ-coded β-actin ( Figure 3—figure supplement 7 ) . While statistically significant , these coding sequence-dependent differences were small , and thus it is unclear if they can prominently contribute to the cells’ phenotype . Cell spreading and polarization are critically determined by their adhesion to the substrate and correlate with their migratory behavior ( Kim and Wirtz , 2013 ) . To test whether focal adhesion changes in our cell populations are accompanied by changes in cell spreading and polarization , we used Celltool ( Pincus and Theriot , 2007 ) to analyze shape distribution of single cells stably expressing eGFP-β-actin , eGFP-γ-actin , and their codon-switched variants . Using images of live single cells , the shape space was parameterized into various shape modes , and shape modes 1 and 2 accounted for ~60% of variance in shapes across all cells ( Figure 3—figure supplement 8A , inset ) . The first mode roughly captures the variation in the size of the cell footprint on the substrate and accounts for ~40% of the variance in shape , while the second mode captures cell polarization and accounts for ~20% of the variance in shape . Using these shape modes to analyze images of cells expressing different actin isoforms’ coding sequences , we found that cells expressing eGFP-β-actin had a larger footprint ( Figure 3—figure supplement 8A , clustered to the left of the y-axis ) and had more variance in their polarization ( Figure 3—figure supplement 8A , spread across the y-axis ) , while cells expressing eGFP-γ-actin exhibited the opposite trends ( Figure 3—figure supplement 8A , clustered mostly in the top right quadrant ) . Expression of the codon-switched actin variants , β-coded γ-actin , and γ-coded β-actin showed that the footprint size depended on the actin isoform coding sequence , while the polarization variance appeared to be amino acid sequence-dependent . Changes in the area of the cell footprint can arise due to either reduced cell spreading or reduced overall cell size . To distinguish between these possibilities , we quantified the area of trypsinized near-spherical cells ( pre-spreading ) , which directly reflects cell size and volume . Cells expressing the γ-actin coding sequence were slightly smaller than those expressing β-actin coding sequence ( Figure 3—figure supplement 8B , left ) . This ~6% difference in cell size was far less prominent than the difference in spread cell area ( Figure 3—figure supplement 8B , right ) , which accounted for a greater than 80% change in the size of cell footprint . Thus , cells expressing γ-actin are less spread on the substrate , and this difference in spreading is coding sequence-dependent . In search of an underlying mechanism that could link actin isoforms’ coding sequence to their intracellular properties , we turned to our previous study that used computational predictions of the mRNA secondary structures for β-actin and γ-actin . This study suggested that the coding region of β-actin mRNA forms a more relaxed secondary structure than that of γ-actin , predicting potential differences in translation elongation rates ( Zhang et al . , 2010 ) . Such differences , if prominent enough , could in principle lead to changes in cells’ ability to form focal adhesions and migrate . To test this prediction , we first compared the rates of overall protein accumulation of eGFP-β- and eGFP-γ-actin , by comparing FRAP of the total eGFP signal in the cell after whole-cell photobleaching . We reasoned that this would serve as a proxy for estimation of newly synthesized β- and γ-actin ( Figure 4A ) . Notably , the recovery observed in these FRAP experiments within a 10-min imaging window arises from the folding and maturation of already synthesized eGFP fused to actin ( since the eGFP maturation rate in vivo has been estimated to be approximately 14 min [Balleza et al . , 2018; Iizuka et al . , 2011] ) ; given the constant time delay , this recovery rate directly reflects the rate of de novo synthesized actin accumulation within the imaging window . Photobleaching was calibrated to ensure that the cells remained healthy and visually normal during the experiment ( Figure 4A , bottom ) . The recovery rate was significantly faster for eGFP-β-actin compared to eGFP-γ-actin ( Figure 4A , top right ) . Thus , newly synthesized β-actin accumulates in cells faster than γ-actin . To directly estimate β- and γ-actin translation elongation rates , we performed single-molecule imaging of nascent peptide synthesis ( SINAPS ) for these two actin isoforms using the SunTag system ( Wu et al . , 2016 ) . Similarly to the constructs used for generating eGFP-actin stable cell populations , we ensured that the coding sequence was the only variable , flanked by otherwise identical upstream and downstream elements , including the promoter of the polyubiquitin gene ( UbC ) for constitutive expression , the N-terminal 5′ SunTag fusion to visualize the nascent peptide , the C-terminal auxin-induced degron to degrade fully synthesized polypeptides and reduce the background signal , the β-actin 3′UTR for cell periphery targeting , and MS2 repeats in the non-coding region to visualize mRNA via constitutively expressed MS2 coat-binding protein ( MCP ) fused to a HaloTag ( Figure 4—figure supplement 1 ) . Simultaneous imaging of SunTag-bound superfolder GFP ( sfGFP ) and MS2-bound HaloTag in fixed cells ( labeled with JaneliaFluor 646 ) enabled us to estimate the number of nascent peptides ( NAPs ) per mRNA , a measure of the ribosome load and , by proxy , the translation elongation rate . Assuming similar translation initiation rates for both constructs ( given their identical 5′ sequences at and around the translation initiation sites ) , fewer elongating ribosomes in this assay arise from their faster translocation over the mRNA , leading to weaker sfGFP signals per mRNA . Thus , differences in ribosome load per mRNA ( and thus , the number of NAPs per mRNA ) would directly indicate differences in translation elongation . γ-actin coding sequence showed a nearly twofold higher level of NAPs/mRNA than that of β-actin ( Figure 4B , Figure 4—figure supplement 2 ) , indicating that the elongating ribosomes had a twofold higher load , and thus slower translocation rate , over γ-actin mRNA compared to β-actin . This is the first characterization of different ribosome loads on individual mRNAs of two protein isoforms that have the same coding sequence length . Next , we estimated the real-time translation elongation rate of β- and γ-actin using FRAP of individual translation sites . To minimize mRNA movement , we tethered SINAPS-β-actin and SINAPS-γ-actin mRNA to focal adhesions , by co-transfecting cells with vinculin-MCP-HaloTag fusion . Fluorescence recovery rate of individual translation sites directly reflects the rate of translation elongation to generate new nascent peptides bearing new sfGFP bound to the SunTag peptides . This recovery rate was approximately twofold slower for γ-actin compared to β-actin ( Figure 4C ) , confirming that γ-actin translation elongation is indeed slower than that of β-actin , in agreement with the NAP/mRNA measurements . Using both sets of data , we conclude that the translation elongation rate of the two actin isoforms differs by approximately twofold—faster for β-actin compared to γ-actin ( Figure 4D ) . During cell migration , the initial formation of nascent focal adhesions critically depends on local actin subunit supply rate . Many studies assume that this subunit supply rate is not a limiting factor in vivo , due to high concentrations of G-actin at the cell leading edge ( Raz-Ben Aroush et al . , 2017 ) . However , it is likely that the actin bundles forming at focal adhesion sites must compete with the actin meshwork at the lamellipodium for polymerization competent actin . There is increasing evidence suggesting such a competition between various actin-driven processes in vivo ( Faust et al . , 2019; Suarez and Kovar , 2016 ) . In addition , it is possible that at a given moment , some , or most , of the free actin can be sequestered , for example , by monomer-binding proteins ( Skruber et al . , 2018 ) , forcing the elongating leading edge filaments to depend on de novo synthesized actin . In support , actin mRNA targeting to the cell leading edge is essential for cell migration , suggesting that local actin synthesis at the cell leading edge must be important ( Katz et al . , 2012 ) . Furthermore , local actin translation bursts have been observed in neurons ( Buxbaum et al . , 2014 ) . It is possible that these bursts , regardless of the overall actin concentrations , are required for locally supplying actin subunits at the focal adhesions during cell migration . If so , replacing the faster translationally elongating β-actin with the slower elongating γ-actin at these sites could potentially limit this supply and make a difference in focal adhesion anchoring , leading to shorter actin bundles at the focal adhesions , poorer spreading , and faster migration seen in γ-actin-expressing cells . Since measuring local polymerization-competent actin in a cell is impossible experimentally , we used the computational model of active networks , MEDYAN ( Popov et al . , 2016 ) . We simulated actin bundle growth at the focal adhesions at different subunit supply rates , in the presence of non-muscle myosin II motors and α-actinin as crosslinkers , which are critical for actin filament bundling in cells ( Chandrasekaran et al . , 2019; Figure 5A and Figure 5—figure supplement 1A ) . In these simulations , filaments elongate by incorporating newly supplied actin monomers , and then bundle together through the action of myosin motors and crosslinkers . During 10-min simulations , the time window typically sufficient for establishment of robust focal adhesions , varying actin subunit supply rate resulted in pronounced differences in the length of the actin bundle growing from the focal adhesion site ( Videos 5 and 6 ) . A twofold decrease in subunit supply rate resulted in over a twofold decrease in actin bundle length ( Figure 5A , right and Figure 5—figure supplement 1B ) . To test this prediction experimentally , we measured the length of eGFP-actin-decorated bundles emanating from paxillin-positive focal adhesion patches in cells stably expressing different eGFP-actin isoforms , using both the migrating cells at the edge of the wound and single cells ( Figure 5B–D ) . In both types of cultures , actin bundles associated with the focal adhesion sites were markedly longer in β-actin-expressing cells , compared to those expressing γ-actin ( Figure 5B–D ) . Moreover , these trends followed the actin coding sequence , rather than the amino acid sequence ( Figure 5B , C ) . Thus , the slower subunit supply dictated by differences in translation elongation rates of β- and γ-actin coding sequences during the initial events of focal adhesion formation and maturation bears direct consequences to cell adhesion and migration . Our study follows up on the recent discovery of the essential role of nucleotide , rather than amino acid , sequence in non-muscle actin isoform function and demonstrates for the first time that actin coding sequence , uncoupled from other gene elements , can directly affect cell behavior . We found that differences in β- and γ-actin coding sequences result in different ribosome elongation rates during their translation , leading to changes in cell spreading , focal adhesion anchoring , and cell migration speed . This study constitutes the first direct comparison of translation rates of two closely related proteins and the first demonstration that these translation rates can mediate their functions in vivo . On the surface , it appears to be surprising that expression of the slower-translating γ-actin can make the cells move faster than the faster-translating β-actin . However , this result fits well into the context of the previously proposed localized bursts of β-actin translation , implicated in cell spreading and focal adhesion formation ( Condeelis and Singer , 2005; Katz et al . , 2012 ) . Our data show that faster translating β-actin is required for generating stable focal adhesions , while the slower translating γ-actin leads to faster focal adhesion turnover . Importantly , our data also show that the effects mediated by slower translating γ-actin on cell migration manifest only when the γ-actin mRNA is targeted to the cell periphery via the β-actin zipcode sequence . Our study on codon profile-mediated actin isoform-specific translation differences , along with studies that showed actin isoform UTRs conferring isoform-specific mRNA localization with unique functional consequences ( Kislauskis et al . , 1993; Moradi et al . , 2017 ) , establish physiological roles to nucleotide elements in determining cellular phenotypes of isoactins . We propose that slower translation at the leading edge makes γ-actin less capable of supporting and sustaining strong focal adhesions . Given that rapid polymerization of actin is required for lamellipodium protrusion and focal adhesion formation , one plausible role of localized fast actin translation at focal adhesions is to balance the competition between actin monomer pools required for protrusion and adhesion formation . In support , in γ-actin-expressing cells , most focal adhesions , while larger in area , do not appear to be visibly anchored by prominent actin bundles . In contrast , cells expressing β-actin contain long actin cables emanating from most of the focal adhesion sites . This difference impairs spreading without significantly changing protrusion dynamics . We propose that the faster migration rate is thus caused by weaker cell-substrate attachment in cells expressing γ-actin ( Figure 6 ) . Notably , this increase in cell speed is only evident in a wound-healing assay , where cells are collectively stimulated to migrate directionally , rather than randomly move around as typical for these cells in single-cell cultures of MEFs . It is likely that actin behavior in response to the strong signals for cells to polarize and move directionally during wound healing depends on local actin translation more critically than during random migration , where cells can change directions or remain stationary for extended periods of time and are not constrained in the direction of their polarization and motility . These constraints could potentially involve tension generated in dense cultures during collective migration ( Trepat et al . , 2009 ) , as well as other forms of signaling in dense cultures . While the local supply rate of actin subunits to the forming focal adhesion sites during cell migration is nearly impossible to measure experimentally with the currently available methods , the use of modeling and simulation enables us to vary this parameter and estimate the subunit supply rate that could make a difference in this process . It appears surprising that a twofold difference in translation elongation rate found in our study could exert such a pronounced effect on the length of the actin bundles forming locally at the focal adhesion sites , especially given the fact that polymerization-competent actin should exist everywhere in the lamellipodia . Our results suggest that the concentration of this polymerization-competent actin pool may be far lower than previously estimated , potentially due to the competition between different actin pools undergoing rapid polymerization , as well as the action of monomer sequestering proteins and/or posttranslational modifications that may prevent actin from incorporating into filaments ( Skruber et al . , 2018 ) . It is also possible that , even with a higher actin concentration in the lamellipodia , focal adhesions compete with the monomer pool , thus requiring local translation of β-actin . Notably , these differences are expected to be even higher with the native non-eGFP-fused actin isoforms , which likely differ in their translation initiation rates in addition to the difference in elongation rates we observed . These questions , and the exact interplay between newly synthesized and diffusible actin in cells , constitute an exciting direction of future studies . Our previous work has shown that actin coding sequence leads to differential arginylation of β- and γ-actin ( Zhang et al . , 2010 ) , and this arginylation is important for directional cell migration ( Karakozova et al . , 2006 ) . In the present study , the use of N-terminal eGFP fusions likely excludes arginylation as a variable in our cell populations , since arginylation is believed to require an exposed β-actin N-terminus . The use of GFP-actin fusions also likely limits the types of effects we can observe , since these fusions are not fully functionally equivalent to the native actin and are impaired , for example , in formin nucleation ( Chen et al . , 2012 ) . Thus , our constructs cannot substitute for all aspects of normal actin in cells , and this could explain the fact that the strongest effects we observe are related to cell adhesion and migration , the processes that are likely able to fully utilize eGFP-actin . Notably , in our cells , the endogenous β- and γ-actin are still present and still able to support cell migration , likely compensating for these types of functions , and potentially diminishing the observed phenotypes . Thinking of the nucleotide sequence , rather than the amino acid sequence , as a determinant of actin function is a novel view that opens up many exciting questions . The present study demonstrates that coding sequence alone can play a significant role in cell behavior , but it does not exclude the possibility that other nucleotide-based elements of the actin gene also contribute to actin isoforms’ global role in organism survival . For example , it has been shown that a unique intron in the γ-actin gene can contribute to its function ( Lloyd and Gunning , 1993 ) . Studies of the interplay of nucleotide- and amino acid-based determinants in actin functions constitute exciting future directions in the field . Constructs were generated using the eTC GFP beta-actin full-length plasmid ( Addgene plasmid # 27123 ) and eTC GFP beta-actin ΔZip ( Addgene plasmid # 27124 ) , which were gifts from Robert Singer ( Rodriguez et al . , 2006 ) . Plasmid #27123 encodes the human β-actin promoter , followed by TC-eGFP , a five-amino-acid ( GSTSG ) linker , and the full-length human β-actin complementary DNA ( cDNA ) containing β-actin 3′UTR ( eTC GFP beta-actin full-length plasmid ) , followed by the transcription terminator bGH polyA , which carries the typical AAUAAA sequence responsible for transcription termination and polyadenylation of the mRNA . Plasmid #27214 is identical , except that it does not contain the beta-actin 3′UTR ( eTC GFP beta-actin ΔZip ) . No 5′UTR or any other non-coding elements from the β-actin gene besides the promoter are included in these plasmids . To generate actin isoform-encoding constructs for this study , we used plasmid #27213 or #27214 as the backbone for the construct with and without β-actin 3′UTR , respectively . To obtain constructs expressing actin isoforms in this study , β-actin coding sequence ( starting with the first ATG and ending with the terminator codon ) was replaced with the corresponding sequence encoding mouse β-actin or mouse γ-actin ( generating eGFP-β-actin and eGFP-γ-actin from plasmid #27213 and eGFP-β-actin Δ3′UTR and eGFP-γ-actin Δ3′UTR from plasmid #27214 ) . For the codon-switched constructs , point mutations were introduced into the coding sequence of the actin isoforms as shown in Figure 1 and described in Vedula et al . , 2017 to generate eGFP-β-coded γ-actin and eGFP-γ-coded β-actin . pUbC-FLAG-24xSuntagV4-oxEBFP-AID-baUTR1-24xMS2V5-Wpre was a gift from Robert Singer ( Addgene plasmid # 84561 ) and was used to generate the actin isoform SINAPS reporters ( Wu et al . , 2016 ) . The β- and γ-actin coding sequences were cloned in place of the oxEBFP sequence in the original construct . For fixed-cell imaging of ribosome load per mRNA , phage UbC NLS HA stdMCP stdHalo , a gift from Jeffrey Chao ( Addgene plasmid # 104999 ) ( Voigt et al . , 2017 ) , was used . For constructing the mRNA tether in live-cell real-time imaging of translation dynamics , pUbC-nls-ha-stdMCP-stdGFP , a gift from Robert Singer ( Addgene plasmid # 98916 ) ( Wu et al . , 2016 ) , was used to clone mouse vinculin sequence upstream of stdMCP , and stdGFP was replaced by HaloTag . pHR-scFv-GCN4-sfGFP-GB1-NLS-dWPRE , a gift from Ron Vale ( Addgene plasmid # 60906 ) ( Tanenbaum et al . , 2014 ) , was used for the NAP sensor . pBabe TIR1-9myc , a gift from Don Cleveland ( Addgene plasmid # 47328 ) ( Holland et al . , 2012 ) , was used for degrading the fully synthesized SINAPS construct . Spontaneously immortalized MEFs used in this project were obtained in the lab from E12 . 5-E16 . 5 mouse embryos and immortalized by continuous passaging in culture , using Dulbecco’s modified Eagle’s medium ( DMEM ) ( Gibco ) supplemented with 10% fetal bovine serum ( FBS ) ( Gibco ) as the tissue culture medium . These cells were produced and maintained in the lab and have not been independently authenticated , but they were continuously observed to maintain characteristic morphology of mouse embryonic fibroblasts . All mycoplasma tests conducted in the lab were negative . To obtain the stable cell cultures described in this study , these cells were transfected with the linearized EGFP-actin constructs described above . Following G418 selection , GFP-positive cells were sorted using fluorescence activated cell sorting ( FACS ) and cultured . Live-cell imaging was carried out in FluorBrite DMEM ( Life Technologies ) culture media supplemented with 10% FBS ( Sigma ) and L-glutamine ( Gibco ) . HEK-293T cells were cultured in DMEM ( Gibco ) supplemented with 10% FBS ( Gibco ) . Lipofectamine 2000 ( Life Technologies ) was used to transfect these cells with plasmids for generating either lentiviral particles , pMD . G , pPAX , and plasmid containing gene of interest , or retroviral particles , pCL10A and pBabe TIR1-9myc . Virus-containing medium was harvested and used to infect immortalized MEFs in the following order: first , either UbC-NLS-HA-stdMCP-stdHalo for fixed-cell imaging of number of NAPs/mRNA or Vinculin-stdMCP-Halo for live-cell dynamics of translation elongation , second , TIR1-9myc followed by puromycin selection of infected cells , third , scFv-GCN4-sfGFP-GB1-NLS , and lastly , either SINAPS-β-actin- or SINAPS-γ-actin-containing lentivirus . These polyclonal cells were used for imaging the number of NAPs on each of β- and γ-actin constructs . Indole-3-acetic acid was used at 500 µg/ml to induce degradation of fully synthesized SINAPS constructs . Janelia Fluor 646-tagged Halo ligand ( Promega ) was used at 200 nM final concentration to label SINAPS-mRNA in cells prior to fixation/imaging . Cell migration was stimulated by making an infinite scratch wound . The cells were allowed to recover for a period of 2 hr before imaging . Images were acquired using a X10 phase objective on a Lecia DMI 4000 equipped with a Hamamatsu ImagEM EMCCD camera . Images were captured every 5 min for 10 hr . Migration rates were measured as the area covered by the edge of the wound in the field of view per unit time using Fiji ( NIH ) . For TIRF wound-healing experiments , cells were imaged on a Nikon Ti with a X100 , 1 . 49 NA objective using the 488 nm laser and an Andor iXon Ultra 888 EMCCD . For all FRAP experiments , imaging was carried out on a Nikon Ti inverted microscope equipped with either an Andor iXon Ultra 888 EMCCD camera ( 0 . 13 µm/pixel—for imaging TIRF-FRAP and widefield whole-cell eGFP-actin FRAP ) or an Andor iXon Ultra 897 EMCCD camera ( 0 . 16 µm/pixel—for imaging SINAPS-FRAP using a Yokogowa CSU X1 spinning disc confocal ) . Photobleaching was carried out with a Bruker miniscanner equipped with XY Galvo mirrors . The region of interest for photobleaching was defined using a freehand Region of Interest ( ROI ) manager in Nikon Instruments NIS elements software: an elliptical region encompassing an actin patch at the cell periphery for TIRF-FRAP , the entire cell for widefield whole-cell FRAP , and a single-pixel spot containing the translation site for SINAPS-FRAP . eGFP-actin-expressing cells were seeded on Matek glass bottom dishes and allowed to spread overnight . For photobleaching , the 488 nm laser was set to 80% power and used to bleach a defined eGFP-actin patch at the cell periphery with a dwell time of 400 µs/pixel . Images were acquired in the TIRF mode with the 488 nm laser set to 50% power and 200 ms exposure and an electron-multiplying ( EM ) gain of 200 . Images were acquired at 3 s intervals for 12 s pre-bleach and 6 min post-bleach . The change in fluorescence intensity in a circle within an actin patch that was not bleached was used as a reference to account for photobleaching during acquisition . The change in intensity within a circle of the same area within the bleached actin patch was used to calculate the recovery curve . The obtained values were normalized to one at pre-bleach , and the resulting post-bleach curves were fit using non-linear regression to a single exponential fit in GraphPad PRISM . For whole-cell eGFP-actin FRAP , the whole cell was outlined . For photobleaching , the 488 nm laser was set to 70% power with a dwell time of 70 µs/pixel . Acquisition was carried out using a 488 nm LED illumination from Spectra/Aura with 10% illumination intensity and 200 ms exposure with an EM gain of 300 . Images were acquired every 10 s for a total of 10 min after bleaching . The recovery curves obtained were fit using a linear regression model in GraphPad PRISM . For live-cell SINAPS-FRAP , cells expressing SINAPS-actin constructs were tethered using Vinculin-stdMCP-Halo ( see sections 'Constructs' and 'Generation of polyclonal stable cell populations' above ) . For photobleaching , the 488 nm laser was set to 80% power with a dwell time of 1 ms/pixel . Images were acquired in the spinning disc confocal mode with the 488 nm laser set to 30% power and 35 ms exposure with an EM gain of 300 . Images were acquired at 700 ms intervals for 30 s pre-bleach and 7 min post-bleach . FishQuant ( Mueller et al . , 2013 ) was used to detect NAP and mRNA signals . Spots in the NAP channel that were within 300 nm of a spot in the mRNA channel were considered bonafide NAPs and were used for estimating the integrated fluorescence intensity in both channels . Airlocalize ( Lionnet et al . , 2011 ) was used to fit the signal from tethered NAPs . The integrated signal was recorded pre-bleach and post-bleach . These values were used to calculate the translation elongation rates of the two actin isoforms . Assuming that beta-actin has the same elongation rate as Suntag , AID , and linkers , following the theoretical derivation of Wu et al . , 2016 , it is straight forward to show that the proportion of beta-actin contribution to recovery time and NAP/mRNA is ( L+ ( N + 1 ) /2 ) / ( S + L + ( N + 1 ) /2 ) , in which N is the total number of Suntags , S is the beta-actin length in the unit of 1 Suntag , and L is the length of AID and linkers in the unit of 1 Suntag , shown as the gray bar in Figure 4B and C . It is not surprising to see from those figures that the ratios of recovery time to NAP/mRNA are similar for beta- and gamma-actin , since the initiation rates for both constructs should be the same , given the identical N-termini . Therefore , we calculated the variance-weighted geometric average of the two ratios , which is T ~ 26 . 9 s , and used it to combine the data from recovery time and NAP/mRNA to calculate the beta-actin elongation rate: Rb= ( S + L + ( N + 1 ) /2 ) /t or Rb= ( S + L + ( N + 1 ) /2 ) /n/T , where t is recovery time and n is NAP/mRNA , for each data point , followed by geometric averaging . The contribution from Suntag , AID , and linkers to recovery time is T0 = ( L+ ( N + 1 ) /2 ) /Rb , which is used to calculate the gamma-actin elongation rate: Rg = S/ ( t-T0 ) or Rg = S/ ( n*T-T0 ) for each data point , followed by geometric averaging . The results are shown in Figure 4D . To quantify the amount of actin polymer , cells were seeded on coverslips in six-well plates at 20 , 000 cells/well overnight and fixed in 4% ( w/v ) paraformaldehyde ( PFA ) at room temperature for 30 min . Cells were then stained with phalloidin conjugated to AlexaFluor 594 ( Molecular Probes ) . Images were acquired on Leica DM6000 at X40 and the total intensity of phalloidin per cell was measured using Fiji ( NIH ) . To analyze focal adhesions in single cells , eGFP-actin-expressing cells were seeded on coverslips and allowed to adhere and spread overnight . Cells were then fixed in 4% ( w/v ) PFA at room temperature for 30 min followed by 0 . 5% Triton-X 100 treatment for 5 min . Cells were incubated with mouse anti-paxillin monoclonal antibody ( BD Biosciences ) , followed by AlexaFluor 555-conjugated goat-anti-mouse secondary antibody ( Life Technologies ) . Cells were imaged with Citifluor ( Cytoskeleton Inc ) anti-bleaching agent . To analyze cell spreading and cell area , Celltool was used to outline cell shapes and classify them and extract shape modes . The shape modes that captured 60% of the overall variability in the shape model were used to assess the distribution of cell shapes in a principle component analysis ( PCA ) plot . Additionally , a kernel density estimate of the marginal was used to plot the area of focal adhesions . FISH eGFP mRNA probes ( conjugated to Quasar 670 dye ) were purchased from LGC Biosearch Technologies ( VSMF 1015–5 ) and FISH was carried out as per manufacturers’ protocol . Briefly , cells were seeded onto coverslips in six-well plates at 20 , 000 cells/well overnight and fixed in 4% ( w/v ) PFA at room temperature for 30 min followed by treatment with 70% alcohol at 4°C for 1 hr . Cells were incubated with 125 nM probes at 37°C overnight . Cells were stained with 4′ , 6-diamidino-2-phenylindole ( 5 ng/ml ) and mounted using Prolong Diamond ( Life Technologies ) . Images were acquired using Leica DM6000 at X40 . Z-stacks were acquired , and blind deconvolution was carried out using Leica LAS X software . Cells were seeded onto 10-cm culture dishes and grown to confluence . RNA was isolated using RNeasy mini kit ( Qiagen ) and cDNA was synthesized using oligo dT primers with a first strand cDNA synthesis kit ( Applied Biosystems ) . After standard curves were obtained , quantitative PCR ( qPCR ) was carried out using SybrGreen ( Applied Biosystems ) and the following primer sets . PCR was carried out on QuantStudio Flex 6 Real Time PCR system ( Applied Biosystems ) . ΔΔCt method was used to estimate the relative expression levels of mRNA using Tbp as the reference transcript . Computational simulations of actin bundle growth from focal adhesions to predict the bundle length at different β- and γ-actin local supply rates were performed using a recently developed software MEDYAN ( Popov et al . , 2016 ) . In brief , MEDYAN simulates actin networks by integrating the stochastic diffusion-reaction dynamics and mechanical relaxation of the cytoskeletal network . Diffusing molecular species , including actin monomers , unbound myosin motors , and unbound crosslinkers , are contained in a solution phase . Stochastic chemical reactions such as actin ( de ) polymerization and ( un ) binding of motors and linkers follow mass-action kinetics , and change the mechanical energy of the actin network . The net forces are then periodically relaxed using conjugate-gradient mechanical equilibration . This step also updates reaction rates of motor walking , motor unbinding , and linker unbinding , based on residue tension after minimization . We used a 1×1×4 μm3 simulation box , containing non-muscle myosin II motors , alpha-actinin crosslinkers , actin monomers , and actin filaments growing from the bottom focal adhesion region . The focal adhesion region was presented as a hemisphere with 30 actin filaments attached . We tested and found that the actin filaments never grow longer than 4 μm in the z-direction , and thus , no length constraints on the actin bundles factored into the simulations . Actin filaments were only allowed to elongate at one end ( the barbed end ) , while the elongation rate constant was averaged over filament polymerization rates and depolymerization rates of both the barbed end and the pointed end . The filament elongation was driven by the addition of actin monomers to the system , simulating the synthesis of actin monomers near the focal adhesion region . Multiple actin supply rates were tested at 50% increments based on the experimental measurements of the differences between β- and γ-actin synthesis rates . The simulations were run for 10 min to match the timescale of the experiments . The starting concentration of actin at the attachment site was assumed to be ~2 μM locally , creating an initial bundle at around 0 . 1 μm long . In the simulation , the majority ( more than 90% ) of actin for the filament growth was assumed to arise from the de novo subunit addition . The concentrations of myosin mini-filaments ( 0 . 012–0 . 021 μM ) and alpha-actinin crosslinkers ( 1 . 25 μM ) were chosen to ensure proper bundling of filaments ( all model parameters are listed in Table 1 ) . To determine the actin bundle length , we measured the F-actin distribution along the Z-axis and defined the actin bundle length as the width of central 80% of the F-actin distribution ( Figure 5—figure supplement 1A ) . Although the length measured in simulations was much shorter than that in the experiments , the beta-actin bundles were ~50–80% longer than gamma-actin bundles , in agreement with the experimental measurements .	Most mammalian cells make both β- and γ-actin , two proteins which shape the cell’s internal skeleton and its ability to migrate . The molecules share over 99% of their sequence , yet they play distinct roles . In fact , deleting the β-actin gene in mice causes death in the womb , while the animals can survive with comparatively milder issues without their γ-actin gene . How two similar proteins can have such different biological roles is a long-standing mystery . A closer look could hold some clues: β- and γ-actin may contain the same blocks ( or amino acids ) , but the genetic sequences that encode these proteins differ by about 13% . This is because different units of genetic information – known as synonymous codons – can encode the same amino acid . These ‘silent substitutions’ have no effect on the sequence of the proteins , yet a cell reads synonymous codons ( and therefore produces proteins ) at different speeds . To find out the impact of silent substitutions , Vedula et al . swapped the codons for the two proteins , forcing mouse cells to produce β-actin using γ-actin codons , and vice versa . Cells with non-manipulated γ-actin and those with β-actin made using γ-actin codons could move much faster than cells with β-actin . This suggested that silent substitutions were indeed affecting the role of the protein . Vedula et al . found that cells read γ-codons – and therefore made γ-actin – much more slowly than β-codons: this also affected how quickly the protein could be dispatched where it was needed in the cell . Slower production meant that bundles of γ-actin were shorter , which allowed cells to move faster by providing a weaker anchoring system . Overall , this work provides new links between silent substitutions and protein behavior , a relatively new research area which is likely to shed light on other protein families .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2021	Different translation dynamics of β- and γ-actin regulates cell migration
Neuropathic pain following peripheral nerve injury is associated with hyperexcitability in damaged myelinated sensory axons , which begins to normalise over time . We investigated the composition and distribution of shaker-type-potassium channels ( Kv1 channels ) within the nodal complex of myelinated axons following injury . At the neuroma that forms after damage , expression of Kv1 . 1 and 1 . 2 ( normally localised to the juxtaparanode ) was markedly decreased . In contrast Kv1 . 4 and 1 . 6 , which were hardly detectable in the naïve state , showed increased expression within juxtaparanodes and paranodes following injury , both in rats and humans . Within the dorsal root ( a site remote from injury ) we noted a redistribution of Kv1-channels towards the paranode . Blockade of Kv1 channels with α-DTX after injury reinstated hyperexcitability of A-fibre axons and enhanced mechanosensitivity . Changes in the molecular composition and distribution of axonal Kv1 channels , therefore represents a protective mechanism to suppress the hyperexcitability of myelinated sensory axons that follows nerve injury . Following traumatic nerve injury spontaneous activity develops initially in myelinated and subsequently in unmyelinated sensory axons ( Wall and Gutnick , 1974; Kajander and Bennett , 1992; Boucher et al . , 2000; Michaelis et al . , 2000; Wu et al . , 2001; Liu et al . , 2000a ) . The onset of this spontaneous activity is associated with the emergence of pain-related sensory changes in animal models and is critical for the maintenance of peripheral neuropathic pain ( Haroutounian et al . , 2014 ) in patients where selective blockade suggests the involvement of myelinated axons ( Campbell et al . , 1988 ) . Ectopic activity is particularly prominent in myelinated afferents and peaks within the first few days post injury and then declines over subsequent weeks ( Kajander and Bennett , 1992; Liu et al . , 2000a; 2000b; Han et al . , 2000 ) . Such ectopic activity arises both at the neuroma site and also at the level of the dorsal root ganglion ( DRG ) ( Han et al . , 2000; Liu et al . , 2000b; Amir et al . , 1999; 2005; Wall and Devor , 1983 ) . Altered expression , function and trafficking of voltage-gated ion channels are key determinants of these excitability changes . Shaker type voltage-gated potassium channels ( Kv1 channels ) are important determinants of neuronal excitability . They are formed by heteromultimers of α and β subunits ( MacKinnon , 1991 ) . The characteristics of the outward currents they carry depend on subunit composition . Sensory neurons are known to express Kv1 channels and functionally these channels have been shown to limit excitability of sensory neurons: For instance Kv1 . 2 suppresses excitability at the level of the sensory neuron cell body ( Gold et al . , 1996; Rasband et al . , 2001; Zhao et al . , 2013; Everill et al . , 1998 ) and Kv1 . 1 acts as a ‘brake’ on mechanosensitivity at the terminals of C-mechano-nociceptor and Aβ-mechanoreceptors ( Hao et al . , 2013 ) . Kv1 channels also act as excitability brakes for cold thermal sensitivity in intact and damaged axons of primary sensory neurons ( many of such fibres are also mechano-sensitive ) ( Roza et al . , 2006; Madrid et al . , 2009 ) . Kv1 channels are known to be expressed in the juxtaparanodal region of myelinated sensory axons . An unexplored issue , however , is whether the distribution of these channels changes under pathological neuropathic states . Saltatory conduction in myelinated fibres depends on the molecular organization of channel domains within the axon ( Chang and Rasband , 2013 ) : voltage-gated sodium channels ( Nav ) are clustered at the node of Ranvier . Nodes are flanked by the paranode , which is an important point of attachment between the axon and the terminal loops of the Schwann cell . Just inside the innermost axoglial junction of the paranode is the juxtaparanode a domain enriched in Kv1 channels Kv1 . 1 and 1 . 2 . The localisation of Kv1 . 1 and 1 . 2 to the juxtaparanode is dependent on the formation of a molecular scaffold , which includes the adhesion molecules caspr2 and TAG-1 ( Poliak et al . , 2003 ) . In the naïve state in adulthood , the juxtaparanodal Kv1 channels are thought not to have a major influence on axon conduction properties of peripheral myelinated axons ( Poliak et al . , 2003; Chiu and Ritchie , 1980; Sherratt et al . , 1980; Rasband et al . , 1998 ) , probably because they are electrically insulated from the node of Ranvier under the myelin sheath . However during development ( Vabnick et al . , 1999 ) and following primary demyelination ( Rasband et al . , 1998 ) ( during which myelin is removed but the axon remains intact ) , Kv1 . 1 and 1 . 2 become more widely distributed to include the paranode and even the node ( Arroyo et al . , 2004 ) , and can act to suppress excitability . Although Kv1 . 1 and 1 . 2 expression within the soma is known to be down-regulated following axon transection , and this leads to hyperexcitability at the soma ( Rasband et al . , 1998; Ishikawa et al . , 1999; Park et al . , 2003 ) , the distribution of these channels at the nodal complex and damaged nerve terminal ( in the neuroma that forms ) has not been examined . Furthermore , little is known regarding the distribution of other members of the shaker type Kv1 channels family such as Kv1 . 4 and 1 . 6 following nerve injury . Here we show that within a neuroma , expression levels of Kv1 . 1 and 1 . 2 are markedly reduced but over time Kv1 . 4 and 1 . 6 expression increases within juxtaparanodes and paranodes . At sites remote from injury , there is also a gradual redistribution of Kv1 channels to the paranode . Electrophysiological and behavioural experiments suggest that changes in subunit expression and redistribution of Kv1 channels act a ‘brake’ on the hyperexcitable state that arises in myelinated axons following traumatic nerve injury . To investigate the role of Kv1 channels in hypersensitivity after nerve injury we used a model of complete sciatic nerve transection followed by positioning of the proximal stump superficially under the skin of the leg [modified version of Dorsi et al . ( 2008 ) ] . This model enables us to study both the expression of Kv1 channels within the neuroma and undertake behavioural analysis using specific blockers of Kv1 channels . To study how the localisation of Kv1 channels changes within the nodal complex , we used a pan-Nav channel antibody to label the node of Ranvier , a Caspr antibody to label the paranode and Kv1 . 2 and Caspr2 antibodies to label the juxtaparanode . In the naïve axon , we observed that over 90% ( mean 91 ± SEM 2 . 9% , n = 4 animals ) of the nodes presented a characteristic morphology with Nav clustering in the centre , surrounded by caspr at both sides and Kv1 . 2 clustered within the juxtaparanode . Of other members of the Kv1 channels family , Kv1 . 4 and 1 . 6 are expressed by DRG cells ( Everill et al . , 1998; Thakur et al . , 2014; Chiu et al . , 2014 ) . We found that in the naïve state Kv1 . 4 was only expressed within a very small proportion of nodes ( 5 . 5 ± 4 . 6% , n = 4 animals ) within the juxtaparanode and Kv1 . 6 was not present within the nodal complex ( Figure 1 ) . 10 . 7554/eLife . 12661 . 003Figure 1 . Kv1 channels expression in the naïve nerve and 21 days after sciatic nerve axotomy ( note that the samples were collected from the neuroma site ) . ( a ) Representative images of longitudinal nerve sections immunostained with Kv1 channels in green ( Kv1 . 2 , Kv1 . 4 and Kv1 . 6 respectively ) , a panNav antibody in red ( to identify the node ) , and caspr in blue ( to identify the paranode ) . Kv1 . 2 is expressed in the juxtaparanode in naïve nerves but it is not present at 21 days after nerve injury . Kv1 . 4 and kv1 . 6 are not present in uninjured nerve but are expressed after nerve injury . Note that when Kv1 . 4 and Kv1 . 6 are expressed , they are not confined to the juxtaparanode only but invade the paranode . ( b ) Western blots showing expression of Kv1 channels in the naïve nerve , at 7 and 21 days after axotomy . Kv1 . 1 and Kv1 . 2 are expressed in the naïve nerve and down-regulated after axotomy , while Kv1 . 4 and Kv1 . 6 have a low/null expression in the naïve nerve and are up-regulated after injury ( p<0 . 05 , p<0 . 001 , one Way ANOVA , n=6 per group ) . Scale bars = 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12661 . 00310 . 7554/eLife . 12661 . 004Figure 1—source data 1 . Source data for Figure 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 12661 . 004 At the site of the neuroma ( day 7 and 21 ) , only half of the nodes demonstrated this typical morphology ( day 7 = 47 . 5 ± 5 . 1% , day 21 = 46 ± 6 . % , n = 4 animals per group , 36–64 nodes per animal ) ; the rest were split ( day 7 = 23 . 3 ± 5 . 9% , day 21 = 13 . 5 ± 2 . 7% ) , presented as heminodes ( caspr at one side only , day 7 = 21 . 8 ± 4 . 4% , day 21 = 28 . 9 ± 5 . 5% ) , or were ‘naked’ ( Nav clusters alone , with no caspr , day 7 = 7 . 2 ± 3% , day 21 = 11 . 4 ± 3 . 2% , Figure 2 ) . This is in accordance with previous literature examining the localisation of voltage-gated sodium channels ( Henry et al . , 2006; Thakur et al . , 2014 ) . At day 7 after injury , Kv1 . 2 channels were not located strictly not only in the juxtaparanodal regions but also overlapped with paranodal proteins . To objectively measure this , we quantified the distance between the Nav channels staining and the distal end of the caspr staining , and the distance between the Nav channels staining and the proximal end of Kv1 channels . The difference between these two distances was indicative of the level of overlap between Kv1 channels and paranodal proteins ( note that ‘naked’ nodes were not included in the analysis of the spatial distribution of Kv1 channels because by definition these only consist of Nav clusters without paranodal and juxtaparanodal proteins ) . In naïve axons the distance between Nav channels staining and the end of caspr was 3 . 8 ± 0 . 2 μm , and the distance between Nav channels staining and the start of Kv1 . 2 staining was 4 . 2 ± 0 . 2 μm , resulting in a relatively small , albeit positive , difference between these distances ( 0 . 5 ± 0 . 08 μm , n = 4 animals , 25–40 nodes per animal ) , indicating there was no overlap . At day 7 after injury , the distance between Nav channels staining and the end of caspr was 4 ± 0 . 1 μm , and the distance between Nav channels staining and the start of Kv1 . 2 staining was 3 . 2 ± 0 . 2 μm , giving a negative value for the difference between both distances ( -0 . 8 ± 0 . 1 μm , n = 3 animals , 32–35 nodes per animal ) , which indicates that the Kv1 channels were co-localised with caspr staining and moving closer to the node ( Figure 3 ) . Note that the distance between Nav channels staining and the end of caspr staining remained unchanged after injury , while the distance between Nav channels staining and the start of Kv1 . 2 staining was significantly reduced . Contactin-associated protein-like 2 ( Caspr2 ) forms a complex with Kv1 channels at the juxtaparanode ( Chiu et al . , 2014 ) . We evaluated if caspr2 moves closer to the node together with Kv1 channels . We measured the distance between the Nav channels staining and the distal end of the caspr staining , and the distance between the Nav channels staining and the proximal end of caspr2 . In naïve axons , the distance between Nav channels staining and the end of caspr was 3 . 8 ± 0 . 3 μm , and the distance between Nav channels staining and the start of caspr2 staining was 4 . 3 ± 0 . 0 μm , resulting in a small difference between these distances ( 0 . 49 ± 0 . 09 μm , n = 4 animals , 25–30 nodes per animal ) , indicating there was no overlap . At day 7 after injury , the distance between Nav channels staining and the end of caspr was 4 . 1 ± 0 . 2 μm , and the distance between Nav channels staining and the start of caspr2 staining was 2 . 8 ± 0 . 2 μm , giving a negative value for the difference between both distances ( -1 . 2 ± 0 . 1 μm , n = 3 animals , 30 nodes per animal ) , which indicates that the caspr2 co-localised with caspr staining and had moved closer to the node together with Kv1 channels ( Figure 3 ) . 10 . 7554/eLife . 12661 . 005Figure 2 . Nav channel expression . Representative sections of longitudinal nerves immunostained with Kv1 . 2 in green , a panNav antibody in red ( to identify the node ) , and caspr in blue ( to identify the paranode ) from neuroma day 21 . ( a ) A typical pattern of Nav expression localized at the node of Ranvier and flanked by caspr staining is shown . The altered forms of Nav channel accumulations seen in the injured nerve included ( b ) split nodes: These were nodes that had two distinct Nav channels accumulations , separated by a gap in the Nav channels staining within the same fibre and with each Nav channels accumulation flanked on one side with caspr staining , or ( c ) naked nodes: those Nav channel accumulations that lacked an association with caspr ( d ) heminodes: nodes where the caspr staining was located on only one side of a contiguous Nav channel accumulation . ( e ) Quantification of different types of sodium cluster accumulation in the naïve state and after nerve injury is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 12661 . 00510 . 7554/eLife . 12661 . 006Figure 2—source data 1 . Source data for Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 12661 . 00610 . 7554/eLife . 12661 . 007Figure 3 . Relocalization ok Kv1 . 2 and caspr2 at 7 days after neuroma . ( a ) Representative longitudinal sections of nerves immuno-stained with Kv1 . 2 in green , a panNav antibody in red ( to identify the node ) and caspr in blue ( to identify the paranode ) . Kv1 . 2 is expressed in the juxtaparanode in naïve nerves but it also co-localized with caspr staining at 7 days after injury . Representative longitudinal sections of nerves immuno-stained with caspr2 in green and caspr in red . Caspr2 is confined to the juxtaparanode in naïve nerve but co-localized with caspr at 7 days after injury . ( b ) We quantified the distance between the sodium channel staining ( Nav ) and the end of the caspr staining , distance between the sodium channel staining ( Nav ) and the start of the Kv1 . 2/1caspr2 staining , and difference between these distances . A negative value represents an overlap of paranodal and juxtaparanodal proteins . Note that he distance between the sodium channel staining ( Nav ) and the end of the caspr staining remains unchanged after nerve injury , while the distance between the sodium channel staining ( Nav ) and the start of the Kv1 . 2/caspr2 staining is significantly shortened after nerve injury , indicating co-localization of Kv1 . 2 and caspr2 with caspr ( n = 5 animals , 20–41 nodes per animal ) , p<0 . 001 , one way ANOVA Tukey post hoc tests ) . We analyzed uninjured ( naïve ) nerve , nerve at the site of the neuroma ( day 7 ) , and nerve 1 cm proximal to the neuroma ( day 7 ) . The effect on Kv1 . 2 re-localization remains the same at the site far from the neuroma . The effect on caspr2 re-localization is slightly smaller at the site 1 cm proximal to the neuroma compared with the neuroma site , but it is still significantly different from the naïve p<0 . 001 , p<0 . 05 , PRN = paranode , JXP = juxtaparanode . Scale bars = 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12661 . 00710 . 7554/eLife . 12661 . 008Figure 3—source data 1 . Source data for Figure 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 12661 . 008 The redistribution of the channels seen could simply be a reflection of direct injury at the site of axotomy . We therefore , studied a site proximal to the neuroma ( 1 cm ) and compared it to the neuroma site . The effect on Kv12 re-localization remains the same at the site far from the neuroma: The difference between Nav channels staining and the end of caspr distance and Nav channels staining and the start of Kv1 . 2 distance was -1 ± 0 . 09 μm , at the site close to the neuroma and -1 . 1 ± 0 . 09 μm , at the site far from the neuroma ( p = 0 . 6 , n = 5 animals , 24–26 nodes per animal ) . The effect on caspr2 re-localization is slightly smaller at the site 1cm proximal to the neuroma compared with the site close to the neuroma , but it is still significantly different from the naïve ( p<0 . 001 ) : The difference between Nav channels staining and the end of caspr distance and Nav channels staining and the start of caspr2 distance was -1 ± 0 . 1 μm , at the site close to the neuroma and -0 . 5 ± 0 . 1 μm , at the site far from the neuroma ( p = 0 . 06 , n = 5 animals , 21–26 nodes per animal , Figure 3c ) . This change on juxtaparanode proteins localisation could be due to a disorganisation of the paranodal axo-glial junctions ( paranode loops ) . Therefore , we examined their ultrastructural anatomy using electron-microscopy and measured the distance between the axon and the paranodal loops ( axo-glial distance ) , the number of detached and everted loops and the minimal and maximal distance between loops . We analysed sciatic nerves from sham-operated and neuroma animals and we observed very few detached or everted loops , with no differences between groups . The close apposition between the axon and the paranodal loop was unchanged as the axo-glial distance was not significantly changed ( sham = 9 . 7 ± 0 . 4 nm , neuroma = 11 . 4 ± 0 . 7 nm , n = 4–5 animals per group , 6–9 nodes per animal one way ANOVA p = 0 . 08 ) . We observed a small but significant increase in the maximal distance between loops ( sham = 12 . 5 ± 1 nm , neuroma = 18 . 9 ± 1 . 4 nm , one way ANOVA , p = 0 . 005 ) . In summary , these results suggest that although there was no major disruption of the septate axoglial junctions there was a small but significant increased separation between the paranodal loops ( Figure 4 ) . Note that the axonal diameter at the node did not change ( Figure 4d ) . 10 . 7554/eLife . 12661 . 009Figure 4 . Ultrastructural anatomy of the node of Ranvier within the sciatic nerve following axotomy . We used electron microscopy to look at the ultrastructure anatomy of the node . ( a ) Shows a diagram of the node , paranode and juxtaparanode and a low magnification section of this area in a sham-operated nerve . The red box denotes the area that was used for quantification as seen in b . ( b ) High magnification views of the paranodal loops are shown in the sham and 21 days following axotomy ( magnification 135 000x ) ( c ) We quantified different aspects of the attachment of the Schwann cell paranodal loops to the axon . This is illustrated in right panel which denotes the different parameters measured: The maximal and minimal distance between interloops , the distance between the glia and the axon , the number of detached loops and the number of everted loops . We found a significant increase in the maximal distance between loops in the neuroma compared to sham nerves ( one way ANOVA , p = 0 . 005 ) . There were no significant differences in any of the other measurements . ( d ) We quantified the diameter of the axons at the site of the node and found no difference between the uninjured and injured axons . Scare bars: 200 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 12661 . 00910 . 7554/eLife . 12661 . 010Figure 4—source data 1 . Source data for Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 12661 . 010 βII spectrin is a cytoskeletal protein that has recently been shown to be essential for the localization of Kv1 channels to the juxtaparanode and is proposed to form a sub-membranous barrier to lateral diffusion of Kv1 channels into the paranode ( Zhang et al . , 2013 ) . Using IHC , we looked at βII spectrin in the neuroma and found that it is expressed at the paranode and juxtaparanode . We quantified the staining at the paranodal domain and found that the expression of this protein was reduced by more than half compared to naive ( immunofluorescence normalised to naïve: 0 . 4 ± 0 . 02 , p<0 . 001 , t-test , n = 50–83 heminodes , a–b ) . βII spectrin is also expressed in the sub-membranous regions of Schwann cells where it does not appear to change with nerve injury . To further quantify the change of expression of this protein in the neuron , we performed Western blotting in the soma of sensory neurons ( dorsal root ganglia- DRG ) and observed a 30% decrease following nerve injury compared to naive ( expression relative to naïve: 0 . 7 ± 0 . 08 , p = 0 . 04 , t-test , n = 4 , Figure 5c–d ) . 10 . 7554/eLife . 12661 . 011Figure 5 . βII spectrin expression in naïve and neuroma nerves . ( a ) Representative sections of longitudinal nerves immunostained with βII spectrin in green , a panNav antibody in red ( to identify the node ) , and caspr in blue ( to identify the paranode ) . βII spectrin is expressed both in the surface of Schwann cells ( arrows ) and in the axon at the paranodal and juxtaparanodal region ( arrow heads ) in naïve nerves . At 21 days after axotomy ( neuroma ) , βII spectrin can be only seen in the Schwann cell ( arrows ) but not in the axonal domains . ( b ) Quantification of βII spectrin immunofluorescence in the paranode ( identified by caspr staining ) showing a significant reduction in neuroma versus naïve ( immunofluorescence normalised to naïve: 0 . 4 ± 0 . 02 , p<0 . 001 , t-test , n = 50–83 heminodes ) . ( c ) Western blots showing expression of βII spectrin in the DRG of naïve and day 21 neuroma . ( d ) Quantification of WBs . Expression of βII spectrin in the DRG was reduced by 30% after nerve injury . PGP9 . 5 was used as a loading control ( expression relative to naïve: 0 . 7 ± 0 . 08 , p = 0 . 04 , t-test ) . *p<0 . 001 , p<0 . 05 . Error bars denote SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 12661 . 01110 . 7554/eLife . 12661 . 012Figure 5—source data 1 . Source data for Figure 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 12661 . 012 At day 21 after injury , in marked contrast with the naïve axons , very few of the nodes at the neuroma site ( day 21 ) showed Kv1 . 2 immunostaining ( 8 . 3 ± 0 . 8% in neuroma vs . 86 . 1 ± 4 . 4% in naive , n = 4 animals per group , 25 nodes per animal p<0 . 001 t-test ) . Conversely , most of the nodes at the neuroma site ( day 21 ) showed intense Kv1 . 4 immunostaining ( 73 . 3 ± 12% in injured vs . 5 . 5 ± 4 in naive , n = 4 animals per group , 30 nodes per animal p<0 . 001 t-test ) and Kv1 . 6 immunostaining ( 66 . 6 ± 14% vs . in injured vs . none in naive , n= 4 animals per group , 30 nodes per animal p<0 . 001 t-test ) ( Figure 1 ) . ( Note that naïve nerves were comparable in terms of quantification to nerves from sham-operated animals [p = 0 . 48] ) . We used Western blotting to quantify the expression of the different α subunits and found that Kv1 . 1 and Kv1 . 2 expression were significantly reduced at days 7 and 21 following nerve injury , while Kv1 . 4 and Kv1 . 6 were significantly upregulated at the neuroma site ( Figure 1 ) . We looked at Kv1 channel expression in the DRG using IHC and we observed that Kv1 . 2 expression is reduced after injury , while Kv1 . 4 and Kv1 . 6 expression remains unchanged ( Figure 6a ) ( this is at the DRG soma , although expression could be seen to increase within paranodes/juxtaparanodes after injury ) . We also quantified protein expression within the DRG over the same time course using Western blot analysis . The expression of Kv1 . 2 was significantly reduced following nerve injury ( Figure 6b–e ) consistent with previous findings , ( Everill et al . , 1998; Ishikawa et al . , 1999; Kim et al . , 2002; Yang et al . , 2004 ) , and there was a trend for a reduction in Kv1 . 1 although this did not reach significance . The expression of Kv1 . 4 and 1 . 6 within the DRG did not significantly change following injury ( one Way ANOVA , n = 6 per group ) suggesting that increased expression within the juxtaparanode and paranode at the neuroma site is a likely consequence of altered trafficking of these proteins rather than global changes in expression . 10 . 7554/eLife . 12661 . 013Figure 6 . Expression of Kv1 channels in the DRG in the naïve state , 7 and 21 days after axotomy ( neuroma ) . ( a ) Representative sections of naive and neuroma day 21 DRG immunostained with Kv1 . 2 , Kv1 . 4 , and Kv1 . 6 . Note that Kv1 . 2 expression in DRG cells and axonal juxtaparanodes ( arrow heads ) is reduced after injury , while Kv1 . 4 and Kv1 . 6 expression remains unchanged in DRG cells , and it is present in axonal juxtaparanodes ( arrow heads ) after injury . In each panel ( b–e ) , a representative blot is shown for each time point with GAPDH as a loading control . Quantification of 6 animals per condition is shown below ( b , d , e ) Expression of Kv1 . 1 , Kv1 . 4 and Kv1 . 6 within the DRG does not significantly change after axotomy . ( c ) Kv1 . 2 expression is significantly decreased after axotomy . ( p<0 . 05 , one Way ANOVA , n = 6 per group ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12661 . 01310 . 7554/eLife . 12661 . 014Figure 6—source data 1 . Source data for Figure 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 12661 . 014 We next looked into human nerve tissue to see if these changes were relevant to patients with neuropathic pain We collected 6 control samples ( from subjects having their sural nerves removed to use as a bridge for hand reconstructive surgery ) and 6 samples obtained from patients undergoing removal of Morton’s neuroma ( interdigital nerve entrapment neuropathy ) . IHC ( n = 3 per group , 8–10 nodes per patient ) showed that only Kv1 . 2 is expressed in the juxtaparanode of healthy subjects ( 90 ± 10% of nodes were Kv1 . 2 positive ) with absent Kv1 . 4 and 1 . 6 staining as observed in the rat . However , in neuroma Kv1 . 2 expression in the juxtaparanode was minimal ( 13 . 3 ± 8 . 1% ) whilst Kv1 . 4 and Kv1 . 6 were expressed in most of the nodes ( 92 . 5 ± 7 . 4% for Kv1 . 4; 73 . 5 ± 8 . 8 for Kv1 . 6 ) ( Figure 7a–b ) . We used western blotting to quantify Kv1 channels proteins in the nerves of the patients ( n = 6 per group ) and found that Kv1 . 2 expression was significantly decreased in neuroma compared to control nerve ( to 0 . 48 ± 0 . 1 of the control , Mann-Whitney U-Test , p 0 . 005 ) . In contrast , expression of Kv1 . 4 and Kv1 . 6 were significantly increased in neuroma compared to controls ( to 6 . 3 ± 3 . 5 , and 9 . 4 ± 6 . 6 of the control respectively , Mann-Whitney U-Test p = 0 . 005 both ) ( Figure 7c–d ) . 10 . 7554/eLife . 12661 . 015Figure 7 . Kv1 channels expression in the sural nerve of healthy volunteers ( control ) and from patients with painful Morton neuroma . ( a ) Representative sections of longitudinal nerves immunostained with Kv1 channels in green ( Kv1 . 2 , Kv1 . 4 and Kv1 . 6 respectively ) , a panNav antibody in red ( to identify the node ) , and caspr in blue ( to identify the paranode ) . Kv1 . 2 is expressed in the juxtaparanode in control nerves but it is not present in the injured nerve . Kv1 . 4 and kv1 . 6 are not present in control nerve but are expressed in neuroma within the juxtaparanode and encroaching on the paranode nodes . ( b ) Quantification of the percentage of Kv1 . 2 , Kv1 . 4 , and Kv1 . 6 positive nodes in control and neuroma nodes ( n = 3 per group , one way ANOVA ) . ( c ) Western blots showing expression of Kv1 channels in control and neuroma nerve . ( d ) Quantification of WBs ( n = 6 per group , one way ANOVA ) . Kv1 . 2 is expressed in the control nerve and down regulated after axotomy , while Kv1 . 4 and Kv1 . 6 have a low/null expression in the control nerve and are up-regulated in neuroma . PGP9 . 5 was used as a loading control . Error bars denote ( e ) Patients and control subjects demographic data . The female/male ratio is different in patients with Morton neuroma and controls ( patients with traumatic lesion of the hand ) reflecting the F/M ratio of these different conditions . The mean age of patients with Morton neuroma is slightly higher than in control subjects , although it is not significant ( t-test ) . All patients with Morton neuroma presented with pain ( mean NRS 4 . 6 ) , while control presented no pain in the area innervated by the sural nerve ( Mann Whitney test ) . NRS: numerical rate score . Error bars denote SEM . Scale bars = 5 μm , p<0 . 001 , p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 12661 . 01510 . 7554/eLife . 12661 . 016Figure 7—source data 1 . Source data for Figure 7 . DOI: http://dx . doi . org/10 . 7554/eLife . 12661 . 016 We investigated the localisation of Kv1 channels at a site distant from the injury site . To do so , we used a model of L5 spinal nerve transection ( SNT ) ( Kim and Chung , 1992 ) and studied the dorsal roots ( ie . proximal to the DRG ) . We used this model instead of the neuroma model to have certainty that all the dorsal root axons studied had their peripheral terminals injured . In the dorsal roots from naïve animals , the distance between Nav channels staining and the end of caspr was 3 . 5 ± 0 . 2 μm , and the distance between Nav channels staining and the start of Kv1 . 2 staining was 4 ± 0 . 2 μm , resulting in a small positive difference between these distances ( 0 . 5 ± 0 . 1 μm , n = 4 animals , 25–32 nodes per animal ) , indicating there was no overlap . Seven days after transection of L5 spinal nerve this distance was still positive ( 0 . 24 ± 0 . 05 μm , n = 4 animals; distance Nav-end caspr 3 . 7 ± 0 . 2 μm , distance Nav-start Kv1 . 2 3 . 9 ± 0 . 1 μm; 30–40 nodes per animal ) . However , at 21 days after nerve injury , we noted a significant overlap between the end of caspr staining and the start of the Kv1 . 2 staining ( -1 . 4 ± 0 . 3 μm , n = 4 animals , 38–40 nodes per animal , one way ANOVA , p<0 . 001; distance between Nav-end caspr 3 . 9 ± 0 . 4 μm , distance between Nav-start Kv1 . 2 2 . 5 ± 0 . 4 μm ) . We observed this novel localisation of Kv1 . 2 in the paranode which ( in contrast to the neuroma site ) is still clearly present after nerve injury within the dorsal root . We also observed novel expression of Kv1 . 4 and 1 . 6 in the dorsal root of injured animals , and these were localised to the paranode in addition to the juxtaparanode ( the distance between end of caspr and start of Kv1 . 4 and 1 . 6 staining was -1 . 7 ± 0 . 6 μm and -1 . 07 ± 0 . 2 μm respectively; for Kv1 . 4 distance between Nav-end caspr 3 . 6 ± 0 . 4 μm , distance between Nav-start Kv1 . 4 1 . 9 ± 0 . 6 μm; for Kv1 . 6: distance between Nav-end caspr 3 . 4 ± 0 . 4 μm , distance between Nav-start Kv1 . 6 1 . 6 ± 0 . 4 μm n = 4 animals per group , 30–35 nodes per animal Figure 8 ) . 10 . 7554/eLife . 12661 . 017Figure 8 . Kv1 channels expression in the dorsal roots of naïve animals and 21 days after spinal nerve transection ( SNT ) . ( a ) Representative sections of longitudinal dorsal roots immunostained with Kv1 channels in green ( Kv1 . 2 , Kv1 . 4 and Kv1 . 6 , respectively ) , a panNav antibody in red ( to identify the node ) , and caspr in blue ( to identify the paranode ) . Kv1 . 2 is expressed only in the juxtaparanode in naïve nerves but after injury it invades the paranode . Kv1 . 4 and kv1 . 6 are not present in uninjured nerve but are expressed within the juxtaparanode after nerve injury and also invade the paranode . ( b ) Re-localization of caspr2 at 21 days after spinal nerve transection ( SNT ) . Representative sections of longitudinal L5 dorsal roots immuno-stained with caspr2 in red , Kv1 . 2 in green , and caspr in blue . Kv1 . 2 and caspr2 are expressed in the juxtaparanode in naïve nerves but co-localized with caspr staining at 21 days after injury . ( c ) Quantification of: distance between the sodium channel staining ( Nav ) and the end of the caspr staining , distance between the sodium channel staining ( Nav ) and the start of the Kv1 . 2/1 . 4/1 . 6/caspr2 staining , and difference between these distances . A negative value in this difference represents an overlap of paranodal and juxtaparanodal proteins . Note that the distance between the sodium channel staining ( Nav ) and the end of the caspr staining remains unchanged after nerve injury , while the distance between the sodium channel staining ( Nav ) and the start of the Kv1 . 2/1 . 4/1 . 6/caspr2 staining is significantly shortened after nerve injury . Kv1 . 4 and Kv1 . 6 were absent in naive ( n = 5 animals/4 sections per animal , p<0 . 05 , p<0 . 001 ) . Scale bars = 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12661 . 01710 . 7554/eLife . 12661 . 018Figure 8—source data 1 . Source data for Figure 8 . DOI: http://dx . doi . org/10 . 7554/eLife . 12661 . 018 Contactin-associated protein-like 2 ( Caspr2 ) is normally localized at the juxtaparanode and associates with K+ channels ( Chiu et al . , 2014 ) . Interestingly , we observed that caspr2 is mobilized into the paranode regions in a similar way to Kv1 channels; the difference between the Nav-caspr staining distance and the Nav-caspr2 distance is minimal in naïve roots ( 0 . 5 ± 0 . 1 μm; distance between Nav-end caspr 3 . 8 ± 0 . 3 μm , distance between Nav-start caspr2 4 . 3 ± 0 . 4 μm , 30–38 nodes per animal 4 animals ) , and after SNT , this distance becomes negative ( -0 . 88 ± 0 . 1 μm , n = 5 animals; distance between Nav-end caspr 3 . 4 ± 0 . 3 μm , distance between Nav-start caspr2 2 . 5 ± 0 . 3 μm 35–40 nodes per animal p<0 . 001 , t-test ) indicating an overlap between caspr and caspr2 immuno-labeling ( Figure 8 ) . Spontaneous activity in naïve axons was present in less than 5% of A-fibres ( 4 . 5% ) . We assessed the effect of blocking the Kv1 channels using α-DTX , a toxin isolated from black and green mamba snakes which is a selective and effective blocker of Kv1-containing oligomers composed of Kv1 . 1 , Kv1 . 2 , or Kv1 . 6 subunits ( Harvey , 2001 ) . We applied the toxin at the neuroma site ( or , in control animals , acutely cut sciatic nerve stump ) and to the L5 DRG and recorded from sensory axons in thin strands dissected from the dorsal root . In the naïve situation ( n = 222 neurons , 10 animals , Figure 9a ) , the incidence of spontaneous activity did not significantly change after α-DTX ( 5 . 4 and 9 . 8% of myelinated afferents when the toxin was applied to nerve stump or L5 DRG , respectively ) . Two days after transecting the sciatic nerve ( n = 291 neurons , 4 animals ) , spontaneous activity at the injured nerve increased to 22% of myelinated afferents , and it was similar with or without toxin ( toxin applied to neuroma 26 . 1% , toxin applied to the L5 DRG 26% ) . However , at 7 days after nerve injury ( n = 241 neurons , 7 animals ) the proportion of spontaneously active afferents had decreased to 6 . 2% , and application of the toxin now induced a significant increase in proportion of spontaneously firing myelinated afferents ( 11 . 2% toxin at the neuroma p = 0 . 07 , and 15 . 6% toxin to L5 DRG , p = 0 . 002 , chi-square test ) . At day 21 after injury ( n = 237 neurons , 7 animals ) spontaneous activity has decreased to baseline levels ( 2 . 5% of afferents within the dorsal root ) , but application of the toxin to both neuroma site or L5 DRG significantly increased the proportion of myelinated afferents demonstrating spontaneous activity ( 7 . 2% p = 0 . 03 and 17 . 7 ±% p<0 . 001 respectively , chi-square test ) ( Figure 9a ) . In summary , we observed an acute increase in spontaneous activity following nerve injury , which was reversed with time . However , at this later time , blockade of Kv1 channels ( which had no effect in the naïve state ) could reinstate spontaneous activity almost to levels seen acutely after nerve injury . Therefore , Kv1 channels appear to have a role in the recovery from increased excitability following nerve injury ( Figure 9b–d ) . 10 . 7554/eLife . 12661 . 019Figure 9 . Local application of α-DTX reinstates primary afferent hyperexcitability at later time points following nerve injury . ( a ) Schematic illustration of 3-chamber recording system . 1 ) Recording chamber , 2 ) middle chamber , 3 ) stimulating chamber . The toxin was applied in chambers 1 or 2 , respectively . ( b ) Following sciatic nerve transection , there is a large increase in the proportion of primary afferents demonstrating spontaneous activity at day 2 , which is suppressed at days 7 and 21 post injury; Local application of α-DTX to the neuroma and L5 DRG at these later time points ( days 7 and 21 ) significantly increases the proportion of afferents , which are spontaneously active ( total proportions per group , chi-square tests , p<0 . 05 , p<0 . 01 , p<0 . 001 ) ( c ) neuroma application and ( d ) DRG application of α-DTX . Both recordings were carried out 21 days post-surgery . ( e ) In the presence of α-DTX , significantly more neurons respond to mechanical stimulation at the neuroma site using a 15 g von Frey filament ( p<0 . 05; total proportions per group , chi-square tests , all neuroma day 21 ) . ( f ) Representative traces showing greater responsiveness to mechanical stimulation with von Frey filaments after local α-DTX application . DOI: http://dx . doi . org/10 . 7554/eLife . 12661 . 01910 . 7554/eLife . 12661 . 020Figure 9—source data 1 . Source data for Figure 9 . DOI: http://dx . doi . org/10 . 7554/eLife . 12661 . 020 Because Kv1 channels have been shown to have a crucial role in mechano-sensitivity ( Hao et al . , 2013 ) , we tested their role in hypersensitivity following nerve injury . In the presence of α-DTX , significantly more neurons responded to mechanical stimulation using a 15g von Frey filament at the neuroma site at day 21 ( with 4 g: pre 17 . 9% post 21 . 8%; with 8 g: pre 22 . 1% post 26%; with 15 g: pre 24 . 7% , post 30 . 5% , p = 0 . 006 , chi-square test; n = 259 neurons; Figure 9e–f ) . We subsequently used behavioural measures to examine mechanical sensitivity after nerve injury . For this purpose , we used the sciatic neuroma model used before in which the sciatic nerve of rats is transected and the proximal end was sutured superficially below the skin on the animal’s leg . We applied von Frey filaments of increasing forces to the site of the skin covering the neuroma . To test the role of Kv1 channels on hypersensitivity , we applied α DTX subcutaneously at the site of neuroma . At baseline , the withdrawal threshold was high and did not change with the application of α DTX ( vehicle: 142 . 9 ± 13 g , toxin 150 ± 17 . 7 g , n = 9 animals per group ) . Three days after injury , this threshold dropped to 8 . 2 ± 0 . 6 g and was not changed by applying the toxin ( 9 . 7 ± 1 . 1 g ) ( n = 9 animals per group ) . However , with time this threshold began to normalise reaching 12 . 9 ± 0 . 6 g at 7 days ( n = 8 ) and 44 . 8 ± 1 . 6 g at 21 days ( n = 8 ) . When we injected αDTX , this recovery was not seen and thresholds stayed low ( day 7: 6 . 9 ± 0 . 9 g , p = 0 . 003 , n = 9; day 21: 13 ± 2 . 1 g , p<0 . 001 , n = 7 , RM two way ANOVA Figure 10 ) . 10 . 7554/eLife . 12661 . 021Figure 10 . Mechanical hypersensitivity is restored by blocking Kv1 channels . Mechanical withdrawal thresholds were assessed by applying a range of Von Frey hairs to the skin over the neuroma site ( labelled with a suture ) . Animals were randomised to receive either subcutaneous αDTX or saline 30 min before testing . Hypersensitivity after nerve injury is very pronounced until day 7 , when it slowly starts recovering . At this time point , perineuromal application of αDTX reversed this early recovery . At 3 weeks after nerve injury hypersensitivity is much recovered and perineuromal injection of αDTX restored mechanical hypersensitivity to levels seen acutely after injury ( RM two way ANOVA , p<0 . 05 , *p<0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12661 . 02110 . 7554/eLife . 12661 . 022Figure 10—source data 1 . Source data for Figure 10 . DOI: http://dx . doi . org/10 . 7554/eLife . 12661 . 022 The time point when the mechanical hypersensitivity began to recover in injured animals coincides with the time when Kv1 . 1 and Kv1 . 2 are reduced but Kv1 . 4 and Kv1 . 6 are being expressed . α-DTX is a selective blocker for Kv1 . 1 , Kv1 . 2 and Kv1 . 6 but has little activity against Kv1 . 4 . CP 339818 hydrochloride however , which is a selective blocker of Kv1 . 3 and Kv1 . 4 ( Nguyen et al . , 1996 ) , did not have any effect on mechanical hypersensitivity at early or late time-points post injury suggesting that Kv1 . 4 is dispensable for the suppression of hyperexcitability . ( baselines: saline 133 ± 21 g , CP339818 164 ± 38 g; neuroma day 3: saline 51 ± 2 g , CP339818 44 ± 5 g; neuroma day 7: saline 41 ± 3 g , CP339818 39 ± 4 g; neuroma day 21: 73 ± 13 g , CP339818 91 ± 4 g , RM two way ANOVA , p>0 . 05 , n = 8–7 for saline at day 21] ) . It has previously been documented that within a neuroma nodes of Ranvier become disorganised ( Levinson et al . , 2012 ) and we confirm that here with the majority nodes showing abnormalities such as being elongated , split , heminodes or showing Nav in the absence of caspr . Voltage-gated sodium channels are known to accumulate particularly within axon tips and on denuded axons of the neuroma ( England et al . , 1996 ) . Much less is known regarding the distribution of Kv1 channels changes within the neuroma . This is an important issue as such channels could potentially act as ‘brakes’ on excitability . At an early time point after injury , we found that Kv1 . 2 is no longer confined to the juxtaparanode but extended into the paranode . As this could be only a reflection of direct injury , we also looked into a site in the injured nerve 1 cm proximal to the neuroma and found similar changes in distribution , suggesting this is likely to reflect widespread changes within the axon/axoglial signalling ( further supported by changes within the dorsal root and discussed below ) . At a later time-point following injury , we found a striking reduction in the expression of Kv1 . 1 and 1 . 2 which are normally localised to the juxtaparanode . In contrast Kv1 . 4 and 1 . 6 which are present at a low level in the naïve state are up-regulated and are present both in the paranode and the juxtaparanode . We found broadly similar changes in rodent and human neuroma . This altered expression and localisation is likely to partially reflect the altered relationship between axons and myelinating Schwann cells . Within the neuroma new nodes of Ranvier will be formed as a consequence of myelination of new axon sprouts and remyelination of denuded axons ( Dorsi et al . , 2008; Dyck et al . , 1985 ) . During developmental myelination and remyelination following primary demyelination ( in which the axon remains intact ) Kv1 . 1 and 1 . 2 can at early time points be observed in other regions apart from the juxtaparanode ( at the node of Ranvier and the paranode ) before being restricted to the juxtaparanode as the nodal complex matures ( Rasband et al . , 1998; Vabnick et al . , 1999; Poliak et al . , 2001 ) . Kv1 . 4 and 1 . 6 expression has not to our knowledge been examined during myelination/remyelination . We also studied the dorsal root , which is remote from the injury site to establish whether there were changes in Kv1 channels composition of the juxtaparanode that reflect the general response of the axon to injury rather than local effects such as inflammation and remyelination at the injury site . We also found striking changes within the nodal complex of sensory axons within the dorsal root . Kv1 . 2 was no longer down-regulated as had been noted at the neuroma site but its localisation changed following injury: They could be observed in the paranode as well as the juxtaparanode . In the naïve state , very little Kv1 . 4 and 1 . 6 expression was noted in the juxtaparanode of the dorsal roots . This is in agreement with previous studies that reported a low frequency of Kv1 . 4 immunoreactive juxtaparanodes ( Everill et al . , 1998 ) and no expression of Kv1 . 6 ( Utsunomiya et al . , 2008 ) . We found however that nerve transection led to markedly increased expression of these α-subunits and they were localised both to juxtaparanode and paranode . What factors are responsible for the altered distribution of Kv1 channels within the juxtaparanode ? We used ultrastructural examination of the juxtaparanode in the dorsal root to examine whether structural changes within the paranode could explain the movement of Kv1 channels into this region ( from which they are normally excluded ) . The transverse bands are important points of attachment between the axon and the paranodal loops of the Schwann cell . These axoglial septate-like junctions are formed by the interaction of caspr ( Bhat et al . , 2001 ) and contactin ( Boyle et al . , 2001 ) on the axolemma binding with the 155Kd isoform of Neurofascin expressed on the Schwann cell paranodal loops ( Tait et al . , 2000 ) . These junctions act as diffusion barriers between the nodal and juxtaparanodal membrane . Mice lacking caspr ( Bhat et al . , 2001 ) , contactin ( Boyle et al . , 2001 ) or NF155 ( Pillai et al . , 2009; Sherman et al . , 2005 ) have absent transverse bands , increased distance between the axon and Schwann cell membrane , disorganisation of the paranodal loops and probably as a consequence of the loss of this lateral diffusion barrier Kv1 channels are noted to extend into the paranode . Similarly in mice lacking ceramide galactosyl transferase in which all-putative adhesion components of the paranodal junction are lacking , Kv1 . 2 and caspr2 are also mis-localised to the paranodes ( Poliak et al . , 2001; 2003 ) . On ultrastructural examination of the paranodes within the sciatic nerve following injury , we did not see major structural changes and there was no increase in the distance between the axon and the Schwann cell membrane at the site of attachment of the paranodal loops . We noted an increase in the maximum distance between paranodal loops , which is unlikely to alter the ability of molecules to passively diffuse between the membrane domains of the juxtaparanode and paranode ( however it will increase the diameter of the helical pathway between paranodal loops connecting the extracellular space to the axonal internode . One potential consequence of which would be reduced passive resistance to current flow between the node and voltage-gated potassium channel ( VGKC ) in the juxtaparanode and paranode , which could then have a greater influence on nodal excitability ( Shroff et al . , 2011 ) . A recent publication has demonstrated that in mice lacking βII spectrin expression in axons Kv1 channels were no longer restricted to the juxtaparanode but could also be observed in the paranode even though the structural integrity of axoglial junctions was intact ( Zhang et al . , 2013 ) . βII spectrin contributes to the sub-membranous cytoskeleton of the axon-linking membrane proteins to actin and appears to act as a barrier limiting the lateral diffusion of membrane proteins . We found that paranodal expression of βII spectrin was reduced following axotomy and a reduction in this sub-membranous barrier is compatible with the lateral movement of Kv1 channels into this region that we observed . As well as Kv1 channels we also see caspr2 overlapping with paranodal markers following nerve injury and again such paranodal localisation of caspr2 was also reported in mice in which axonal βII spectrin is conditionally ablated . Caspr2 complexes with and is important for the correct localisation of Kv1 channels ( Poliak et al . , 1999; 2003 ) suggesting that this whole protein complex is mis-localised following nerve injury . Although loss of a sub-membranous barrier to diffusion of the VGKC-complex is one explanation for their paranodal localisation , we do not yet have a full understanding of the regulation of Kv1 channels trafficking . Phosphorylation events may have a role to play as Kv1 . 2 can undergo phosphorylation , which impacts on surface expression/localisation ( Gu and Gu , 2011; Yang et al . , 2007 ) . Following axonal injury sensory axons become hyper-excitable and this is important in driving and maintaining neuropathic pain ( Han et al . , 2000 ) . A recent study showed that myelinated sensory fibres are key in maintaining mechanical allodynia in several neuropathic pain models ( Xu et al . , 2015 ) . Spontaneous activity and mechanical stimulus evoked activity has been recorded in myelinated afferents innervating neuroma using microneurography ( Nyström and Hagbarth , 1981 ) . The role of Kv1 channels has mainly focussed on their importance in suppressing excitability at the soma rather than the axon following injury . The expression of a number of α Kv1 channels sub-units has been documented to decrease following peripheral axotomy including Kv1 . 1 , 1 . 2 ( Everill et al . , 1998; Hao et al . , 2013; Ishikawa et al . , 1999; Kim et al . , 2002; Yang et al . , 2004 ) and in some reports Kv1 . 4 ( we did not see a reduction in Kv1 . 4 using western blot analysis of DRG lysate following sciatic axotomy , however , this is a less proximal lesion compared to spinal nerve ligation [Everill et al . , 1998] ) . Correspondingly , the K currents mediated by such channels are reduced when measured at the soma ( Yang et al . , 2004 ) both in small and large diameter DRG cells . The focus has therefore been on the loss of K currents within the soma which normally act as a ‘break’ on excitability , and combined with increased excitatory drive for instance due to the dysfunction of voltage-gated sodium channels and hyperpolarization-activated cyclic nucleotide-gated ( HCN ) channels this leads to ectopic activity ( Waxman and Zamponi , 2014 ) . While changes at the soma are undoubtedly important , ectopic impulses also arise along the axon . There has been much less focus on the distribution and function of Kv1 channels within the axon following peripheral nerve injury . The function of Kv1 channels is critically dependent on their targeting to specific neuronal compartments ( Trimmer , 2015 ) . The expression of Kv1 channels within the DRG soma and the axon should not be assumed to be the same: For instance the expression of Kv1 . 6 remains stable within the DRG both at the level of mRNA and protein ( see Kim et al . ( 2002 ) , Yang et al . ( 2004 ) and our own data ) but as we show here expression markedly increases in axons within peripheral nerve and dorsal root . Altered α subunit composition and localisation of Kv1 channels is likely to have functional implications . In the naïve state , the Kv1 channels Kv1 . 1 and 1 . 2 are located within the juxtaparanode , below the insulating myelin sheath and at least in peripheral nerves have little functional impact on nodal excitability and conduction ( Poliak et al . , 2003; Chiu and Ritchie , 1980; Sherratt et al . , 1980; Rasband et al . , 1998 ) . During development when Kv1 channels are observed in the node and paranode using specific blockers of this K current suggests that these Kv1 channels prevent re-entrant excitation in motor axons ( Vabnick et al . , 1999 ) . Following primary demyelination , the re-distribution of Kv1 channels to the paranode acts to suppress continuous conduction in demyelinated axons ( Rasband et al . , 1998 ) . In certain contexts therefore and especially when Kv1 channels begin to encroach on the paranodal regions there is evidence that these channels can suppress excitability of the axon . As has been previously demonstrated we have found that the rate of ectopic activity within myelinated axons is very high in the first week and then decreases the longer the time elapsed following the initial injury ( Campbell et al . , 1988 ) . Understanding adaptive mechanisms to suppress such hyper-excitability will potentially provide insight as to why in certain patients such mechanisms fail leading to chronic pain states . Over a similar time period as this reduction in spontaneous activity we have observed increased expression of Kv1 . 4 and 1 . 6 as well as redistribution of Kv1 channels to the paranode and juxtaparanode domains . Selective inhibition of Kv1 channels with α-DTX reinstates a higher level of ectopic activity , increases mechanosensitivity of afferents innervating the neuroma and on behavioural testing also exacerbates mechanical hypersensitivity , which had begun to normalise at 3 weeks post injury . α-DTX blocks Kv1 . 1 , Kv1 . 2 and Kv1 . 6 . As expression of Kv1 . 1 and 1 . 2 are decreased while Kv1 . 6 is increased , most probably the effect seen with the toxin is through Kv1 . 6 . Selective inhibition of Kv1 . 4 did not recapitulate these events emphasising the role of Kv1 . 6 ( and subunits with which it complexes ) in suppressing hyperexcitability . Neuronal Kv1 proteins form heterotetramerization of α subunits , which also associate with auxiliary Kvβ subunits ( Jan and Jan , 2012 ) , adding complexity in ascribing function to individual α subunits . α subunits confer particular pharmacological and biophysical properties on these channels and in addition there may be interactions between subunits . For instance Kv1 . 4 , usually shows N-type rapid inactivation through an N-terminal inactivation ball however this can be over-ridden if associated with a Kv1 . 6 α subunit ( Roeper et al . , 1998 ) , due to its NIP ( N-type inactivation prevention ) domain . In conclusion , we have found major changes in Kv1 channels subunit composition and distribution within the axolemma of myelinated axons following traumatic nerve injury . In contrast to the soma in which Kv1 channels expression is reduced this increased availability of Kv1 channels within the paranodes and altered subunit composition appears to fulfil an adaptive role in suppressing excessive excitability in myelinated afferents . Adult male Sprague-Dawley rats were used in accordance with UK Home Office and Pontificia Universidad Catolica’s regulations ( animals in the UK were purchased from Charles-River UK , animals from Chile were purchased onsite ) . Rats were group housed and placed on a 12 hr-light 12 hr-dark cycles . Two different nerve injury models were used: the neuroma model and the L5 spinal nerve transection ( SNT ) model . The neuroma model of neuropathic pain was based on the TNT model ( Dorsi et al . , 2008 ) , but performed with some modifications . Briefly , the sciatic nerve was dissected free of adjacent tissue , ligated with a suture , and cut proximal to its bifurcation . The needle from the suture was passed through a subcutaneous tunnel to the lateral aspect of the hindlimb where it was pushed through the skin . The nerve was drawn into the tunnel until the ligature is adjacent to the skin . The suture was cut , and the incision closed . The suture tied to the distal end of the sciatic nerve was visible through the skin and served as the target for mechanical stimuli . An analogous site served as the target on the contralateral hindlimb . For the L5 SNT model ( Kim and Chung , 1992 ) , one-third of the L6 transverse process was removed and the L5 spinal nerve was identified and dissected free from the adjacent L4 spinal nerve and then tightly ligated using 6–0 silk and then transected distally to the suture . Sham-operated animals served as a control . We used these two different models as the neuroma model is the most adequate for performing behavioural tests as the injured nerve can be directly stimulated , while the L5 SNT model has the advantage that it gives certainty that all the dorsal root axons studied had their peripheral terminals injured . For both models animals were deeply anaesthetised with a mix of isofluorane and oxygen . Postoperative analgesia was given for the first 5 days postop ( tramadol 50 mg/kg/day p . o ) . Animals were checked every day after surgery to check for self-mutilation behaviour ( autotomy ) , which prompted us to sacrifice the animal . Calculation of the sample size needed was done for each experiment as described below . Experimental protocols were reviewed and approved by 'Coordinación de Ética , Bioética y Seguridad de las investigaciones UC' ( experiments done in Chile ) and were performed in accordance with the UK Home Office regulations ( experiments done in the UK ) . We report this study in compliance with the ARRIVE guidelines ( Kilkenny et al . , 2010 ) ( 20 points checklist ) . The study was conducted at Hospital Clinico UC-Christus in Santiago , Chile . Morton’s neuroma patients that were due surgery for resection of painful neuromas were recruited for donating a small sample of the tissue resected during surgery . Control samples were obtained from subjects undergoing hand reconstructive surgery in where the sural nerve is harvested and used as a bridge to connect disrupted ends of motor nerves in the hand . A small sample for these healthy sural nerves was collected to use as control in this study . We used the Numeric Rating Scale ( NRS; which is a self-reporting scale where 0 is no pain and 10 is the worst imaginable pain ) to assess for pain before surgery . Informed consent was obtained from all subjects before surgery . The study protocol was assessed and approved by the Ethics Scientific Committee of the School of Medicine Pontificia Universidad Catolica de Chile ( reference number 14–389 ) . The sample sizes were calculated using a power of 80% and an α error of 0 . 05% , assuming a 2 times increase or decrease in Kv channel expression with a variance of 0 . 6 from the mean , which resulted in a sample needed of 3 patients per group . After a defined survival time ( 7 and 21 days ) , animals were terminally anaesthetized with pentobarbital and transcardially perfused with 0 . 9% heparinised saline . The L5 DRG , the L5 spinal nerve , and the sciatic nerve were removed . We dissected the sciatic nerve free from connective tissue and collected 5 mm from the site of the neuroma and 5 mm from a site 1 cm proximal . Tissue for immunohistochemistry was post fixed in 4% paraformaldehyde ( PFA ) in 0 . 1 M phosphate buffer ( PB ) for 30 min and cryoprotected in 20% sucrose for 3 days . Tissue obtained from patients was fixed immediately after resection in 4% PFA for 30 min and then cryoprotected in 20% sucrose for 3 days . The samples were embedded in OCT , cryostat cut ( 8 µm ) and thaw-mounted onto glass slides . Sections were pre-incubated in buffer ( PBS , pH 7 . 4 , containing 0 . 2% Triton X-100 and 0 . 1% sodium azide ) containing 10% normal donkey serum for 30 min and then incubated with primary antibodies overnight at room temperature . Primary antibodies used are shown in Table 1 . Following primary antibody incubation , sections were washed and incubated for 2 hr with secondary antibody solution ( donkey anti-rabbit Cy3 1:400; goat anti guinea pig AMCA 1:100 , donkey anti mouse FITC 1:400; all from Stratech , UK ) . Slides were washed with PBS , cover-slipped with Vectashield mounting medium ( Vector Laboratories , UK ) and visualised under a Zeiss Axioplan 2 fluorescent microscope ( Zeiss , UK ) . All quantification of different IHC parameters was done with the investigator blinded to the identity of the group to which the animals belonged . Nodal quantification was done by assessing on average 31 nodes per animal , and using 4–5 animals per condition . For quantification of βII spectrin the intensity of the immunofluorescence of the axonal paranodal area ( identified by caspr staining ) was measured and the background of each section was subtracted . Then , measurements were normalised against the mean of the controls ( naïve axons ) . The sample sizes were calculated using a power of 80% and an α error of 0 . 05% , assuming a 2 times increase or decrease in expression with a variance of 0 . 5 from the mean , which resulted in a sample needed of 4 animals per group . 10 . 7554/eLife . 12661 . 024Table 1 . Different antibodies used in the study . DOI: http://dx . doi . org/10 . 7554/eLife . 12661 . 024AntibodyConcentration used IHC WBCompanyRabbit anti Pan voltage gated sodium channel ( Cat No . S6936 ) 1:1000Sigma-AldrichMouse anti Kv1 . 2 ( K14/16 . 2 ) 1:100 1:500UC Davis/NIH NeuroMab FacilityMouse anti Kv1 . 1 ( K36/15 . 1 ) 1:100 1:200UC Davis/NIH NeuroMab FacilityMouse anti Kv1 . 4 ( K13/31 ) 1:100 1:200UC Davis/NIH NeuroMab FacilityMouse anti Kv1 . 6 ( K19/36 ) 1:100 1:500UC Davis/NIH NeuroMab FacilityGuinea Pig anti Caspr1:1000 1:1000From Dr Manzoor Bhat - UT Health Science Center San Antonio ( Bhat et al . , 2001 ) Rabbit anti Caspr2 ( ab105581 ) 1:500 1:400AbcamRabbit anti Pan Neurofascin1:500Gift from Prof Peter Brophy- University of Edinburgh ( Pomicter et al . , 2010 ) Mouse anti βII spectrin ( Clone 42 ) 1:500 1:1000BD BioscienceGAPDH1:10000AbcamPGP 9 . 51:5000UltracloneIHC: Immunohistochemistry; WB: Western Blot analysis . We quantified sodium channel clusters following the following criteria: ( 1 ) typical nodes were nodes where the Nav channels fill the gap at the node of Ranvier as identified by the paranodal staining of caspr on both sides of the node , ( 2 ) split nodes were nodes that had two distinct Nav channels accumulations , separated by a gap in the Nav channel staining within the same fibre and with each Nav channels accumulation flanked on one side with caspr staining , or ( 3 ) heminodes were nodes where the caspr staining located on only one side of a contiguous Nav channel accumulation , ( 4 ) while those Nav channel accumulations lacked an association with caspr were classified as ‘naked’ accumulations . Tissue was collected , quickly frozen in liquid nitrogen and was homogenized in NP40 lysis buffer ( 20 mM Tris , pH 8 , 137 mM NaCl , 10% glycerol , 1% NP-40 , 2 mM EDTA ) , 20 μM leupeptin , 5 mM sodium fluoride , 1 mM sodium orthovanadate , 1 mM PMSF and protease inhibitor cocktail ) . The lysate were spun at 13 , 000 rpm at 4°C for 15 min and the protein concentration of supernatant was determined using a BCA Protein Assay kit ( Thermo Scientific ) . 50 µg of each sample was separated using 8% or 10% SDS-PAGE , and transferred to nitrocellulose membranes . Membranes were then blocked in 10% skimmed milk in PBS-T ( PBS+ 0 . 1% Tween 20 ) for 1 hr at room temperature . Membranes were incubated with primary antibody ( anti mouse Kv1 . 1 , Kv1 . 2 , Kv1 . 4 , Kv1 . 6 , GAPDH , PGP9 . 5 and anti-rabbit Caspr2 as shown in Table 1 ) , diluted in PBS-T at 4°C overnight . After washing with PBS-T for 6 times and5 min each time , membranes were incubated with sheep anti-mouse or donkey anti rabbit HRP-conjugated secondary antibody ( 1:10 , 000–1:20 , 000; ECL , GE Healthcare , Amersham , UK ) at room temperature for 1 hr . After several PBS-T washes as described above membranes were revealed using ECL-prime reagent ( GE Healthcare ) for 5 min for detection by autoradiography . For WB of Kv1 . 4 and Kv1 . 6 in rat tissue ( sciatic nerve and DRG ) the membranes were cut in three pieces; the top piece was probed with Kv1 . 4 antibody , the middle one was probed with Kv1 . 6 antibody and the bottom one probed with GAPDH antibody . For WB of Kv1 . 1 and for Kv1 . 2 the membranes were cut in 2 pieces: the top one was probed with either Kv1 . 1 or Kv1 . 2 antibody and the bottom one was probed with GAPDH antibody . The 2 or 3 pieces of the membranes were lined up as a single membrane before exposing it to the film so that the molecular weight can be calculated by measuring the running distance of the molecular weight marker and the target bands . This could be done as the bands labelled by the antibodies have quite different molecular weights . This allowed us to optimize the use of the tissue obtained from animals and reduce the number of animals needed ( in accordance with our obligations under animal licensing procedures ) . For Western Blots analysis , films were scanned with Cannon Scanner ( LiDE 210 ) , and the intensity of specific bands was quantified using Quantity One software ( Bio-Rad ) . The same size rectangle was drawn around each band to measure intensity , and the background was subtracted . Target band detected was normalized against loading control GAPDH or PGP9 . 5 correspondingly for analysis . The sample sizes were calculated using a power of 80% and an α error of 0 . 05% , assuming a 2 times increase or decrease in expression with a variance of 0 . 5 from the mean , which resulted in a sample needed of 4 animals per group ( we used 6 animals per group in case we had to put any animal down due to autotomy ) . Sciatic nerves were dissected at the site of the neuroma and were processed for resin embedding as previously described ( Huang et al ) . Briefly nerves were post fixed in 3% glutaraldehyde at 4°C overnight , washed in 0 . 1 M PB , osmicated , dehydrated , and embedded in epoxy resin ( TAAB Embedding Materials , UK ) . Longitudinal sections 1 μm thick were cut on a microtome and stained with toludine blue before being examined on a light microscope . Ultrathin sections were cut on an ultramicrotome and stained with lead uranyl acetate . Sections were mounted on unsupported 100 mesh grids . Sections were visualised on a PHILIPS TECNAI 12 BIOTWIN transmission electron microscope at the Unidad de microscopia avanzada , Pontificia Universidad Catolica de Chile . We measured the diameter of the axons at the site of the node , the maximal and minimal distance between interloops , the distance between the glia and the axon , the number of detached loops , and the number of everted loops using Image J ( NIH , USA ) and a 135000x magnification . We quantified between 8 and 14 nodes per animal , and we used 5 animals per condition ( sample sizes were calculated using a power of 80% and an α error of 0 . 05% , assuming a change in distance of 50% with a variance of 0 . 4 , which resulted in a sample needed of 4 animals per group , however due to the difficulty in the technique we included one more animals in each group ) . The investigator was blinded to the treatment group of each specimen , however , this was sometimes difficult to conceal as the anatomy in the injured nerves was much more disrupted than in naive nerves . Recordings were performed under anaesthesia ( urethane , 1 . 5 g/Kg , i . p . ) on naïve rats ( n = 10 , 222 neurons ) , or after sciatic nerve ligation at 2 days ( n = 4 , 291 neurons ) , at 7 days ( n = 7 , 241 neurons ) , and at 21 days ( n = 7 , 237 neurons ) . A tracheotomy was performed and the L5 dorsal roots and DRGs were exposed via laminectomy . Sciatic nerve neuroma with proximal nerve ( 5–6 mm long ) and contralateral uninjured sciatic nerve were exposed . The contralateral sciatic nerve was acutely cut to disconnect from the periphery just before recording . The entire site was covered in agarose gel and four chambers created by removing blocks of this gel . These were 1 ) neuroma chamber , containing ipsilateral neuroma and part of sciatic nerve which is subjected to stimulation; 2 ) acutely cut nerve end chamber , containing contralateral sciatic nerve proximal end; 3 ) DRGs chamber , containing L5 DRGs from both sides; 4 ) spinal recording chamber , containing part of L5 dorsal roots from both sides near entry zone to spinal cord . The neuroma chamber and nerve cut end chamber were filled with mineral oil during stimulation , and the oil was replaced with αDTx ( 100 nM in saline ) during toxin application . The DRGs chamber was filled with saline or αDTx/saline solution , and the recording chamber was always filled with mineral oil . The pool temperatures were not controlled , but as animals were warmed using an infrared lamp from the back , the pool was therefore heated , and typically was at 34–35°C . Just before recording , the L5 dorsal root was cut near entry zone , a filament was teased out and hooked up for recording . Each filament was stimulated electrically with increasing current to recruit sequentially each conducting axon in that filament . The conduction velocity of each conducting axon was noted . Thus , the number of functioning axons in each filament was determined ( typically , 6–10 ) . Spike discrimination was used to detect the number different axons firing spontaneously in each filament ( typically 0–3 ) during a pre-treatment baseline and under 3 different treatment conditions: 1 ) no αDTx in any of the chambers; 2 ) αDTx in neuroma or nerve cut end chamber; 3 ) αDTx in neuroma or nerve cut end chamber and DRGs chambers . The αDTx was applied for at least 20 min before recording . An independent investigator prepared the drugs individually and labelled them for each animal according to the randomization schedule . Data analysts were blinded as the conditions under which all recording were made . Signals were amplified with an AC-coupled amplifier ( Neurolog NL104A with headstage NL100AK ) , then high-pass- and low-pass filtered ( Neurolog NL125 ) at 500 Hz and 5 KHz frequencies . The filtered signals were passed through a Humbug 50 Hz noise eliminator ( Quest scientific , Vancouver , BC , Canada ) , further amplified ( Neurolog NL 106 ) , fed to an analog-to-digital converter PowerLab , and sampled at 20 KHz with Labchart software ( ADinstruments , UK ) . Stimulation ( 200 µs square-wave pulses ) was delivered from a stimulus isolator ( Neurolog , NL800A ) . The filter settings used strongly favours recordings from A-fibres and not C-fibres . All the fibres recorded to nerve stimulation conducted in the A fibre range ( >2 m/sec ) . The size of the filaments recorded was also unfavourable for detecting clear single unit C fibre activity . Three minutes baseline was recorded to examine spontaneous activity . The percentage of spontaneously firing units was calculated as the number of spontaneously active units divided by the number of conducting fibres determined in recruitment recording . The firing rate was calculated as the total number of spikes during recording divided by the time recorded . The mean firing rate per unit was the firing rate divided by the total number of different units recorded in each treatment group . For the axonal mechanosensitivity experiments ( n = 4 , 259 neurons ) , mechanical stimulation was applied to the neuroma using increasing forces of von Frey filaments ( 4 , 8 , and 15 g ) , and the number of distinct spikes ( neurons ) firing in response were counted following spike discrimination . The total number of conducting axons in each filament was determined by incremental electrical stimulation of the sciatic nerve . The percentage of mechanosensitive units was calculated as the number of different neurons responding by firing action potentials upon mechanical stimulation , divided by the number of conducting fibres ( which was determined in the same way as for the spontaneous activity experiments ) . Axonal mechanosensitivity was assessed before and after toxin application at 21 days after axotomy . Mechanosensitivity experiments were carried out on separate animals to spontaneous activity experiments to ensure that any spontaneous activity encountered was not caused acutely by the repeated mechanical stimulation of the neuroma . Data was analyzed using software Labchart . Statistics comparing proportions of neurons exhibiting either spontaneous activity or mechanosensitivity were performed using chi-square test with Yates correction . Values were reported as percentages , calculated from the proportions . Mechanical withdrawal thresholds were assessed by applying a range of Von Frey hairs ( Somedic , Sweden ) to the skin over the neuroma site ( labelled with a suture as previously described ) . Animals were randomised to receive either subcutaneous αDTX ( 0 . 5 ml at 100 nM in saline , Alomone , UK ) or saline ( which was administered locally at the site of the neuroma ) using a computer-generated random sequence . The sample sizes were calculated using a power of 80% and an α error of 0 . 05% , assuming a 60% decrease in withdrawal threshold with a variance of 25% from the mean , which resulted in a sample needed of 7 animals per group . Experimental groups were the following: baseline with vehicle , baseline with toxin; day 3 after surgery with vehicle , day 3 with toxin; day 7 with vehicle , day 7 with toxin; day 21 with vehicle , day 21 with toxin . To reduce the amount of animals of the study the animals that received saline only were used again for the consecutive time-points . The animals that received toxin had to be sacrificed after testing , as the toxin is irreversible . The toxin or saline were injected 30 min before testing . Autotomy after nerve injury ( especially neuroma model ) appears at around 10 days after injury . Therefore , we allocated 2 extra animals for the saline group , and 2 extra for toxin day 21 . We had to sacrifice 1 animal from the saline group at day 6 , and 2 animals from the toxin group day 21 ( at day 15 and 17 after injury respectively ) , due to self-mutilating behaviour . For testing , rats were gently restrained using a towel on a table . Calibrated von Frey hairs were applied to the skin covering the neuroma until the fibre bent . Withdrawal of the limb by the animal was recorded as a response . The 50% withdrawal threshold was determined using the up-down method ( Dixon , 1980 ) . An independent investigator prepared the drugs individually and labelled them for each animal according to the randomization schedule . Operators and data analysts were blinded throughout the study . The data were distributed normally and the differences between groups was analysed using a 2 way ANOVA repeated measures . Values were reported as mean ± SEM . This experiment was repeated for testing CP339818 ( Kv1 . 4 blocker; #C-115 , 0 . 5 ml at 300 nM in saline Alomone Labs UK ) . We randomly allocated 8 animals per group , and we had to put one animal from the saline group down due to autotomy at day 12 .	Around 20% of the world’s population experiences long-lasting “chronic” pain , which often results in poor sleep , depression and anxiety . One of the most disabling forms of chronic pain is called neuropathic pain , which results from injuries to sensory nerves . Pain or discomfort is felt in response to touches that are not normally painful . Neuropathic pain is difficult to treat as we do not fully understand the molecular mechanisms that cause it . Stimulating a nerve causes it to produce action potentials . At a molecular level , these action potentials are generated by ions moving into and out of the neuron through proteins called ion channels . The movement of sodium ions into a neuron triggers an action potential , and the movement of potassium ions out of the neuron returns it to a resting state . After a sensory nerve is cut or otherwise damaged it becomes hyperactive and produces spontaneous electrical activity that the brain interprets as pain signals . However , it is not fully understood how cutting a nerve affects the ion channels in a way that generates this hyperactivity . Different types of ion channel are found in different regions of the nerve cell; for example , type 1 potassium channels are normally found in a region called the juxtaparanode at the axon of the neuron . Calvo et al . have now tracked what happens to type 1 potassium channels after nerve injury in rats . Soon after nerve damage occurred , nearly all of these ion channels disappeared from the juxtaparanode . At the same time , electrical activity in the cut nerve increased , and the recovering animals responded in ways that suggested they were hypersensitive to the nerve being touched . Three weeks after the injury , most rats lost their hypersensitivity and the electrical activity in the cut nerve returned to near-normal levels . Calvo et al . found that the recovering nerves contained new subtypes of type 1 potassium channels . These potassium channels did not just appear in the juxtaparanode: they also invaded the ‘fence’ region that normally separates potassium channels from sodium channels . The same was observed to happen in the nerves of patients that suffer from neuropathic pain due to a nerve injury . At this late time point after nerve injury , blocking the activity of potassium channels produced the same abnormal increase in the nerve’s electrical activity as seen immediately after the nerve had been cut . The rats’ hypersensitivity to touch also returned . This suggests that the appearance of the new potassium channel subtypes might be a protective mechanism that reduces the activity of a damaged nerve to decrease pain . These findings suggest new ways of treating neuropathic pain . Further studies are now needed to investigate whether drugs that can activate the new potassium channel subtypes could stop pain from an injured nerve becoming a long-term problem .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Altered potassium channel distribution and composition in myelinated axons suppresses hyperexcitability following injury
The origins and functions of kidney macrophages in the adult have been explored , but their roles during development remain largely unknown . Here we characterise macrophage arrival , localisation , heterogeneity , and functions during kidney organogenesis . Using genetic approaches to ablate macrophages , we identify a role for macrophages in nephron progenitor cell clearance as mouse kidney development begins . Throughout renal organogenesis , most kidney macrophages are perivascular and express F4/80 and CD206 . These macrophages are enriched for mRNAs linked to developmental processes , such as blood vessel morphogenesis . Using antibody-mediated macrophage-depletion , we show macrophages support vascular anastomoses in cultured kidney explants . We also characterise a subpopulation of galectin-3+ ( Gal3+ ) myeloid cells within the developing kidney . Our findings may stimulate research into macrophage-based therapies for renal developmental abnormalities and have implications for the generation of bioengineered kidney tissues . Macrophages are professional phagocytes with roles in tissue development , immunity , regeneration , remodelling , and repair ( Bosurgi et al . , 2017; Eom and Parichy , 2017; Godwin et al . , 2013; Stamatiades et al . , 2016 ) . In the adult kidney , macrophages can cause , prevent , and repair damage ( Cao et al . , 2015; Rogers et al . , 2014 ) . During organogenesis , evidence suggests that macrophages promote kidney growth ( Alikhan et al . , 2011; Rae et al . , 2007 ) , but details regarding their developmental localisation and functions are lacking . Characterising the roles for kidney macrophages may lead to the development of macrophage-based strategies to enhance renal maturation in cases of developmental abnormalities and premature birth . Kidney development begins when the caudal Wolffian duct is induced to branch , forming a ureteric bud that invades the metanephric mesenchyme by embryonic day ( E ) 11 in the mouse ( Pichel et al . , 1996; Saxén and Sariola , 1987 ) . The ureteric bud undergoes rounds of iterative branching to form the adult collecting duct tree , while cells endogenous to the metanephric mesenchyme generate nephrons and stroma ( Grobstein , 1955; Kobayashi et al . , 2008; Epelman et al . , 2014; Shakya et al . , 2005; Short et al . , 2014 ) . Other kidney cell types , such as monocytes , macrophages , and most endothelial cells , derive from extra-renal sources ( Epelman et al . , 2014; Hoeffel et al . , 2015; DeFalco et al . , 2014; Sequeira-Lopez et al . , 2015; Sims-Lucas et al . , 2013 ) . In embryonic development , macrophage progenitors arrive in organs in waves . The first wave is generated from yolk sac-derived primitive macrophage progenitors ( Palis et al . , 1999; Schulz et al . , 2012 ) , the second from yolk sac-derived erythro-myeloid progenitors ( EMPs ) that migrate and colonise the fetal liver ( Hoeffel et al . , 2015; Rantakari et al . , 2016 ) , and the third from haematopoietic stem cells ( HSCs ) that emerge in the aorta-gonad-mesonephros region ( Medvinsky and Dzierzak , 1996; Sheng et al . , 2015 ) . Unlike yolk sac-derived primitive macrophage progenitors , both HSCs and EMPs in the foetal liver are thought to pass through a monocytic intermediate phase before differentiating into mature macrophages ( Hoeffel et al . , 2015; Hoeffel and Ginhoux , 2018; Schulz et al . , 2012 ) . Lineage tracing studies in mice suggest that kidney macrophages are initially derived from the yolk sac but are derived almost exclusively from foetal monocytes after birth ( Epelman et al . , 2014; Hoeffel et al . , 2015; Munro and Hughes , 2017; Sheng et al . , 2015 ) . In the adult , macrophages and endothelial cells form anatomical and functional units that function to clear immune complexes from the renal blood ( Stamatiades et al . , 2016 ) . In development , kidney macrophages closely associate with the epithelial tubules of nephrons ( Rae et al . , 2007 ) , but their interactions with vascular endothelial cells have not been examined . In recent years , kidney vascularisation has been a focus of renal developmental research ( Daniel et al . , 2018; Halt et al . , 2016; Hu et al . , 2016; Munro et al . , 2017a; Munro et al . , 2017b ) ; however , possible roles for macrophages in this process remain unexplored . We show that macrophages in the embryonic kidney restrict the early domain of nephron progenitor cells , frequently interact with blood vessels , are enriched for mRNAs linked to vascular development , and promote endothelial cross-connections . To characterise macrophage distribution as the metanephric kidney’s component parts emerge , we immunostained and optically cleared whole-mount E9 . 5–10 . 5 mouse embryos . At these stages , all macrophages in the embryo-proper are derived from the yolk sac and express colony-stimulating factor one receptor ( Csf1r ) , fractalkine receptor ( Cx3cr1 ) , and EGF-like module receptor 1 ( F4/80 ) ( Mass et al . , 2016; Schulz et al . , 2012 ) . Csf1r and Cx3cr1 mark both mature macrophages and their precursors , whereas F4/80 marks only mature macrophages ( Frame et al . , 2016; Mass et al . , 2016 ) . At E9 . 5 , bilateral nephrogenic cell populations , which expressed sine oculis-related homeobox 2 ( Six2 ) , were arranged as cords that lay medial to Wolffian ducts and lateral to the dorsal aorta in the caudal part of the mouse embryo ( Figure 1a–c; Video 1 ) . No F4/80+ macrophages were near to the Six2+ nephrogenic cords at this stage ( Figure 1c; Video 2 ) . This finding is consistent with previous studies using Cx3cr1GFP embryos that showed GFP-expressing cells arrived in the embryo-proper only from E9 . 5 onwards ( Mass et al . , 2016; Gomez Perdiguero et al . , 2013; Schulz et al . , 2012; Stremmel et al . , 2018 ) . By E10 . 5 , as ureteric bud outgrowth started , Six2+ metanephric cells had accumulated at the caudal end of the Wolffian duct ( Figure 1d; Figure 1—figure supplement 1 ) ( as previously described by Wainwright et al . , 2015 ) . Immunostaining for a pan-endothelial marker ( CD31 ) demonstrated that the metanephric cell populations were avascular at this stage ( Figure 1e; Video 2 ) . CD31 also marked primitive germ cells and intra-aortic hematopoietic clusters containing HSCs in the caudal part of the E10 . 5 mouse embryo ( Figure 1—figure supplement 1; Gomperts et al . , 1994; Wakayama et al . , 2003; Yokomizo and Dzierzak , 2010 ) . F4/80+ macrophages were now present in high numbers in the caudal part of the embryo ( Figure 1e–f ) . These macrophages largely avoided the metanephric mesenchyme ( Figure 1e–f; Video 2 ) but localised in high numbers alongside rostral clusters of Six2+ cells ( Figure 1g–h ) . The localisation of macrophages at E10 . 5 suggested to us that they may clear rostral Six2+ cells as the metanephric mesenchyme assembles at the caudal aspect of the Wolffian duct . To test this hypothesis , we depleted macrophages in vivo by crossing transgenic mice , expressing codon-optimised Cre ( iCre ) under the control of the Csf1r promoter ( Csf1ricre ) , with transgenic mice that had a floxed-STOP cassette and the diphtheria toxin A subunit ( DTA ) knocked into the Rosa26 locus ( RosaDTA ) ( Figure 1—figure supplement 2 ) . In this system , DTA expression is specifically induced in iCre+ cells , resulting in cell death by inhibiting protein synthesis ( Breitman et al . , 1987; Collier , 2001; Plummer et al . , 2017 ) ( Figure 1—figure supplement 2 ) . Consistent with the hypothesis that macrophages clear rostral nephrogenic progenitors , macrophage-depleted E11 . 5 Csf1ricreicre+RosaDTA embryos had elongated metanephric mesenchyme populations and larger rostral clusters of Six2+ cells compared to somite pair-matched littermate controls ( Figure 1i–l ) . Moreover , Six2+ nuclei of some rostral nephrogenic progenitors were observed within the cell bodies of macrophages , suggesting that active phagocytosis occurs at this site ( Figure 1j; Figure 1—figure supplement 3 ) . As Six2+ nephrogenic progenitors secrete signals that stimulate ureteric bud outgrowth ( Sainio et al . , 1997 ) , we next examined whether persistence of rostral nephrogenic progenitors in the absence of macrophages resulted in ectopic rostral ureteric bud outgrowth . In all macrophage-depleted embryos , ureteric bud outgrowth had occurred only at the normal anatomical position ( in 9/9 kidneys analysed ) ; however , its morphological development was delayed compared to littermate controls in E11 . 5 and E12 . 5 embryos ( Figure 1—figure supplement 4 ) . Vascular organisation appeared normal in macrophage-depleted embryos relative to the developmental stage of the kidney ( Figure 1i; Figure 1—figure supplement 4 ) . Collectively , these data show that macrophages arrive near to the metanephric mesenchyme as it condenses at the caudal aspect of the Wolffian duct at E9 . 5-E10 . 5 . These early macrophages arrange alongside rostral nephrogenic cells but avoid caudal ones . In the absence of macrophages , rostral nephrogenic cell clearance and ureteric bud development are delayed . We next sought to characterise macrophage arrival and localisation in the early kidney after ureteric bud outgrowth . The ureteric bud had invaded the metanephric mesenchyme by E11 and bifurcated by E11 . 5 ( Figure 2a–b ) . At E11-E11 . 5 , numerous F4/80+ macrophages and blood vessels were present in the peri-Wolffian mesenchyme ( situated between the metanephric mesenchyme and Wolffian duct ) , but relatively few were present within the metanephric mesenchyme ( Figure 2a–c ) . The few F4/80+ macrophages within the metanephric mesenchyme were most often at its border nearest the peri-Wolffian mesenchyme ( Figure 2b ) , in agreement with macrophage localisation in Csf1rEGFP kidneys ( Figure 2—figure supplement 1 ) ( Rae et al . , 2007 ) . Yolk sac-derived macrophage progenitors are trafficked into the embryo-proper via the bloodstream ( Stremmel et al . , 2018 ) . Upon reaching the embryo-proper , these primitive macrophages exit the vascular system and can invade newly forming organs , such as the brain , via extravascular tissue migration ( Cuadros et al . , 1993; Fantin et al . , 2010; Herbomel et al . , 2001 ) . Up until E12 . 5 , kidney macrophages are yolk sac-derived ( Hoeffel et al . , 2015 ) . Whole-mount immunostaining demonstrated that F4/80+ macrophages within the E11-E11 . 5 peri-Wolffian mesenchyme were always extravascular ( Figure 2a–b ) , which led us to hypothesise that these macrophages travel via trans-tissue migration . Indeed , time-lapse imaging of cultured E11 . 5 Csf1rEGFP kidneys indicated that these macrophages migrate via extravascular routes and regularly interact with other macrophages as well as the abluminal surfaces of isolectin B4-labelled blood vessels ( Video 3 ) . Blood vessels begin entering the kidney from the peri-Wolffian mesenchyme between E11 . 5-E12 when interstitial tissue regions first form ( Munro et al . , 2017b ) . We hypothesised that macrophages follow similar routes ( in time and space ) to colonise the kidney . To examine this , we immunostained and imaged whole-mount kidneys at relevant time points . At E11 . 5 , Six2+ nephron progenitor cells were present throughout the metanephric mesenchyme as a single population , which contained very few macrophages ( Figure 2d ) . By E12 . 5 , the Six2+ population had split ( Figure 2e ) , resulting in the ‘capping’ of each ureteric bud tip ( Herring , 1900; Short et al . , 2014 ) . As Six2+ nephron progenitor populations split , interstitium forms in the fissures and becomes vascularised ( Airik et al . , 2006; Daniel et al . , 2018; Munro et al . , 2017b ) . Whole-mount imaging of E12 . 5 kidneys demonstrated that F4/80+ macrophages had occupied these interstitial fissures and commonly wrapped around the interstitial blood vessels ( Figure 2e–f ) . Collectively , these data suggest that macrophages and blood vessels regularly interact as they enter the early kidney at the same time . Static images of whole-mount kidneys indicate that this occupation occurs between E11 . 5 and E12 . 5 as the nephron progenitor population splits ( Figure 2g ) . At this stage , blood vessels and most macrophages reside within the renal interstitium , rather than the ‘cap mesenchyme’ regions , which are rich in Six2+ nephron progenitor cells . We next investigated whether the close association between macrophages and endothelial cells continued throughout later kidney development . First , we confirmed that F4/80 specifically marked myeloid cells in the developing kidney by co-staining Csf1rEGFP kidneys with anti-GFP and anti-F4/80 . At E14 . 5 , all detectable F4/80+ cells co-expressed Csf1r ( 100 ± 0% , mean ±SEM ) ( Figure 3a; Video 4 ) . We then immunostained and optically cleared E13 . 5 , E15 . 5 , and E18 . 5 kidneys to explore organ-wide macrophage localisation . F4/80 , CD31 , and Gata3 co-staining demonstrated that macrophages were present throughout the medullary and cortical portions of developing kidneys and were often localised around the renal vasculature ( Figure 3—figure supplement 1; Video 5 ) . The type of association of macrophages with blood vessels depended on whether those vessels were anatomically large- or small-calibre . Macrophages often arranged parallel to major renal blood vessels , such as segmental arteries , but did not typically interact with their endothelial cells ( Figure 3b; Figure 3—figure supplement 1 ) . In contrast , macrophages directly interacted with endothelium of small calibre vessels , such as the newly forming cortical nephrogenic zone blood vessels ( Figure 3c–d ) . Important organogenetic processes occur in cycles within the cortical nephrogenic zone of the developing kidney: ureteric bud tips branch , nephron progenitor populations split , nephrogenesis initiates , and interstitial vascular plexuses form ( Herring , 1900; Lindström et al . , 2018b; Munro et al . , 2017b; Short et al . , 2014 ) . Due to the significance of this zone in renal organogenesis , we next focused on macrophages within this region . Using the E17 . 5 peripheral nephrogenic zone as a representative example , we determined that 89 . 6 ± 1 . 45% ( mean ±SEM ) of macrophages localised within the interstitium , the remainder being within the cap mesenchyme ( nephron progenitor populations; Figure 3e; Figure 3—figure supplement 2 ) . Of the interstitial macrophages , 78 . 5 ± 2 . 1% ( mean ±SEM ) were in contact with the abluminal surface of blood vessels , and we defined these cells as perivascular macrophages ( Figure 3e; Figure 3—figure supplement 2 ) . The alignment of perivascular macrophages , which we defined based on their longest axis , strongly correlated with the alignment of their associated blood vessel ( Figure 3f ) . This tendency of macrophages to localise around the nephrogenic zone vasculature was qualitatively consistent from E13 . 5 and throughout prenatal kidney development ( Figure 3—figure supplement 3 ) . Nephrogenic zone blood vessels are at a vascular front in the developing kidney ( i . e . a site of neovascularisation; Munro et al . , 2017b ) . In this region , the extent of macrophage interaction with the vasculature and vascular basement membrane varied: a single macrophage may have peri- , trans- , extra- , and intra-vascular regions ( Figure 3g–g’ ) . We did not detect any entirely intravascular F4/80high kidney macrophages , indicating that these cells either arrived via extravascular routes , that precursor cells gained F4/80high status only after exiting the vasculature , or a combination thereof . Although most macrophages were interstitial , they never expressed the renal cortical interstitial markers Meis1/2 ( Brunskill et al . , 2008 ) ( Figure 3h ) , consistent with the documented exogenous origins of kidney macrophages ( Hoeffel et al . , 2015; Kobayashi et al . , 2014 ) . Within the cortical nephrogenic zone , F4/80high macrophages typically co-stained for the mannose receptor , CD206 , a marker associated with perivascular/mature macrophage status ( Figure 3i–j ) ( Faraco et al . , 2016; Galea et al . , 2005 ) . Even at E11 . 5 , before the nephrogenic zone formed , F4/80+ macrophages typically co-expressed CD206 ( Figure 3i ) . Immunostaining confirmed that CD206+ macrophages were often perivascular ( Figure 3—figure supplement 4 ) and demonstrated that CD206 predominantly localised in intracellular vesicle membranes , while F4/80 localised in the plasma membrane ( most highly at sites of lamellipodia and filopodia ) ( Figure 3k–k’ ) . A previous study demonstrated a tight relationship between macrophages and nephrons ( Rae et al . , 2007 ) , the latter of which develop and mature deep within the nephrogenic zone ( Figure 3—figure supplement 5 ) . In accordance with this finding , we observed macrophages in contact with the basement membrane of developing nephrons ( Figure 3—figure supplement 5; Video 6 ) . Macrophages also frequently interacted with blood vessels that wrapped around the developing nephron tubules ( Figure 3—figure supplement 5; Video 6 ) . These data show that F4/80+CD206+ macrophages in the cortical nephrogenic zone are mainly perivascular throughout the embryonic period of renal organogenesis . These macrophages align with and wrap around newly forming cortical interstitial blood vessels . In the developing lung , a subpopulation of galectin3+ ( Gal3+ ) myeloid cells are dispersed alongside yolk sac-derived F4/80high macrophages ( Tan and Krasnow , 2016 ) . Gal3+ cells colonise the lung from E12 . 5 onwards and are speculated to be foetal liver-/monocyte-derived macrophages ( Tan and Krasnow , 2016 ) . We stained Csf1rEGFP embryos with anti-Gal3 and demonstrated that the developing kidney also contained a subpopulation of Gal3+Csf1rEGFPEGFP+ myeloid cells ( Figure 4a ) . Gal3+ cells were largely distinct from CD206+ macrophages , although a small percentage of cells co-expressed both proteins ( Figure 4b–d ) . The proportion of myeloid cells that were Gal3+ increased substantially later in development , while the relative proportion of CD206+ cells decreased ( Figure 4d; Figure 4—figure supplement 1 ) . At E17 . 5 , 14 . 4% ± 2 . 2% ( mean ±SEM ) of Gal3+ myeloid cells in the cortical nephrogenic zone were being carried within blood vessels at the point of fixation ( Figure 4e–f’; Figure 4—figure supplement 2 ) . Gal3+ cells were morphologically spherical in comparison to the irregular morphology of the CD206+ macrophages ( Figure 4g ) . In later kidney development , tubular lumens also strongly stained for Gal3 ( Figure 4—figure supplement 1 ) , in agreement with its expression pattern in the developing human kidney ( Winyard et al . , 1997 ) . As the localisation and morphology of Gal3+ macrophages differed from F4/80+CD206+ macrophages , we next used single-cell RNA sequencing data to explore the heterogeneity between their transcriptional landscapes . We gathered single-cell RNA sequencing data from E18 . 5 kidney cells using the Smart-seq2 protocol ( Picelli et al . , 2013 ) . We identified kidney macrophages using principle component analyses ( PCA ) based on the expression of 48 genes chosen based on relevant literature ( including macrophage markers ) ( Supplementary file 1 ) . By performing PCA specifically on these macrophages , distinct clusters of F4/80highCD206highGal3low and Gal3highF4/80lowCD206low macrophage were identified ( Figure 4—figure supplement 3 ) , substantiating our immunofluorescence observations . The mRNA signatures of F4/80highCD206high cells , but not of Gal3high cells , were consistent with their being mature macrophages ( signatures based on Mass et al . , 2016 ) ( Figure 4h–i ) . PANTHER gene over-representation testing ( http://www . pantherdb . org/ ) demonstrated that genes associated with biological processes such as immune response ( false discovery rate [FDR] corrected p=1 . 22×10−4 ) and defence response ( FDR corrected p=1 . 25×10−4 ) were enriched in the top 1% of genes expressed by Gal3high cells ( Figure 4j; Supplementary file 2 ) , suggesting that Gal3high macrophages are primed for pathogenic invasion . The top 1% of genes expressed by F4/80highCD206high cells were over-represented in biological processes such as pinocytosis ( FDR corrected p=3 . 8×10−3 ) , endocytosis ( FDR corrected p=1 . 6×10−8 ) , endosomal transport ( FDR corrected p=2 . 1×10−3 ) , and phagocytosis ( FDR corrected p=1×10−2 ) ( Figure 4k; Supplementary file 3 ) . The top 1% of genes expressed in both F4/80highCD206high cells and Gal3high cells were also over-represented for several cellular components of the endosomal-lysosomal system , consistent with their roles as professional phagocytes ( Supplementary file 4–5 ) . Notably , biological processes relating to tissue development were also enriched in the F4/80highCD206high macrophages ( Supplementary file 3 ) . Of these processes , several related to vascularisation , such as vasculature development ( FDR corrected p=2 . 7×10−2 ) and blood vessel morphogenesis ( FDR corrected p=3 . 9×10−2 ) . Gene expression by Gal3high cells was not associated with development-related processes ( Supplementary file 2 ) . These results suggest that F4/80highCD206high macrophages may act as pro-developmental cells during kidney development , consistent with a previous report ( Rae et al . , 2007 ) . We next explored whether distinct populations of CD206+ and Gal3+ immune cells were also present in the developing human kidney by analysing publicly available single-cell RNA sequencing data on GUDMAP ( gestational week 14–18; Lindström et al . , 2018a ) . In agreement with our results in the mouse , CD206 ( MRC1 in the human ) and Gal3 ( LGALS3 in the human ) were differentially expressed in immune cell types of the human embryonic kidney ( Figure 5a–i ) . 90% of MRC1+ cells were macrophages ( group 11 ) ( Figure 5h ) , whereas LGALS3+ immune cells were distributed between the macrophage cluster ( group 11; 58% of LGALS3+ immune cells ) and a distinct leukocyte cluster ( group 10; 42% of LGALS3+ immune cells ) ( Figure 5i ) . The group 10 immune cells specifically expressed L-selectin ( SELL ) ( Figure 5g ) , which functions to tether leukocytes to endothelium ( Stein et al . , 1999 ) . F4/80 ( EMR1 in the human ) was not expressed by macrophages in the foetal human kidney ( Figure 5j ) , consistent with a previous report showing EMR1 is not expressed by human macrophages ( Hamann et al . , 2007 ) . Collectively , these results show that the developing kidneys of the mouse and human contain heterogeneous populations of immune cells . In the developing mouse kidney , Gal3high myeloid cells often travel via the renal vasculature and intermingle with mature F4/80highCD206high macrophages . Both Gal3high and F4/80highCD206high cells were enriched for mRNAs associated with professional phagocytes , but only F4/80highCD206high cells were enriched for mRNAs associated with vascular development . Given that many macrophages in the kidney interact with the interstitial vasculature , we hypothesised that macrophage localisation in the interstitium depended on the presence of blood vessels . To test this , we pharmacologically inhibited vascular development in cultured E12 . 5 kidneys ( using three different pan-VEGF inhibitors: sunitinib , vatalanib , and semaxanib ) and compared macrophage localisation . Vascular density was markedly reduced in the treated kidneys ( Figure 6a–c ) , but macrophage localisation ( in/out of the interstitium ) did not differ between groups ( one-way ANOVA; p=0 . 7142; n = 4–9 kidneys per group ) ( Figure 6d ) . Notably , macrophage density was reduced in vascular-depleted kidneys ( Figure 6e ) and was positively correlated with vascular density when assessing all groups ( r = 0 . 71; p<0 . 0001 ) ( Figure 6f ) and even when assessing only control groups ( r = 0 . 74; p=0 . 004 ) ( Figure 6g ) . Vascular-depletion did not alter other aspects of kidney development ( Figure 6—figure supplement 1 ) and an assistant counting blind-coded samples verified the macrophage density count differences between groups ( Figure 6—figure supplement 1 ) . These data suggest that endothelial-derived signals are not required for macrophage navigation in the renal interstitium . Renal macrophages engulf cells in the developing mouse and human kidney ( Camp and Martin , 1996; Erdosova et al . , 2002 ) ; however , the identities of the phagocytosed cells are unknown . We observed F4/80+ macrophages engulfing erythrocytes and dying endothelial cells in the kidney ( Figure 7a–b; Video 7 ) , demonstrating that , as well as spatially associating with the vasculature , kidney macrophages directly interact with blood and vascular components during development . Based on kidney macrophage localisation and gene expression ( Figures 1–2; Video 5; Figure 7c–d ) , as well as their known abilities in regulating vascular formation ( Picelli et al . , 2013; Fantin et al . , 2010; Rymo et al . , 2011 ) , we hypothesised that macrophages promote renal vascularisation . To test whether macrophages are required for the renal vasculature to develop normally , we depleted macrophages in cultured kidney explants using an anti-Csf1r mAb blocking antibody ( M279 ) . Csf1r is essential for macrophage proliferation , migration , and survival ( Dai et al . , 2002; Mouchemore and Pixley , 2012 ) . E12 . 5 kidneys were treated for 72 hr with either 20 µg/ml of anti-Csf1r or anti-rat IgG as a control , and macrophages were robustly depleted ( Figure 7e–g ) . Consistent with previous studies , anti-Csf1r treated kidneys were smaller than controls ( p=0 . 0046; Figure 7h; Alikhan et al . , 2011; Rae et al . , 2007; Sauter et al . , 2014 ) . The area covered by CD31+ endothelium per field of view did not significantly differ between groups ( p=0 . 077; Figure 7i ) ; however , macrophage-depleted kidneys had higher numbers of isolated , unconnected endothelial structures ( p=0 . 011; Figure 7j ) and the average size of CD31+ structures were reduced relative to controls ( p=0 . 046; Figure 7k ) . These data demonstrate that macrophages support production and/or maintenance of endothelial cross-connections in the developing kidney . In this report , we characterised macrophage arrival , localisation , and heterogeneity in the developing mouse kidney and show that macrophages support the assembly of the kidney and its vasculature . Kidneys depleted of macrophages contained discontinuous endothelial structures , while control-treated kidneys had characteristic net-like vascular plexuses . These data are consistent with macrophages having the capacity to promote vascular anastomoses , as has been described in other biological settings ( Fantin et al . , 2010; Liu et al . , 2016 ) . Blood vessels provide developing organs with oxygen and essential nutrients to promote growth ( Chung and Ferrara , 2011 ) . Studies have linked foetal macrophages with enhanced branching morphogenesis , nephrogenesis , and growth during kidney development ( Alikhan et al . , 2011; Rae et al . , 2007; Sauter et al . , 2014 ) and our results showing macrophages promote renal vascularisation may partly explain these links . While we demonstrate a close relationship between macrophages and developing renal blood vessels , important questions remain to be addressed . Future studies should examine the specific mechanisms by which macrophages mediate kidney vascularisation and should explore other potential roles for renal perivascular macrophages in the embryo . In the adult kidney , perivascular macrophages survey vascular transport for small immune complexes ( Stamatiades et al . , 2016 ) . As the embryo is thought to be relatively sterile ( Tissier , 1900 ) , there are likely few immune complexes transported via the vasculature during development , but perivascular macrophages may nevertheless survey the bloodstream for factors such as maternal-derived antibodies that have crossed the placental barrier ( Morphis and Gitlin , 1970 ) . Further , we demonstrated that macrophages can engulf erythrocytes in the developing kidney , suggesting a possible role for renal perivascular macrophages in clearing cells carried via the vasculature . One way that macrophages shape developing tissues is via their roles in cell clearance ( Wood and Martin , 2017 ) ; for example , macrophages clear excess neurons in the developing brain and spinal cord ( Cuadros et al . , 1993 ) and engulf dying cells in embryonic footplates to facilitate digit formation ( Wood et al . , 2000 ) . Our findings indicate that macrophages play a similar role at the beginning of kidney organogenesis , as they prune the rostral domain of the early nephron progenitor population . Our study failed , however , to expose the precise mechanism ( s ) by which rostral nephron progenitors are cleared . We observed Six2+ nuclei of rostral nephron progenitors within the cell bodies of some macrophages , suggesting active phagocytosis of these cells , but it is unclear whether other mechanisms also contribute to their clearance; for instance , macrophages may generate signals and/or act as cellular chaperones to promote the caudal migration of these cells . A trophic role for macrophages in kidney development was first suggested by Rae et al . , 2007 , who demonstrated that treating cultured kidney explants with Csf1 , a macrophage mitogen , results in enhanced renal growth and branching morphogenesis associated with increased macrophage numbers . Along with this finding , studies in the postnatal mouse revealed that increased activation of the Csf1/Csf1r pathway results in increased kidney weight and volume ( Alikhan et al . , 2011 ) , whereas blockade of this pathway in the adult results in reduced kidney weight ( Sauter et al . , 2014 ) . Our results add to these findings by showing that , in the absence of macrophages in vivo , the beginning of kidney development is delayed . Collectively , these studies suggest that macrophages act to enhance growth and maturation of renal tissue across the lifespan . The growth and maturation of renal organoids using current differentiation protocols is limited . Since blood vessels and phagocytic cells are important for renal organogenesis ( Alikhan et al . , 2011; Rae et al . , 2007; Sauter et al . , 2014 ) , our results may have significant implications for the generation of kidney organoids . Endothelial cells , monocytes , and macrophages are thought to arrive in the developing kidney from extra-renal sources ( Hoeffel et al . , 2015; Kobayashi et al . , 2014 ) and future studies might usefully compare organoid maturation when macrophages and blood vessels are exogenously added to these systems . In conclusion , we show that foetal kidney macrophages are a multifunctional cell type that frequently interact with newly forming renal blood vessels and encourage organogenesis . These results may inform future macrophage-based strategies for the prevention and treatment of neonatal and adult kidney diseases . Wild-type embryonic tissues that were used for descriptive studies and kidney explant culture experiments were obtained from outbred CD-1 mice killed by qualified staff of a UK Home Office–licensed animal house following guidelines set under Schedule 1 of the UK Animals ( Scientific Procedures ) Act 1986 . Experiments were performed in accordance with the institutional guidelines and regulations as set by the University of Edinburgh . The morning of vaginal plug discovery was considered as embryonic day ( E ) 0 . 5 and staging was confirmed based on kidney and limb bud morphology . Transgenic mice used in this study: MacGreen ( Csf1rEGFP ) reporter mice have previously been described ( Sasmono et al . , 2003 ) . Transgenic Csf1rEGFPEGFP+ embryos were identified based on limb bud GFP fluorescence using a Zeiss Axioscope A1 microscope . Csf1ricre and RosaDTA mice were bred onto the C57BL/6JOlaHsd genetic background and maintained as heterozygotes . Staging of ~E11 . 5 Csf1ricreicre+RosaDTA embryos was based on somite pair ( sp ) counting and 44–49 sp embryos were used for analyses in Figure 1 . At E11 . 5 , Csf1ricreicre+RosaDTA embryos were identified using flow cytometry to identify CD45+ cells and immunofluorescence to identify F4/80+ cells . Csf1rEGFP , Csf1ricre , and RosaDTA transgenic mice were maintained at the University of Edinburgh according to locally approved procedures . E18 . 5 Six2EGFP/Cre mouse embryos were used for mouse single-cell RNA sequencing experiments ( for details of this mouse line , see Kobayashi et al . ( 2008 ) ) . These mice were maintained at Bar-Ilan University according to locally approved procedures . Kidneys were isolated using previously described methods ( Davies , 2010 ) . Kidney explants were cultured on sterile 24 mm polyester membrane inserts with 0 . 4 µm pores ( Transwells; Corning 3450 ) in 1 . 5 ml of kidney culture medium ( KCM; Minimum Essential Medium Eagle [Sigma , M5650] supplemented with 1% penicillin/streptomycin [Sigma , P4333] and 10% foetal calf serum [Invitrogen , 10108165] ) per well in 6-well plates . Kidneys were grown at 37°C in a 5% CO2 environment . Kidneys were cultured for indicated times and the medium was changed every 48 hr , unless otherwise stated . Whole-mount samples were fixed with 4% paraformaldehyde ( PFA ) for 10–60 mins ( depending on sample size ) . After PFA fixation , samples were washed in 1x phosphate buffered saline ( PBS; 1 × 30 mins ) and then dehydrated in a methanol: dH2O serial dilution ( 20% , 40% , 60% , 80% , and then 100%–15 mins per step ) . After dehydration , samples were either stored at −20°C or directly processed . Samples were rehydrated in a methanol: dH2O serial dilution ( 80% , 60% , 40% , 20% , and then 0%–15 mins per step ) . Kidneys were rinsed in PBS ( 3 × 30 mins ) and blocked with 1 x PBS with 5% bovine serum albumin ( BSA; Sigma , A9647 ) and 10% donkey serum ( Sigma , D9663 ) for 1 hr at room temperature or overnight at 4°C . Kidneys were then incubated with primary antibodies , which were diluted in 50% blocking buffer with 50% 1x PBS , overnight at 4°C . Kidneys were rinsed in 1x PBS ( 3 × 1 hr or overnight at 4°C ) and were then incubated with secondary antibodies ( in 50% blocking buffer with 50% 1x PBS ) overnight at 4°C . For details regarding the antibodies used , see Supplementary file 6 . Following incubation with secondary antibodies , kidneys were washed in 1x PBS ( 4 × 1 hr ) and mounted onto glass slides in mounting medium ( Vectashield; Vectorlabs , H1000 ) . Samples were covered by cover-slips , which were stuck in place on the slide using nail varnish . Samples were optically cleared when deep imaging through thick samples was required . Samples were first fixed in 4% PFA for 1–2 hr . They were washed in 1x PBS ( 2 × 1 hr ) and then dehydrated in a methanol: dH2O serial dilution ( 20% , 40% , 60% , 80% , and then 100%–15 mins per step ) . After a 100% methanol wash for 1 hr , samples were incubated with Dent’s bleach ( methanol: dimethyl sulfoxide [DMSO]: 30% hydrogen peroxide; 4:1:1 ) for 2 hr , then stored in 100% methanol at −20°C . To continue the protocol , samples were rehydrated in a methanol: dH2O serial dilution ( 80% , 60% , 40% , 20% , 0%–15 mins per step ) . They were washed with 1x PBS with 0 . 2% Triton X-100 ( 2 × 30 mins ) . They were then permeabilised for 4 hr using 1x PBS with 0 . 2% Triton X-100 , 300 mM glycine , and 20% DMSO at room temperature . Samples were then blocked with 1x PBS with 0 . 2% Triton X-100 , 3% donkey serum , and 10% DMSO overnight at 4°C . They were then incubated with primary antibodies , which were diluted in 50% blocking buffer with 50% 1x PBS at 4°C for 1–5 days on a rocker . They were then washed in 1x PBS-Tween for 3 × 2 hr . Samples were incubated in secondary antibodies for 24 hr at 4°C diluted in 50% blocking buffer with 50% 1x PBS . Subsequently , samples were washed with 1x PBS-0 . 2% Tween for 3 × 2 hr and then left in 1x PBS overnight at 4°C . They were dehydrated using a methanol: dH2O serial dilution ( 15-mins per step; 20% , 50% , 75% , and 100% methanol ) . They were placed in a glass vial containing 50% benzyl alcohol/benzyl benzoate ( BABB ) with 50% methanol until they sank to the bottom of the vial and were then cleared in 100% BABB until transparent . After clearing , samples were placed on a slide in a drop of BABB and covered with a glass cover-slip prior to imaging . Erythrocytes were observed in BABB cleared tissue as this clearing method does not remove heme from haemoglobin and because heme acts as a major chromophore under visible light ( Horecker , 1943; Munro et al . , 2017b; Yokomizo et al . , 2012 ) . At E9 . 5 and E10 . 5 , numbers of F4/80+ macrophages were manually counted in individual z-planes through cleared caudal regions of mouse embryos using plugin → analyse → cell counter in ImageJ . To calculate macrophage density in the caudal region of embryos , regions devoid of tissue and regions that included nephrogenic cells were traced using the Freehand selections tool on ImageJ and were subtracted from the overall area per field of view . Macrophage density ( F4/80 cells per mm2 ) was calculated by dividing macrophage number by the area of the caudal tissue region in each field of view . Macrophage density within the Six2+ nephrogenic populations were scored in the same way , with the area covered by Six2+ cells on each field of view being calculated using the Freehand selections tool . To calculate macrophage density along the rostro-caudal axis of the metanephric mesenchyme at E10 . 5 , we used the Straight-line tool on ImageJ to draw a line along the length of the main population of Six2+ cells and measured the length of this line . We divided this line by four to define each quarter of the population along its rostro-caudal axis and counted macrophage numbers per quarter . We then related macrophage numbers ( calculated using plugin → analyse → cell counter in ImageJ ) to the area of each quarter to calculate macrophage density ( F4/80 cells per mm2 ) . We also quantified macrophage density within isolated rostral populations of Six2+ cells at E10 . 5 . Macrophage density was represented on a linear 3-colour scale , with blue indicating the lowest density , black indicating values in the middle , and red indicating the highest density values . Macrophage density was calculated in the E11 . 5 peri-Wolffian mesenchyme and metanephric mesenchyme using ImageJ . The metanephric mesenchyme was defined and drawn using the freehand selection tool by a kidney development researcher blinded to the experimental purpose . To eliminate potential bias based on macrophage or vasculature localisation , only the Gata3 channel ( showing ureteric bud ) was shown to the blinded researcher . Peri-Wolffian mesenchyme and metanephric mesenchyme areas were calculated and macrophage numbers counted ( using plugin → analyse → cell counter in ImageJ ) to calculate relevant macrophage densities ( F4/80 cells per mm2 ) . To characterise the total rostro-caudal length of Six2+ populations in macrophage-depleted and control embryos we analysed confocal z-planes of caudal part embryos stained with anti-Six2 . The segmented line tool in ImageJ was used to calculate the total length of the Six2+ population from the most caudal to the most rostral Six2+ cell ( using analyse → measure in ImageJ ) . To characterise the area of rostral Six2+ clusters , we used the Freehand selections tool in ImageJ to draw around these cell clusters and used analyse → measure to define the area of these populations . Csf1r and F4/80 co-staining: Csf1rEGFPEGFP+ kidneys were fixed , immunostained , and cleared . An anti-GFP antibody was used to label Csf1rEGFPEGFP+ cells . Co-localisation of GFP+ and F4/80+ cells was quantified using plugin → analyse → cell counter in ImageJ . Quantification was based on assessments of 8 z-planes from an E14 . 5 Csf1rEGFPEGFP+ kidney . F4/80 and CD206 co-staining: Quantification was performed as described above using antibodies against F4/80 and CD206 . Quantifications were made at E11 . 5 , E14 . 5 , and E17 . 5; at E14 . 5 and E17 . 5 quantifications were based on macrophages within the cortical nephrogenic zone . Gal3 and CD206 co-staining: Quantification was performed as described above using antibodies against Gal3 and CD206 . Quantifications were made at E11 . 5 , E14 . 5 , and E17 . 5; at E14 . 5 and E17 . 5 quantifications were based on cells within the cortical nephrogenic zone . All images were generated using the Zeiss LSM800 confocal microscope , except from the time-lapse images , which were generated using the Nikon A1R confocal microscope . Objectives of 5-63x were used . Objective lenses were oil-immersed from 40x upwards . Images were analysed and processed using ImageJ ( FIJI ) and IMARIS ( version 8 . 3 . 1 ) . E11 . 5 Csf1rEGFPEGFP+ kidneys were explanted and cultured in Transwells as described above . A 6-well plate containing the Transwells was placed in an imaging chamber and the kidneys were grown at a temperature of 37°C in a 5% CO2 environment . Isolectin-B4 ( I32450; ThermoFisher ) was added to the culture medium at a concentration of 1:1000 . For the video showing the entire kidney , images were taken every 15 mins . For the video showing a small region of the kidney , images were taken every five mins . Macrophage localisation was quantified in the E17 . 5 nephrogenic zone by co-staining for F4/80 , collagen IV , and Six2 . Collagen IV marks blood vessel basement membranes in the developing kidney ( Munro et al . , 2017b ) . To quantify macrophage localisation , we analysed macrophage location in 3D by scrolling through z-planes . Quantifying macrophage localisation in individual z-planes was necessary because macrophages sitting above the cap mesenchyme in 3D may have appeared within the cap mesenchyme in flattened 2D images - the same was true for macrophage localisation around blood vessels . The alignment of perivascular macrophages was measured by drawing a line along the longest axis of the macrophage using the straight-line tool in ImageJ . The angle of this line was the measured using analyse → measure in ImageJ . The same measurement was performed for the underlying blood vessel , with the straight line being drawn along the blood vessel up to the site of its splitting . The sphericity of Gal3+ and CD206+ cells in the E14 . 5 kidney was measured using IMARIS ( version 8 . 3 . 1 ) . Cells were surface-rendered and non-cellular rendered objects manually removed . IMARIS automatically calculated the sphericity of each rendered cell . E12 . 5 kidneys were cultured on Transwell filters in 1 . 5 ml KCM for 3 days with either 20 µg/ml anti-Csf1r mAb blocking antibody ( M279 ) or 20 µg/ml anti-rat IgG ( as a control ) . To retain anti-Csf1r and anti-rat IgG ( control ) within the medium , KCM was not refreshed during the 3 day culture period . After 3 days , cultured kidneys were directly fixed in methanol and processed for immunofluorescence ( as described in the ‘Whole-mount immunofluorescence’ section , after the fixation step ) . Macrophage numbers were quantified using the add spots tool on the F4/80 channel in IMARIS ( version 8 . 3 . 1 ) . Spot number per kidney was defined as the total macrophage number per kidney . To calculate kidney area , kidneys were drawn using the Freehand selections tool in ImageJ and the area covered by each kidney was measured . To calculate the CD31+ area per field ( % ) , the CD31 channel was prepared for thresholding using process → filters → median [radius = 2 pixels] then image brightness/contrast was adjusted in ImageJ . Images were thresholded using image → adjust → threshold ( default thresholding ) . After thresholding , the percentage of the thresholded area per field of view was measured and defined as the vascular density ( % CD31+ area per field ) . The numbers of isolated CD31+ structures per field were then quantified by a blinded counter that was given code samples using plugin → analyse → cell counter in ImageJ . The number of CD31+ structures per field was divided by the total CD31+ area per field to give the average % of the field covered by each CD31+ structure . E12 . 5 kidneys were cultured on Transwell filters for 3 days in either 1 . 5 ml of KCM-only ( control ) , KCM with vascular development inhibitors , or KCM with DMSO ( AppliChem , A3672-0100; vehicle control ) . Inhibitor concentrations were calculated based on half-maximal inhibitory concentration ( IC50 ) values against Vegfr2 ( using values from the IUPHAR/BPS Guide to Pharmacology; http://www . guidetopharmacology . org/ ) . Vatalinib had previously been used at a concentration of 1 µM to inhibit vascular development in cultured kidneys ( Halt et al . , 2016 ) ; therefore , we used 1 µM of vatalinib in our experiments . 1 µM of vatalinib is at a concentration that is 47 . 62 times greater than its IC50 against Vegfr2 . To calculate concentrations to use for the other inhibitors , we multiplied sunitinib’s and semaxanib’s IC50s against Vegfr2 by 47 . 62 ( for consistency ) . Based on these calculations , the inhibitor concentrations used in the KCM were 1 µM of vatalinib , 1 . 076 µM of sunitinib , and 9 . 524 µM of semaxanib . As a vehicle control , DMSO was added at the same volume as the highest volume used for the inhibitors . After 3 days of culture , cultures were stopped , and kidneys were directly fixed in methanol and processed for immunofluorescence . Data are mean ± standard error of the mean . Data that passed normality testing were analysed using parametric tests . Data that did not pass normality tests were analysed using non-parametric tests ( specific tests used are indicated in relevant text ) . When two experimental groups were being compared , t-tests or the non-parametric equivalent were used . When more than two groups ( which were normally distributed ) were compared , one-way ANOVAs were performed with post-hoc testing then being used to compare differences between individual groups . All p-values were based on two-tailed comparisons . GraphPad ( version 5 ) was used for statistical testing and graph preparation . IMARIS ( version 8 . 3 . 1 ) and Adobe Premiere Pro CC ( 2015 ) were used to prepare videos and Adobe Illustrator CC ( 2015 ) was used to prepare figures .	The kidneys clean our blood by filtering out waste products while ensuring that useful components , like nutrients , remain in the bloodstream . Blood enters the kidneys through a network of intricately arranged blood vessels , which associate closely with the ‘cleaning tubes’ that carry out filtration . Human kidneys start developing during the early phases of embryonic development . During this process , the newly forming blood vessels and cleaning tubes must grow in the right places for the adult kidney to work properly . Macrophages are cells of the immune system that clear away foreign , diseased , or damaged cells . They are also thought to encourage growth of the developing kidney , but how exactly they do this has remained unknown . Munro et al . therefore wanted to find out when macrophages first appeared in the embryonic kidney and how they might help control their development . Experiments using mice revealed that the first macrophages arrived in the kidney early during its development , alongside newly forming blood vessels . Further investigation using genetically modified mice that did not have macrophages revealed that these immune cells were needed at this stage to clear away misplaced kidney cells and help ‘set the scene’ for future development . At later stages , macrophages in the kidney interacted closely with growing blood vessels . As well as producing molecules linked with blood vessel formation , the macrophages wrapped around the vessels themselves , sometimes even eating cells lining the vessels and the blood cells carried within them . These observations suggested that macrophages actively shaped the network of blood vessels developing within the kidneys . Experiments removing macrophages from kidney tissue confirmed this: in normal kidneys , the blood vessels grew into a continuous network , but in kidneys lacking macrophages , far fewer connections formed between the vessels . This work sheds new light on how the complex structures in the adult kidney first arise and could be useful in future research . For example , adding macrophages to simplified , laboratory-grown ‘mini-kidneys’ could make them better models to study kidney growth , while patients suffering from kidney diseases might benefit from new drugs targeting macrophages .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2019	Macrophages restrict the nephrogenic field and promote endothelial connections during kidney development
A 3 . 3 MDa macromolecular cage between two Escherichia coli proteins with seemingly incompatible symmetries–the hexameric AAA+ ATPase RavA and the decameric inducible lysine decarboxylase LdcI–is reconstructed by cryo-electron microscopy to 11 Å resolution . Combined with a 7 . 5 Å resolution reconstruction of the minimal complex between LdcI and the LdcI-binding domain of RavA , and the previously solved crystal structures of the individual components , this work enables to build a reliable pseudoatomic model of this unusual architecture and to identify conformational rearrangements and specific elements essential for complex formation . The design of the cage created via lateral interactions between five RavA rings is unique for the diverse AAA+ ATPase superfamily . Virtually every aspect of cellular function relies on a AAA+ ATPase machine as a key player ( Erzberger and Berger , 2006; Snider et al . , 2008 ) . The name of this superfamily , ‘ATPase Associated with Diverse Cellular Activities’ , reflects this remarkable versatility . Widespread among bacteria and archea , members of the MoxR family of AAA+ ATPases are important modulators of multiple stress tolerance pathways ( Snider et al . , 2006; Wong and Houry , 2012 ) . The best characterized MoxR representative , the RavA protein , has been recently associated to oxidative stress , antibiotic resistance and iron-sulfur cluster assembly in Escherichia coli ( Babu et al . , 2014; Wong et al . , 2014 ) . Furthermore , it is involved in acid stress and nutrient stress responses via its interaction with the acid stress-inducible lysine decarboxylase LdcI , which buffers the bacterial cytoplasm by transforming lysine into cadaverine while consuming intracellular protons and producing CO2 ( Sabo et al . , 1974; Soksawatmaekhin et al . , 2004; Snider et al . , 2006; Wong and Houry , 2012 ) . We recently demonstrated that RavA prevented binding of LdcI to its potent inhibitor , the stringent response alarmone , ppGpp ( El Bakkouri et al . , 2010 ) . Thus the LdcI–RavA complex maintains LdcI activity even if the bacterium experiences both acid stress and starvation , as it is often the case in the host stomach through which enterobacteria transit before reaching the bowel where the pathogenesis typically occurs ( El Bakkouri et al . , 2010 ) . The crystal structures of both individual components of the complex provided important insights into their function ( El Bakkouri et al . , 2010; Kanjee et al . , 2011 ) . However , the structural principles of the LdcI–RavA interaction are puzzling . Indeed , RavA is a 300 kDa lily-shaped hexameric ring ( El Bakkouri et al . , 2010 ) , whereas LdcI is a 800 kDa toroid composed of two pentameric rings stacked together back-to-back ( Kanjee et al . , 2011 ) . But how can a hexamer bind a double pentamer ? The strength of the interaction ( Kd of about 20 nM ) between these two symmetry mismatched proteins and the resulting mass of 3 . 3 MDa inferred from analytical ultracentrifugation came out as a surprise ( Snider et al . , 2006; El Bakkouri et al . , 2010 ) . The ∼35 Å resolution of our first negative stain electron microscopy ( EM ) reconstruction was however so low that the assembly strategy of the complex , seemingly composed of two LdcI decamers and five RavA hexamers , appeared even more enigmatic ( Snider et al . , 2006 ) . We biochemically showed that the foot of RavA , inserted into the discontinuous triple helical bundle protruding as a leg from the AAA+ ATPase core , is necessary and sufficient for LdcI binding , albeit with a three orders of magnitude lower affinity . We therefore called this foot domain LARA for ‘LdcI associating domain of RavA’ ( El Bakkouri et al . , 2010 ) . Yet , although speculations were tempting , correct interpretation of the structural bases of the LdcI–RavA interaction was impossible . Hence it was imperative to provide higher resolution insights into the design principles of this intriguing architecture . Here we present an 11 Å resolution cryo-electron microscopy ( cryoEM ) map of the LdcI–RavA complex and a 7 . 5 Å resolution cryoEM map of the LdcI-LARA complex , which together enable unambiguous flexible docking of the crystal structures , finally fitting together the pieces of the jigsaw . The astounding LdcI–RavA assembly is reminiscent of a symmetrical floral pattern , featuring two parallel five-petal blossoms of LdcI festooned by a garland of five RavA lilies . The entire complex possesses the dihedral fivefold symmetry of the lysine decarboxylase and a large central cavity of 3 × 106 Å3 ( Figure 1 , Figure 1—figure supplements 1 and 2; Video 1 ) . Upon complex formation , the RavA hexamer loses its sixfold circular symmetry . The compact outward-pointing hub formed by the six AAA+ modules remains virtually unchanged and the ATP binding sites preserved , whereas the legs massively rearrange in order to comply with the dihedral symmetry of the whole assembly ( Figure 2A , Figure 2—figure supplement 1; Video 2 ) . Four protomers of each of the five RavA hexamers are involved in the interaction with the LdcI , while the remaining two legs of each hexamer mediate the RavA–RavA interaction at the equator of the complex ( Video 2 ) . 10 . 7554/eLife . 03653 . 003Figure 1 . Cage-like architecture of the LdcI–RavA complex ( 11 Å resolution ) . ( A ) Top view with LdcI facing the reader , ( B ) Side view with RavA facing the reader . For this RavA hexamer , LARA domain positions are indicated by ellipses ( solid black for the two LARA domains interacting with the inner LdcI rings from above , dotted black for the two LARA domains interacting with the outer LdcI rings from underneath and invisible from this orientation , solid dark blue for the LARA domains interacting equatorially with the triple helical domains of adjacent RavA monomers ) . ( C ) Side cut-away view . Complex dimensions are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 03653 . 00310 . 7554/eLife . 03653 . 004Figure 1—figure supplement 1 . Structures of the individual RavA hexamer and the LdcI decamer . ( A ) Pseudoatomic model of the RavA hexamer based on rigid fitting of the crystal structure of the RavA monomer into the negative stain EM map of the RavA-ADP hexamer ( El Bakkouri et al . , 2010 ) . Each monomer is colored differently . One of the RavA protomers is colored to highlight its domain composition . ( B ) Crystal structure of the D5-symmetrical LdcI decamer ( Kanjee et al . , 2011 ) . ppGpp is shown as black spheres . DOI: http://dx . doi . org/10 . 7554/eLife . 03653 . 00410 . 7554/eLife . 03653 . 005Figure 1—figure supplement 2 . CryoEM analysis of the LdcI–RavA cage . ( A ) CryoEM micrograph of LdcI–RavA displaying particle boxing . ( B ) Correspondence between projections of the 3D model and particle class averages , illustrating the reliability of the 3D reconstruction . ( C ) Angular distribution for particles used for the final 3D reconstruction shown for the asymmetric unit . ( D ) Gold-standard FSC curves calculated for unmasked ( blue ) and soft shaped masked ( red ) maps , indicating the resolution of 14 Å and 11 Å respectively according to FSC = 0 . 143 criterion . ( E ) Comparison between the initial negative stain EM map ( Snider et al . , 2006 ) , the cryoET sub-tomogram average map used as an initial model for refinement in this study and the final 11 Å cryoEM map . DOI: http://dx . doi . org/10 . 7554/eLife . 03653 . 00510 . 7554/eLife . 03653 . 006Video 1 . Overview of the LdcI–RavA cage-like structure . DOI: http://dx . doi . org/10 . 7554/eLife . 03653 . 00610 . 7554/eLife . 03653 . 007Figure 2 . The structural organization of the LdcI–RavA cage . ( A ) The RavA hexamer is represented as two triskelia . The pseudoatomic model of the RavA hexamer in the context of the LdcI–RavA complex is superimposed with the isolated RavA negative stain EM map ( 25 Å resolution , El Bakkouri et al . , 2010 ) to show the conformational changes of the RavA legs induced by LdcI binding . ( B–G ) The RavA loop at the beginning of the LARA domain ( amino acids 329–360 ) is shown as a broken line . ( B ) Schematics of LdcI–RavA interaction with RavA . ( C ) CryoEM map and pseudoatomic model of LdcI–RavA ( 11 Å resolution ) . This particular orientation of the complex illustrates the origins of the close-up views ( E ) and ( G ) surrounded with a solid rectangle and a broken line rectangle , respectively . ( D ) Top view of the cryoEM map and pseudoatomic model of the LdcI-LARA complex ( 7 . 5 Å resolution ) . ( E ) Close-up view of the LdcI–RavA complex ( 11 Å resolution ) showing the LdcI-LARA interaction and arising from the bold rectangle in ( C ) . ( F ) Close-up view of LdcI-LARA ( 7 . 5 Å resolution ) in the same orientation as in ( E ) . The higher resolution of this 3D reconstruction enables a more precise fitting of individual crystal structures . ( G ) Close-up view of the equatorial RavA–RavA interaction via the triple helical bundles and the foot–leg interaction ( arising from the broken line rectangle in ( C ) ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03653 . 00710 . 7554/eLife . 03653 . 008Figure 2—figure supplement 1 . Conformation rearrangements of RavA induced by LdcI binding . ( A ) Pseudoatomic model of RavA from the LdcI–RavA complex superimposed with the negative stain EM map of RavA ( 25 Å resolution , El Bakkouri et al . , 2010 ) to show three different conformations of the RavA leg: the bent conformation induced by the binding to LdcI outer ring in blue/dark blue , the extended conformation induced by the binding to LdcI inner ring in yellow/red , and the intermediate conformation induced by equatorial RavA–RavA binding in green/dark green . The reorientation of the LARA domains upon interaction with LdcI is clearly visible . ( B–D ) Orthogonal views of the RavA crystal structure monomer ( gray , black ) superimposed with each of the three different conformations of RavA from the LdcI–RavA pseudoatomic model . Colors as in ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03653 . 00810 . 7554/eLife . 03653 . 009Figure 2—figure supplement 2 . CryoEM analysis of LdcI-LARA . ( A ) LdcI-LARA cryoEM micrograph displaying boxed particles . ( B ) Correspondence between projections of the 3D model and particle class averages showing the reliability of the 3D reconstruction . ( C ) Angular distribution of particles used for the final 3D reconstruction shown for the asymmetric unit . ( D ) Gold-standard FSC curves calculated for unmasked ( blue ) and soft shaped masked ( red ) maps , indicating the resolution of 8 . 8 Å and 7 . 5 Å respectively according to FSC = 0 . 143 criterion . DOI: http://dx . doi . org/10 . 7554/eLife . 03653 . 00910 . 7554/eLife . 03653 . 010Figure 2—figure supplement 3 . Insights into the LdcI-LARA interaction in LdcI–RavA and LdcI-LARA maps . ( A–C ) Superimposition of LdcI–RavA and LdcI-LARA maps highlighting the conservation of the LdcI-LARA structure . ( A ) Top view , ( B ) Close-up top view , ( C ) Close-up side view . ( D and E ) Fitting of the LdcI and LARA crystal structures into the LdcI-LARA map . The LARA domain is fitted as a rigid body , the LdcI monomer is fitted either ( D ) rigidly or ( E ) flexibly , which involved only subtle movements . ( F ) The difference between the rigid and the Flex-EM fit ot the LdcI monomer into the LdcI-LARA map presented in D and E is illustrated here in terms of the FSC curves between the corresponding model and the map ( blue for the rigid fit and red for the Flex-EM model ) . ( G ) Quality of the Flex-EM model of the LdcI–RavA complex is illustrated by FSC curves indicating differences between the LdcI–RavA map and ( i ) dark blue–a rigid fit of the crystal structure of LdcI ( Kanjee et al . , 2011 ) and of the symmetric pseudoatomic model of the RavA hexamer ( El Bakkouri et al . , 2010 ) into it , ( ii ) red—the present Flex-EM model of the entire LdcI–RavA complex , ( iii ) green—the present Flex-EM model of the LdcI-LARA region ( based on the LdcI-LARA map ) , ( iv ) light blue–the present Flex-EM model of the AAA+ domain core and the triple helical domains of RavA . DOI: http://dx . doi . org/10 . 7554/eLife . 03653 . 01010 . 7554/eLife . 03653 . 011Figure 2—figure supplement 4 . Mapping the interaction surfaces between RavA and LdcI . ( A ) List of designed mutations . Proteins , mutations , expected effects of mutations and aims of the experiments are indicated . Mutations are colored or highlighted with yellow stars , triangles and rectangles in a consistent way through the figure . ( B ) Sequence of the flexible region ( residues 329–360 ) located between the second helix of the triple helical bundle ( blue cylinder ) and the first β-strand of the LARA domain of RavA ( blue arrow ) . The designed mutations described in ( A ) are indicated . ( C ) Size exclusion chromatography analysis of the interaction between the different mutants probing LdcI–RavA interaction . The elution position of the molecular weight standards is shown on top . ( D ) Mutations presented in ( C ) localized on a cut-out of the 3D map and the pseudoatomic model of the LdcI-LARA complex . RavA mutations are highlighted as in ( A ) and LdcI mutations critical for the RavA binding ( E634K and Y697S ) are shown in green . LdcI is in light orange and the LARA domain in dark blue . The triple helical bundle of the entire RavA ( light blue ) is shown for clarity . ( E ) Size exclusion chromatography analysis of mutations probing the RavA–RavA interactions along the equator of the LdcI–RavA complex . ( F ) Mutations presented in ( E ) localized on a cut-out of the 3D map and the pseudoatomic model of the LdcI–RavA complex . DOI: http://dx . doi . org/10 . 7554/eLife . 03653 . 01110 . 7554/eLife . 03653 . 012Figure 2—figure supplement 5 . Comparison between triskelia of RavA and clathrin . ( A ) Two RavA triskelia are colored as in Figure 2 . ( B ) A clathrin trikelion is colored in blue with the tip domain in dark blue . Since no 3D EM map of the 28 triskelia clathrin mini-coat that has approximately the same size as the LdcI–RavA complex is available , the structure of the bigger hexagonal D6 barrel ( EMD-5119 ) is shown for comparison . DOI: http://dx . doi . org/10 . 7554/eLife . 03653 . 01210 . 7554/eLife . 03653 . 013Video 2 . Sequential representation of the LdcI–RavA complex formation . Morphing of an isolated RavA hexamer into the double triskelion conformation in the context of the LdcI–RavA complex . DOI: http://dx . doi . org/10 . 7554/eLife . 03653 . 013 Specifically , the RavA hexamers morph into two axially superimposed triskelia rotated 180° with respect to each other ( Figure 2B , C ) . The triskelion configuration and the loop ( amino acids 329–360 ) at the beginning of the LARA domain allow the optimal positioning of each LARA domain for interaction with LdcI subunits . Indeed , in the context of the complex , the two rings of each LdcI particle are not placed equivalently—one ring of each LdcI faces the exterior of the complex , while the second faces the cavity . Remarkably , one leg of the RavA triskelion stretches its foot out in order to present the LARA domain to the subunits of the inner ring of LdcI from above , while in the second leg the N-terminus of the loop sharply bends , allowing the foot to interact with the outer ring of LdcI from underneath ( Figure 1B , Figure 2B , C , E , Figure 2—figure supplement 1 ) . As a result of these intricate rotations , both feet appear to interact with the inner or the outer rings of LdcI in exactly the same way . This surprising finding is confirmed by the 7 . 5 Å resolution cryoEM map of the LdcI-LARA complex where each LdcI protomer binds one LARA domain , and which can be perfectly superimposed with the 3D reconstruction of the entire LdcI–RavA complex ( Figure 2D , Figure 2—figure supplement 2 , Figure 2—figure supplement 3 ) . At this resolution , the crystal structures of LdcI and LARA can be docked into the EM density unambiguously because the secondary structure elements are clearly resolved ( Figure 2D , Figure 2—figure supplement 3 ) . The LdcI decamer can be considered as barely affected by the LARA domain binding and all LdcI-LARA interaction sites are indeed equivalent ( Figure 2E , F , Figure 2—figure supplement 3; Video 3 ) . Noteworthy , it is the loop 329–360 of RavA which is responsible for the LdcI-LARA interaction . 10 . 7554/eLife . 03653 . 014Video 3 . From the LdcI–RavA cage structure to the interaction between the LdcI and the LARA domains in the higher resolution LdcI-LARA complex . Colors as in Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 03653 . 014 Compared to the minimal LdcI-LARA complex , the scaffold provided by the AAA+ hub and the triple helical bundles of the entire RavA increases the stability of the LdcI–RavA cage . Two legs of each triskelion place the LARA domains on the LdcI . The third legs interact laterally with their counterparts from the neighboring RavA via their triple helical bundles while the above-mentioned loop 329–360 of their LARA domain feet packs against the triple helical domain of a neighboring RavA hexamer , further stabilizing the complex architecture ( Figure 1B , Figure 2B , C , G ) . Mutational analysis was carried out to further probe the interaction between LdcI and RavA , and in particular the surface of the LdcI involved in the LARA domain binding , the versatile LARA domain loop and the equatorial interaction between triple helical bundles of RavA . The results are consistent with the resulting pseudoatomic model and identify residues essential for complex stability ( Figure 2—figure supplement 4 ) . Thus , critical residues in LdcI required for RavA binding were found to be E634 and Y697 , both part of a C-terminal β-sheet ( Figure 2—figure supplement 4A , C , D ) . The charged residues R347 and R348 and the hydrophobic residues I343 and F344 at the tip of the LARA domain loop are required for RavA binding to LdcI ( Figure 2—figure supplement 4A–D ) . Hence , the data suggest that the interaction between the two proteins is mediated both by charged and hydrophobic residues . Moreover , shortening the loop by deleting residues 329–335 , or constraining its N-terminus by introducting two prolines ( S331P , D332P ) , weakens or abolishes the LdcI–RavA complex formation ( Figure 2—figure supplement 4A–D ) . Finally , mutating residues R315 and E452 in the first and second helices of the RavA triple helical bundle , respectively , destabilized complex formation ( Figure 2—figure supplement 4A , E , F ) suggesting that the RavA–RavA leg–leg interaction is mediated by these two helices . To summarize , the same loop at the N-terminus of the LARA domain appears involved in ( 1 ) positioning the foot in respect to the rest of the leg so that to allow for equivalent interaction with both the inner and the outer LdcI rings , ( 2 ) binding to LdcI , and ( 3 ) binding to the triple helical bundle of the adjacent RavA ( Figure 2B , E–G ) . This loop therefore clearly emerges as the main determinant of the LdcI–RavA cage formation . While the 7 . 5 Å resolution of the current LdcI-LARA reconstruction precludes building a reliable pseudoatomic model of the 329–360 LARA domain loop based on the experimental EM density , its structural plasticity is obviously essential to fullful this role ( Figure 2—figure supplement 4 ) . The triskelion representation of the RavA hexamer inside the LdcI–RavA assembly brings out a parallel between the LdcI–RavA complex and the eukaryotic vesicle coats , in particular clathrin coats ( Harrison and Kirchhausen , 2010; Faini et al . , 2013; Figure 2—figure supplement 5 ) . Indeed , clathrin is a triskelion-shaped trimer with a compact hub jutting out three alpha-superhelical legs tipped with a domain that interacts with adapters and cargo molecules . Clathrin triskelia assemble in vitro into cages of pentagons and hexagons , the hubs representing the cage vertices and the helical legs intertwining between vertices in a two-fold symmetrical manner to create the edges . In the case of the LdcI–RavA cage , the LARA domain feet appear to be specifically evolved for the LdcI binding ( El Bakkouri et al . , 2010 ) . The equatorial vertices of the cage are provided by the hub made by the AAA+ domains of RavA , whereas the edges are created by the triple helical legs related by a twofold symmetry and intertwining between two adjacent AAA+ domain vertices . The fact that LdcI–RavA complex and eukaryotic vesicle coats have certain architectural principles in common is striking , because the LdcI–RavA architecture is remarkably different from all AAA+ ATPase assemblies described so far . Indeed , to our knowledge it is the only complex ( out of 30 released EM maps in EMDB and 43 oligomeric crystal structures in PDB in July 2014 ) composed of several laterally interacting AAA+ ATPase rings , and the only one enclosing a central cavity other than the cavity framed by the AAA+ modules assembled into a ring or a spiral . Why would the cell create such an exquisite and elaborate architecture simply to prevent ppGpp interaction with LdcI ? One may posit an explanation based on the markedly higher stability of the entire LdcI–RavA complex compared to the LdcI-LARA complex: the scaffold provided by the AAA+ hub and the triple helical legs of RavA is required to glue the LARA domain in place and effectively preclude ppGpp binding . Noteworthy , the LARA foot binds ∼30 Å away from the closest ppGpp binding pocket . Thus , contrary to our earlier prediction based on the low resolution negative stain EM map ( Snider et al . , 2006; El Bakkouri et al . , 2010 ) , the LARA domain does not appear to directly block the access of ppGpp to its binding pocket but rather to induce a local conformational change in this pocket reducing ppGpp affinity . Based on the fact that RavA is a MoxR AAA+ ATPase and that the MoxR family is known to have chaperone-like functions important for maturation or assembly of specific protein complexes , and taken into account the cage design of the LdcI–RavA complex , it would be tempting to suggest that this architecture may also fulfill yet another role . Along these lines , an involvement of RavA and its binding partner , a VWA-domain protein ViaA , in the Fe–S cluster assembly and particular respiratory pathways ( Babu et al . , 2014; Wong et al . , 2014 ) and a physical interaction of RavA , ViaA ( Wong et al . , 2014 ) and LdcI ( Erhardt et al . , 2012 ) with specific subunits of the highly conserved respiratory complex I ( Erhardt et al . , 2012; Wong et al . , 2014 ) were recently documented . Our structure of the exquisite LdcI–RavA cage substantiates a hypothesis that it may act as a chaperone , protecting substrates from denaturation or disassembly under acid stress conditions inside the central cavity , and spurs on further functional investigation . Here we unraveled the layout principles of the unique LdcI–RavA edifice , elucidated conformational rearrangements and specific elements essential for complex formation , and made a step towards general understanding how Nature elegantly solves the problem of the symmetry mismatch between individual components of protein complexes . RavA mutants were generated via the QuickChange method ( Stratagene , La Jolla , California ) . Proteins were expressed from p11 plasmid ( Zhang et al . , 2001 ) which encodes an N-terminal 6xHis-tag followed by the Tobacco Etch Virus ( TEV ) protease cleavage site . Plasmids were transformed into BL21- ( DE3 ) Gold pLysS strain and overexpression induced by addition of 0 . 5 mM Isopropyl β-D-1-thiogalactopyranoside ( IPTG ) at 30°C for 4 hr . All RavA mutants were purified as previously described ( El Bakkouri et al . , 2010 ) . Briefly , proteins were first purified by Ni-NTA affinity chromatography , then incubated with TEV protease overnight to remove the N-terminal His-Tag , and further purified via ion exchange chromatography using a MonoS column and by size exclusion chromatography on a Superdex S200 size exclusion column . LdcI mutants were also generated via the QuickChange method . Wild type and mutant proteins were expressed from pET22b plasmid which encodes a C-terminal uncleavable 24 amino acid linker containing 6xHis-tag . The plasmids were transformed into CF1693 strain bearing the pT7-322-Tetr plasmid to express the T7 polymerase . CF1693 contains a deletion of relA and spoT genes , which renders the strain not capable of generating ppGpp ( Xiao et al . , 1991 ) . Protein overexpression was induced by addition of 1 mM Isopropyl β-D-1-thiogalactopyranoside ( IPTG ) at 18°C for 20 hr . Proteins were purified on Ni-NTA resin , followed by ion exchange chromatography using a MonoQ column , and then by size exclusion chromatography on a Superdex S200 column . The isolated LARA domain was expressed and purified as described ( El Bakkouri et al . , 2010 ) . The rationale behind the mutants was provided by sequence alignment between three species of enterobacteria with the closest LdcI and RavA—Escherichia coli , Salmonella enterica and Enterobacter aerogenes—and by the pseudoatomic models produced in this study . For example , the fit of the crystal structures of the LdcI and the LARA domain of RavA into the LdcI-LARA reconstruction clearly shows that the β-strands 632–636 and 696–698 of the LdcI are involved in the interaction with the loop 329–360 of the LARA domain . In the case of the LdcI , both β-strands being strictly conserved between the three bacterial species compared , a charged ( E634 ) and a hydrophobic ( Y697 ) residue were chosen to probe the interaction and mutated into an oppositely charged ( E634K ) and a polar uncharged ( Y697S ) residue . Mutating the LdcI residues D638 and E681 , situated further away from the indicated β-sheet did not affect the RavA–LdcI interaction ( data not shown ) . In the case of the LARA domain of RavA , the loop 329–360 could not be modeled based on the current resolution of the EM map . Moreover , this loop appeared to be the main determinant of the LdcI–RavA interaction and to undergo major conformational rearrangements in comparison to the crystal structures of the RavA monomer ( see the main text and the methods below ) . Therefore the length of the loop ( deletion of amino acids 329–335 ) , the flexibility of its N-terminus ( S331P , D332P ) , as well as four residues in the middle of the loop which might be involved in the interaction ( two charged , R347 and R348 , and two hydrophobic , I343 and F344 ) were chosen for mutations ( Figure 2—figure supplement 4 ) . The four latter residues are shown only in the Figure 2—figure supplement 4A but not in the Figure 2—figure supplement 4D to avoid overinterpretation of the fit provided that their precise position cannot be predicted . To test complex formation between the various LdcI and RavA mutants , gel filtration chromatography was performed on a Superose 6 column . LdcI and RavA proteins were mixed at a ratio of 2:3 and incubated in a buffer containing 25 mM TrisHCl pH 7 . 9 , 200 mM NaCl , 5% glycerol , 10 mM MgCl2 , 0 . 1 mM PLP , 2 mM ATP , and 1 mM DTT for 30 min prior to injection onto the column . Protein samples were collected in 1 ml fractions and resolved on SDS-PAGE gels . All gels were then silver stained . For LdcI-LARA complex formation , 0 . 6 mg/ml of LdcI was mixed with 1 mg/ml of LARA for 15 min at room temperature in a buffer containing 50 mM Mes pH 6 . 5 , 100 mM NaCl , 0 . 2 mM PLP , 1 mM DTT and 0 . 01% glutaraldehyde ( glutarahaldehyde initially buffed in 800 mM Mes pH 6 . 5 ) . The LdcI:LARA ratio was 1:10 ( 7 . 3 µM of LdcI for 73 µM of LARA ) . The glutaraldehyde cross-linking reaction was stopped by adding a final concentration of 62 mM TrisHCl pH 7 . 4 µl of sample was applied to glow-discharged quantifoil grids ( Quantifoil Micro Tools GmbH , Germany ) 400 mesh 2/1 , excess solution was blotted during 2 s with a Vitrobot ( FEI ) and the grid frozen in liquid ethane ( Dubochet et al . , 1988 ) . Data collection was performed on a FEI Polara microscope operated at 300 kV under low dose conditions . CryoEM micrographs were collected on a CCD Ultrascan Gatan USC 4000 ( 4 k × 4 k ) using FEI EPU automatization software at a magnification of 102 . 413× , giving a pixel size of 1 . 464 Å . The contrast transfer function ( CTF ) for each micrograph was determined with CTFFIND3 ( Mindell and Grigorieff , 2003 ) and corrected by phase flipping using bctf ( Heymann et al . , 2008 ) . Defocus ranged between 1 . 5 and 2 . 7 µm . For initial 3D model determination , a subset of 11 , 102 particles was picked manually using EMAN boxer routine ( Ludtke et al . , 1999 ) . Band-pass filtered particles were centered against a rotationally averaged total sum and classified using multivariate statistical analysis ( MSA ) as implemented in IMAGIC ( van Heel et al . , 1996 ) . A subset of ∼10 class averages was selected ( based on the visual match between the class average and the individual particles ) as references for multi-reference alignment ( MRA ) . After three rounds of MRA/MSA , a 3D model was calculated by angular reconstitution . As two and fivefold symmetries were clearly visible in class averages and eigenimages , D5 symmetry was imposed for 3D calculations . Particle orientations were refined by multiple cycles of MRA , MSA and angular reconstitution , gradually incorporating more particles . The resulting initial model of LdcI-LARA ( ∼18 Å resolution ) was used for particle picking using an automated particle picking procedure based on the Fast Projection Matching algorithm ( Estrozi and Navaza , 2008 ) . The resulting dataset of 26 , 165 LdcI-LARA particles was then used to refine the initial model by projection matching ( AP SH command in SPIDER , D5 symmetry imposed ) ( Frank et al . , 1996; Shaikh et al . , 2008 ) . When the alignments had stabilized , more than 95% of the images aligned to the same references in consecutive rounds of alignment . The final 3D reconstruction comprised 23 , 540 particles . The resolution was estimated based on the gold-standard FSC = 0 . 143 criterion ( Scheres and Chen , 2012 ) by dividing the data in two independent halves and refining them iteratively against the angular reconstitution model low pass filtered to 50 Å resolution . The resolution estimated based on the unmasked FSC curve was 8 . 8 Å , whereas the FSC curve obtained with soft shaped masks ( dilated 5 pixels and with a fall–off profile of a cosine half-bell of 4 pixel width ) yielded the resolution of 7 . 5 Å ( Figure 2—figure supplement 2D ) . The final map was sharpened with EMBfactor ( Fernández et al . , 2008 ) using calculated B-factor of 463 Å2 . Conditions of optimal LdcI–RavA complex formation were initially tested by negative stain EM on a JEOL 1200 EX microscope . Different pHs ( ranging from 6 to 8 ) , salt concentrations ( ranging from 100 to 300 mM ) , protein ratios ( ranging from 1:1 . 5 to 1:3 LdcI monomer:RavA monomer ) , ADP concentrations ( from 0 to 5 mM ) were tested . No cross-linking agent was used to stabilize this high affinity complex . The most promising conditions were further analyzed by cryoEM on a FEI CM200 microscope . The final optimal condition chosen for data collection was: 1 . 26 mg/ml of LdcI mixed with 0 . 94 mg/ml of RavA for 15 min at room temperature in a buffer containing 25 mM Mes pH 6 . 5 , 200 mM NaCl , 3 mM ADP , 0 . 8 mM PLP and 1 mM DTT . The resulting ratio of LdcI:RavA was 1:2 ( in the complex , the ratio is LdcI–RavA ratio is 1:1 . 5 , so there was 1 . 3 molar excess of RavA ) . Quantifoil grids were flash frozen in liquid ethane ( Dubochet et al . , 1988 ) using Vitrobot and a blotting time of 2 to 3 s . Initial 3D model was determined using cryo-electron tomography ( cryoET ) and subtomogram averaging as follows . Tomograms were recorded on a FEI Polara microscope between −65 and 65° with 2° angular step on a CCD Ultrascan Gatan USC 4000 at a nominal magnification of 51160× giving a pixel size of 2 . 932 Å . The total electron dose for each tomogram was around 40 electrons/Å2 . The tomograms were aligned with the IMOD suite using 5 nm gold fiducial beads ( BBInternational , UK ) for frames alignment . 95 subvolumes were extracted in IMOD and aligned using PEET ( Kremer et al . , 1996 ) . The subvolume average was consistent with the previously determined negative stain EM map of LdcI–RavA ( Snider et al . , 2006; Figure 1—figure supplement 2 ) and clearly displayed a D5 symmetry . A high resolution dataset was then collected on a FEI Polara microscope operated at 300 kV under low dose conditions . Micrographs were recorded on Kodak SO-163 film at 59 , 000 magnification , with defocus ranging from 1 . 3 to 3 . 3 μm . Films were digitized on a Zeiss scanner ( Photoscan ) at a step size of 7 µm giving a pixel size of 1 . 186 Å . After CTF determination and correction ( performed as for the LdcI-LARA dataset ) , particles were automatically picked using the Fast Projection Matching algorithm with projections of the cryoET reconstruction as a template . The resulting dataset of 35 , 443 particles was then used to refine the initial cryoET model by projection matching in SPIDER with D5 symmetry imposed ( Frank et al . , 1996; Shaikh et al . , 2008 ) . 60% of the best correlating particles were used to refine the 3D reconstruction . When the alignments had stabilized , more than 95% of the images aligned to the same references in consecutive rounds of alignment . The final map comprised 21 , 265 particles with an even view distribution around the equatorial axis . The resolution was estimated based on the gold-standard FSC = 0 . 143 criterion ( Scheres and Chen , 2012 ) by dividing the data in two independent halves and refining them iteratively against the cryoET model low pass filtered to 50 Å resolution . The resolution estimated based on the unmasked FSC curve was 14 Å , whereas the FSC curve obtained with soft shaped masks ( dilated 5 pixels and with a fall–off profile of a cosine half-bell of 4 pixel width ) yielded the resolution of 11 Å ( Figure 1—figure supplement 2D ) . For LdcI-LARA fitting , LdcI decamer crystal structure ( pdb 3N75 ) was fitted into the LdcI-LARA cryoEM map using the fit-in-map module of UCSF Chimera ( Pettersen et al . , 2004 ) , unambiguously identifying extra densities corresponding to LARA domains . The density of the LARA domain ( residues 361–424 , pdb 3NBX ) was extracted and the LARA atomic model reliably rigidly fitted using SITUS ( Wriggers , 2012 ) . Subtle domain movements being noted in the LdcI structure as compared to the crystal structure ( Figure 2—figure supplement 3D , E ) , the LdcI density was extracted and used for flexible fitting with Flex-EM ( Topf et al . , 2008 ) . LdcI crystal structure was divided into three rigid bodies: the wing domain ( residues 1–128 ) , an intermediate domain ( residues 131–142 , 145–382 , 386–397 , 402–495 , 527–562 ) and a C-terminal domain ( residues 501–521 and 566–710 ) , leaving inter-domains flexible . The Cα RMSD between the initial pdb of the LdcI monomer model and the final Flex-EM model is 0 . 65 Å . The cross correlation values between the cryoEM map and either the rigidly fitted monomer or the final Flex-EM model are 0 . 75 and 0 . 84 respectively . The FSC curves between the cryoEM map and either the rigid fit or the Flex-EM model give an FSC = 0 . 5 at 11 Å resolution and 8 . 3 Å resolution respectively ( Figure 2—figure supplement 3F ) . For LdcI–RavA fitting , superposition of LdcI–RavA and LdcI-LARA maps ( Figure 2—figure supplement 3A–C ) , demonstrated that LdcI and LARA positions were conserved in both maps ( cross-correlation of 0 . 86 between the two maps filtered at 11 Å resolution ) . In order to take advantage of the higher resolution of the LdcI-LARA map , the LdcI-LARA fit was kept fixed for the LdcI–RavA fitting . Thus , only the AAA+ ATPase domain and the triple helical domains of RavA ( pdb 3NBX ) remained to be fitted . The initial rigid fit of these domains using UCSF Chimera fit-in-map module indicated that the AAA+ ATPase domain position was conserved compared to the isolated RavA structure , maintaining the integrity of the ATPase active sites . The rigid fit also revealed that RavA binding to LdcI required movements of the triple-helix domains to bring the LARA domains in their LdcI-interacting position , whereas the RavA legs involved in RavA–RavA equatorial interactions barely moved in comparison to the isolated RavA ( Figure 2—figure supplement 1 ) . In order to propose a pseudoatomic model of RavA in its LdcI-binding state , RavA flexible fitting was thus performed using Flex-EM . The ATPase domains ( residues 1–268 ) , the triple helical bundles ( residues 274–329 , 444–496 ) and the LARA domains ( residues 361–424 ) were considered as rigid , leaving inter-domains flexible . The LARA domains were given as initial position their position in LdcI-LARA maps , and the initial positions of the AAA+ ATPase domains were those of the rigid body fit . 25 cycles of Flex-EM led to improved position of triple-helix domains leaving the AAA+ ATPase and LARA domain fit quasi unchanged . The versatile 330–360 loop of the LARA domain was omitted from the models because it clearly undergoes major rearrangements that cannot be reasonably modeled based on the current quality of the EM maps . The FSC curves between the cryoEM map of the LdcI–RavA complex and either the rigid fit of the LdcI decamer and the RavA hexamer , or the Flex-EM model obtained as described , give an FSC = 0 . 5 at 26 Å resolution and 20 Å resolution respectively ( Figure 2—figure supplement 3G ) . Interestingly , the FSC curve between the LdcI-LARA part of the LdcI–RavA map and its Flex-EM model has an FSC = 0 . 5 at 11 Å resolution , whereas the FSC curve between the equatorial region of the map ( corresponding to the AAA+ domain core and the triple helical domains of RavA ) and its Flex-EM model has an FSC = 0 . 5 at 20 Å resolution . This difference in the local quality of the Flex-EM model can be explained by ( i ) the better quality of the Flex-EM model of the LdcI-LARA region ( because this part of the model is based on the higher resolution LdcI-LARA map ) , and ( ii ) the better definition of the LdcI-LARA region of the LdcI–RavA map in comparison to the equatorial region corresponding to the AAA+ domain core and the triple helical domains of RavA . CryoEM maps and Cα traces of the corresponding fitted atomic structures have been deposited in the Electron Microscopy Data Bank and Protein Data Bank , respectively , with accession codes EMD-2679 and PDB-4upb for LdcI–RavA , EMD-2681 and PDB-4upf for LdcI-LARA .	Bacteria inhabit most habitats on Earth , ranging from our bodies to the deepest depths of oceans . In order to thrive in these diverse surroundings , bacteria have developed sophisticated systems that enable them to adapt to changes in their environment . For instance , the bacteria that live in our stomach are exposed to acidic conditions which would normally kill other living organisms , so they have evolved an ‘acid stress response’ to protect themselves . A variety of systems are responsible for the acid stress response and in E . coli , the most common bacterium in our body , one of these systems relies on two proteins: RavA is a protein that is thought to help other proteins to assemble correctly , while LdcI decreases acidity . These two proteins bind to each other in order to carry out their function . It is known from previous work that RavA is a symmetric ring-shaped structure made of six equal parts ( that is , it is a hexamer ) , whereas LdcI is made up of two stacked rings , each composed of five equal parts ( that is , it is a double pentamer ) . But how can a hexamer fit together with a double pentamer ? This puzzle , which is encountered throughout Nature , is known as ‘symmetry mismatch’ . Malet et al . have now used a technique known as cryo-electron microscopy to work out how RavA and LdcI fit together . These experiments show that the part of RavA that makes contacts with LdcI re-arranges itself to form a structure that matches the fivefold symmetry of LdcI . The result is a novel cage-like structure composed by two double pentamers of LdcI , which are parallel to each other and linked together by five hexamers of RavA . Malet et al . propose that this unique structure protects specific proteins from getting destroyed by acids .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "physics", "of", "living", "systems" ]	2014	Assembly principles of a unique cage formed by hexameric and decameric E. coli proteins
RecQ helicases promote genomic stability through their unique ability to suppress illegitimate recombination and resolve recombination intermediates . These DNA structure-specific activities of RecQ helicases are mediated by the helicase-and-RNAseD like C-terminal ( HRDC ) domain , via unknown mechanisms . Here , employing single-molecule magnetic tweezers and rapid kinetic approaches we establish that the HRDC domain stabilizes intrinsic , sequence-dependent , pauses of the core helicase ( lacking the HRDC ) in a DNA geometry-dependent manner . We elucidate the core unwinding mechanism in which the unwinding rate depends on the stability of the duplex DNA leading to transient sequence-dependent pauses . We further demonstrate a non-linear amplification of these transient pauses by the controlled binding of the HRDC domain . The resulting DNA sequence- and geometry-dependent pausing may underlie a homology sensing mechanism that allows rapid disruption of unstable ( illegitimate ) and stabilization of stable ( legitimate ) DNA strand invasions , which suggests an intrinsic mechanism of recombination quality control by RecQ helicases . RecQ helicases are a family of DNA helicases that play essential roles in maintaining genomic integrity through extensive involvement in DNA recombination , replication , and repair pathways ( Bachrati and Hickson , 2003; Bennett and Keck , 2004; Chu and Hickson , 2009 ) . Escherichia coli RecQ ( Ec RecQ ) helicase is the founding member of the family ( Nakayama et al . , 1984 ) and plays roles in both suppressing illegitimate recombination and facilitating various steps of DNA recombinational repair ( Hanada et al . , 1997; Ryder et al . , 1994; León-Ortiz et al . , 2018 ) . RecQ helicases are highly conserved from bacteria to humans and eukaryotic RecQ helicases have been shown to play similar pro- and anti-recombination functions . Most unicellular organisms , such as E . coli and yeast , express a single RecQ homolog , whereas multi-cellular organisms often possess multiple RecQ helicases specialized to different roles in genome maintenance processes . The fundamental conserved activity of RecQ helicases is the ATP-dependent unwinding of double-stranded DNA ( Nakayama et al . , 1984 ) . All RecQ members possess two evolutionarily conserved RecA-like helicase domains with an ATP binding and hydrolysis site located in a cleft between them ( Bennett and Keck , 2004; Chu and Hickson , 2009; Bachrati and Hickson , 2008 ) . Similar to other superfamily ( SF ) one and SF2 helicases , RecQ members also contain N- and C-terminal accessary domains that provide additional or specialized functionalities ( Fairman-Williams et al . , 2010 ) . The RecQ C-terminal domain ( RQC ) comprises zinc binding and winged-helix ( WH ) sub-domains associated with protein structural integrity and duplex DNA binding , respectively . Although less conserved , many RecQ-family members , including Ec RecQ and multiple human RecQ homologs , possess an accessory single-stranded ( ss ) DNA-binding module termed the helicase-and-RNAseD-C-terminal ( HRDC ) domain ( Bernstein and Keck , 2005; Vindigni and Hickson , 2009 ) . The HRDC , while generally dispensable for helicase activity , is critical for certain recombination intermediate processing steps , such as disruption of displacement strand ( D-loop ) invasion and double Holliday junction resolution ( Rezazadeh , 2012; Singh et al . , 2012; Chatterjee et al . , 2014; Harami et al . , 2017 ) . Biochemical studies have established that full length RecQ has a higher ssDNA binding affinity than RecQ constructs lacking the HRDC , which is consistent with the findings that the interaction between the HRDC and ssDNA contributes to DNA substrate specificity of RecQ helicases ( Bernstein and Keck , 2005; Vindigni and Hickson , 2009 ) . Recently , we provided evidence that HRDC interactions contribute to DNA substrate-geometry dependent binding orientation and unwinding by RecQ , and demonstrated that these HRDC-mediated interactions play a role in suppressing illegitimate recombination in E . coli ( Harami et al . , 2017 ) . Whereas these findings indicate that the HRDC strongly favors binding of RecQ to D-loop structures in an orientation that promotes disruption of the invading DNA strand ( Harami et al . , 2017 ) , it is not clear how RecQ can subsequently discriminate between homologous and non-homologous strand invasions; once correctly oriented on the D-loop , RecQ can unwind any invading strand and indiscriminately disrupt all D-loop formations . In this work we identify a potential solution to this quandary , suggested by the observation that HRDC-dependent pausing during hairpin DNA unwinding is not random but occurs repeatedly at distinct positions on the DNA hairpin . We reason that if the frequency or duration of the unwinding pauses is related to the degree of DNA homology , then the more than 10-fold decrease in average unwinding rate due to pausing can provide a mechanism of homology sensing . Thus , if pausing is correlated with homology , then the resulting modulation of the average unwinding rate of an oriented RecQ helicase will result in discrimination of legitimate versus illegitimate D-loops . To test this theory , we set out to determine the origin of HRDC-mediated pausing by investigating the unwinding mechanism of E . coli RecQ and HRDC-induced pausing using single-molecule magnetic tweezers ( MT ) -based assays and rapid transient kinetic assays . We found that long-lived HRDC-induced pauses of wild type RecQ ( RecQwt ) and shorter-lived pauses of RecQ core domain ( HRDC deletion mutant; RecQ-dH ) are sequence dependent and both correlate with DNA duplex stability . Sequence-dependent pausing is a direct consequence of the unique DNA unwinding mechanism: RecQ unwinds one base-pair per ATP hydrolysis cycle but releases the nascent ssDNA only after unwinding ~5 bp . The translocation kinetics arising from this 5 bp kinetic step depend on the duplex stability , which results in sequence-dependent pausing of the core RecQ that is further stabilized by the HRDC binding to the displaced ssDNA . Kinetic modeling indicates that the affinity of the HRDC for ssDNA is enhanced at pause sites , rather than remaining constant . The HRDC thus acts as a non-linear amplifier of the transient sequence-dependent pauses of the core enzyme . Our study demonstrates that the coupling between the core unwinding mechanism and the HRDC-ssDNA interactions dramatically alter the mode of unwinding in a sequence dependent manner , and , in conjunction with previous work , potentially implicates a mechanistic basis for recombination quality control provided by RecQ helicases . Single-molecule measurements of RecQ helicase unwinding activity were performed with 174- or 584-base pair DNA hairpins using an MT apparatus ( Figure 1A ) . DNA hairpin substrates were attached to the flow-cell surface and to a 1- or 2 . 8 µm magnetic bead via a 1 . 1 kbp double-stranded DNA handle and 60-nucleotides of single-stranded poly-dT , respectively ( Figure 1A ) . Measurements with the DNA hairpin were performed at a constant force of 8 pN under which the hairpin did not open spontaneously . In the presence of RecQ helicase ( 20–100 pM ) , unwinding activity was monitored in real-time by tracking the three-dimensional position of a tethered bead at 60 or 200 Hz . Trajectories of the bead extension as a function of time were analyzed by fitting with a T-test based step finding algorithm to obtain the mean unwinding rate , the ‘step’ unwinding rate between pauses , the pause positions , and the pause durations ( Harami et al . , 2017; Seol et al . , 2016 ) . As described previously ( Harami et al . , 2017 ) , frequent pausing and strand-switching by WT RecQ ( RecQWT ) is caused by the HRDC as the HRDC deletion mutant ( RecQ-dH ) shows significantly less pausing during DNA hairpin unwinding ( Figure 1B ) . Pausing is attributed to transient binding of the HRDC domain to the displaced single-stranded DNA behind RecQ . Since both the displaced and the translocation strands of ssDNA are under tension in the hairpin substrate , binding of the HRDC to the displaced strand will prevent forward motion of the helicase . HRDC binding to either duplex DNA ahead of the helicase , or to the translocation strand of ssDNA behind the helicase , are ruled out by the lack of pauses during the unwinding of a ‘gapped’ DNA substrate in which the displaced strand is not constrained ( Harami et al . , 2017 ) . Given the mechanical origin of the pausing associated with transient binding of the HRDC , the pause positions would be expected to be random , dependent on the stochastic kinetics of the interaction between HRDC and the displaced ssDNA . Interestingly , the dwell-time histogram as a function of position for RecQWT unwinding traces exhibits peaks at distinct positions along the hairpin ( Figure 2A; top ) . The peaks in the dwell-time histogram of unwinding traces arise from long and/or frequent pauses at specific positions during DNA hairpin unwinding by RecQ helicase ( Figure 2—figure supplement 1 ) . To identify the sequence context of the pauses , the extension change associated with DNA hairpin opening was converted to base-pairs via the worm-like chain ( WLC ) model of DNA ( Manosas et al . , 2010 ) . Each unwound base pair resulted in the increase of the molecular extension by two ssDNA nucleotides , which at an applied force of 8 pN corresponds to ~0 . 8 nm assuming a 1 nm persistence length and a 0 . 65 nm inter-phosphate distance . With this conversion factor , the extension change for the fully open hairpin was 174 bp , consistent with the actual DNA hairpin size ( 174 bp ) . To determine if pausing is related to DNA base-pair energy , we compared the unwinding dwell time histogram ( Figure 2A; top ) with the DNA base-pair stability calculated by performing a running average ( 6 bp window ) of the exponential of the DNA base-pair energy for the 174 bp DNA hairpin sequence based on the nearest neighbor base-pair energy model ( Patten et al . , 1984; SantaLucia , 1998; Huguet et al . , 2010 ) . We found that the peak locations of pausing and duplex stability were highly correlated ( Figure 2A; bottom ) . The exact locations of peaks were identified by globally fitting the dwell-time histogram and the exponential of the average DNA melting energy with the sums of Gaussian distributions ( Figure 2—figure supplement 2 ) . The relationship between pausing during unwinding and the peaks in the dwell-time histogram is explained in Figure 2—figure supplement 2B ( top ) . Consistent with this observation , pause positions from the dwell time histogram of RecQWT were linearly correlated with peak positions from the DNA base-pair energy profile ( Figure 2B; top ) with a slope of 0 . 99 ± 0 . 03 , linear correlation coefficient ( Pearson’s r ) of 0 . 97 , and χ2 = 0 . 85 , indicating a strong linear correlation . The sequence around the peak positions ( ±4 bp ) contained a high percentage of GC ( ~70% ) , consistent with the finding that the pause positions are related to the duplex stability of the DNA . This finding raises the question of how HRDC-dependent pausing is correlated with DNA base-pair melting despite the fact that the HRDC itself does not play a role in unwinding DNA or exhibit sequence-specific ssDNA binding . We hypothesized that the HRDC may amplify or stabilize transient pauses associated with RecQ core domain ( RecQ-dH ) encountering regions of increased duplex stability ( high GC content ) . To test this hypothesis , we determined if the transient pausing positions of RecQ-dH correlated with the peaks in the DNA base-pair stability curve ( Figure 2B ) . The pause positions for RecQ-dH were obtained from dwell time histograms following the same procedure used for RecQWT ( Figure 2—figure supplement 2C ) and plotted as a function of the peak positions of DNA base-pair stability ( Figure 2B ) . The pause positions of RecQ-dH were linearly correlated with the duplex stability peaks , returning a slope of 0 . 96 ± 0 . 06 , Pearson’s r = 0 . 95 , and χ2 = 1 . 1 . Moreover , the pause positions of RecQ-dH are statistically identical to those of RecQWT , confirming that HRDC-dependent pausing likely originates from stabilization of sequence-dependent unwinding kinetics of RecQ helicase . Sequence-dependent pausing by RecQ-dH reveals important mechanistic insights into the unwinding and translocation mechanism . If the enzyme unwinds one base pair per each kinetic step , the largest energy difference for a single base-pair opening ( G/C vs A/T ) is ~2 . 0 kBT so the pause duration ratio of G/C to A/T will be a maximum of ~7 fold . However , the roughly 20-fold difference in the time the enzyme requires to unwind DNA at the longest pause duration sites in comparison to the average unwinding rate , suggests that more than a single base pair is being opened by the enzyme during each kinetic step . Following this simple analysis , we suggest that pausing is governed by a combination of the DNA base-pair stability and the number of base pairs melted by the helicase during each kinetic step . During processive unwinding , this melting step is the rate limiting step that determines the unwinding rate and pause durations . To distinguish among possible models for the unwinding mechanism of RecQ-dH based on its pausing behavior , we simulated unwinding trajectories comprising a series of pauses and translocations ( Figure 3A ) . Based on previous studies of helicases ( Manosas et al . , 2010; Cheng et al . , 2007; Neuman et al . , 2005; Cheng et al . , 2011; Lin et al . , 2017; Myong et al . , 2007 ) , we considered two scenarios for RecQ-dH unwinding with an n-bp kinetic step size: either the enzyme unwinds n base-pairs simultaneously then rapidly translocates along the unwound DNA ( simultaneous melting model ) , or it sequentially unwinds n base-pairs then releases the newly melted ssDNA ( delayed release model ) ( Figure 3A ) . We exclusively simulated RecQ-dH unwinding and pausing kinetics rather than RecQWT due to the significantly more complex behavior of the RecQWT unwinding trajectories ( Figure 1B ) . In the simultaneous melting model , the pause duration , τ is related to the sum of n base-pair energies at the position of the ith kinetic step , ( 1 ) τp ( i ) = ARecQexp⁡∑s=1nG1bpi-1+s Here G1bp is the free energy required for 1 base-pair melting at the ith position calculated using the nearest-neighbor energy parameters ( Patten et al . , 1984; SantaLucia , 1998; Huguet et al . , 2010 ) , s is a step index ranging from 1 to n , and ARecQ is a pre-factor used to adjust the simulation to give the same average unwinding ( 46 nt/s ) and translocation rate ( ~100 nt/s ) as the RecQ-dH construct ( Manosas et al . , 2010 ) . In the delayed release model , τpi is the sum of the pause times associated with melting each of n base pairs at the ith kinetic step , ( 2 ) τp ( i ) =∑s=1nARecQexp⁡ ( G1bp ( ( i−1 ) +s ) ) Stochastic simulations of both models were run with different step-sizes , n . For each value of n , the pre-factor ARecQ was adjusted to match the measured average rate of RecQ-dH , and the single-strand DNA translocation rate was 100 bp/s ( Manosas et al . , 2010; Bagchi et al . , 2018 ) . Simulated unwinding traces were generated for different kinetic step-sizes for the two different models ( 100 traces per each condition , example traces are shown in Figure 3—figure supplement 1 ) . Simulated traces were analyzed with a T-test based step finding algorithm with the same parameters used for experimental data analysis . Pause durations were binned over 5 bp intervals for simulation and experimental traces and the mean pause duration for each bin was calculated ( Figure 3B ) . Simulation results were compared with experimental data for RecQ-dH by calculating the reduced χ2 ( χν2 ) between the simulated and experimental traces ( Figure 3C ) . For the delayed release model , χν2 reached a minimum around 5 bp ( χν2 = 0 . 9x10−1 ) , lower than the minimum for simultaneous melting model that reached a minimum at 2 bp ( χν2 = 1 . 4x10−1 ) . This suggests that a delayed release scenario may describe the unwinding mechanism of core RecQ . To confirm this finding , we investigated how the RecQ-dH unwinding rate was affected by Na+ concentration and compared the results with the two unwinding models . As DNA base-pair melting energy increases with Na+ concentration ( SantaLucia , 1998; Huguet et al . , 2010 ) , the average unwinding rate predicted by the simultaneous melting model should decrease more rapidly than that predicted by the delayed release model ( Figure 3D ) . We varied the Na+ concentration from 25 to 500 mM while maintaining Mg2+ at 5 mM under otherwise identical buffer conditions . The unwinding rate of RecQ-dH decreased with increasing Na+ concentration . The relative decrease in unwinding rate was much better described by the delayed release model with a 5 bp kinetic step . than the simultaneous 2 bp DNA melting model ( Figure 3D ) . The small deviation between the delayed release model and the measured Na+ concentration dependence of the unwinding rate suggests that although the duplex unwinding remains the rate-limiting step , the Na+ concentration effects other aspects of unwinding such as protein-DNA interactions , which are beyond the scope of the simple model . Thus , to further test and confirm the delayed release unwinding model , we performed two additional experiments as explained below . The significant DNA sequence dependence of the RecQ-catalyzed DNA unwinding rate and pausing characteristics detected in MT single-molecule experiments should be reflected in ensemble unwinding kinetic measurements , which are suitable for the determination of the kinetic step size and the macroscopic dsDNA unwinding rate ( Lucius et al . , 2003 ) . In these experiments unwinding kinetics are monitored via the appearance of fully unwound reaction products . Thus , ensemble unwinding experiments complement MT experiments , in which individual unwinding steps are monitored . Importantly , these techniques together should allow determination of the microscopic unwinding mechanism of RecQ helicase constructs based on the proposed base-pair energy dependent unwinding models . To test this idea , we performed single-turnover unwinding kinetic experiments in which we rapidly mixed complexes of RecQWT or RecQ-dH with forked DNA substrates of varying GC content with ATP and excess unlabeled ssDNA traps in a quenched-flow instrument and monitored the time course of fluorescently-labeled ssDNA generation via gel electrophoresis of reaction products ( Figure 4A ) . Forked DNA substrates used in the experiments comprised two 21-nt ssDNA arms and a 33 bp dsDNA segment containing 12 ( gc36 ) , 16 ( gc48 ) or 26 GC ( gc79 ) bps ( sequences described in Supplementary file 1Table S1 ) . Unwinding traces comprised a short ( ~0 . 1 s ) initial lag , followed by a biphasic appearance of the labeled ssDNA reaction product ( Figure 4B ) . The rapid rise originated from single unwinding runs of initially DNA-bound helicase molecules . The slow rise phase originates from premature dissociation , followed by slow rebinding , of the enzyme to the DNA substrate ( hindered but not totally inhibited by the ssDNA trap strand ) that eventually led to full unwinding of the DNA fork ( Harami et al . , 2017 ) . To obtain parameters of unwinding , data were analyzed with a modified version of a previously described n-step kinetic model ( Lucius et al . , 2003 ) . In its simplest form the model assumes that DNA unwinding occurs as a result of n consecutive rate limiting steps that have a uniform rate constant . This model is generally suitable for the determination of the macroscopic dsDNA unwinding rate , the kinetic step size and the number of intermediates in the unwinding reaction ( Lucius et al . , 2003 ) . Using a modified version of the n-step model ( Figure 4C ) , global fitting of the unwinding kinetics of the gc36 , gc48 and gc79 substrates using an integer series of n ranging from 1 to 7 revealed smallest χv2 values for an apparent kinetic step size of 5 bp for both RecQ and RecQ-dH ( Figure 4B–D ) with all DNA substrates , similar to that suggested by our MT results ( Figure 3A–B ) and by previous findings ( Lin et al . , 2017; Harami et al . , 2015 ) . However , the n-step model does not consider the sequence dependence of the rates of elementary unwinding steps , precluding the distinction between different microscopic mechanisms producing the same kinetic step size . Therefore , we used the same physical framework as described for the MT experiments ( Equations 1 and 2 ) and performed global kinetic fitting to all transients of a given helicase construct ( RecQWT or RecQ-dH ) unwinding the different forked DNA substrates , based on the DNA sequence-dependent simultaneous melting and delayed release unwinding models ( Figure 4B–C ) . For both models , fitting was done using an integer series of n ranging from 1 to 7 . In agreement with the results of the MT analysis ( Figure 3C ) , the smallest χv2 value was obtained for the delayed release model with a kinetic step size of 4 bp for RecQ-dH and 5 bp for RecQWT ( Figure 4B and C , other parameters are listed in Supplementary file 1 Table S2 ) . If RecQ-dH takes a certain kinetic step size , it could in principle be directly observed in the single-molecule unwinding traces . However , the enzyme unwinds DNA too rapidly at high ATP concentrations for steps to be routinely and accurately detected , given the spatial resolution limits of the measurement . Under our experimental conditions , the average baseline noise was ~14 nm at 200 Hz data collection rate . Thus , in order to observe , for example , a 4 bp step ( i . e . a 3 nm change in DNA extension ) , the average pause duration should be >300 ms or the unwinding rate should be less than 13 bp/s ( ~3 fold slower than 42 bp/s ) . We tried three different conditions to decrease the unwinding rate of RecQ: lowering the ATP concentration ( Figure 5—figure supplement 1 ) and including non-hydrolysable ATP analogues , ATPγS or AMP-PNP , in the assay ( Figure 5—figure supplement 2 ) . We found that decreasing the ATP concentration ( sufficiently lowering the unwinding rate ) resulted in frequent and extensive enzyme backsliding ( observable as rapid partial rezipping of the hairpin during an unwinding event ) , which complicates kinetic step size measurements ( Figure 5—figure supplement 1 ) . AMP-PNP showed extremely slow dissociation kinetics from RecQ-dH that were inappropriate for unwinding assays ( Figure 5—figure supplement 2 ) . On the other hand , ATPγS , showed a comparable binding affinity to ATP with a significantly shorter binding time ( ~1 s ) than AMP-PNP ( Figure 5—figure supplement 2 ) . In addition , ATPγS binding transiently locks RecQ in the strong DNA-binding ATP bound state without backsliding , leading to long duration pauses that effectively increased the spatial resolution by permitting longer averaging times ( Figure 5A ) . We measured the unwinding activity of RecQ-dH at different fractions of ATPγS ( 0 . 05–0 . 5 mM ) while keeping the total combined concentration of ATP and ATPγS constant at 1 mM . The unwinding rate decreased with increasing ATPγS fraction ( Figure 5A ) . We reason that when the concentration of ATPγS is such that it is bound at least once per kinetic step , then the predominant physical step-size measured in the hairpin unwinding trajectories will correspond to the kinetic step-size . Step-sizes were estimated with two different step finding algorithms: a step finding program originally developed by Kerssemakers and coworkers ( Kerssemakers et al . , 2006 ) and the T-test based step finding analysis ( Seol et al . , 2016 ) . To determine the average kinetic step-size for each condition , the estimated step-sizes were histogrammed and fit with Gaussian distributions ( Figure 5B and C ) . We found that the estimated step size of RecQ-dH from both step-finding algorithms were comparable , converging from ~8 bp at a low ATPγS fraction to 5 bp at higher ATPγS fractions , suggesting that the average kinetic step size of RecQ-dH is 5 bp [T-test: 5 . 3 ± 0 . 1 ( center ) ; 3 . 0 ± 0 . 6 ( Standard Deviation ) ; Kerssemakers: 5 . 2 ± 0 . 1 ( center ) ; 2 . 1 ± 0 . 1 ( Standard Deviation ) , errors correspond to the standard deviations from Gaussian fitting] . The broad step-size distribution could reflect the stochastic nature of ssDNA release by RecQ . Also , it is likely that the two ssDNA strands are released asynchronously by RecQ . In line with this , we occasionally observed a 2 . 5 bp kinetic step at 500 µM ATPγS , and the step-size distribution at lower ATPγS fractions included peaks at 7 . 5 , 10 , and 12 . 5 bp , consistent with a fundamental step-size of 2 . 5 bp corresponding to the release of one ssDNA strand of 5 nt ( Figure 5—figure supplement 3 ) . The prolonged pause state due to ATPγS binding instead of ATP at the cleft between two RecA domains of RecQ enabled us to probe the mechano-chemical coupling , C , of RecQ helicase that is a measure of the number of ATP hydrolyzed per kinetic step . For C = m/n ( m ATP hydrolysis per n kinetic step size ) , the average number of bound ATPγS , l , can be estimated based on the binomial probability distribution . ( 2 ) l=∑i=0mim ! im-i ! Pi1-Pm-i ( 3 ) P=kATPγS[ATPγS]kATPATP+kATPγS[ATPγS] P is the probability of ATPγS binding per each cycle . kATP is the ATP on-rate , kATPγS is the ATPγS on-rate , and [ATPγS] and [ATP] are the concentrations of ATPγS and ATP , respectively . The hydrolysis rate of ATPγS by RecQ in the presence of excess dT45 was measured by monitoring thiophosphate production ( Saran et al . , 2006 ) and estimated to be < 0 . 2/s ( Appendix 1 and Figure 5—figure supplement 1 ) . This is significantly slower than the measured pause escape rate ( 2 . 8 ± 0 . 1/s ) at 500 µM ATPγS , suggesting that koffATPγS is much faster than the rate of ATPγS hydrolysis . Thus , we could simplify the mean pause duration per kinetic step , τ and the mean unwinding rate , v as , ( 4 ) τ=l/kATPγSoff+1/kstep ( 5 ) v=nτ where kstep is the mean kinetic stepping rate without ATPγS . We obtained the average pause durations for different fractions of ATPγS from 5 to 50 % and globally fitted the pause durations and the average unwinding rates as a function of ATPγS concentration with Eq . 4 and 5 respectively ( Figure 5D ) . From this global fitting , we found that C = 1 . 0 ± 0 . 2 bp/ATP , kATP/ kATPγS = 1 . 2 ± 0 . 2 , and 1/koffATPγS = 0 . 4 ± 0 . 1 s suggesting a tight mechano-chemical coupling in agreement with previous ensemble measurements ( Harami et al . , 2015; Sarlós et al . , 2012 ) . We note that rebinding of ATPγS was not taken into account for simplicity in Equations 2 and 3 , which is reasonable as ATPγS concentration is lower than ATP except for 50% ATPγS , and because the relative on rate ( kATP[ATP] vs kATPγS[ATPγS] ) of ATPγS is lower than that of ATP . To ensure that this simplification is reasonable , we simulated how many ATPγS molecules are bound instead of ATP per base pair based on the fitting parameter , kATP/kATPγS = 1 . 2 at 50% ATPγS . We found that the average is less than one ATPγS per base pair at this condition indicating that repetitive ATPγS binding at the same site is rare . The sequence-dependent unwinding mechanism of RecQ consisting of a 5 bp kinetic step results in transient pauses that are further stabilized by the HRDC , which results in the long-lived sequence-dependent pausing of RecQWT ( Figure 2 ) . In addition to the sequence dependence , HRDC-dependent pausing exhibits two interesting features: occasional repetitive rezipping and unwinding ( shuttling ) around the pause position and significantly prolonged pausing durations for certain pausing positions ( Figure 2 and Figure 2—figure supplement 1 ) . It appears that HRDC-binding triggers this shuttling behavior in which 5–10 bp are repetitively unwound and rezipped at the relatively long-lived ( >0 . 14 s ) intrinsic pause positions . Shuttling activity repeats until RecQ passes the sequence-dependent roadblock . This complex shuttling behavior was significantly enhanced at those regions where long pauses of RecQ-dH are clustered , such as at 55 , 90 , and 120 bps ( Figure 2—figure supplement 1 ) resulting in the apparent high dwell probabilities at these sites ( Figure 2A ) . Since these positions also exhibit the highest base-pair stabilities ( Figure 2A ) , the average dwell time of RecQWT is strongly correlated with the base-pair stability . Indeed , the average dwell times for RecQWT are highly non-linearly correlated with the base-pair stability ( Figure 6A ) . In contrast , the dwell-times for RecQ-dH scale linearly and much less dramatically with the base-pair stability ( Figure 6A ) , indicating that the HRDC-stabilized pausing can be described as a non-linear amplifier of the intrinsic sequence-dependent unwinding kinetics . We found that a simple kinetic competition model in which HRDC binding is in kinetic competition with the forward motion of the helicase ( Appendix 1 ) cannot reproduce the dramatic changes in pause probability observed for RecQWT hairpin unwinding ( Figure 6—figure supplement 1 ) . In line with this , the pausing duration distribution for RecQwt is better described by a double rather than single exponential distribution ( Figure 6—figure supplement 1B ) . Both pause lifetimes ( 1 . 8 ± 0 . 3 s and 0 . 4 ± 0 . 1 s ) are longer than the average pause duration ( 0 . 14 ± 0 . 03 s , see SI ) for the core RecQ ( RecQ-dH ) , indicating that there are multiple HRDC-dependent pause states . To test the proposal that base-pair stability-dependent RecQWT pausing underlies a potential mechanism of homology sensing , we investigated the effect of introducing single mismatches at high probability pause sites . We tested hairpin substrates containing 1 , 2 , or three single mismatches: ( i ) a mismatch introduced at 90 bp , ( ii ) mismatches introduced at 90 bp and 104 bp and ( ii ) mismatches introduced at 90 bp , 104 bp , and 124 bp . All mismatches were generated by changing G to T on the displaced strand ( detailed sequence information is in Appendix 1 ) . We found that pausing of RecQWT around the 90 bp unwound hairpin position was significantly reduced compared to intact 174 bp hairpin when a mismatch was present at 90 bp and additional mismatch at 124 bp further suppressed pausing around 120 bp ( Figure 6B and Figure 6—figure supplement 2 ) . The effect of mismatches on pausing of RecQWT can be clearly demonstrated by comparing the dwell-time histograms of three DNA substrates ( Figure 6B ) . The prominent peaks in the dwell time histogram of the intact hairpin DNA were diminished one by one with the introduction of mismatches at the corresponding positions confirming the correlation between pausing and homology ( Figure 6B and Figure 6—figure supplement 2 ) . The fundamental activity of helicases is the unwinding of duplex nucleic acids . In general , the unwinding mechanism has been classified as either passive or active depending on the degree to which the enzyme ‘actively’ destabilizes the duplex rather than ‘passively’ waiting for a thermal fluctuation to expose ssDNA ( Betterton and Jülicher , 2005; Lohman et al . , 2008 ) . For a purely passive helicase , the enzyme does not provide external work to destabilize duplex DNA and translocates only when ssDNA is exposed by thermal fluctuations . On the other hand , an active helicase is actively involved in disrupting the DNA duplex , and in principle , is less sensitive to base-pair energy or sequence . In previous single molecule experiments , E . coli RecQ helicase was identified as an active helicase based on the minimal force and DNA sequence dependence of duplex unwinding ( Manosas et al . , 2010 ) . In that study , following theoretical work by Betterton et al ( Betterton and Jülicher , 2005 ) , RecQ unwinding was modeled as one base-pair melting followed by 1–2 bases translocation . However , we found that the pause durations were generally longer than would be expected for melting of 1 base pair when we compared our results with simulations . We considered two different scenarios: RecQ either destabilizes multiple base-pairs ( ≥2 bp ) during each kinetic step similar to NS3 helicase ( Cheng et al . , 2007 ) or delays releasing of multiple unwound base-pairs similar to speculative models suggested in previous studies ( Cheng et al . , 2011; Lin et al . , 2017; Myong et al . , 2007; Ma et al . , 2018 ) . However , the minimal dependence of the unwinding rate on Na+ concentration in addition to the sequence-dependent pauses cannot be explained by multi-base-pair melting . Rather , we found that an alternative scenario in which RecQ delays the release of nascent single-strand DNA ( delayed release ) was a better fit to the pause duration and Na+-dependent unwinding rate data , though the associated kinetic step size ( number of bp unwound prior to release ) was not uniquely constrained by the pause duration or Na+-dependent unwinding rate measurements ( Figure 3 ) . This finding is consistent with previous studies revealing ‘asynchronous’ release of nascent ssDNA ( Lin et al . , 2017; Ma et al . , 2018 ) . Nonetheless , the mechanism of delayed release of newly melted nucleotides remains unclear . Previous results suggest that a putative electrostatic interaction between newly melted ssDNA and RecQ sequesters several nucleotides of ssDNA . We consider a similar possibility in which RecQ releases the nascent ssDNA only when the accumulated torsion or tension on bound ssDNA is high enough to disrupt the interaction ( Myong et al . , 2007 ) . We further refined the delayed release model by directly measuring a 5 bp kinetic step size for DNA unwinding using ATPγS , which sufficiently slows down the unwinding rate without inducing the frequent back-sliding observed at reduced ATP concentrations ( Figure 5 and Figure 5—figure supplement 1 ) ) . Recent single molecule fluorescent studies showed 2–4 bp kinetic step ( Lin et al . , 2017; Ma et al . , 2018 ) . This smaller and more random nature of the kinetic step size is likely due to the low ATP concentration ( 2–5 µM ) , at which ATP binding likely becomes the dominant rate-limiting step slower than or on the same order as the intrinsic off-rate of the nascent DNA . Consistent with this model , the study found a correlation between the ATP concentration and the measured kinetic step size . The mechano-chemical coupling and kinetic analysis of ssDNA translocation of RecQ have been studied in detail ( Sarlós et al . , 2012; Rad and Kowalczykowski , 2012 ) . Our unwinding kinetic step is consistent with a recent a study in which a five nucleotide kinetic step for RecQ translocation was estimated ( Rad and Kowalczykowski , 2012 ) . Other helicases display multi base-pair kinetic unwinding steps under sufficient ATP concentrations ( above KM ~ 20 µM ) ( Lohman et al . , 2008 ) . The mechano-chemical coupling is a measure of how many chemical cycles an enzyme completes to take one mechanical step . In the case of RecQ or other helicases , it corresponds to how many ATP molecules are consumed per one base translocation ( or base-pair unwound for unwinding ) . For translocation , RecQ shows a tight coupling close to one nucleotide step per ATP hydrolysis ( Sarlós et al . , 2012 ) . Our study reveals that the mechano-chemical coupling for unwinding maintains one base-pair melting per ATP hydrolysis ( Figure 4C ) , which is also supported by the results of a recent single-molecule florescence study of RecQ unwinding ( Lin et al . , 2017 ) . The proposed kinetic model based on our ATP dependent kinetic analysis ( Appendix 1; Figure 5—figure supplement 1 ) suggests that DNA melting precedes ATP hydrolysis . In this model , ATP binding stabilizes the DNA-RecQ interaction and facilitates DNA melting presumably coupled to an ATP binding-dependent conformational change of RecQ such as rotation of the helicase domains relative to one another , which explains more frequent backsliding under lower ATP concentration ( Bernstein et al . , 2003; Manthei et al . , 2015; Pike et al . , 2009 ) . Recent structural results suggest that RecQ binding may melt two base-pairs of DNA before ATP binding ( Manthei et al . , 2015 ) . This may occur at the initial binding of RecQ ( or rebinding ) as the initiation of unwinding , but not the unwinding rate , is highly dependent on Na+ concentration . Whereas we establish that pausing arises from the stability of DNA duplex , recent work by Voter et . al suggests an alternative mechanism for sequence-dependent pausing . In their work , they identify a ‘Guanine binding pocket’ located in the helicase domain that specifically interacts with guanine bases to destabilize G-quadruplex structures . It is possible that these interactions could also slow down the unwinding rate at clusters of guanine bases in the translocation strand by inducing short pauses ( Voter et al . , 2018 ) . However , the translocation sequence at the strong pause locations of our DNA hairpin is mixture of G and C bases , suggesting that the pauses we observed originate from the duplex stability . Nevertheless , we cannot entirely rule out the possibility that these specific guanine interactions contribute slightly to the pausing of RecQ core over and above the dominant effect of DNA duplex stability . One of the essential aspects of RecQ is that it processes diverse , non-canonical , DNA substrates in which the HRDC plays an important role in modulating substrate-specific unwinding of RecQ . It has been shown that the HRDC regulates the binding orientation of RecQ core to promote disruption of D-loop structures , early homologous recombination intermediates ( Harami et al . , 2017 ) . However , it was not clear how it can regulate unwinding of RecQ to selectively disrupt illegitimate or non-homologous invading DNA strands since the HRDC presumably cannot directly sense DNA sequence homology ( Harami et al . , 2017 ) . Our present study reveals that the HRDC-ssDNA interactions are strongly coupled to DNA sequence-dependent pausing of the RecQ helicase core: ssDNA binding by the HRDC is not random but occurs at DNA sequences where the helicase core pauses due to the high duplex stability ( Figure 2 ) . On the other hand , either a low homology ( base-pair mismatches ) or low duplex stability ( low GC regions ) strongly reduces RecQ pausing ( Figures 2 and 6B and Figure 6—figure supplement 2 ) . Importantly , this feature can support discrimination between legitimate and illegitimate recombination events by RecQ helicases , in accordance with the increased illegitimate recombination frequencies detected in vivo upon compromising RecQ HRDC function ( Harami et al . , 2017; Wang et al . , 2016 ) . Recombination events proceed through the formation of a displacement loop ( D-loop ) flanked by genomic DNA , which , due to the limited mobility of these large DNA domains , mimics the hairpin geometry of the magnetic tweezers experiments in which the displaced DNA strand is constrained ( Figure 6C and D ) . Previously we showed that the HRDC both targets RecQ to D-loop intermediates and orients the enzyme in a configuration favoring D-loop disruption ( Harami et al . , 2017 ) . The results obtained here provide a mechanistic basis for the subsequent discrimination between legitimate and illegitimate recombination based on the length and stability of the D-loop structure . RecQ-catalyzed unwinding of long and stable D-loops will be frequently interrupted by HRDC-stabilized pauses that drastically decrease the average unwinding rate . This slow average unwinding rate potentially permits the initiation of down-stream recombination processes associated with DNA synthesis resulting in extension and further stabilization of the D-loop . Conversely , RecQ unwinding of short and/or unstable D-loops will proceed rapidly ( 60–80 bp/s ) resulting in the efficient disruption of the D-loop before it can be further extended . Our study reveals that the strategic location of the HRDC relative to the core domain , combined with sequence-dependent DNA unwinding , enable RecQ helicase to control pausing and shuttling in a substrate-dependent manner and expand its biological activity beyond simple duplex DNA unwinding . Whereas this study focused exclusively on E . coli RecQ , the homology sensing mechanism we propose is potentially applicable to the suppression of illegitimate , or so called ‘homeologous recombination’ by BLM ( Wang et al . , 2016 ) . Another biological role of HRDC domain-mediated pausing and shuttling could be linked to the role of RecQ helicases in G-quadruplex secondary DNA structure processing . G-quadruplex structures were shown to act as replication road blocks and these regions were shown to be recombinational hot spots ( van Wietmarschen et al . , 2018; Rhodes and Lipps , 2015 ) . RecQ helicases can efficiently unwind G-quadruplex structures , possibly to aid DNA replication , suppress genome instability and to influence transcription of various genes ( Voter et al . , 2018; Mendoza et al . , 2016 ) . Prolonged shuttling at G-quadruplex sites could ensure that these secondary structures remain unfolded until further steps of replication or DNA repair can proceed . In line with this idea , the HRDC domain of human BLM helicase was shown to be essential for efficient , repetitive unwinding of G-quadruplexes Chatterjee et al . , 2014 ) . In this study , we focused on elucidating the sequence-dependent unwinding and pausing mechanism of RecQ helicase in vitro with purified proteins . Whereas our results indicate a possible mechanism for homology sensing by RecQ helicases , the translocation and pausing kinetics on which the model is based could be modulated in vivo due to the interactions with other DNA binding and processing enzymes . For example , single-strand binding protein ( SSB ) would likely compete with the HRDC for ssDNA binding . However , we recently demonstrated that SSB is displaced by RecQ despite the much higher apparent binding affinity of SSB for ssDNA ( Mills et al . , 2017 ) . Furthermore , the high local concentration of the HRDC , which is tethered to the RecQ core by a flexible linker , likely results in the HRDC out-competing other ssDNA binding proteins for the newly melted ssDNA . Nonetheless , as is often the case , RecQ helicases play diverse roles in DNA processing through the interaction with other proteins , thus , future experiments in the presence of other proteins that interact with RecQ in vivo including , SSB , RecJ , RecA , and Topoisomerase III are warranted to test our homology model in a context that more closely approximates physiological conditions . The production of RecQWT and RecQ-dH were previously described in detail ( Seol et al . , 2016 ) . Forked DNA substrates ( described in Supplementary file 1 Table S1 ) were generated and single-turnover unwinding experiments were performed as in ref ( Harami et al . , 2015 ) . Global fitting kinetic analysis was performed using KinTek Global Kinetic Explorer 4 . 0 . The magnetic tweezers and the experimental set-up were previously described ( Seol and Neuman , 2011 ) . A mixture of DNA hairpin ( 3 pmol ) and anti-digoxigenin ( 0 . 5 µg ) was incubated in phosphate buffered solution ( PBS , pH 7 . 5 ) for 20 min and introduced into the sample chamber , which was incubated overnight at 4°C . The sample chamber was washed with 1 ml of wash buffer ( WB , 1X PBS , 0 . 02 % v/v Tween-20 , and 0 . 3 % w/v BSA ) to remove unbound DNA molecules and 40 µl of magnetic beads ( MyOne , Invitrogen ) were introduced to form DNA hairpin tethers . Correct DNA hairpins were identified by the sharp DNA extension change upon DNA hairpin unfolding under high force ( ~19 pN ) . Upon finding a proper DNA substrate , the chamber was washed with 200 µl of RecQ buffer ( 30 mM Tris pH 8 , 50 mM NaCl , 5 mM MgCl2 , 0 . 3 % w/v BSA , 0 . 04 % v/v Tween-20 , 1 mM DTT , and 1 mM ATP ) . After washing , RecQ was added at a concentration of 20–100 pM in 200 µl RecQ buffer . DNA unwinding measurements were done by tracking a DNA tethered magnetic bead in real-time with custom written routines in Labview . During the measurement , a 1 µm polystyrene stuck bead was tracked to correct sample cell drift by adjusting the sample cell position using 3-D piezo stage ( Physik Instrumente ) to compensate for the drift . The unwinding traces were analyzed with a custom-written T-test based algorithm in Igor Pro 6 ( Wavemetrics ) and the Kerssemakers step finding program in MatLab ( Seol et al . , 2016; Kerssemakers et al . , 2006; Carter and Cross , 2005 ) .	Molecules of DNA carry instructions for all of the biological processes that happen in cells . Therefore , it is very important that cells maintain their DNA and quickly repair any damage . DNA molecules are made of two strands that twist together to form a double helix . The most reliable way to repair damage affecting both DNA strands involves a process known as homologous recombination . In this process , one of the strands of the broken DNA joins up with a strand of an identical or similar DNA molecule to make a triple-stranded structure known as a D-loop . This allows the cell to rebuild the damaged DNA using the intact DNA as a template . To ensure that the DNA is repaired correctly , enzymes known as RecQ helicases bind to and unwind D-loops if the strand pairs are poorly matched , whilst not disrupting pairs that are correctly matched . It remains unclear , however , how these enzymes are able to distinguish whether DNA strands in D-loops are a good or bad pair . To address this question , Seol , Harami et al . measured how individual RecQ helicases from a bacterium known as Escherichia coli unwind DNA . The experiments showed that the enzymes were better able to unwind sections of double-stranded DNA that were less stable than other sections of DNA ( indicating the two strands may be a bad match ) . This causes the helicase to pause at stable sections of the DNA as it unwinds the double helix of the D-loop . Further experiments showed that a region of the helicase known as the HRDC domain increased the duration of these pauses , leading to a dramatic decrease in the unwinding speed . Seol , Harami et al . propose that this difference in unwinding speed prevents RecQ from unwinding legitimate matching D-loops while permitting rapid disruption of illegitimate D-loops that could lead to damaged DNA being repaired incorrectly . Mutations in the human versions of RecQ helicases lead to Bloom’s syndrome and Werner’s syndrome in which individuals are predisposed to developing cancer . Understanding how cells repair DNA may ultimately help to treat individuals with these and other similar conditions .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2019	Homology sensing via non-linear amplification of sequence-dependent pausing by RecQ helicase
Photosynthetic starch reserves that accumulate in Arabidopsis leaves during the day decrease approximately linearly with time at night to support metabolism and growth . We find that the rate of decrease is adjusted to accommodate variation in the time of onset of darkness and starch content , such that reserves last almost precisely until dawn . Generation of these dynamics therefore requires an arithmetic division computation between the starch content and expected time to dawn . We introduce two novel chemical kinetic models capable of implementing analog arithmetic division . Predictions from the models are successfully tested in plants perturbed by a night-time light period or by mutations in starch degradation pathways . Our experiments indicate which components of the starch degradation apparatus may be important for appropriate arithmetic division . Our results are potentially relevant for any biological system dependent on a food reserve for survival over a predictable time period . Organisms must control the rate of consumption of their stored food reserves to prevent starvation during periods when food acquisition is not possible . A classic example of this requirement is provided by the response to the light/dark cycle in plants . During the day , plants utilize solar energy for carbon assimilation through photosynthesis . During the night , when solar energy is unavailable , plants utilize stored carbohydrate—usually starch—to allow continued metabolism and growth . In the model plant Arabidopsis thaliana this phenomenon is essential for productivity: mutants with defects in either the accumulation or the degradation of starch have reduced productivity and exhibit symptoms of starvation ( Usadel et al . , 2008; Yazdanbakhsh et al . , 2011 ) . The leaf starch content of Arabidopsis increases approximately linearly with time during the day: more than half of the carbon assimilated via photosynthesis may be stored as semi-crystalline starch granules inside chloroplasts . At night , starch content decreases approximately linearly with time such that 95% of starch is utilized by dawn ( Gibon et al . , 2004; Graf et al . , 2010 ) . This pattern of utilization is extremely robust , and is achieved even when darkness comes unexpectedly early ( Graf et al . , 2010 ) . It is also likely to be optimal for the efficient utilization of carbohydrate over the light/dark cycle ( Gibon et al . , 2004; Smith and Stitt , 2007; Graf and Smith , 2011; Stitt and Zeeman , 2012; Feugier and Satake , 2013 ) . However , despite the high importance for plant productivity of precise control of starch degradation , the way in which such dynamics are generated is very poorly understood . One intriguing possibility is the existence of a mechanism that dynamically measures the starch content and the expected time to dawn , then arithmetically divides these two quantities to compute the appropriate starch degradation rate . Such a mechanism could ensure complete utilization of available starch reserves at dawn despite variation in both the starch content at the onset of darkness and the subsequent duration of darkness . Consistent with this idea , we have recently shown that computation of the appropriate starch degradation rate in a normal night requires the circadian clock ( Graf et al . , 2010 ) , which indicates how information about the expected time to dawn is obtained . Although levels of transcripts for starch-degrading enzymes undergo large daily changes , levels of the enzymes themselves do not ( Smith et al . , 2004; Lu et al . , 2005; Kötting et al . , 2005; Yu et al . , 2005; and our unpublished data ) . Therefore , it is likely that an arithmetic division mechanism would control flux through starch degradation at a post-translational level . Post-translational control would also permit swifter modulation of the catalytic capacity of these abundant proteins than would be possible via transcription/translation . Accordingly , in this paper we propose appropriate analog division mechanisms that operate through post-translational chemical kinetics . More generally , we also consider analog chemical kinetic schemes that generate addition , subtraction and multiplication operations . To the best of our knowledge the starch degradation system constitutes the first concrete realization of such arithmetic operations in biology . In this context , we therefore introduce two mathematical models which can both implement arithmetic division between the starch content and the expected time to dawn . We then successfully test predictions from the two models by examining the pattern of starch degradation in abnormal light/dark cycles and in a range of mutant plants defective in components of the starch degradation apparatus . Finally , our experiments also indicate which components of the starch degradation apparatus may be important for the appropriate implementation of arithmetic division . We first investigated the robustness of a potential arithmetic division calculation to perturbations in both the numerator ( starch content ) and denominator ( expected time to dawn ) . Previously , we showed that the rate of starch degradation is appropriately adjusted in response to an unexpectedly early night ( imposition of darkness 8 hr after dawn on plants grown in 12-hr light/12-hr dark cycles ) ( Graf et al . , 2010 ) . We found that adjustment also occurs in response to an unexpectedly late night . Plants grown in 12-hr light/12-hr dark cycles were subjected to darkness at either 12 hr or 16 hr after dawn . In both cases starch content decreased approximately linearly with time during the night , but with different slopes such that starch reserves were almost exhausted by dawn ( Figure 1A ) . We also found that similar adjustments could be performed in a cca1/lhy circadian clock mutant which has a free-running period of <24 hr ( Alabadi et al . , 2001 ) . In 12-hr light/12-hr dark cycles , this mutant degrades its starch by approximately 21 hr after dawn , rather than the normal 24 hr ( Graf et al . , 2010; Figure 1B ) . When subjected to an unexpected early night , the starch degradation rate in the mutant was adjusted , such that starch reserves were again exhausted at around 21 hr after dawn ( Figure 1B ) . Appropriate adjustments of the rate of starch degradation also occurred in wild-type plants in which environmental manipulations produced different starch contents at the end of the 12-hr light period . A subset of a uniform batch of plants was transferred to a reduced light level for a single light period , leading to a twofold reduction in the starch content at the onset of darkness . For these and control plants subjected to normal light levels , starch content decreased approximately linearly with time during the night , but with different slopes such that starch reserves were almost exhausted by dawn in both cases ( Figure 1C ) . Appropriate adjustment of starch degradation also occurred in subsets of plants exposed to three different regimes of varying light intensity over a single light period that generated two different starch contents at the onset of darkness ( Figure 1—figure supplement 1 ) . To investigate whether this phenomenon is widespread among plants , we examined the model grass Brachypodium distachyon . Ancestors of Arabidopsis and Brachypodium diverged at least 140 Myr ago . Starch content in Brachypodium leaves increased through the light period . As in Arabidopsis , the approximately linear decrease of starch with time following either a normal night ( 12 hr after dawn ) or an unexpectedly early night ( 8 hr after dawn ) was such that starch reserves were almost depleted by dawn ( Figure 1D ) . 10 . 7554/eLife . 00669 . 003Figure 1 . Starch content levels from experiments with unexpected variation in either starch content at the onset of darkness or the time of onset of darkness . ( A ) Starch turnover in Arabidopsis grown in 12-hr light/12-hr dark , then subject to unexpected early ( 8 hr , n = 6 individual rosettes , circles ) normal ( 12 hr , n = 6 , squares ) or unexpected late ( 16 hr , n = 5 , triangles ) onset of darkness . ( B ) Starch turnover in Arabidopsis cca1/lhy mutant grown in 12-hr light/12-hr dark , then subject to unexpected early ( 9 hr , circles ) , or normal ( 12 hr , squares ) onset of darkness ( n = 6–10 ) . ( C ) Starch turnover in Arabidopsis exposed to different daytime light levels: 90 µmol quanta m−2 s−1 ( open squares ) or 50 µmol quanta m−2 s−1 ( filled squares ) ( both n = 5 , previously all plants grown in 12-hr light/12-hr dark with 90 µmol quanta m−2 s−1 ) . ( D ) Starch turnover in Brachypodium grown in 12-hr light/12-hr dark , then subject to unexpected early ( 8 hr , circles ) or normal ( 12 hr , squares ) onset of darkness ( both n = 6 ) . Error bars are standard error of the mean throughout . DOI: http://dx . doi . org/10 . 7554/eLife . 00669 . 00310 . 7554/eLife . 00669 . 004Figure 1—figure supplement 1 . Starch content levels in Arabidopsis plants exposed to different regimes of varying light level over a single light period . Three sets of plants ( each n = 5 individual rosettes ) were grown in 12-hr light , 12-hr dark and were then subject to different light regimes during a single day . One set ( squares ) was exposed to normal light levels ( 180 μmol quanta m−2 s−1 ) , the other two were shaded to about 55% of normal light level ( 100 μmol quanta m−2 s−1 ) for either the first 6 hr ( circles ) or the second 6 hr ( triangles ) of the 12-hr light period , with the normal light level for the other 6-hr period . Error bars are standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 00669 . 004 Overall , these results demonstrate that the control of starch degradation at night to achieve almost complete consumption at the expected time of dawn can accommodate unexpected variation in the time of onset of darkness , starch content at the start of the night , and patterns of starch accumulation during the preceding day . Although the rate of degradation is different in a circadian clock mutant with an altered period from that in the wild-type , the capacity to adjust starch degradation in response to an unexpectedly early night is not compromised . It is also likely that the mechanism underlying this control is present in evolutionarily-distant species of plants . Robust computation of the appropriate starch degradation rate in response to perturbations of both the expected time to dawn and starch content is clearly consistent with the implementation of arithmetic division . Since it is conceptually unclear how such a computation might be performed , we turned to mathematical modeling to generate possible mechanisms . The fact that starch exists as large polymers that are mostly inaccessible within granule matrices means that a separate measure is likely required to provide information about the total amount of starch present at any given time . Hence , we assume the existence of a soluble molecule S whose concentration is proportional to the amount of starch in a granule . Since plants are able to adjust the rate of starch degradation according to variations in two independent quantities ( the expected time to dawn and the amount of starch present ) , two separate species of molecule are clearly required . Therefore , we further assume the existence of a molecule T whose concentration encodes information about the expected time to dawn . In our first model the T molecule concentration is proportional to the expected time to dawn , except during a period after dawn when its value must be reset ( Figure 2B ) . To compute the appropriate degradation rate , the S and T molecule concentrations must therefore be arithmetically divided . We propose that computations of this form can be carried out most simply using analog chemical kinetics . As shown in Figure 2A , it is straightforward to perform addition , subtraction and multiplication operations . Subtraction can be implemented through efficient sequestration and multiplication by a two-species chemical reaction . Division is slightly more intricate , but can be implemented using the model introduced in Figure 2D ( other conceptually similar models are discussed in the ‘Materials and methods’ ) . Here , S molecules associate with the starch granule surface , where they permit starch degradation ( presumably in combination with other elements , as shown in Figure 2D ) and where the S molecules can also be degraded . T molecules inhibit S molecule and starch degradation by binding to S and causing its detachment from the granule surface . In this way , it can be seen intuitively that a division-like operation can be implemented ( for rigorous calculations , see ‘Materials and methods’ ) . In Figure 2F , H , J , we show the best fit of this model to the data from Figure 1A–C with good results . 10 . 7554/eLife . 00669 . 005Figure 2 . Chemical kinetic models capable of implementing analog arithmetic operations . ( A ) Pictorial summaries of schemes for analog implementation of addition , subtraction and multiplication between the concentrations of two molecules S and T . Square brackets indicate concentrations . ( B ) and ( C ) Schematic behavior of the stromal concentrations of S and T molecules ( [SC] and [TC] respectively ) , in ( B ) first and ( C ) second arithmetic division models . In the first model , the T molecule tracks the time to expected dawn after a reset-time tr . In the second model the T molecule concentration increases with time proportionally to 1/ ( expected time to dawn ) between tr1 and tr2 . ( D ) and ( E ) Pictorial summaries of ( D ) first and ( E ) second analog arithmetic division models ( not all reactions shown in pictures , for full details see ‘Materials and methods’ ) . In the reaction schemes , molecules not attached to the starch granule surface have a ‘C’ subscript . The blue disk represents components of the starch degradation apparatus potentially activated by the S molecule in the first model , and by the ST complex in the second model . Best fits ( full lines ) of first ( F ) , ( H ) , and ( J ) and second ( G ) , ( I ) , and ( K ) arithmetic division models to Arabidopsis data from Figure 1A–C . DOI: http://dx . doi . org/10 . 7554/eLife . 00669 . 005 A second distinct possibility also exists for computing the appropriate degradation rate . We now assume that the T molecule concentration increases as the expected time of dawn approaches , before being reset . If this increase is such that the T molecule concentration is approximately proportional to 1/ ( expected time to dawn ) ( Figure 2C ) then the appropriate degradation rate can be computed by multiplying the S and T concentrations . This is implemented by the reaction scheme shown in Figure 2E: S molecules associate with the starch granule surface , where they recruit T molecules from the stroma . The resulting molecular complex permits the degradation of starch ( again presumably in combination with other elements , as shown in Figure 2E ) and of the S molecule itself . The output of this second model is very similar to that of the first model ( fits to data of Figure 1A–C shown in Figure 2G , I , K ) in that a division computation is still performed , but now the timing information is encoded differently in the T molecule concentration . One potential difficulty is the need to pre-compute the reciprocal of the expected time to dawn . A simple possible scheme to achieve this goal is outlined in the ‘Materials and methods’ . Of course , a combination of the two above models is also possible , involving both multiplication and division by factors dependent on the expected time to dawn , such that overall an appropriate division computation is still performed . However , the additional complexity required for such implementations makes such a combined model less likely . Importantly we assume that the granule surface area does not limit the reaction rate as the granule shrinks , consistent with an approximately linear decrease of starch content with time . Accordingly , in both models the degradation reactions ( for both the starch and the S molecules ) occur only in region ( s ) of granule surface of overall fixed area as each granule shrinks . This would be the case if one or more of the additional components required for starch degradation ( illustrated in Figure 2D , E ) is present at a fixed number on the granule surface as the granule shrinks . Clearly the assumption of a non-limiting surface area cannot remain true if the granule shrinks to very small volumes , as could happen at the end of the night . However , our models still fit the experimental data well even at these times ( see Figure 2F–K ) . Taken together , our modeling results show that arithmetic division ( as well as other arithmetic operations ) can be implemented simply using analog post-translational chemical kinetics . Furthermore , output from both models fits our experimental data well . We note that there is some variation in the fitted parameters for both models ( see Tables 3–6 ) , which arises due to variation between experimental data sets even for a single genotype . Such variation is widely observed for measurements of primary metabolites over the light/dark cycle and could arise from batch to batch variation in expression levels of degradation or clock components . As the fit of the two models to the data is equally good , distinguishing between them is currently challenging . However , we can test two critical predictions common to both models . The first prediction is that the degradation rate is continuously computed via arithmetic division during the night . Such a scheme clearly has advantages in its flexibility and potential to recover from unexpected perturbations . To test whether this prediction is correct , we interrupted a normal night with a period of light , ending 5 hr before the expected time of dawn . During this night-time light period starch accumulated to levels very similar to those present at the end of the day-time light period . Comparison of the rate of starch degradation following the night-time light period with the rate before this period , and with the rate that would have been expected over a normal 12-hr night , allowed a robust assessment of whether the degradation rate had been reset . We confirmed that the night-time light period did not re-entrain the circadian clock , by monitoring expression of the clock gene LHY ( see ‘Materials and methods’ and Figure 3—figure supplement 1 ) . For three independent experiments we found that the rates of starch degradation immediately following the night-time light period ( between 19 hr and 21 hr after dawn ) were significantly greater than both the corresponding actual rates before this period ( between 12 hr and 14 hr after dawn ) and the corresponding rates that would have been expected during a normal 12-hr night , see insets in Figure 3A–C ( p = 1 . 6 × 10−3 and 3 . 8 × 10−5 , respectively , details of the statistical analysis in the ‘Materials and methods’ ) . The latter rates are those that would have ensured the complete depletion of the starch content measured at 12 hr at the expected time of dawn ( 24 hr ) . We obtained the same result by changing the duration and the starting time of the night-time light period ( Figure 3D ) . This increase in starch degradation rate is consistent with our prediction that the rate is continuously computed during the night , rather than set only once at the first onset of darkness . In Figure 3 we show the best fits of the first model to the data , with good results . As before , the second model produced very similar fits ( see Figure 3—figure supplement 2 ) . 10 . 7554/eLife . 00669 . 006Figure 3 . Starch content levels from experiments incorporating night-time light period . Arabidopsis plants grown in 12-hr light/12-hr dark were subjected to onset of darkness at 12 hr , followed by an unexpected period of light , followed by extended darkness . ( A ) – ( C ) Three data sets ( n = 12 individual rosettes , except n = 10 for C ) , in which the unexpected period of light was between 14 hr and 19 hr after dawn . ( D ) In the fourth dataset ( n = 12 ) the period of light was between 16 hr and 20 hr after dawn . Full lines are best fits to the first division model . The second model produces very similar fits ( see Figure 3—figure supplement 2 ) . The insets show the respective starch degradation rates computed from the 12-hr and 14-hr experimental time points ( dark grey bars ) compared to those computed from the 19-hr and 21-hr experimental time points in panels ( A–C ) or the 20-hr and 22-hr time points in panel ( D ) ( light grey bars ) . The white bars are the expected starch degradation rates in a normal 12 hr night , that is rates that would have ensured the complete depletion of the starch content measured at 12 hr at the time of expected dawn ( 24 hr ) . Error bars are standard error of the mean throughout . DOI: http://dx . doi . org/10 . 7554/eLife . 00669 . 00610 . 7554/eLife . 00669 . 007Figure 3—figure supplement 1 . Transcript levels of LHY from experiment incorporating night-time light period . LHY transcript levels ( relative to ACT2 ) measured in Arabidopsis plants kept in continuous darkness after a normal night ( squares ) , or subjected to a 5-hr night-time light period between 14 hr and 19 hr after dawn , and then kept in continuous darkness ( circles ) , as in Figure 3A–C . Data for the night-time light period are from the same plants as in Figure 3B . n = 5 individual rosettes , error bars are standard error of the mean . The night-time light period is shown on top of graph . DOI: http://dx . doi . org/10 . 7554/eLife . 00669 . 00710 . 7554/eLife . 00669 . 008Figure 3—figure supplement 2 . Best fits ( full lines ) of the second division model to starch content data from experiments incorporating night-time light period . Error bars are standard error of the mean throughout . DOI: http://dx . doi . org/10 . 7554/eLife . 00669 . 008 The second prediction from the models concerns the effects on starch degradation of perturbations in components of the arithmetic division or degradation apparatus . To reproduce the normal pattern—in which starch is almost completely degraded by the expected time of dawn—our models require that a factor involved in specifying the relative degradation rates of the starch and S molecules must be fine-tuned to one ( see ‘Materials and methods’ ) . If instead this factor is <1 , so that the starch is degraded too slowly , then rather than complete degradation the models predict that only a certain percentage of the starch will be consumed during the night regardless of the starch content at the end of the preceding light period ( for full calculations , see ‘Materials and methods’ ) . Accordingly , we predict that perturbations to parts of the arithmetic division or degradation apparatus might result in degradation of only a fraction of starch during the night . Interestingly , several mutants defective in proteins involved in , or related to , starch degradation exhibit approximately linearly decreasing starch content with time during the night , but have an elevated starch content at the end of the night . These include beta-amylase mutants bam3 and bam4 , the debranching enzyme mutant isa3 , and mutants lacking phosphoglucan water dikinase ( pwd , also called gwd3 ) , a glucan phosphate phosphatase ( sex4 ) and a glucan phosphate phosphatase-like protein ( lsf1 ) ( Smith , 2012 ) . We showed previously that this abnormal pattern of starch turnover was rapidly regained in lsf1 mutants that were transferred to normal light/dark cycles after starch was reduced to very low levels by prolonged darkness ( Comparot-Moss et al . , 2010 ) . We re-analyzed these data and also performed a similar experiment on the sex4 mutant . For both mutants , we found that the fraction of end-of-light period starch content degraded during the night was approximately the same on successive nights ( around 30% for the sex4 mutant and 45% for the lsf1 mutant ) , regardless of the starch content at the end of the respective preceding light period . This resulted in a progressive increase in the end-of-night starch content to an approximately constant value over 3–4 days after return to normal light/dark cycles ( Figure 4A , B ) . Thus the pattern of starch degradation in these mutants is as predicted from the models for a situation in which the above factor is incorrectly set to a value <1 . 10 . 7554/eLife . 00669 . 009Figure 4 . Starch content levels in mutant Arabidopsis plants defective in components of the starch degradation apparatus . ( A ) Starch content in wild-type ( WT ) plants and lsf1 and sex4 mutant plants during four days of 12-hr light/12-hr dark following 5 days of continuous darkness , where plants were transferred back into the light ( at time 0 hr on the x-axis ) 132 hr after the end of the previous light period ( n = 6 individual rosettes ) . Data for wild-type and lsf1 plants are from ( Comparot-Moss et al . , 2010 ) . ( B ) The percentage of starch degraded during each of the four nights in ( A ) . ( C ) – ( E ) Starch content in lsf1 , sex4 and pwd mutant plants grown in 12-hr light/12-hr dark cycles then subject to unexpected early ( 8 hr , circles ) or normal ( 12 hr , squares ) onset of darkness ( n = 5 ) . The continuous and dashed lines are linear fits to the normal and early night datasets respectively . ( F ) For each of the labeled genotypes , R is the ratio between the starch degradation rates ( each normalized by their respective end-of-light period starch content and as determined from the linear fits ) during the normal and early nights . The dashed line shows the expected value of R for wild-type ( WT ) plants , that is , ratio of rates that would ensure the complete depletion of the starch content in all cases at the time of expected dawn ( 24 hr ) . See ‘Materials and methods’ for details about the linear fitting and the calculation of R . Error bars are standard error of the mean throughout . Figure 4—figure supplement 1 shows the datasets used to calculate R for WT , bam3 , bam4 and isa3 . DOI: http://dx . doi . org/10 . 7554/eLife . 00669 . 00910 . 7554/eLife . 00669 . 010Figure 4—figure supplement 1 . Starch content levels during unexpectedly early night in wild-type , bam3 , bam4 , isa3 mutant plants . Starch content in wild-type ( WT ) , bam3 , bam4 , isa3 mutant Arabidopsis plants grown in 12-hr light , 12-hr dark cycles then subject to unexpected early ( 8 hr , circles ) or normal ( 12 hr , squares ) onset of darkness ( n = 6 individual rosettes for WT , n = 5 for mutants; the WT dataset analyzed here is the one already shown in Figure 1A ) . The continuous and dashed lines are linear fits to the normal and early night datasets respectively . Error bars are standard error of the mean throughout . DOI: http://dx . doi . org/10 . 7554/eLife . 00669 . 010 We next attempted to gain experimental insight into how the arithmetic division mechanism modulates flux through the starch degradation pathway . In our models the computation of the starch degradation rate is the result of interactions between the molecules encoding the information about starch content and time to dawn ( S and T ) and the starch degradation apparatus . If this idea is correct , we expect that mutants lacking components of the degradation apparatus involved in these interactions may also lack the ability to adjust the rate of starch degradation in response to variation in starch content or time of onset of darkness . To look for a mutant in this category , we imposed an unexpectedly early night on six mutants , each lacking a protein involved in starch degradation ( lsf1 , sex4 , bam3 , bam4 , isa3 , pwd ) . All six mutants show approximately linear decreases in starch content with time during the night , but they have higher end-of-night starch contents than wild-type plants . This mutant collection covers the majority of currently-known components of the chloroplastic starch degradation apparatus . For five of the six mutants , the rates of starch degradation ( as determined from linear fits ) were lower during an unexpectedly early night ( starting 8 hr after dawn ) than in a normal night ( starting 12 hr after dawn ) , as is the case in wild-type plants ( Figure 4C , D and Figure 4—figure supplement 1 ) . To quantify the adjustment of the starch degradation rates in the mutants , we calculated the ratio R between the degradation rates ( normalized by the respective end-of-light period starch content ) during the normal and the early night . For wild-type plants , the expected value of R is 16/12 ≈ 1 . 33 , that is , the ratio between the expected lengths of the early and the normal night . In Figure 4F we show that the values of R for five of the six mutants are consistent with the wild-type value . However for the sixth mutant , pwd , the rate of starch degradation was not adjusted in response to an unexpectedly early night ( Figure 4E , F ) , as we found an R which was significantly different from the wild-type value ( p = 0 . 01 , details of the statistical analysis in the ‘Materials and methods’ ) . This finding indicates that PWD—phosphoglucan , water dikinase—may be a node at which information about expected time to dawn and starch content is integrated to set an appropriate rate of starch degradation . PWD contributes to a cycle of phosphorylation and dephosphorylation of glucosyl residues within starch polymers that is essential for normal starch degradation . Two enzymes—glucan water dikinase ( GWD ) and PWD—add phosphate groups to starch polymers at the granule surface , and two further enzymes SEX4 and LSF2—remove them . All four enzymes are necessary for normal rates of starch degradation , and loss of GWD almost completely prevents degradation at night ( Smith , 2012 ) . The phosphate groups are thought to disrupt the ordered packing of the starch molecules at the granule surface , allowing access to starch degrading enzymes ( beta-amylases and isoamylase 3 ) . Subsequent removal of the phosphate groups is essential for full degradation of the starch polymers because the phosphates block the action of the exo-acting beta-amylases ( Baunsgaard et al . , 2005; Kötting et al . , 2005; Ritte et al . , 2006; Hejazi et al . , 2009 ) . The cycle as a whole is thus an attractive candidate for integration of factors that modulate flux through starch degradation . Because the phosphorylation/dephosphorylation cycle modifies the granule surface , it is a potential candidate not only for flux modulation but also for the storage of information about starch content . To discover whether phosphate groups may provide information about starch content , we measured the amount of granule-bound phosphate ( on the 6-position of glucosyl residues , representing about 80% of the total phosphate ) over the light/dark cycle . If phosphorylation simply tracks starch polymer synthesis during the day , phosphate levels per unit mass of starch will be constant and will thus contain no information about starch content . Surprisingly , however , we found a large increase/decrease in the level of phosphate per unit mass of starch over the light/dark cycle ( Figure 5 ) . This discovery implies that the S molecule may be a modulator of activities of enzymes of the phosphorylation/dephosphorylation cycle , thus generating a daily pattern of change in the accessibility of the granule surface to hydrolytic enzymes that approximately tracks starch content . 10 . 7554/eLife . 00669 . 011Figure 5 . Daily change in starch phosphate content ( measured as glucose 6-phosphate , G6P ) in Arabidopsis leaves . Results are normalized by total amount of glucose ( Glc ) in starch at each time point . Starch was extracted from rosettes of 26-day-old plants . n = 3 pools of 10 rosettes except at 24 hr time point , with n = 2 pools of 15 rosettes . Error bars represent the range ( i . e . , error bar edges correspond to highest and lowest values measured ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00669 . 011 Overall , our work provides a new framework and perspective for understanding the control of reserve utilization in plants at night . Our experiments provide strong support for an implementation of arithmetic division in night-time starch degradation . We used mathematical modeling to generate two simple mechanisms capable of analog implementation of such an operation . Predictions from the models were then verified and a potential point identified at which the division computation could be integrated into the starch degradation pathway . We also showed that the phosphorylation state of the starch granule surface could provide information about starch content through the day . Our analysis may also be relevant to a broader class of biological processes , where food reserves accumulated in advance of periods of predictable length without further food intake are just sufficient to permit survival to the expected end of the period . For example , migrating little stints ( Calidris minuta ) arriving at their Arctic summer breeding grounds after a 5000-km journey have sufficient remaining lipid reserves for an average of only 0 . 6 days ( Tulp et al . , 2009 ) . During the 4-month fast period of egg-incubating male emperor penguins ( Aptenodytes forsteri ) , lipid reserves are used such that they reach a critical depletion level at approximately the point at which the females are expected to return . Unexpected extension of the fast period leads to catabolism of protein and abandonment of offspring in favor of hunting for food ( Groscolas and Robin , 2001 ) . As in Arabidopsis leaves , the rate of reserve utilization in these examples can potentially be computed by arithmetically dividing the reserve levels by the anticipated time of fasting . It is a longstanding idea that cells are able to use proteins to store and process information through networks of interactions ( Bray , 1995 ) . Understanding how such biochemical networks work and what kind of computations they perform is an ongoing challenge ( see Deckard and Sauro , 2004; Lim et al . , 2013 ) . Our analysis here has underlined the utility of analog chemical kinetics in performing arithmetic computations in biology . Importantly , we have for the first time provided a concrete example of a biological system where such a computation is of fundamental importance . This contrasts to previous work where elegant theoretical implementations of arithmetic operations lacked specific biological applications ( Cory and Perkins , 2008; Buisman et al . , 2009 ) . Analog chemical kinetic approaches may potentially also be useful for calculations in synthetic biology applications ( Benenson , 2012 ) , where they are likely to prove much simpler to implement than alternative schemes based on much more complex digital circuitry . Arabidopsis thaliana ( in the Col0 background , except for cca1/lhy and its wild-type which were in the Ws background ) and Brachypodium distachyon ( Bd21 ) plants were grown as in Graf et al . ( 2010 ) on soil in 12-hr light/12-hr dark with 200 µmol quanta m−2 s−1 illumination at a constant temperature of 20°C for 21 days . Plants were harvested and extracted in dilute perchloric acid for analysis of starch , which was then quantified enzymatically as previously described ( Graf et al . , 2010 ) . RNA was extracted from plant material and qPCR was performed as in Graf et al . ( 2010 ) . Oligonucleotide primer sequences were as follows:PrimerSequence 5′ to 3′LHY-F AT1G01060GACTCAAACACTGCCCAGAAGALHY-RAT1G01060CGTCACTCCCTGAAGGTGTATTTACT2-F AT3G18780ACTTTCATCAGCCGTTTTGAACT2-R AT3G18780ACGATTGGTTGAATATCATCAG Starch granules were prepared as in Ritte et al . ( 2000 ) and used immediately without drying . Granules were resuspended in two pellet volumes of water and boiled for 15 min then digested with 20 U amylogucosidase ( Roche ) and 2 U α-amylase ( Megazyme ) in 100 mM Na acetate pH 4 . 8 for 9 hr at 37°C . Glucose was assayed enzymatically following ( Hargreaves and ap Rees , 1988 ) and glucose 6-phosphate was measured enzymatically using the fluorimetric assay of ( Zhu et al . , 2009 ) . We now describe in full mathematical detail the models introduced in the main text . In this section we describe how starch contents during the degradation process were calculated using the equations previously derived . For Model 1 , the values of the following parameters are required: γ , ( βkST1 ) −1 , ρ0 , tr; and for Model 2: γ , ρ0 , tr1 and tr2 . In order to fit the data from the night-time light period experiments , we also considered the possibility that the T dynamics given by Equations 1 and 2 ( see also Figure 2B , C ) can be phase shifted by a time t0 . The addition of t0 as an extra fitting parameter for the night-time light period experiments was justified by our data on transcripts of the clock gene LHY ( Figure 3—figure supplement 1 ) , showing that the night-time light period may induce a phase shift in the expression of some genes . The phase shift of the T dynamics determines a shift in the time of expected dawn , that , in Model 1 , is defined as the time at which [TC] falls to zero ( see Equation 1 and Figure 2B ) and which in Model 2 is the time at which [TC] would diverge without a reset ( see Equation 2 and Figure 2C ) . Therefore , if the T dynamics are phase shifted by t0 , in our models the starch degradation rates are adjusted in such a way so as to deplete the starch reserves at ( 24 + t0 ) hr after the previous dawn , instead of the normal 24 hr . We found this phenotype in cca1/lhy plants , which run out of starch earlier than 24 hr after the previous dawn . Hence , to reproduce the phenotype of cca1/lhy in our models , t0 was also used as a fitting parameter . A more extensive discussion about cca1/lhy can be found below in the ‘Mutant phenotypes’ section . A full list of the system parameters is shown in Tables 1 and 2 . 10 . 7554/eLife . 00669 . 012Table 1 . Full list of the system parameters for Model 1DOI: http://dx . doi . org/10 . 7554/eLife . 00669 . 012SymbolDefinitionρ0Starch content at the beginning of the dark period . γNormalization variable . ( βkST1 ) −1kST1 is the ratio of the reaction parameter associated with reaction rST1 with the backward rate of reaction rS . β is the proportionality constant between [TC] and Δt . trTime at which [TC] levels finish being reset at the beginning of the day ( see Figure 2B ) . t0Phase shifting parameter of the [TC] dynamics given by Equation 1 . The next dawn is expected to come ( 24 + t0 ) hr after the previous one . 10 . 7554/eLife . 00669 . 013Table 2 . Full list of the system parameters for Model 2DOI: http://dx . doi . org/10 . 7554/eLife . 00669 . 013SymbolDefinitionρ0Starch content at the beginning of the dark period . γNormalization variable . tr1Time at which [TC] levels finish being reset at the beginning of the day ( see Figure 2C ) . tr2Time at which [TC] levels start being reset at the end of the day ( see Figure 2C ) . t0Phase shifting parameter of the [TC] dynamics given by Equation 2 . The next dawn is expected to come ( 24 + t0 ) hr after the previous one . The best fit to a given dataset was found by minimizing the function:L ( θ→ ) =∑i=1N ( ρStheory ( ti , θ→ ) −ρSexp ( ti ) ρSexp ( ti ) ) 2 , where ρSexp ( ti ) are the N mean values of starch contents measured at times ti . ρStheory ( ti , θ→ ) is the starch value predicted by the model at time ti with parameter values θ→ . The set of parameters θ→ that minimizes L corresponds to the maximum-likelihood estimates of the parameters under the assumption that the experimental measurements are normally distributed around the theoretical values with a constant relative error . Note that , in the fits of the data from the cca1/lhy mutant plants for the early and the normal night , we did not consider the two data points closest in time to 24 hr from the datasets , as they are characterized by very low values of starch content and , therefore , are likely to be affected by higher relative errors compared to the other data points . For the same reason , for the linear fits shown in Figure 4—figure supplement 1A we did not consider the data point at t = 24 hr in the normal night datasets and the data points at t = 22 hr and t = 24 hr in the early night datasets . A simulated annealing algorithm was used for the minimization of L ( θ→ ) . The parameter values of the best fits of the models are given in Tables 3–6 , along with the ranges in which the parameters were allowed to vary . 10 . 7554/eLife . 00669 . 014Table 3 . The values of the parameters of the Model 1 best fits shown in Figure 2DOI: http://dx . doi . org/10 . 7554/eLife . 00669 . 014Model 1WT early night ( panel F ) WT normal night ( panel F ) WT late night ( panel F ) cca1/lhy early night ( panel H ) cca1/lhy normal night ( panel H ) WT low light level ( panel J ) WT normal light level ( panel J ) ρ0 ( mg g−1 FW ) within 10% of the measured value8 . 511 . 011 . 74 . 25 . 13 . 66 . 1γ ( 0 . 7–3 . 0 ) 1 . 81 . 81 . 91 . 21 . 21 . 51 . 3 ( βkST1 ) −1 ( 1 . 5–5 . 0 ) hr5 . 05 . 05 . 01 . 61 . 72 . 11 . 5tr ( 9 . 0–12 . 0 ) hr9 . 0Any value in the specified rangeAny value in the specified range11 . 711 . 7Any value in the specified rangeAny value in the specified ranget0 ( −5 . 0–5 . 0 ) hr for the cca1/lhy data , t0 = 0 for WT0 . 00 . 00 . 0−4 . 2−2 . 50 . 00 . 0With each parameter the range used in the best fit search is indicated . WT: wild-type plants . 10 . 7554/eLife . 00669 . 015Table 4 . The values of the parameters of the Model 2 best fits shown in Figure 2DOI: http://dx . doi . org/10 . 7554/eLife . 00669 . 015Model 2WT early night ( panel G ) WT normal night ( panel G ) WT late night ( panel G ) cca1/lhy early night ( panel I ) cca1/lhy normal night ( panel I ) WT low light level ( panel K ) WT normal light level ( panel K ) ρ0 ( mg g−1 FW ) within 10% of the measured value8 . 311 . 012 . 14 . 25 . 13 . 55 . 8γ ( 0 . 7–3 . 0 ) 1 . 11 . 01 . 11 . 31 . 21 . 20 . 9tr1 ( 9 . 0–12 . 0 ) hr11 . 012 . 012 . 011 . 011 . 410 . 510 . 2tr2 ( 20 . 0–23 . 0 ) hr20 . 120 . 020 . 021 . 821 . 421 . 422 . 5t0 ( −5 . 0–5 . 0 ) hr for the cca1/lhy data , t0 = 0 for WT0 . 00 . 00 . 0−2 . 4−1 . 30 . 00 . 0With each parameter the range used in the best fit search is indicated . WT: wild-type plants . 10 . 7554/eLife . 00669 . 016Table 5 . The values of the parameters of the Model 1 best fits shown in Figure 3DOI: http://dx . doi . org/10 . 7554/eLife . 00669 . 016Model 1Panel APanel BPanel CPanel Dρ0 ( mg g−1 FW ) within 10% of the measured value5 . 49 . 07 . 74 . 2γ ( 0 . 7–3 . 0 ) 2 . 41 . 41 . 11 . 0 ( βkST1 ) −1 ( 1 . 5–5 . 0 ) hr5 . 02 . 62 . 12 . 0tr ( 9 . 0–12 . 0 ) hr9 . 09 . 010 . 312 . 0t0 ( −5 . 0—5 . 0 ) hr4 . 33 . 32 . 4−0 . 5With each parameter the range used in the best fit search is indicated . 10 . 7554/eLife . 00669 . 017Table 6 . The values of the parameters of the Model 2 best fits shown in Figure 3—figure supplement 2DOI: http://dx . doi . org/10 . 7554/eLife . 00669 . 017Model 2Panel APanel BPanel CPanel Dρ0 ( mg g−1 FW ) within 10% of the measured value5 . 59 . 17 . 74 . 2γ ( 0 . 7–3 . 0 ) 1 . 51 . 30 . 70 . 7tr1 ( 9 . 0–12 . 0 ) hr12 . 09 . 09 . 09 . 0tr2 ( 20 . 0–23 . 0 ) hr20 . 020 . 922 . 122 . 1t0 ( −5 . 0–5 . 0 ) hr5 . 05 . 02 . 40 . 0With each parameter the range used in the best fit search is indicated . By combining our results from the night-time light period experiments shown in Figure 3A–C , we can show that the evidence in favor of a re-calculation of the starch degradation rate after the night-time light period is statistically significant . In the experiments shown in Figure 3A–C , plants were subjected to 5 hr of light in the middle of the night ( between 14 hr and 19 hr after the previous dawn ) . The starch content at each time point was measured by averaging the starch content of n = 10–12 individual rosettes and the standard error of the mean was also calculated . The starch degradation rate before the night-time light period in the i-th experiment was measured as:μBi=ρSi ( 12 hr ) −ρSi ( 14 hr ) 14 hr−12 hr , that is , as the difference between the starch content at 12 hr and 14 hr divided by the time interval . Similarly , the starch degradation rate after the night-time light period for the i-th experiment was:μAi=ρSi ( 19 hr ) −ρSi ( 21 hr ) 21 hr−19 hr , These rates were averaged over the three experiments , and the difference between the two averages was calculated:μA−μB=1N∑i=1N ( μAi−μBi ) , where N = 3 was the number of experiments . The error on this quantity was estimated by propagating standard errors of the mean . We found μA − μB = 0 . 43 ± 0 . 14 ( mg g−1 FW ) /hr which is significantly greater than 0 ( one-tailed Welch’s t-test , p = 1 . 6 × 10−3 ) . We also calculated the average difference between μAi andμnormali=ρSi ( 12 hr ) 12 hr , where μnormali is the rate that would have been expected over a normal 12-hr night . We found that this average isμA−μnormal=1N∑i=1N ( μAi−μnormali ) =0 . 48±0 . 11 ( mgg−1 FW ) /hr , where the error was again estimated by propagating the standard errors of the mean . This value is again significantly greater than 0 ( one-tailed Welch’s t-test , p = 3 . 8 × 10−5 ) . These results provide strong evidence against the hypothesis that a fixed degradation rate is set only once at the first onset of darkness , and is instead compatible with our prediction that the rate is continuously re-computed throughout the night . In the following , we show how the models can explain the mutant phenotypes discussed in the main text ( see Figures 1B and 4 and Figure 4—figure supplement 1 ) . LHY and CCA1 are central components of the clock . The cca1/lhy mutant has a free-running period of significantly <24 hr under continuous light . This mutant is characterized by too high a rate of starch degradation ( see Figure 1B ) . Indeed , the starch reserve is exhausted around 21–22 hr after the previous dawn , instead of 24 hr as in the wild type . Interestingly , if these mutant plants are given an early night , the starch degradation rate is adjusted such that all the starch is again degraded around 21–22 hr after previous dawn ( see Figure 1B ) . The models we discussed previously can straightforwardly explain such a phenotype by assuming that in this mutant , the time of expected dawn is shifted to a time t < 24 hr after the previous dawn . There are different perturbations of the T dynamics that can produce this effect in our models , and still reproduce the data from cca1/lhy equally well . For instance , for cca1/lhy in Model 1 , [TC] after being reset , could decrease more steeply than in the wild-type , and drop to zero at a time t < 24 hr after the previous dawn . Then [TC] could remain at low levels , before rising again around 24 hr . Another perturbation appropriate for both Models 1 and 2 , would be to phase shift the T dynamics given by Equations 1 and 2 , respectively by a time t0 < 0 hr , in such a way that the time of expected dawn becomes ( 24 + t0 ) hr < 24 hr , as discussed above in the ‘Parameters’ section . For the sake of simplicity , we fitted the cca1/lhy data by assuming that the latter perturbation takes place , and accordingly we used t0 as a fitting parameter . lsf1 and sex4 mutant plants were kept in the dark for 132 hr after the end of the previous light period , then exposed to normal 12-hr light/12-hr dark cycles and the starch content measured . As Figure 4A shows , during the days following the prolonged dark period the mutant plants failed to degrade their entire starch reserve at night . Instead , the total starch content degraded by the end of each night , expressed as a percentage of the starch content at the end of the respective preceding light period , was approximately constant and much lower than that of wild-type plants ( Figure 4B ) . Interestingly , the same mutants could also adjust their starch degradation rate in response to an unexpected early night . In order to quantitatively verify this observation , we performed linear fits of the data from the unexpected early and normal nights:ρSearly ( t ) =−μearly⋅ ( t−tearly ) +ρS , 0early , ρSnormal ( t ) =−μnormal⋅ ( t−tnormal ) +ρS , 0normal , where ( μearly , ρS , 0early ) and ( μnormal , ρS , 0normal ) are the fitting parameters , with tearly = 8 hr , tnormal = 12 hr the times of onset of darkness respectively for the unexpected early and normal night . From these fits , we calculated the ratio R between the starch degradation rates in the normal and the unexpected early night , normalized by the respective starch content at the time of onset of darkness:R=μnormal/ρS , 0normalμearly/ρS , 0early . If the degradation rate is adjusted as happens in wild-type plants , during the unexpected early night relative to a normal night , the same fraction f of the initial starch content should be degraded by the time of expected dawn . Therefore , μnormal≈fρS , 0normaltday−tnormal and μearly≈fρS , 0earlytday−tearly; hence , R≈tday−tearlytday−tnormal=1612≈1 . 33 . As Figure 4F shows , the values of R found for lsf1 and sex4 are compatible with 1 . 33 ( dashed line ) , therefore supporting the hypothesis that these mutants are able to appropriately adjust their starch degradation rate in response to an unexpected early night . We also found the same type of rate adjustment in bam3 , bam4 and isa3 mutants ( see Figure 4F and Figure 4—figure supplement 1 ) , which also failed to exhaust their starch reserves by the end of the night . The models make a precise prediction about how such a phenotype can be produced . We will focus on Model 1 , but analogous arguments also hold true for Model 2 . We showed that for Model 1 , in order to obtain a linearly decreasing starch content with time , and with the appropriate degradation rate to ensure complete starch depletion at the time of expected dawn , two normalization conditions must hold:fD1AdkSkST1V0β=γ=1 , mSfD2α fD1=χ=1 . The first condition ensures that the number of S molecules , and therefore the starch content , decreases linearly with time during the night . The second condition is needed to keep the number of S molecules and the amount of starch strictly proportional , so that both these quantities are fully degraded by the end of the night . We now assume that the second of the two conditions is not valid , and , in particular , that χ < 1 . In this case , the starch degradation proceeds at a slower rate , breaking the proportionality between the number of S molecules and the amount of starch . By using Equations 4 and 8 , and the initial condition at the start of the dark period , ΔStot=αNStot , we find that a fraction of starch approximately equal to χ only is degraded by the time of expected dawn . Such a phenotype is clearly compatible with the disruptions caused by the above mutations . Indeed by assuming that the second condition above is not valid , our models can be well fit to the full night-time starch profiles in all the above mutants ( data not shown ) . One way to perturb only the second of the two normalization conditions shown above ( and not the first ) , would be to alter fD2 , which is the starch degradation reaction parameter . This reasoning fits well with the known functions of the above genes , all of which play roles , directly or indirectly , in starch degradation . The pwd mutant lacks the glucan water dikinase responsible for the addition of the phosphate groups to the 3-position of glucose moieties in starch . This mutation generated an approximately linear decrease of starch content with time during the night , but with a rate which was too low to ensure the complete utilization of the starch reserve by the time of expected dawn . Yet , as opposed to the previously discussed mutations , this mutant did not have the ability to adjust the starch degradation rate in response to an unexpected early night ( see Figure 4E , F ) . Indeed , for this mutant we found that the ratio between the degradation rates ( normalized by the respective end-of-light period starch content ) during the normal and the early night was R = 1 . 10 ± 0 . 10 , which is significantly less than the wild-type value ( R = 16/12 , one-tailed Z-test , p = 0 . 01 ) . This finding indicates that PWD is potentially a key node where information about starch content and expected time to dawn from the circadian clock are integrated to control starch degradation dynamics . This makes PWD an obvious target for future experiments that aim to elucidate the identities of the S and T molecules . The models detailed above use two distinctly different methods of implementing the division operation . In one case , exemplified by Model 1 , the S and T molecule concentrations are divided , with the T molecule concentration tracking the time to expected dawn . In Model 2 , the S and T concentrations are multiplied , with the T molecule concentration tracking the reciprocal of the time to expected dawn . The precise implementation of these two methods of performing arithmetic division could , however , vary , with slightly different reaction schemes but the same underlying principles . To illustrate this point , we briefly outline other models related to Model 1 above , where the TC molecule again has the behavior given in Equation 1 . We first consider:rS:SC↔SrT:TC↔TrT2:T+T↔T2rST:S+T↔STrST2:S+T2→ST2rSCT2:ST2→SC+T2rD1:ST→T rD2:ST+Starch→ST . Here both S and T molecules can directly and reversibly associate to the granule surface ( reactions rS and rT ) . Once bound to the surface , S and T can form a complex ST ( reaction rST ) that permits S molecule and starch degradation ( through reactions rD1 and rD2 respectively ) . The T molecule can also hinder starch degradation by forming dimers ( reaction rT2 ) that are able to associate with S molecules ( reaction rST2 ) and induce them to detach from the granule surface ( reaction rSCT2 ) . By a similar analysis to that carried out for Model 1 and Model 2 , and for similar reasons , it can be shown that this model can also implement arithmetic division between the S and T molecule concentrations . More complex models can also easily accommodate additional molecules , which , for instance , could recruit S and T molecules to the granule surface and be part of the starch degradation apparatus . For example , in the set of reactions:rM:MC↔MrS:SC+M→SMrT:TC+M→TMrSM:SM→SC+MCrTM:TM→TC+MCrD1:SM→MCrD2:SM+Starch→SM , M molecules which reversibly bind to the granule surface ( reaction rM ) , recruit the S and T molecules from the surrounding compartment ( reactions rS and rT ) , to form the SM and TM complexes , which can then dissociate from the surface through the reactions rSM and rTM . The SM complex can permit S molecule and starch degradation ( reactions rD1 and rD2 ) . Since the T molecule hinders starch degradation by binding to M molecules and preventing them from binding to S , it can be shown that this model can also implement arithmetic division between the S and T molecule concentrations . Exposure of Arabidopsis plants to light during the normal night could potentially lead to the early , light-induced expression of clock genes , and under some circumstances might re-entrain the clock . Such re-entrainment might affect the level of the T molecule . To discover whether our experimental conditions gave rise to such problems , we investigated the behavior of transcript levels of LHY , a central clock gene , by qPCR analysis . For the experiment shown in Figure 3B , LHY transcript levels were slightly elevated during the night-time light period , then peaked about 2 hr later than in plants that were not exposed to the light period ( Figure 3—figure supplement 1 ) . This result indicates that the night-time light period has only minor effects on the clock , which is not re-entrained: if this were the case the peak of LHY expression observed with the night-time light period would not be expected to occur at around 24 hr after the previous dawn .	Many organisms build up reserves of food when it is available so that metabolism and growth can continue when food is no longer available . Plants , for example , use energy from the sun to produce starch during the day , which is then used as a source of energy during the night . In some plants the amount of starch increases linearly with time during the day , and declines linearly with time during the night , so that the plant contains very little starch when the sun rises the following day . Although a plant is likely to starve if it cannot store or consume starch effectively , very little is known about the mechanisms that plants use to ensure that they store enough starch and do not use it up too quickly . Plants are able to track the time to dawn using their internal circadian clock so , as Scialdone et al . point out , if they can also track how much starch they have stored , they might somehow be dividing the amount of starch by the time to dawn to work out the rate at which starch can be consumed so that it lasts until sunrise . But could plants actually perform such calculations ? To gain some insight into this puzzle , Scialdone et al . constructed mathematical models in which information about the size of the starch store and the time until dawn was encoded in the concentrations of two kinds of molecules ( called S for starch and T for time ) . In one model , they propose that the S molecules stimulate starch consumption , and the T molecules prevent this from happening , with the rate of starch consumption being equal to the concentration of S molecules divided by the concentration of T molecules . These models were able to reproduce the results of previous experiments , including experiments in which dawn arrived unexpectedly early or unexpectedly late . Scialdone et al . then performed experiments which confirmed predictions of the models for the pattern of starch consumption in plants lacking relevant enzymes . These experiments also revealed that an enzyme called PWD may be the point at which results of the division computation are integrated into the starch consumption pathway . More generally , this work shows that sophisticated arithmetic computations can be important in biology . Moreover , whereas computers rely on digital logic , Scialdone et al . show that arithmetic computations can also be performed by exploiting the analogue dynamics that take place between molecules .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "plant", "biology" ]	2013	Arabidopsis plants perform arithmetic division to prevent starvation at night
Aggregation of Cu–Zn superoxide dismutase ( SOD1 ) is implicated in the motor neuron disease , amyotrophic lateral sclerosis ( ALS ) . Although more than 140 disease mutations of SOD1 are available , their stability or aggregation behaviors in membrane environment are not correlated with disease pathophysiology . Here , we use multiple mutational variants of SOD1 to show that the absence of Zn , and not Cu , significantly impacts membrane attachment of SOD1 through two loop regions facilitating aggregation driven by lipid-induced conformational changes . These loop regions influence both the primary ( through Cu intake ) and the gain of function ( through aggregation ) of SOD1 presumably through a shared conformational landscape . Combining experimental and theoretical frameworks using representative ALS disease mutants , we develop a ‘co-factor derived membrane association model’ wherein mutational stress closer to the Zn ( but not to the Cu ) pocket is responsible for membrane association-mediated toxic aggregation and survival time scale after ALS diagnosis . The aggregation of SOD1 is believed to be one of the chief causative factors behind the lethal motor neuron disease , amyotrophic lateral sclerosis ( ALS ) ( Shaw and Valentine , 2007 ) . Although more than 140 SOD1 mutations have been reportedly associated with ALS , there is no correlation between the stability ( and aggregation ) of these mutations and their disease manifestations . SOD1 aggregation has been investigated extensively in vitro by altering the solution conditions , such as temperature , pH , the presence of metal chelators , and the reduction of disulfide linkages ( Bush , 2002; Niwa , 2007; Rodriguez et al . , 2005 ) . The results of these studies clearly suggest that the aggregation of SOD1 is heterogeneous containing multiple steps , which is presumably the reason behind the lack of a structural understanding of aggregation processes ( Pasinelli et al . , 2004; Tomik et al . , 2005 ) . WT SOD1 contains Cu2+ ( Cu ) and Zn2+ ( Zn ) as cofactors . It has been established that Cu is responsible for the primary function of SOD1 ( the dismutase activity ) , and cell membrane acts as a scaffold in the process of Cu transfer to apo-SOD1 ( metal free non-functional protein ) through a Cu delivery chaperone ( CCS ) ( Culotta et al . , 1997 ) . Previous studies have found noticeable presence of SOD1 in human serum lipoproteins , mainly in LDL and HDL , hinting at a possible protective role of SOD1 against the lipid peroxidation ( Mondola et al . , 2016 ) . It has also been noted that SOD1 has a physiological propensity to accumulate near the membranes ( Ilieva et al . , 2009 ) of different cellular compartments , including mitochondria , endoplasmic reticulum ( ER ) , and Golgi apparatus ( Manfredi and Kawamata , 2016 ) . In addition , computational studies have shown that the electrostatic loop ( loop VII , residues 121–142 ) and Zn-binding loop ( loop IV , residues 58–83 ) promote membrane interaction of apo-SOD1 initiating the aggregation process ( Chng and Strange , 2014 ) . Membrane binding induced aggregation of SOD1 has also been shown experimentally both in vitro and inside cells ( Hervias et al . , 2006; Yamanaka et al . , 2008; Choi et al . , 2011 ) . Inclusions of SOD1 have been detected in the inter-membrane space of mitochondria originating from the spinal cord ( Mondola et al . , 2016 ) . The above results can be reconciled by suggesting that cell membrane can play crucial roles not only in shaping up the primary function of the protein , but also in defining its aggregation process of generating fibrillar and non-fibrillar aggregates ( the gain of function ) , with the loops IV and VII contributing critically to both processes . We hypothesize that ( 1 ) the induction of metal cofactors for the stabilization of loop IV and VII , membrane interaction , and SOD1 aggregation would be some of the crucial elements in defining the overlapping folding-aggregation landscape of SOD1; ( 2 ) metal pocket perturbation by mutational stresses ( as in disease variants ) would modulate membrane association and facilitate aggregation , and ( 3 ) the difference in aggregate morphology as a result of differential membrane interaction may contribute to the variation in cellular toxicity observed in ALS . In this paper , we investigated the above hypothesis by studying how different structural elements ( i . e . co-ordination of individual metals , membrane association , and the location of mutations ) attenuate the toxic gain of function of SOD1 . An effective understanding of the role of individual metals ( Cu and Zn ) would require studying SOD1 variants containing only one metal ( Cu or Zn ) in addition to a variant that contains none . We have therefore prepared an apo ( metal free ) protein , which serves the latter purpose . For the former , we have generated two single metal containing mutants of SOD1 , viz . H121F ( only Zn , no Cu ) and H72F ( only Cu , no Zn ) , which are situated near the key loop VII ( H121F ) and loop IV ( H72F ) at the protein structure ( Figure 1a ) . Using computational analysis based on a statistical mechanical model and detailed in vitro experiments , we propose here a ‘Co-factor derived membrane association model’ of SOD1 aggregation and its possible implication in ALS . We demonstrate that differential metal binding and membrane assisted conformational changes can work in concert to attenuate the rate and propensity of aggregation . While apo ( no metal ) protein and H72F mutant ( no Zn ) experience strong membrane interaction , the WT ( both metals ) protein and H121F ( no Cu ) mutant do not show significant binding . We further find that membrane-induced aggregates of H72F and apo protein showed significantly higher toxicity in terms of cell death and model membrane deformation when compared to WT and H121F mutant . We finally check the validity of this model to ALS using computational and experimental studies . For the computational validation , we show , using 15 ALS disease mutants , that the distance between the mutation site and Zn correlates well with the membrane binding energy and patient survival time after disease diagnosis , while Cu site does not seem to have any prominent role . For the experimental study , we use two well-studied disease mutants ( G37R where mutational site is close to the Cu pocket and I113T where mutational site is near Zn pocket ) to show that the model accounts well for their membrane binding/aggregation , correlating well with their disease onset phenotypes . This model puts forward a mechanism that Zn pocket destabilization ( either by metal content variation or by mutational stress near Zn center ) is the driving force behind the toxic gain of function of SOD1 mediated by the process of membrane association . The large size of SOD1 ( 151 residues ) precludes a detailed characterization of the conformational landscape , the role of ions in determining the stability-folding mechanism , and the effect of numerous mutations via all-atom simulation methods . To overcome this challenge and to obtain a simple physical picture of how the energetics of folding is governed by metal ions , we resort to constructing the folding landscape of reduced SOD1 variants through the statistical mechanical Wako–Saitô–Muñoz–Eaton ( WSME ) model ( see Materials and methods for model description and parametrization ) ( Wako and Saitô , 1978; Muñoz and Eaton , 1999 ) . Here , we employ the bWSME model where stretches of three consecutive residues are considered as a block ( b ) that reduces the total number of microstates from 42 . 7 million to just ~450 , 000 ( Gopi et al . , 2019 ) . The model , however , incorporates contributions from van der Waals interactions , simplified solvation , Debye–Hückel electrostatics , excess conformational entropy for disordered residues , and restricted conformational freedom for proline residues ( Naganathan , 2012; Rajasekaran et al . , 2016 ) . The predicted average folding path of SOD1 WT ( with both Cu and Zn bound ) highlights that the folding is initiated around the metal binding regions with early folding of the loop IV ( nucleated by Zn ) in the unfolded well and aided by flickering structure in the electrostatic loop ( loop VII , nucleated by Cu ) . The rest of the structure coalesces around this initial folding site leading to the native state . This folding mechanism is very similar to that proposed earlier via detailed kinetic studies ( Leinartaite et al . , 2010 ) . In the absence of metal ions , the apo variant folds through an alternate pathway wherein the folding is nucleated through the first three strands ( residues 1–40 ) following which the rest of the structure folds , recapitulating the results of single-molecule experiments ( Figure 1; Sen Mojumdar et al . , 2017 ) . It is important to note that the first three strands exhibit higher aggregation propensity as predicted from different computational servers ( Figure 1—figure supplement 1– ) . Interestingly , the folding mechanism of the Zn-bound SOD1 ( with no Cu bound ) is similar to that of the WT hinting that Zn coordination promotes proper folding . On the other hand , the folding mechanism of the Cu-bound SOD1 ( in the absence of Zn ) is similar to that of apo-SOD1 with additional folding probability in the region around the electrostatic loop . Taken together , the statistical modeling highlights how the absence of metals and particularly the absence of Zn ( or mutations that affect Zn binding and not Cu binding ) alters the folding mechanism by populating partially structured states involving beta strands in the unfolded well thus possibly increasing the chances of aggregation . Importantly , the model provides multiple testable predictions on the differential roles of Zn and Cu , which we address below via experiments . We have recently shown that the mutants H121F and H72F contain negligible Cu and Zn , respectively , while the apo protein is completely devoid of metal ( Chowdhury et al . , 2019 ) . We validated this further using atomic absorption spectroscopy ( Table 1 ) and activity measurements ( Figure 2—figure supplement 1 ) . Guanidinium-induced equilibrium unfolding transitions of H121F and H72F were found to be similar ( Figure 2—figure supplement 2 ) . We used steady-state tryptophan fluorescence , far UV CD , and FTIR spectroscopy to characterize these different proteins . SOD1 is a single tryptophan protein ( Trp32 ) , in which the tryptophan residue has been shown to be partially buried ( Muneeswaran et al . , 2014 ) . The role of Trp32 within the sequence segment ( Lomize et al . , 2012 ) KVWGSIKGL ( Gohil and Greenberg , 2009 ) of high aggregation propensity has been investigated before ( Taylor et al . , 2007 ) . We found that the formation of apo form resulted in a large shift in Trp32 emission maximum ( 332 nm for WT protein and 350 nm for apo protein ) ( Figure 2a ) . In contrast , other two mono-metallated variants ( H121F and H72F ) exhibited fluorescence emission maxima at wavelengths , which were intermediate between the WT and apo proteins ( 342 nm for H121F and 345 nm for H72F ) ( Figure 2a ) . Next , we performed acrylamide-quenching experiments to measure the solvent surface exposure of Trp32 for all variants . The values ( Table 2 ) of the Stern−Volmer constant ( Ksv ) were determined using a straight line fit , as shown in Figure 2—figure supplement 3 . Ksv for WT ( 6 . 8 ± 0 . 1 M−1 ) was significantly lower than that of apo SOD1 ( 14 . 3 ± 0 . 1 M−1 ) . Steady-state fluorescence maxima in combination with acrylamide quenching data suggested an appreciable conformational alteration in going from the WT to the apo form . Interestingly , Zn-starved H72F mutant showed higher Ksv compared to Cu-starved H121F mutant ( Figure 2—figure supplement 3 , Table 2 ) . Far-UV circular dichroism ( CD ) spectra for WT and apo protein were in line with earlier observations ( Figure 2—figure supplement 4; Banci et al . , 2007 ) . Specifically , we found a slight broadening in the far UV CD spectrum of the apo protein when compared to the WT . In agreement with steady-state fluorescence data , the far UV-CD spectrum of the Zn-deficient H72F protein was found to be similar to the apo variant , while the WT- and Cu-deficient H121F variant displayed similar spectra . We then used FT-IR spectroscopy to complement far-UV CD results and to obtain a preliminary estimate of the secondary structure contents of the protein variants , using amide-I FTIR spectral region . The carbonyl ( C = 0 ) stretching vibrations at amide-I region provides information related to the secondary structure ( beta sheet 1633–1638 , alpha helix 1649–1656 , disorder and turns and loops 1644 and 1665–1672 cm−1 ) . The analyses of the FT-IR data were carried out using published method using two steps ( Yang et al . , 2015; Kong and Yu , 2007; Bandyopadhyay et al . , 2021 ) . First , the peak positions were assigned using the double derivatives of the FT-IR data for all protein variants ( Figure 2—figure supplement 5 shows the representative double derivative plot of WT SOD1 in the absence of lipid ) . The peak positions were selected from the minima of the secondary derivatives of the FT-IR absorbance data . In the second step , selected peak positions thus determined were used for the fitting of the FT-IR raw data using Gaussian distributions analyses . Analysis of the secondary structure of WT protein ( Figure 2b ) showed the presence of 10% alpha helix , 38% beta sheet , and 52% turns and loops including disordered stretches . The percentage of the secondary structure determined from the FT-IR analysis was found to be consistent with the data obtained from the crystal structure ( PDB 4BCY with 11% alpha helix , 40% beta sheet , and 49% turns and loops ) , thus validating our method ( Danielsson et al . , 2013 ) . FT-IR data showed a decrease in beta sheet content ( from 38% to 31% ) as apo protein ( Figure 2c ) formed . In contrast , the behavior of H72F mutant ( Figure 2—figure supplement 6a , beta sheet content of 32% ) was found to be similar to the apo protein , while H121F mutant ( Figure 2—figure supplement 6b , beta sheet content of 37% ) remained similar to WT protein . The percentage of secondary structure elements of all protein variants are shown in Figure 2d . To obtain a preliminary understanding of the possible membrane binding sites of SOD1 , we resorted to computational techniques using ‘Orientation of protein in membrane’ tool ( Lomize et al . , 2012 ) , which predicted weak interaction of WT on membrane surface ( Figure 2—figure supplement 7a ) . In contrast , the same calculation predicted higher binding affinity of apo protein with the membrane ( Figure 2e ) . When we used ITASSER-modeled structures , the computed values of ΔGtransfer ( free energy change of protein transfer from bulk to the membrane ) was found to be substantially higher for the apo ( −2 . 6 kcal mol−1 ) when compared to the WT protein ( −1 . 2 kcal mol−1 ) . When we used the crystal structure of the WT protein , ΔGtransfer calculation modeling yielded similar results for the WT protein ( −0 . 9 kcal mol−1 ) . To probe protein-lipid binding constants experimentally ( Ka , M−1 ) , we used fluorescence correlation spectroscopy ( FCS ) . FCS monitors diffusional and conformational dynamics of fluorescently labeled biomolecules at single-molecule resolution ( Chattopadhyay et al . , 2002 ) . For FCS experiments , we labeled the cysteine residues of all the SOD1 variants using Alexa-488-maleimide . Figure 2f shows a schematic diagram of how the labeled proteins and protein-lipids complex would behave inside the confocal volume . Using FCS we determined the correlation functions using 50 nM Alexa488Maleimide protein in the presence of increasing concentration of DPPC small unilamellar vesicles ( SUVs ) ( Figure 2g showed the typical correlation functions of alexa labeled apo SOD1 in absence and presence of 100 nM and 500 nM DPPC SUVs ) . We fit the correlation functions using a two component diffusion model and the goodness of the fit was established using the randomness of the residual distribution . In this model , the fast and slow diffusing components corresponded to the free ( with rH113 . 5 Å ) and lipid bound protein ( rH2170 Å ) respectively ( Figure 2h ) . With increasing DPPC SUV concentration , the percentage of slow component increased ( Figure 2f ) , which occurred at the expense of the fast component , and a sigmoidal fit of either of these components yield the values of Ka , which showed that the binding affinities followed the trend: apo ≥H72F>H121F>WT ( Figure 2i , Table 3 ) . Since FCS experiments required the use of labeled proteins in which the presence of bulky fluorescence dye can potentially influence the results , we complemented FCS binding data by measuring the tryptophan fluorescence of the SOD1 variants with increasing concentrations of DPPC SUVs . From the gradual enhancement of tryptophan fluorescence due to lipid binding , we calculated the binding affinities of all the protein variants towards membrane which showed comparable binding constants as obtained from our FCS experiments ( Figure 2—figure supplement 7b , c ) . We then measured the Stern–Volmer constants using acrylamide quenching experiments of Trp32 fluorescence with protein variants in the absence ( Ksv ) and presence of ( Ksvm ) membrane . The parameter Ksv/Ksvm was found maximum for the apo protein , and minimum for WT ( Figure 2—figure supplement 7d , Table 2 ) . H121F and H72F variants behaved like WT and apo protein , respectively . As observed by FT-IR , DPPC binding resulted in no or minimum change in conformation for WT and H121F proteins ( Figure 2—figure supplement 7e , f ) , while a large decrease in beta sheet content with simultaneous rise in non-beta content , specifically alpha helical content , was observed for the apo protein and H72F mutant ( Figure 2j , k Figure 2—figure supplement 7g ) . Aggregation kinetics of WT , apo , and the mutant SOD1 in their TCEP reduced states were studied systematically both in the absence and in the presence of DPPC . A typical protein membrane ratio of 1:2 was maintained for all measurements involving membranes . For the initial assessment of the aggregation kinetics , the fluorescence intensity enhancement of amyloid marker Thioflavin T ( ThT ) was monitored . ThT is known to bind to protein aggregates with cross beta structure giving rise to a large increase in its fluorescence intensity . From the ThT fluorescence assay , we found that the WT protein does not aggregate , both in the absence or in the presence of membrane ( Figure 3a , b ) . For the H121F variant in the absence of membrane , we found a small and slow enhancement of ThT fluorescence and the profile remained unchanged when we added the membrane ( Figure 3a , c ) . In contrast , for apo and H72F variants , ThT assay showed large fluorescence increase and the kinetics followed typical sigmoidal patterns . The addition of membrane increased the rate of aggregation for both variants and a large decrease in the lag times . When compared between the apo and H72F variants , we found that the rate of aggregation is higher ( i . e . with less lag time ) for the apo protein ( Figure 3a , b , c Table 4 ) . We then imaged using atomic force microscopy ( AFM ) the aggregates collected from the plateau regions of the aggregation kinetics ( at a time point when the fluorescence of ThT was maximum [saturated] and did not change ) . Protein ( P ) aggregates will be denoted by Pagg , and Paggm to indicate if they are formed in the absence or presence of membranes respectively . For example , the aggregates of WT in the absence and presence of membranes would be denoted by WTagg , and WTaggm , respectively . AFM imaging also showed that in the absence of membrane , WT and H121F did not form aggregates , fibrillar , or otherwise ( Figure 3—figure supplement 1 ) , while large fibrillar aggregates were found to form with apo ( Figure 3d ) and H72F mutant ( Figure 3—figure supplement 1 ) . The average size of the fibrillar apoagg was found to be 1 . 8–2 μm with an average height of 20 nm . H72Fagg showed similar morphology ( Figure 3—figure supplement 1 ) . Significant morphological differences were noticed for the aggregates of apo and H72F variants , in the absence ( Figure 3d , e , Figure 3—figure supplement 1 ) and the presence of the membrane . The apoaggm appeared to exhibit network of thin aggregates ( the average size was found to be 700–800 nm with an average height of 6–8 nm ) which were found to be connected by the spherical DPPC vesicles ( Figure 3e , inset; Figure 3f ) . To understand the effect of curvature on membrane binding and aggregation , we performed the binding experiments with H72F mutant using liposomes of different curvatures . We used DPPC SUVs ( diameter ~78 nm ) , LUVs ( diameter ~140 nm ) and GUVs ( ~20 μm ) for this study . Our results showed that with increasing curvature of the DPPC lipid vesicles , binding , and aggregation of H72F increased that is fibril formation rate and extents were found to be highest in the presence of SUVs and moderate for LUVs and lowest for GUVs ( Figure 3—figure supplement 2a , b ) . To investigate the effect of H72F mutant toward curvature induction in GUVs , we studied the aggregation of H72F in the presence of GUVs and the final point aggregates were imaged through a transmission electron microscope . We found that as a result of the incubation with H72F , the size of the GUVs reduced many folds starting from an average size of 5–10 μm to about 1–2 μm . Numerous smaller vesicles are observed in the background that suggest a possible vesiculation of the GUV induced by H72F during the long incubation period ( Figure 3—figure supplement 2c , d ) . Such vesiculation has also been reported earlier in some cases of other amyloid structures ( Meker et al . , 2018 ) . These results suggested possible membrane remodeling induced by Zn-deficient H72F mutant , an aspect we would like to study in further detail . To determine the effect of membrane composition , we studied DPPG , which is a widely used negatively charged membrane with chain length identical to DPPC . Interestingly , our results showed that both binding and the rate of aggregation increased in the presence of a negatively charged membrane ( Figure 3—figure supplement 3 ) . We investigated the effect of the aggregates of different protein variants on cellular toxicity ( by measuring cell viability ) in general and on the cell membranes ( using a number of spectroscopy and imaging assays ) in particular . Cell viability was measured by the standard MTT assay . We used aggregates collected at the plateau region of the aggregation kinetics . MTT assay using SHSY5Y cell line showed minimum toxicity in terms of cell death for WTagg , WTaggm , H121Fagg , and H121Faggm ( Figure 4—figure supplement 1 ) . In contrast , apoagg , and H72Fagg , showed significantly higher neuronal dead cell population which increased further with apoaggm and H72Faggm , respectively . Although neuronal cell death is one of the decisive factors of aggregate toxicity , the severity of ALS has been found to depend on the extent of membrane perturbations , which may contribute to multiple events including ( 1 ) mitochondria-associated membrane ( MAM ) collapse and disruption ( Watanabe et al . , 2016 ) , ( 2 ) the synaptic dysfunction due to impaired synaptic vesicles function toward neurotransmission ( Casas et al . , 2016; Song , 2020 ) , and ( 3 ) the prion like spread of toxic aggregates between cells presumably through macropinocytosis ( Yerbury , 2016; McAlary et al . , 2019 ) . Therefore , it is necessarily important to investigate in detail the protein aggregates induced membrane perturbation , which has never been addressed before . We used three different assays for the in vitro studies , viz . ( 1 ) phase-contrast microscopy using a membrane model of giant unilamellar vesicle ( GUV ) to probe how the presence of aggregates change their size and shapes , ( 2 ) a calcein release assay to probe aggregate-induced pore formation , and ( 3 ) FTIR to determine the molecular mechanism of the influence of different aggregates on the structure of lipids . For the imaging assay , we used time based optical microscopic investigation to unveil how the addition of Pagg , and Paggm affects the size and shape of GUVs . GUV is widely used as a model membrane system , providing free-standing bilayers unaffected by support-induced artifacts yet with sufficiently low curvature to well mimic cellular membranes and mitochondrial membrane as well . An advantage of this assay comes from the use of phase contrast without requiring any external fluorophore label . We made the GUVs composed of DOPC:DOPE:PI:DOPS:CL in the ratio 4 . 5:2 . 5:1:0 . 5:1 . 5 , which mimics mitochondrial membrane composition ( Gohil and Greenberg , 2009 ) . Figure 4a is a representative example of how apo-aggregates formed in the absence of membrane ( apoagg ) behaved with the GUVs . In contrast , Figure 4b shows the influence of the apo-aggregates formed in the presence of membrane ( apoaggm ) ( Figure 4—figure supplement 2 ) . Figure 4c shows an example of a control , in which WT samples ( WTagg , which should contain minimum aggregates population ) were added to the GUVs . GUV images of Figure 4a , b clearly demonstrate that the aggregates attach on the surface of GUVs ( shown by arrows ) leading to membrane deformation and change in lamellarity . These two images also show the pore formation , which was further established by the contrast loss . A comparison between Figure 4a , b shows visually that membrane deformation is more prominent ( more damage ) and faster ( occurs at earlier time points ) in the presence of Paggm . All these changes in GUVs were found absent in Figure 4c , in which WTagg , was used . For the quantification of the image data , we determined the difference in refractive indices between the exterior and interior of GUVs ( represented by Iptp , peak-to-peak intensity in Figure 4d; Sannigrahi et al . , 2019 ) . In the case of intact vesicles , the values of Iptp would be high ( Figure 4d , left ) due to the difference in sugar asymmetry between outside and inside of GUVs , which would disappear in the case of membrane deformation ( Figure 4d , right ) . We plotted the time dependence of Iptp to determine quantitatively the membrane deformation kinetics by measuring the deformation rate constants , λ ( Figure 4e , inset ) . The GUV deformation was insignificant and no detectable kinetics were found for WTagg , , WT aggm , H121Fagg , and H121Faggm ( Figure 4e ) . On the other hand , apoagg , and H72Fagg , exhibited significantly high deformation rate ( λapoagg ~ 1 . 8×10−3 s−1 and λH72Fagg ~ 2 . 1×10−3 s−1 ) and both kinetics appeared cooperative ( sigmoidal behavior , Figure 4e ) . It is interesting to note that in the presence of apoaggm and H72Faggm , the deformation rate increased significantly ( λapoaggm ~ 3 . 5×10−3 s−1 and λH72Faggm ~ 3 . 9×10−3 s−1 ) . More interestingly , both apoagg and apoaggm showed a tendency to create attachments between vesicles to generate vesicular assembly and co-operative deformations ( high-contrast images in Figure 4f , which is also shown by a schematic drawing ) . The image analysis and visualization of the GUVs in presence of apoaggm suggested that the co-operative vesicular clustering and deformation presumably occurred through allosteric communications mechanism by the aggregates ( Figure 4—figure supplement 2 ) . We found that apoaggm and H72Faggm are more efficient towards inducing vesicular assembly and deformations ( Figure 4f , Figure 4—figure supplement 3 , Figure 4—figure supplement 4 ) . To rule out the effect of only SUVs ( which are present along with the protein ) , we performed a control experiment by treating the GUVs with similar concentration of DPPC SUVs as employed for the formation of Paggm . Our results showed insignificant changes in Iptp values in the presence of DPPC SUVs ( Figure 4—figure supplement 5 ) , further highlighting that Pagg and Paggm are responsible for the GUV perturbations . Electron microscopic images of the synapses infused with ALS variants , like G85R , showed vacant active zones ( AZs ) and occasional abnormal membranous structures , whereas there occurred no reduction in the synaptic vesicles number in case of WT SOD1 ( Song , 2020 ) . Similar observation was found previously by Wang et al . , 2009 . Using C . elegans as model system , they showed that the neuronal toxicity in ALS appears due to synaptic dysfunction that occurs because of misfolded and aggregated disease mutant driven lowering in the number of organelles including synaptic vesicles and mitochondria . The pore formation and vesicle rupture by aggregates may be a possible reason for the reduced synaptic vesicle population in case of ALS disease mutants . To investigate this issue , we prepared calcein entrapped SUVs composed of DOPE , DOPS , and DOPC at the molar ratio 5:3:2 to mimic the synaptic vesicle composition and curvature ( Fusco et al . , 2016 ) . In this assay , we measured the percentage of calcein leakage from the dye entrapped inside lipid vesicles ( Figure 4g , Figure 4—figure supplement 6 ) . The extent of calcein leakage after the treatment of different protein aggregates followed the trend similar to what was observed by GUV micrographic observation ( Figure 4e ) , which suggested that the membrane rupture is facilitated in significantly higher rate and extent by apoaggm and H72Faggm , while compared with apoagg and H72Fagg , respectively ( Figure 4g , inset , Figure 4—figure supplement 6a ) . In order to induce membrane deformation as observed by previous two assays , the aggregates need to inflict substantial changes in lipid structure . To determine the extent of conformational change the lipid molecules experience by protein variants , we used ATR-FTIR to measure quantitatively the populations of different rotamers in a general planar trans-oriented phospholipid bilayer of DPPC . CH2 wagging band frequency ( 1280–1460 cm−1 ) of the hydrocarbon tail region of the bilayer was carefully monitored for this purpose ( Lewis and McElhaney , 2013; Maroncelli et al . , 1982 ) . The results show a significant increase in the populations of nonplanar kink+gtg/ rotamers ( this band appears at 1367 cm−1 ) when planar lipid bilayer was treated with apoagg/apoaggm and H72Fagg/H72Faggm ( Figure 4g; Figure 4—figure supplement 6b , c ) . We also found that apoaggm/H72Faggm exerted greater effect on lipids than apoagg , /H72Fagg . Interestingly , differences in the populations of non-planar conformers are found to be similar to the variation in the extent of calcein release induced by protein variants ( Figure 4g ) . This observation provides preliminary evidence that both events occur by similar trigger ( presumably the membrane attachment by the aggregates ) . In the previous few sections using WT , apo , H121F , and H72F , we established that the WT ( containing both Zn and Cu ) and the Zn bound H121F protein did not aggregate or induced toxicity through membrane deformation . In contrast , the removal of either Zn ( H72F mutant ) or both metals ( the apo protein ) resulted in strong membrane association , aggregation and induction of cellular toxicity . The data clearly suggest that Cu plays secondary roles in aggregation induced toxicity , while Zn pocket destabilization acts in concert with membrane-induced conformational change resulting in aggregation and toxic gain of function . Since these experiments validate successfully the predictions from the statistical mechanical model , we subsequently wanted to understand this behavior could be generalized in ALS disease mutants . For this purpose , we used both in silico and experimental approaches . For the in silico method , we selected fifteen disease mutants of ALS , whose structures are available in the protein data bank ( PDB ) ( Table 5 ) . From the crystal structures , we determined the distance between the mutation site and Zn ( and Cu ) for all these disease mutants . In addition , we calculated the transfer free energies , that is the theoretical membrane association energies ( ΔGTr ) , of these mutants using the OPM server . We found that the negative values of ΔGTr decreased linearly with the increased distance of Zn site for these mutants , while it remained non-variant with the distance of Cu site ( Figure 5a ) . We also found that the severity of ALS mutants ( defined here as the survival time in year of ALS patients after diagnosis ) correlates nicely with the distance of the Zn sites ( Figure 5b; Wang et al . , 2008 ) . For the experimental validation , we used two representative ALS disease mutants of varying distance between the mutation site and Zn . The mutations , namely G37R ( mutation site-Zn distance 24 Å , less severe ) and I113T ( mutation site-Zn distance 16 Å , more severe ) ( Figure 5b ) are well studied in literature ( Milardi et al . , 2010; Krishnan et al . , 2006; Banci et al . , 2009 ) . Using FCS we also measured the Ka values of these two disease mutants which showed greater binding affinity for I113T ( Figure 5c , Table 3 ) . Structural characterization using FTIR showed greater extent of alpha helical content in I113T compared to G37R . On the other hand , further increase in the alpha helical structure in I113T was observed on interaction with model membrane . In contrast , we did not observe any membrane induced structural change for G37R ( Figure 5—figure supplement 1 ) . To understand the internal structural changes due to the disease mutations , we performed some in silico study using DynaMut webserver ( http://biosig . unimelb . edu . au/dynamut/ ) to predict the effects of mutational stress on the conformational dynamics , stability and flexibility of protein ( Rodrigues et al . , 2018 ) . The results suggested that G37R mutation increased the rigidity of different regions ( 6–16 , 30–42 , 80–86 aa ) of the protein , whereas significant increase in flexibility of the loop IV and VII region ( 48–52 , 130–150 ) was observed in case of I113T mutation ( Figure 5—figure supplement 2 ) . The predicted changes in folding free energy ( ΔΔG , kcal/mol ) for I113T was found to be much more negative than G37R ( ΔΔG ( ΔGWT – ΔGmutant ) ) suggesting that mutation at 113th position destabilized the protein more in comparison to 37th position ( Figure 5—figure supplement 2 ) . The vibrational entropy change ( ΔΔS ) in I113T was found to be +0 . 188 kcal mol−1 K−1 while −0 . 404 kcal mol−1 K−1 was found for G37R indicating the increase in flexibility for I113T and decrease in flexibility for G37R ( Figure 5—figure supplement 2 ) . Thus the increase in alpha helical content in I113T mutant arises due to the increase in flexibility in the membrane interacting loop regions as suggested by the DynaMut server . Subsequently , we used ThT fluorescence to probe the aggregation kinetics of G37R and I113T in the absence and presence of DPPC SUVs . Results in the absence of membrane showed significantly higher aggregation for I113T when compared to G37R . A notable increase in the rate and extent of aggregation was observed for I113T under membrane environment , whereas for G37R , the change was not significant ( Figure 5d ) . AFM imaging showed linear fibrillar aggregate for I113T in the absence of membrane ( average length is ~1 . 5 μm and height is 26 nm ) , whereas fibrilar network was observed when aggregation occurred in the presence of DPPC SUVs ( average height is 3 . 2 nm ) ( Figure 5e , f ) . In contrast , small nonfibrilar aggregates were found for G37R mutant both in the absence and in the presence of membranes ( Figure 5—figure supplement 3 ) . Finally , we studied the toxic effects of the aggregates of G37R and I113T , which formed in the absence and presence of membrane . Using MTT , we showed that I113T aggregates in the absence of membrane were less toxic than I113T aggregates in the presence of membrane ( Figure 5—figure supplement 4 ) . G37R exhibited minimal toxicity both in the absence or presence of membrane ( Figure 5—figure supplement 4 ) . We observed for these mutants a nice correlation between the extent of calcein release and the populations of nonplanar kink+gtg/ conformers ( Figure 5—figure supplement 5 ) . For further validation , we studied the membrane association and aggregation of a third disease mutant G85R . Earlier reports suggested that the survival time for G85R ( 6 years ) remains between G37R ( 17 years ) and I113T ( 4 . 3 years ) ( Wang et al . , 2008 ) . Our computational and experimental binding study using FCS ( Figure 5—figure supplement 6 ) showed that the binding affinity of G85R ( −2 kcal mol−1 ) is intermediate between that of G37R ( −1 kcal mol−1 ) and I113T ( −3 . 2 kcal mol−1 ) . In addition , AFM imaging ( Figure 5—figure supplement 6 ) suggests that G85R aggregation behavior in membrane environment remains between mild G37R and severe I113T mutants . In this work , we effectively bring a large number of studies together in a coherent framework to address the different roles of two metal cofactors ( Cu and Zn ) and membrane binding on the aggregation and induced toxicity of SOD1 . Based on steady-state fluorescence , acrylamide quenching , and FT-IR data , the investigated proteins could be classified as ( 1 ) WT and WT-like mutant , H121F and ( 2 ) apo and apo like mutant , H72F . The battery of experimental approaches clearly validates the prediction from the bWSME model that Zn is largely responsible for the conformational stability of SOD1 . The insertion of Zn at the loop IV region stabilizes the loop while reducing the aggregation propensity of the apo protein . We believe that the stabilization of this long loop may play an important role in SOD1 aggregation biology and its relevance in ALS . It may be noted that shortening of the loop has been shown to increase the stability of the protein ( Yang et al . , 2018 ) . This is also important to point out that SOD1 variants from thermophiles ( like SOD5 from C . albicans SOD5 ) have shorter electrostatic loops compared to human SOD1 ( Gleason et al . , 2014 ) . In contrast , the absence of Cu does not seem to have any significant effect on the stability of the protein , and the Zn containing Cu-deficient protein behaves like WT . We derive a possible maturation/aggregation landscape of SOD1 in the presence of membrane employing data from the current work and available literature ( Figure 6 ) . Since all experiments presented here were carried out using the reduced protein , this scheme does not take dimerization into consideration . SOD1 in its metal free state ( apo protein ) has been shown to be flexible and inherently dynamic . The sequence sites ( 48–80 and 120–140 ) of low folding probability predicted by bWSME calculation ( Figure 1b ) have significant overlap with the sequence sites ( 45–70 and 125–142 ) of high membrane binding as calculated by OPM ( Figure 2e ) . It has been found that truncation at Leu126 of C-terminus resulted in a protein with several transmembrane helices ( Lim et al . , 2015 ) . We found that the apo protein has high membrane binding possibility which is directly supported by membrane binding data ( Figure 2i , Figure 2—figure supplment 7b , c ) . Membrane bound apoSOD1 with an optimized orientation is a requisite for the Cu insertion to take place . Since biological systems have very-low-free Cu salt ( Rae et al . , 1999 ) , Cu coordination to SOD1 occurs through Cu chaperone protein ( CCS ) which transfers Cu to apo SOD1 using membrane as a scaffold ( Pope et al . , 2013 ) . Membrane scaffolding decreases the directionality of metal transport compared to a three-dimensional search of the metal ions . The domain I of CCS binds Cu through two cysteine residues . Since Cu binding affinity of CCS is less than that of SOD1 ( which recruits three or four cysteine residues ) , Cu transport occurs from CCS to SOD1 , and not the other way ( Banci et al . , 2010 ) . Zn binding , which occurs in the next step , stabilizes the loop regions . This event has a few important consequences . First , it has been shown that both CCS and SOD1 exhibit binding to Zn ( Proescher , 2008 ) . While the absence of Zn favors heterodimer formation with SOD1 ( CCS-SOD1 ) , the presence of Zn facilitates the homodimer ( CCS-CCS ) configuration . Second , as shown in this paper , Zn coordinated protein has little or minimum membrane binding and hence Zn bound SOD1 is removed from bi-layer . The third consequence comes from the reduced Zn affinity ( which may come from a Zn compromised mutation or other reason for the sporadic forms of the disease ) , which would result in a misfolded membrane bound protein that is aggregation prone and potentially toxic . As evident , the proposed scheme is somewhat Zn centric in which Cu coordination plays minimum role in the aggregation centric disease process . Partial support of this comes directly from the presented data of the distance dependence between ALS mutations sites and Cu . Also , the mutant proteins without Cu ( but with Zn site coordinated , e . g . H121F or G37R ) do not show aggregation , neither they induce toxicity as studied by our assay systems . Earlier works have also shown that Cu binding has no effect on SOD1 folding ( Bruns and Kopito , 2007 ) . More importantly , it has been shown that CCS knock-out does not affect the onset of the disease or the life span in SOD1 transgenic mice ( Subramaniam et al . , 2002 ) . It may be important to consider that , although more than 140 SOD1 mutants are known in ALS , the disease is predominantly sporadic . Nevertheless , the disease initiation may still be done by a metal free apo ( or Zn free ) configuration , which can be generated from the WT protein as a result of a different trigger ( and not by a genetic factor ) . The results obtained from two ALS mutants ( I113T and G37R ) can be discussed in this context . The present data show that I113T is similar to a zinc deficient protein , with high affinity toward membrane and higher aggregation propensity . In contrast , G37R behaves more like a WT protein . A comparison between the disease phenotype show that I113T is the second most common ALS mutant with average survival of 4 . 3 years , while for G37R , the average survival increases to 17 years . These results are in excellent correlation with the presented scheme . Another interesting correlation of the presented scheme comes from a result that Zn supplement with a moderate dose of 60 mg kg−1 day−1 can increase the days of survival in a transgenic mouse experiment ( Ermilova et al . , 2005 ) . Another approach could be using small molecules which would compete with the metal free protein towards membrane binding , an approach used recently in a Parkinson’s disease model ( Fonseca-Ornelas et al . , 2014 ) . While any drug development initiative targeting ALS and other neurodegenerative diseases suffer from the complications of diseases heterogeneity , poorly understood molecular mechanism and complex delivery avenues , a collaborative and integrative effort involving science and clinic , may be needed to find a successful solution . The molecular mechanism associated with ALS induced toxicity has been extensively studied and widely debated ( Broom , 2012 ) . SOD1 aggregates have been found in ALS patient samples ( Gruzman et al . , 2007; Guareschi et al . , 2012 ) . Mutant SOD1 aggregates have been shown to transfer from cell to cell using a prion like propagation mechanism ( Münch et al . , 2011 ) . It has also been shown that the aggregation induced toxicity of SOD1 variants can occur through its attachment on mitochondrial membrane surface in transgenic ALS mice ( Zhai et al . , 2009 ) . In parallel , pore formation at the membrane and abnormal ion mobility have been found with SOD1 aggregation ( Ray et al . , 2004 ) . Although the above reports directly link a membrane connection with SOD1 aggregation ( and presumably with ALS ) -unlike other neurodegenerative diseases like AD and PD , membrane-induced aggregation studies with SOD1 have been limited ( Choi et al . , 2011 ) . Shahmoradian and colleagues reported that the structure of Lewy bodies in Parkinson’s disease consists of α-synuclein and lipid vesicle clusters instead of the long-assumed amyloid fibril core ( Shahmoradian et al . , 2019 ) . Using correlative light and electron microscopy ( CLEM ) , they show that the vast majority of Lewy bodies actually consist of clusters of various membranous compartments , instead of amyloid fibrils as previously assumed . Assembly of synaptic vesicles has been shown recently by Fusco et al . , 2016 in which two different regions of a protein molecule can be used as a ‘double anchor’ to induce the assembly . We think three particular finding can be discussed in this connection . First , the presented membrane deformation assay using GUV clearly shows the formation of vesicles assemblies ( Figure 4f ) induced by apoaggm and H72Faggm . Second , for both protein variants , relatively small aggregates of the proteins ( formed in the presence of the membrane ) and not the large fibrilar network ( formed in the absence of membrane ) , favored vesicles assemblies . Third , the smaller sized aggregates formed in the presence of membrane seemed to induce more toxicity than the fibrils ( formed in the absence ) . From the above three considerations , it is easy to envision that the anchoring ability of the aggregates toward vesicles assembly would be more efficient for a small-sized aggregates when compared to a fibril . There is an increased interest in the role of the number and nature of membrane lipids in shaping the aggregation landscape of different proteins in neurodegenerative diseases . Given the prion-like propagation in these diseases , our work underscores how characterization of protein–lipid interactions could enhance our molecular understanding of cellular toxicity and enable the identification of therapeutic molecules to mitigate the damage . 1 , 2-Dipalmitoyl-sn-glycero-3-phosphocholine ( DPPC ) , 1 , 2-dioleoyl-sn-glycero-3-phosphoethanolamine ( DOPE ) , phosphoinositol ( PI ) , 1 , 2-dioleoyl-sn-glycero-3-phospho-l-serine ( DOPS ) , and cardiolipin ( CL ) were purchased from Avanti Polar Lipids Inc ( Alabaster , AL ) . 1 , 1′-Dioctadecyl-3 , 3 , 3' , 3'-tetramethylindotricarbocyanine iodide ( DiIC-18 ( 3 ) ) was purchased from Invitrogen ( Eugene , OR ) . All other necessary chemicals were obtained from Aldrich ( St . Louis , MO ) and Merck ( Mumbai , India ) . We employ the Ising-like WSME model ( Wako and Saitô , 1978; Muñoz and Eaton , 1999 ) with the block approximation ( Gopi et al . , 2019 ) to predict the conformational landscape of SOD1 oxidized monomer and its variants using the PDB structure 4FF9 as the reference . Briefly , the model assigns binary variables of 1 or 0 for folded or unfolded status of residues , respectively . We employ a version of the model that accounts of single-stretches of folded blocks ( single-sequence approximation ) , two stretches of folded blocks ( double-sequence approximation [DSA] ) and DSA allowing for interactions across the folded islands if they are interacting in the folded structure . The 151-residue protein SOD1 is therefore reduced to a collection of 49 sequential blocks on assuming a block length of 3 thus reducing the number of microstates from >42 , 700 , 000 ( in the residue-level version of the model ) to 461 , 826 . The energetics of the model involves van der Waals interactions identified with a 6 Å heavy-atom cut-off , all-to-all Debye–Hückel electrostatics , and simplified solvation ( defined by the heat capacity change per native contact of ) ( Naganathan , 2012 ) . Residues identified to be fully folded are assigned an entropic penalty of −13 . 6 J mol−1 K−1 per residue ( ΔSconf ) . The apo form of SOD1 is simulated by assigning an excess conformational entropy of −19 . 7 J mol−1 K−1 per residue ( Rajasekaran et al . , 2016 ) for the stretches of residues 49–82 ( loop IV , Zn binding loop ) and 121–142 ( loop VII , Cu binding loop ) , as reported from NMR order parameter measurements ( Sekhar et al . , 2015 ) . To simulate order in either one or both the loops , the conformational entropy of residues in the loop is modified to −13 . 6 J mol−1 K−1 per residue ( i . e . a lower penalty for folding ) , thus mimicking the variants of SOD1 ( Zn bound , Cu bound , and Holo forms ) . The van der Waals interaction energies are fixed to −35 . 9 , –38 . 2 , −42 . 2 , and −48 . 9 J mol−1 for the Holo , Zn bound , Cu bound , and apo variants , respectively , to simulate iso-stability conditions at 298 K . The heat capacity change per native contact is fixed to −0 . 36 J mol−1 K−1 per native contact . All prolines are assigned an entropic penalty of zero to account for their rigidity . Residue probabilities and folding mechanism as a function of the number of structured blocks are predicted at iso-stability conditions ( i . e . a stability 25 kJ mol−1 at 298 K ) following established protocols by accumulating partial partition functions . To gain an insight as to how the WT and metal starved variants of SOD1 interact with membrane , we resorted to computational approaches . Protein orientations in membranes were theoretically calculated by minimizing a protein’s transfer energy from water to a planar slab that serves as a crude approximation of the membrane hydrocarbon core . For WT SOD1 , we referred to the solved structure 4BCY and for the metal starved forms in vacuo ab-initio models were prepared from Zhang Lab server ( Yang and Zhang , 2015 ) . The membrane binding propensity was calculated submitting the co-ordinate information of the protein forms to OPM server . A protein was considered as a rigid body that freely floats in the planar hydrocarbon core of a lipid bilayer . Accessible surface area is calculated using the subroutine SOLVA from NACCESS with radii of Chothia and without hydrogen ( Lomize et al . , 2006; Lomize et al . , 2007 ) . In OPM , solvation parameters are derived specifically for lipid bilayers and normalized by the effective concentration of water , which changes gradually along the bilayer normal in a relatively narrow region between the lipid head group regions and the hydrocarbon core . Unless otherwise noted , all p-values were calculated by performing a paired or unpaired t-test .	Amyotrophic lateral sclerosis , or ALS , is an incurable neurodegenerative disease in which a person slowly loses specialized nerve cells that control voluntary movement . It is not fully understood what causes this fatal disease . However , it is suspected that clumps , or aggregates , of a protein called SOD1 in nerve cells may play a crucial role . More than 140 mutations in the gene for SOD1 have been linked to ALS , with varying degrees of severity . But it is still unclear how these mutations cause SOD1 aggregation or how different mutations influence the survival rate of the disease . The protein SOD1 contains a copper ion and a zinc ion , and it is possible that mutations that affect how these two ions bind to SOD1 influences the severity of the disease . To investigate this , Sannigrahi , Chowdhury , Das et al . genetically engineered mutants of the SOD1 protein which each contain only one metal ion . Experiments on these mutated proteins showed that the copper ion is responsible for the protein’s role in neutralizing harmful reactive molecules , while the zinc ion stabilizes the protein against aggregation . Sannigrahi et al . found that when the zinc ion was removed , the SOD1 protein attached to a structure inside the cell called the mitochondria and formed toxic aggregates . Sannigrahi et al . then used these observations to build a computational model that incorporated different mutations that have been previously associated with ALS . The model suggests that mutations close to the site where zinc binds to the SOD1 protein increase disease severity and shorten survival time after diagnosis . This model was then experimentally validated using two disease variants of ALS that have mutations close to the sites where zinc or copper binds . These findings still need to be tested in animals and humans to see if these mechanisms hold true in a multicellular organism . This discovery could help design new ALS treatments that target the zinc binding site on SOD1 or disrupt the protein’s interactions with the mitochondria .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics" ]	2021	The metal cofactor zinc and interacting membranes modulate SOD1 conformation-aggregation landscape in an in vitro ALS model
miR-128 , a brain-enriched microRNA , has been implicated in the control of neurogenesis and synaptogenesis but its potential roles in intervening processes have not been addressed . We show that post-transcriptional mechanisms restrict miR-128 accumulation to post-mitotic neurons during mouse corticogenesis and in adult stem cell niches . Whereas premature miR-128 expression in progenitors for upper layer neurons leads to impaired neuronal migration and inappropriate branching , sponge-mediated inhibition results in overmigration . Within the upper layers , premature miR-128 expression reduces the complexity of dendritic arborization , associated with altered electrophysiological properties . We show that Phf6 , a gene mutated in the cognitive disorder Börjeson-Forssman-Lehmann syndrome , is an important regulatory target for miR-128 . Restoring PHF6 expression counteracts the deleterious effect of miR-128 on neuronal migration , outgrowth and intrinsic physiological properties . Our results place miR-128 upstream of PHF6 in a pathway vital for cortical lamination as well as for the development of neuronal morphology and intrinsic excitability . Coordinating functions for microRNAs ( miRNAs ) are rapidly being discovered for each of the steps required for the anatomic and functional construction of the mammalian neocortex , from stem cell proliferation and neurogenesis to neuronal outgrowth and synaptogenesis . miRNAs are short , approximately 22 nucleotide RNA molecules that primarily act as antisense regulators of gene expression . The generation of the active form of miRNAs from their initial nuclear transcripts occurs for the majority of miRNAs via two RNase-mediated processing events ( reviewed in Krol et al . , 2010; Siomi and Siomi , 2010 ) . While still in the nucleus , the primary miRNA transcript ( pri-miRNA ) is cleaved by the concerted action of the DROSHA ribonuclease and the RNA binding protein DGCR8 . DROSHA cleavage releases precursor miRNAs ( pre-miRNAs ) with a size range between approximately 60 and 80 nucleotides that are characterized by a stem-loop secondary structure . After nuclear export , the pre-miRNA is cleaved again to generate the active , ∼22 nucleotide mature miRNA by a second protein complex containing the DICER ribonuclease . Developmental regulation of miRNA expression is known to occur at each step in this biogenesis pathway ( Krol et al . , 2010; Siomi and Siomi , 2010 ) . The global reduction in miRNA levels upon conditional deletion of Dicer or Dgcr8 in neuronal progenitors is associated with early defects in proliferation and migration followed by effects on neuronal morphology including dendritic arborization , spine length , and axonal outgrowth ( reviewed in McNeill and Van Vactor , 2012; Sun et al . , 2013 ) . How individual miRNAs contribute to these phenotypes is rapidly being assessed ( reviewed in Sun et al . , 2013; Rehfeld et al . , 2015; Siegel et al . , 2011; Cochella and Hobert , 2012 ) . Two of the best-studied miRNAs with developmental roles are miR-9 and miR-124 . miR-9 acts alone or together with let-7 and miR-125 to control the timing of cell fate decisions ( Shibata et al . , 2011; Coolen et al . , 2012; La Torre et al . , 2013 ) . Studies on miR-124 exemplify how a single miRNA can influence neuronal specification and function at multiple levels by regulating splicing ( Makeyev et al . , 2007 ) , transcription complexes ( Visvanathan et al . , 2007; Cheng et al . , 2009 ) , and epigenetic modifiers ( Yoo et al . , 2009 ) . Like miR-124 , the brain-enriched miR-128 is highly abundant and upregulated during embryonic mouse brain development . In another parallel to miR-124 , miR-128 was first proposed to act as a developmental regulator of mRNA utilization . By inhibiting the expression of two proteins active in nonsense-mediated mRNA decay ( NMD ) , miR-128 was shown to promote neurogenesis in a cell culture model ( Bruno et al . , 2011 ) . Additional functions for miR-128 were then reported in behavior and memory . In a study on the acquisition and suppression of fear-evoked memory , increased expression of miR-128 correlated with , and was required for , the extinction of a learned fear response ( Lin et al . , 2011 ) . It is presently not known if regulation of NMD mediates the effects on learning , as additional regulatory targets for miR-128 were identified in this context ( Lin et al . , 2011 ) . The mouse genome contains two miR-128 genes , termed miR-128-1 and miR-128-2 , which are positioned within introns of two homologous genes ( respectively , R3hdm1 and Arpp21 , also referred to as R3hdm3 , Rcs or Tarpp ) . The sequence and secondary structures of the precursor miRNAs produced from the two copies of miR-128 differ , but they produce identical ∼21 nt miRNAs after Dicer processing . This arrangement is evolutionarily conserved among vertebrates . Recently , the phenotypes of deletion mutants for the mouse miR-128 genes were reported ( Tan et al . , 2013 ) . The two gene copies were shown to be unequal , with miR-128-2 responsible for approximately 80% of the miR-128 level in the adult forebrain . Deletion of miR-128-2 resulted in hyperactive motor behavior and severe epileptic seizures . Selective ablation of miR-128-2 in post-mitotic neurons in the forebrain was sufficient to cause hyperactivity and seizures that could be rescued by ectopic expression of miR-128 ( Tan et al . , 2013 ) . The phenotype of miR-128 deletion with respect to cortical development has not been determined . To better understand the role of miR-128 in brain development , we have examined the spatial and temporal coordinates of miR-128 expression during mouse corticogenesis and in adult stem cell niches . We present evidence that post-transcriptional regulation restricts the accumulation of miR-128 to post-migratory neurons in the embryonic cortical plate and adult stem cell zones . Premature expression of miR-128 led to deficits in the radial migration and dendritic outgrowth of upper layer cortical neurons that were associated with an increase in intrinsic excitability . In contrast , inhibition of miR-128 during migration led to a shift in final neuronal positioning toward the upper boundary of the cortical plate . We identify the X-linked syndromic intellectual disability gene Phf6 as a significant regulatory target for miR-128 . Co-expression of PHF6 suppressed both the morphological and the physiological aspects of the miR-128 gain-of-function phenotype . As a foundation for the functional analysis of miR-128 , we began by characterizing expression of the two miR-128 genes , miR-128-1 and miR-128-2 in the mouse brain . In agreement with our previous work ( Smirnova et al . , 2005 ) , Northern blots of RNA taken from the mouse cortex at several developmental stages show that the mature , 21 nt miR-128 RNA is upregulated between embryonic day 12 . 5 ( E12 . 5 ) and E18 . 5 and remains high postnatally and in adulthood ( Figure 1A ) . In this experiment , we used a high-sensitivity LNA probe complementary to the mature miRNA that should also allow detection of both miR-128 precursor RNAs ( Figure 1D ) . We detected a single precursor signal present at a low level that , in contrast to the mature form , remained constant at all time points tested ( Figure 1A ) . We next employed precursor-specific probes directed against the divergent sequences of their respective loops ( Figure 1—figure supplement 1A ) . The specificity and efficacy of the two probes was confirmed using RNA from cells transfected with expression constructs for the two isoforms ( Figure 1—figure supplement 1B ) . Using the pre-miR-128-2 specific probe ( see Figure 1D ) , we detected a strong band of the expected size that was present at nearly constant levels throughout embryonic and postnatal development ( Figure 1B ) . Expression of the miR-128-1 precursor was below the limit of detection ( Figure 1—figure supplement 2A ) , indicating that miR-128-2 is more highly expressed than miR-128-1 in the embryonic cortex , consistent with a recent report ( Tan et al . , 2013 ) . Taken together , these results suggest that the dynamic expression of miR-128 in cortical development is achieved at least in part by post-transcriptional regulation of pre-miR-128-2 processing . 10 . 7554/eLife . 04263 . 003Figure 1 . pre-miR-128-2 expression precedes miR-128 . Northern blots of RNA from embryonic and adult mouse brains . RNA from the stages indicated above each lane was hybridized with probes specific for miR-128 ( A ) ; pre-miR-128-2 ( B ) ; and U6 ( C ) as loading control . The position of precursor RNAs is indicated with a filled arrow , the ∼21 nt miRNA with an open arrow . The portion of the filter corresponding to ∼15 to 100 nt is shown . The pre-miR-128-2 sequence is depicted in ( D ) , showing the 21 nt mature sequence that is targeted by the anti-miR-128 LNA probe ( underlined ) and the sequence complementary to the anti-precursor hybridization probe ( red ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04263 . 00310 . 7554/eLife . 04263 . 004Figure 1—figure supplement 1 . Relative activity of pre-miR-128-1-RED and pre-miR-128-2-RED expression constructs . ( A ) The sequences of pre-miR-128-1 ( top ) and pre-miR-128-2 ( bottom ) are depicted , showing the mature 21 nt miRNA sequence as recognized by the anti-miR-128 LNA probe ( underlined ) and the sequence complementary to the anti-precursor hybridization probe ( red ) . ( B ) Northern blots of RNA from HEK-293 cells transfected with Intron-RED empty vector ( Lane 1 ) , pre-miR-128-1-RED ( Lane 2 ) or pre-miR-128-2-RED ( Lane 3 ) . The filter was hybridized with probes specific for mature miR-128 , pre-miR-128-1 , pre-miR-128-2 or U6 RNA as loading control , as indicated above each panel . The position of precursor RNAs is indicated with a filled arrow , the 21 nt miR-128 with an open arrow . The portion of the filter corresponding to approximately 15–100 nt is shown . The pre-miR-128-1-RED expression vector produces less miR-128 than the pre-miR-128-2-RED vector , each probe shows the expected specificity . ( C ) HEK-293 cells were co-transfected with a GFP-based sensor vector containing four perfectly complementary binding sites for miR-128 and either Intron-RED control vector , pre-miR-128-1-RED or pre-miR-128-2-RED expression vectors , as indicated . Both pre-miR-128-RED expression constructs repress the miR-128 sensor but pre-miR-128-1-RED shows less activity than pre-miR-128-2-RED . DOI: http://dx . doi . org/10 . 7554/eLife . 04263 . 00410 . 7554/eLife . 04263 . 005Figure 1—figure supplement 2 . Levels of pre-miR-128-1 are below detection level in Northern blot and in situ hybridization assays . ( A ) Northern blot as in Figure 1A–C , the membrane in this case was hybridized with the pre-miR-128-1 probe ( described in Figure 1—figure supplement 1 ) . Developmental stage of the RNA is indicated above each lane . ( B–B″′ ) In situ hybridization using the pre-miR-128-1 probe ( see Figure 1—figure supplement 1 ) of embryonic day 12 . 5 ( B ) , 16 . 5 ( B′ ) , 18 . 5 ( B″ ) and Adult ( B″′ ) brains . The obtained signal does not exceed background at any time point examined . DOI: http://dx . doi . org/10 . 7554/eLife . 04263 . 005 To gain insight into the temporal and spatial dynamics of miR-128 expression , we performed in situ hybridization studies with probes specific for miR-128 , pre-miR-128-1 , and pre-miR-128-2 at different developmental stages . Comparing the results obtained with miR-128 and pre-miR-128-2 at E12 . 5 , levels of miR-128 barely exceeded the detection limit ( Figure 2A , left ) despite strong precursor staining throughout the dorsal and ventral telencephalon ( Figure 2A , middle ) . The pre-miR-128-1 signal , in contrast , was near or below the detection limit ( see Figure 1—figure supplement 2B ) . These results are consistent with the evidence from Northern blot analysis suggesting that pre-miR-128-2 is the major expressed isoform in the neocortex and that expression of this precursor isoform precedes the accumulation of mature miR-128 . 10 . 7554/eLife . 04263 . 006Figure 2 . Post-transcriptional regulation determines the developmental expression pattern of miR-128 . ( A ) Coronal section at E12 . 5 displaying embryonic telencephalon ( scale bar 500 µm ) . Precursor staining is apparent throughout the dorsal and ventral telencephalon ( middle ) in the absence of miR-128 signal ( left ) . Nissl staining is presented for comparison ( right ) . ( B and C ) Coronal sections at E14 . 5 ( B ) and E16 . 5 ( C ) displaying the developing cortex stained for miR-128 , pre-miR-128-2 , or miR-124 , as indicated . DRAQ5 staining of each section is provided for orientation . miR-128 expression is restricted to the CP at E14 . 5 and E16 . 5 ( left panels ) whereas pre-miR-128-2 is expressed ubiquitously from the MG to the VZ ( middle panels ) . At E16 . 5 miR-128 expression within the CP shows a shallow gradient: stronger in the deep ( D ) compared to the upper layers ( U ) . miR-124 ( right ) expression is detected in the CP and in some cells in the IZ . Nuclear staining is obtained with DRAQ5 . Scale bar 100 µm . ( D ) Quantification of microRNA expression at E14 . 5 in VZ/SVZ and IZ normalized to CP ( as described in ‘Materials and methods’ ) . miR-128 ( gray bars ) expression is highest in the CP with a reduction in the IZ ( fourfold ) and VZ/SVZ ( twofold ) . pre-miR-128-2 ( dark bars ) expression is higher in the VZ/SVZ ( almost threefold ) and in IZ ( 1 . 5-fold ) relative to the CP . miR-124 ( white bars ) is expressed in the CP and in the IZ ( ≈60% of the CP intensity ) , single positive neurons are detectable . ( E ) Quantification of microRNA expression at E16 . 5 as in ( D ) except the CP has been divided into upper ( U ) and deeper ( D ) regions using DRAQ5 . miR-128 ( gray bars ) is expressed mainly within the CP with ninefold lower expression in the VZ/SVZ and IZ and is enriched in deeper compared to upper layer neurons in the CP . pre-miR-128-2 ( dark bars ) expression is higher in the VZ/SVZ and IZ ( 1 . 5-fold ) compared to the CP and it is evenly distributed between upper and deeper layers . Relative distribution of miR-124 is similar to E14 . 5 , with the IZ 10-fold higher than the VZ/SVZ . Representative false color images used for quantification are shown in Figure 2—figure supplement 3 . Three brains per condition were analyzed . One-way ANOVA comparing miR-128 and either pre-miR-128-2 or miR-124 was performed with Bonferroni post-test . p < 0 . 05 , *p < 0 . 001 . MG: marginal zone , CP: cortical plate , IZ: intermediate zone , SVZ: subventricular zone , VZ: ventricular zone , U: upper cortical plate , D: deeper cortical plate . DOI: http://dx . doi . org/10 . 7554/eLife . 04263 . 00610 . 7554/eLife . 04263 . 007Figure 2—figure supplement 1 . Differential expression of miR-128 and pre-miR-128-2 in developing and adult cortex . ( A , B ) Cortical sections from E18 . 5 ( A ) and adult ( B ) brains stained for miR-128 or pre-miR-128-2 as indicated . Nissl staining is provided for comparison . An overview is provided for each section ( left side of panel ) , followed by a representative view of the cortex and higher magnification view of regions of interest ( scale bars 500 µm , 50 µm and 10 µm , respectively ) . Individual cells in the IZ ( A ) and Layer V ( B ) differentially stain for pre-miR-128-2 compared to miR-128 . MG: marginal zone , CP: cortical plate , SP: subplate , IZ: intermediate zone , SVZ: subventricular zone , VZ: ventricular zone . DOI: http://dx . doi . org/10 . 7554/eLife . 04263 . 00710 . 7554/eLife . 04263 . 008Figure 2—figure supplement 2 . Post-transcriptional regulation of miR-128 during embryonic and adult neural migration . ( A , B and C ) Fluorescent LNA probe in situ hybridization of E16 . 5 cortical sections shown in red for miR-128 ( A , left ) pre-miR-128-2 ( B , left ) and miR-124 ( C , left ) and co-stained for the basal progenitor marker Tbr2 in green ( A , B and C middle ) . Merged view with nuclei stained with DRAQ5 in blue is shown for comparison ( A , B and C , right ) . Scale bar represents 100 μm . ( A′ B′ and C′ ) Boxed regions in A , B and C are shown at higher magnification; scale bar represents 10 μm . Staining as for ( A , B and C ) as indicated , red and green channel merge is shown at lower left . Tbr-2+ intermediate progenitors in the SVZ co-stain for pre-miR-128-2 but not miR-128 or miR-124 . miR-128 does not specifically stain the IZ , miR-124+ cells in the IZ are Tbr2− . MG: marginal zone , CP: cortical plate , SP: subplate , IZ: intermediate zone , SVZ: subventricular zone , VZ: ventricular zone . ( D and E ) Merged view of adult brain sagittal sections hybridized as above for pre-miR-128-2 ( D , red ) and miR-128 ( E , red ) and co-stained for the migrating neuroblast marker Doublecortin ( Dcx , green ) . Scale bar 100 μm . ( D′ and E′ ) : Boxed areas from the RMS in D and E are shown in serial magnification . Scale bars represent 50 μm ( first row ) and 10 μm ( second row ) . Individual channels are shown as indicated , with a merged view on the right . Dcx+ neuroblasts stain for pre-miR-128-2 but not miR-128 . DOI: http://dx . doi . org/10 . 7554/eLife . 04263 . 00810 . 7554/eLife . 04263 . 009Figure 2—figure supplement 3 . Differential staining of miR-128 and pre-miR-128-2 in corticogenesis . ( A and B ) . Depicted are the false color renderings of the images in Figure 2B , C used to measure the fluorescence intensity in the VZ/SVZ , IZ and CP ( Figure 2D , E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04263 . 009 The pronounced disparity in the expression domains of miR-128 compared to pre-miR-128-2 was also apparent at later time points . In Figure 2B , we show representative images of in situ hybridizations performed at E14 . 5 with the two miR-128 probes in comparison to the neurogenic miR-124 . To allow a more quantitative comparison , the average signal intensity for each probe within the combined ventricular and subventricular zones ( VZ/SVZ ) , intermediate zone ( IZ ) , and cortical plate ( CP ) was determined and expressed relative to the staining intensity of the cortical plate ( Figure 2D ) . At E14 . 5 mature miR-128 was detected at low levels and preferentially accumulated in the cortical plate compared to the underlying subcortical zones . Staining intensity was approximately twofold ( VZ/SVZ ) to fourfold ( IZ ) lower than the CP ( Figure 2B , left panels and Figure 2D , gray bars ) . The miR-128-2 precursor probe , in contrast , displayed an inverse pattern with almost threefold higher relative staining in the neurogenic VZ and SVZ compared to the CP ( Figure 2B , center panels and Figure 2D , dark bars ) . Consistent with previous reports ( Cheng et al . , 2009 ) , miR-124 was readily detected in the cortical plate but not the VZ or SVZ ( Figure 2B , right panels ) . Within the IZ , an intermediate level of staining was seen ( approximately 60% relative to the CP; Figure 2D , white bars ) . These differential expression patterns were more striking at E16 . 5 ( Figure 2C ) . The staining for mature miR-128 remained highly specific for post-mitotic neurons in the CP ( Figure 2C , left panels ) compared to the widespread presence of pre-miR-128-2 from the VZ to the CP ( Figure 2C , middle panels ) . Like miR-128 , miR-124 displayed uniform , high-level expression in the CP ( Figure 2C , right panels ) . Unlike miR-128 , however , overall levels in the IZ were intermediate compared to the lack of staining in the VZ/SVZ . Individual highly stained miR-124+ cells scattered within the SVZ and IZ may represent migrating neurons , as discussed below . For the quantification at E16 . 5 , the average staining intensities of the upper and lower cortical plate for the three probes were also compared ( Figure 2E ) , to highlight the higher degree of deeper layer compared to upper layer staining we consistently observe using the probe for miR-128 ( Figure 2E , gray bars ) . Comparing miR-128 with pre-miR-128-2 , the difference in relative staining intensities in the VZ/SVZ and in the IZ was highly significant ( Figure 2E , dark bars ) . Similarly , a significant difference in the relative staining of miR-124 compared with miR-128 was observed in the IZ ( Figure 2E , white bars ) . A similar difference in pattern between miR-128 and pre-miR-128-2 was also apparent at E18 . 5: despite uniform expression of pre-miR-128-2 throughout the cortex from the ventricles to the marginal zone , accumulation of miR-128 was restricted to the cortical plate ( Figure 2—figure supplement 1A ) . In the adult , the majority of cortical projection neurons co-express the precursor and mature forms , although pre-miR-128-2+/miR-128− cells can be found scattered in the marginal zone and the subcortical white matter ( Figure 2—figure supplement 1B ) . To better characterize the subcortical cells that express pre-miR-128-2 in the absence of miR-128 during development , we repeated the hybridizations at E16 . 5 using fluorescent detection to allow antibody co-staining . Although many classical marker antibodies are not compatible with the hybridization conditions required for LNA probes ( Silahtaroglu et al . , 2007 ) , we were able to perform co-staining for the intermediate progenitor marker Tbr2 ( Englund et al . , 2005 ) . Within the SVZ , we found that Tbr2+ progenitors stained for pre-miR-128-2 but not miR-128 ( Figure 2—figure supplement 2A , B ) . As expected , Tbr2+ progenitors in the SVZ did not express miR-124 . Unlike miR-128 , however , Tbr2−/miR-124+ cells could readily be detected in the IZ and SVZ , suggesting that miR-124 may be present in migrating cells ( Figure 2—figure supplement 2C ) . The absence of miR-128+ cells in the embryonic subventricular and intermediate zones compared to the post-migratory neurons in the cortical plate suggests that miR-128 is not present in migrating neurons . We were interested in confirming this result in an additional developmental setting and therefore examined whether miR-128 is expressed in the migrating neuroblasts of the adult rostral migratory stream ( RMS ) . To visualize migrating neuroblasts , we performed co-staining with Doublecortin ( Dcx ) . The probe specific for pre-miR-128-2 strongly stained the Dcx+-neuroblasts in the RMS ( Figure 2—figure supplement 2D ) . In contrast , miR-128 was clearly present in the cells surrounding the RMS , but was not detectable in Dcx+-neuroblasts ( Figure 2—figure supplement 2E ) . Similar results were obtained in the neurogenic niche of the adult dentate gyrus . We found that the miR-128-2 precursor was already present in newborn ( Dcx+/NeuN+ ) and mature ( Dcx−/NeuN+ ) granule cells of the dentate gyrus . In contrast , miR-128 was absent in immature neurons ( Dcx+/NeuN+ ) and only present in mature granule cells ( Dcx−/NeuN+ ) ( data not shown ) . In summary , we found that the miR-128-1 isoform is unlikely to contribute significantly to developmental expression of miR-128 , based on the lack of signal in Northern blots or in situ hybridization ( Figure 1—figure supplement 2 ) . Comparing the regulation of pre-miR-128-2 and miR-128 in embryonic corticogenesis suggests that accumulation of miR-128 occurs after the completion of neurogenesis and at the end of radial migration as cortical neurons reach their final position in the cortex and begin their functional and morphological maturation ( Figure 2 and Figure 2—figure supplement 2A–C ) . Similar evidence for post-transcriptional exclusion of miR-128 from migrating neurons was obtained in the adult RMS ( Figure 2—figure supplement 2D , E ) . These results prompted us to test the effects of premature miR-128 expression in embryonic progenitors as they differentiate and migrate to the cortical plate . To gain insight into the biological role of miR-128 , we performed in vivo gain-of-function experiments using in utero electroporation at E15 . 5 to deliver ectopic miR-128 from a plasmid-based expression construct . This allowed us to introduce miR-128 into proliferating and migrating cells that normally do not express the mature miRNA . We used the plasmid vector Intron-RED , which allows precursor miRNA sequences to be expressed from a synthetic intron engineered in dsRed , generating the expression constructs pre-miR-128-1-RED and pre-miR-128-2-RED for the two miR-128 precursors ( see ‘Materials and methods’ for details ) . Comparing the activity of the two constructs in Northern blot and sensor assays revealed that the pre-miR-128-1-RED construct displayed reduced activity compared to pre-miR-128-2-RED ( refer to Figure 1—figure supplement 1B , C ) . The defect in pre-miR-128-1 processing therefore appears to be intrinsic to the precursor and/or flanking sequences and allows the use of the pre-miR-128-1-RED construct as a negative control . To verify that forced miR-128 expression from pre-miR-128-2-RED can overcome the inhibitory mechanism that acts on endogenous pre-miR-128-2 , we stained for mature miR-128 in electroporated brains at E18 . 5 . We could confirm that cells expressing dsRed from the pre-miR-128-2-RED expression vector , but not the control Intron-RED vector , were the sole miR-128+ cells in the IZ ( Figure 3—figure supplement 1 ) . We tested the effect of premature miR-128 expression at P7 , when migration into the cortex is completed . We found that the distribution of control ( Intron-RED ) and pre-miR-128-1 expressing neurons was indistinguishable , with the majority of cells positioned within layers II and III ( Figure 3A ) . In comparison , the majority of pre-miR-128-2 expressing neurons migrated successfully into the cortical plate but their final position was shifted toward the deep layers ( Figure 3A ) . Quantification of the effect on migration confirmed the shift in neuronal position to deeper layers in response to premature expression of miR-128-2 ( Figure 3B ) . These results are the first evidence that miR-128 regulates the process of radial neuronal migration during the establishment of cortical lamination . 10 . 7554/eLife . 04263 . 010Figure 3 . miR-128 misexpression impairs neuronal migration . ( A ) Representative brain sections of P7 mice showing intron-RED control ( left ) , pre-miR-128-1-RED ( middle ) , pre-mir-128-2-RED ( right ) after in utero electroporation at E15 . 5 . Sections were processed for staining with DRAQ5 to reveal nuclei and anti-RFP antibody to reveal electroporated cells . On the right side of each picture the position of the bins used to assess migration is shown ( see ‘Materials and methods’ ) . Scale bars represent 50 µm . ( B ) Percent of total counted neurons present in each bin is plotted . Data are from 3 to 4 mice per condition . Two-way ANOVA with Bonferroni post-test , error bars represent Standard deviation . p < 0 . 05 p < 0 . 01 , p < 0 . 001 . Electroporation of pre-miR-128-2 ( white bars ) but not pre-miR-128-1 ( gray bars ) caused a shift from uppermost layers ( Bin 1 ) to lower layers ( Bin 3 ) compared to control ( black bars ) . ( C ) Quantification of P0 electroporated neurons expressing the upper layer marker Cux1 or the layer V marker Ctip2 . Electroporation of pre-miR-128-2-RED ( gray bars ) does not change the cell fate compared to control ( black bars ) . ( D–E′ ) Representative brain sections of P0 mice , analyzed in ( C ) , stained for dsRed to show pre-miR-128-2-RED electroporated cells ( red , D and D′ ) and Intron-RED ( red , E and E′ ) . In ( D ) and ( E ) sections were co-stained for the layer II-IV marker Cux1 in blue . In ( D′ ) and ( E′ ) sections were co-stained for the layer V marker Ctip2 in blue . Neighboring images show higher magnification views of boxed regions of interest . In ( D ) and ( E ) from top to bottom: pre-miR-128-2 ( red , D ) or control ( red , E ) , Cux1 ( blue ) and merged view . In ( D′ ) and ( E′ ) from top to bottom: pre-miR-128-2 ( red , D′ ) or control ( E′ ) , Ctip2 ( blue ) and merged view . Scale bars 20 μm or 5 μm . Arrowheads in ( D and E ) mark dsRED+/Cux1+ migrating cells . Empty arrowhead in ( D′ and E′ ) marks a dsRED+/Ctip2- cell situated in layer V . ( F ) Representative brain sections of P7 mice showing the control eGFP construct ( left ) and the miR-128 sponge ( right ) after in utero electroporation at E15 . 5 . Sections were processed for staining with DRAQ5 to reveal nuclei and anti-GFP antibody to reveal electroporated cells . On the right side of each picture the position of the bins used to assess migration is shown ( see ‘Materials and methods’ ) . Scale bar represents 50 µm . ( G ) Percent of total counted neurons present in each bin is plotted . Data are from 3 to 5 mice per condition . Two-way ANOVA with Bonferroni post-test , error bars represent Standard deviation *p < 0 . 001 . Electroporation of the miR-128 sponge caused a shift from Bins 2–3 to Bin 1 ( light green bars ) compared to control ( dark green bars ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04263 . 01010 . 7554/eLife . 04263 . 011Figure 3—figure supplement 1 . Ectopic miR-128-2 is processed to miR-128 after in utero electroporation . ( A ) miR-128 in situ hybridization using colorimetric NBT/BCIP detection ( left , false colored in green ) on E18 . 5 brains after electroporation at E15 . 5 with pre-miR-128-2-RED after antibody staining for dsRed ( middle , red ) . Nuclei were stained with DRAQ5 ( blue ) . A merged view of miR-128 expression ( green ) and electroporated neurons ( red ) is on the right side of the panel . ( B ) Magnification of the boxed region in ( A ) . Electroporated neurons ( red ) are the only cells expressing mature miR-128 ( green ) in the IZ . Arrows in ( A ) and ( B ) denote exemplary miR-128+/dsRed+ neurons . ( C ) miR-128 hybridization as in A ( left , false colored in green ) on control E18 . 5 brains after electroporation at E15 . 5 with Intron-RED and antibody staining for dsRed ( red , middle ) . Nuclei were stained with DRAQ5 ( blue ) . A merged view of miR-128 expression ( green ) and electroporated neurons ( red ) is on the right side of the panel . ( D ) Magnification of the boxed region in ( C ) . Control electroporated neurons ( red ) in the IZ do not express mature miR-128 ( green ) . Scale bar 50 µm . U: upper cortical plate , D: deeper cortical plate , SP: subplate . DOI: http://dx . doi . org/10 . 7554/eLife . 04263 . 011 To gain insight into the mechanism of the migration defect , we first tested if premature miR-128 expression affects migration indirectly by interfering with the specification of upper layer neuron identity . The layer II-III neurons targeted by electroporation at E15 . 5 characteristically express the transcription factors Cux1 and Cux2 , while earlier born layer V neurons express Ctip2 ( Nieto et al . , 2004; Arlotta et al . , 2005 ) . Co-staining of electroporated brains at P0 with these layer-specific markers revealed that the majority of Cux1+ cells had reached their destination in the upper layers , but some Cux1+ cells were still present in the deep layers and in the white matter . Regardless of their position in the cortical plate , cells electroporated with pre-miR-128-2-RED co-stained for Cux1 at approximately the same frequency as control cells ( >80% , Figure 3C , D , E ) . Furthermore , dsRed+ cells expressing pre-miR-128-2 in layer V did not express Ctip2 at higher levels than control cells ( <2% , Figure 3C , D′ , E′ ) suggesting that their improper localization was not an indirect consequence of temporal misspecification . Together , these results indicate that the fate of the cells electroporated with premiR-128-2 was not affected despite the defect in neuronal migration . To determine the effect of blocking miR-128 expression on neuronal migration , we repeated the electroporations using a so-called sponge inhibitor . Our sponge inhibitor expresses an eGFP cassette containing an array of 16 high-affinity synthetic miR-128 binding sites within the 3′ UTR under the control of the CAGGS promoter . Upon electroporation of the anti-miR-128 sponge construct at E15 . 5 and analysis at P7 , we observed a significant shift in neuronal position toward the top of the cortical plate in sponge compared to control neurons ( Figure 3F , G ) . The inverse migration phenotypes observed in upper layer neurons upon either increasing ( Figure 3A , B ) or decreasing miR-128 ( Figure 3F , G ) activity suggests that a pathway critical for correct cortical lamination is highly sensitive to the level of miR-128 . Based on their marker expression , manipulation of the onset of miR-128 expression did not affect the temporal identity of the resulting neurons . Careful examination of the electroporated regions , however , revealed differences in the proper bipolar morphology of pre-miR-128-2+ neurons as they migrated radially through the cortical plate ( Figure 4A , B ) . Because migrating neurons change morphology quickly , we analyzed control and pre-miR-128-2 electroporations performed in the same litter to avoid differences due to small variations in mating , electroporation , or sacrifice time . In controls , the majority of the electroporated neurons were already at their correct position in layer II/III , and those still migrating presented long , radially oriented leading processes ( Figure 4A ) . Neurons expressing pre-miR-128-2 were more scattered throughout the cortical plate ( Figure 4B ) , with the leading processes of actively migrating cells frequently branched . To quantify this result , we reconstructed randomly selected neurons located in the deep layers and therefore still in the process of active migration . Control neurons generally had a single , unbranched leading process with occasional short filopodia ( Figure 4C , upper row ) . Neurons expressing pre-miR-128-2 , on the other hand , were consistently more branched and also had more filopodia ( Figure 4C , lower row ) . The morphology of the neurons was quantified using the number of branches and the number of filopodia per neuron as criteria ( see Materials and methods for details ) . Consistent with their overall morphology , ectopic expression of miR-128 in migrating neurons led to an approximately 2 . 5-fold increase in the number of filopodia and a commensurate increase in branch number ( Figure 4D ) . This suggests that the effects of miR-128 on migration are related to a failure in the regulation of cytoskeletal dynamics believed to be responsible for radial movement ( Heng et al . , 2010; Cooper , 2013 ) . Staining of the electroporated area with the radial glia marker Nestin did not revel any obvious changes in the glial scaffold directing migration of these neurons , consistent with a cell autonomous effect ( data not shown ) . 10 . 7554/eLife . 04263 . 012Figure 4 . Neurons misexpressing miR-128 show impaired radial morphology . ( A and B ) P0 sections from littermates electroporated at E15 . 5 with control Intron-RED ( A ) or pre-miR-128-2-RED ( B ) expression constructs . Sections were stained for dsRed to reveal electroporated cells , rendered in black and white . Red lines indicate the boundaries of the deep layers of the cortical plate , as determined by nuclear staining ( not depicted ) . ( C ) Reconstructed migrating neurons sampled from the deep layers ( red lines in A and B ) . Upper row shows Intron-RED control neurons , bottom row shows pre-miR-128-2-RED electroporated neurons . ( D ) Box plot of total branch ( upper graph ) and filopodia ( lower graph ) number per reconstructed neuron . ( 58 neurons from 3 Intron-RED brains and 67 neurons from 5 pre-miR-128-2-RED brains were analyzed , significance determined with an unpaired Student’s t test p < 0 . 05 , *p < 0 . 01 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04263 . 012 To identify regulatory partners for miR-128 that might be responsible for the altered migration , we used prediction algorithms ( TargetScan , Pictar , Diana-microT ) to screen for potential target genes with known or suspected roles in neuronal migration or outgrowth ( Krek et al . , 2005; Friedman et al . , 2008; Maragkakis et al . , 2009 ) . A reporter assay was used to validate sensitivity to miR-128 for the candidate genes Gria3 , Jip3 , Nrp2 , Pard6b , Phf6 , Reelin , and Srgap2 ( Figure 5—figure supplement 1 ) . Of these candidates , Pard6b and Phf6 were also >0 . 5-fold downregulated in a microarray screen of mRNAs affected by miR-128 overexpression in P19 embryocarcinoma cells ( data not shown ) . We concentrated on the Börjeson-Forssmann-Lehmann Syndrome gene Phf6 based on its expression pattern in the embryonic VZ and SVZ ( Voss et al . , 2007; Zhang et al . , 2013 ) and the high degree of similarity between the reported Phf6 migration phenotype to our results with miR-128 ( Zhang et al . , 2013 ) . To allow a direct comparison to the miR-128 expression pattern , we performed in situ hybridizations at E14 . 5 and E16 . 5 for Phf6 mRNA and antibody staining for PHF6 protein ( Figure 5 ) . Using a Phf6-specific LNA probe , at E14 . 5 Phf6 mRNA was detected throughout the cortex with particularly prominent expression in the intermediate zone ( Figure 5A ) . At E16 . 5 Phf6 mRNA was also detected in the intermediate zone , but at a reduced level relative to the cortical plate , ventricular , and subventricular zones ( Figure 5B ) . Antibody staining was consistent with the mRNA expression patterns at both time points , and confirmed the presence of PHF6 protein in the IZ at E16 . 5 ( Figure 5C , D ) . Taken together , these results indicate that PHF6 is expressed throughout the miR-128 negative regions of the VZ , SVZ , and IZ at E14 . 5 and E16 . 5 ( Figure 2B , C; Figure 2—figure supplement 3 ) , and suggest that developmental regulation of PHF6 by miR-128 may occur in the domain of co-expression in the cortical plate . 10 . 7554/eLife . 04263 . 013Figure 5 . Regulation of PHF6 by miR-128 . ( A and C ) Phf6 mRNA ( A ) and protein ( C ) expression domains in E14 . 5 brain are comparable , both the mRNA and the protein are present in the VZ , SVZ , and IZ . The nuclear marker DRAQ5 allows the visualization of the brain subregions . Antibody specificity is documented in Figure 5—figure supplement 3A . ( B and D ) Phf6 mRNA ( B ) and protein ( D ) in E16 . 5 brain section are found in the CP , IZ as well as the SVZ and VZ . mRNA and protein expression patterns are comparable . The nuclear marker DRAQ5 allows the visualization of brain subregions . Scale bar 50 µm . CP: cortical plate , SP: subplate , IZ: intermediate zone , SVZ: subventricular zone , VZ ventricular zone . ( E ) Reporter assay on the Phf6 3′UTR , cloned in an eGFP plasmid . pre-miR-128-RED expression constructs and Intron-RED control were co-transfected with the GFP-Phf6-3′UTR sensor plasmid in HEK-293 cells . The GFP Mean Fluorescent Intensity ( MFI ) of miR-128/Phf6-3′UTR expressing cells is normalized to the GFP MFI of control/Phf6-3′UTR expressing cells . One-Way ANOVA with Bonferroni post-test , error bars represent Standard deviation p < 0 . 01 , *p < 0 . 05 . ( F ) Representative Western blot of extracts from HeLa cells transfected with scrambled control , miR-128 , let-7b or miR-125 synthetic miRNA mimics , as indicated . miR-128 has three , miR- let-7b and miR-125 no predicted binding sites in the Phf6 3′UTR . Upper panel shows signal for endogenous PHF6 protein , lower panel GAPDH as loading control . ( G ) Representative Western blot of extracts from HEK-293 cells transfected with scrambled control , miR-128 , let-7b or miR-124 synthetic miRNA mimics , as indicated . miR-128 has three , miR-124 one and let-7b no predicted binding sites in the Phf6 3’UTR . Upper panel shows signal for endogenous PHF6 protein , lower panel Vinculin as loading control as indicated to the right . ( H ) Quantification of PHF6 protein levels relative to Vinculin , as shown in ( F ) . miR-128 expression reduced PHF6 protein levels approximately 50% compared to the let-7b control ( average of 3 independent experiments , p < 0 . 01 One-Way ANOVA , error bars represent Standard deviation ) . ( I ) qRT-PCR for Phf6 mRNA from staged mRNA samples between E12 . 5 and Adult . Phf6 expression was normalized to the reference mRNA Oaz1 . Average of three independent experiments , error bars show Standard deviation . ( J ) Western blot of PHF6 protein levels in primary cortical neurons cultured for the indicated days in vitro ( DIV ) . ( K ) qRT-PCR for Phf6 mRNA performed on primary cortical neurons , DIV as indicated . Phf6 expression was normalized to the reference mRNA GAPDH . ( Average of three independent experiments , error bars represent Standard deviation ) . ( L ) TaqMan qPCR for miR-128 was performed on the same RNA samples as in Panel J . Expression level was normalized to sno135 RNA ( Average of three independent experiments , error bars represent Standard deviation ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04263 . 01310 . 7554/eLife . 04263 . 014Figure 5—figure supplement 1 . Validation of miR-128 targets using a reporter assay . ( A–F ) Reporter assay using 3′UTR's from putative miR-128 targets , cloned in a modified eGFP plasmid ( GFP-3'UTR ) . miR-128 synthetic miRNA mimic ( A , B , C , D ) or pre-miR-128-RED expression constructs ( E , F ) and Their respective controls were co-transfected with the GFP-3′UTR reporter plasmid in N2A cells ( B , C , D ) , or HEK-293 cells ( A , E , F ) . The GFP mean fluorescent intensity ( MFI ) of miR-128/GFP-3′UTR expressing cells is normalized to the GFP MFI of Control/GFP-3′UTR expressing cells . ( G ) Reporter assay to validate the ability of the miR-128 sponge construct to recruit miR-128 . Synthetic miR-128 or a scrambled negative control miRNA ( Ambion ) were co-transfected with the miR-128 sponge in HEK-293 cells . The GFP mean fluorescent intensity ( MFI ) of miR-128/ miR-128 sponge cells is normalized to the GFP MFI of control miRNA/miR-128 sponge- expressing cells . Average of three independent experiments , p < 0 . 01 , p < 0 . 05 , p < 0 . 001 Student’s T-test ( A , B , C , D ) One-way ANOVA ( E , F , G ) , error bars represent Standard deviation ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04263 . 01410 . 7554/eLife . 04263 . 015Figure 5—figure supplement 2 . Multiple , conserved binding sites for miR-128 in the Phf6 3′UTR . Predicted binding sites for miR-128 in the mouse Phf6 3′UTR , shown in black ( adapted from Diana MicroT-CDS ) . The sequence of mature miR-128 is in red . Watson–Crick pairs are shown with vertical bars and wobble pairs with dots . DOI: http://dx . doi . org/10 . 7554/eLife . 04263 . 01510 . 7554/eLife . 04263 . 016Figure 5—figure supplement 3 . Western blot detection of Phf6 . ( A ) Western blot of HEK-293 transfected with GFP empty vector ( Lane 1 ) or PHF6-GFP plasmid ( Lane 2 ) . The blot was probed with the anti-PHF6 antibody from Bethyl used for immunohistochemistry in Figure 5 . ( 1:4000 ) . An arrow indicates the endogenous PHF6 protein in Lane 1 , detected as a single band . Phf6 overexpression is confirmed in Lane 2 . ( B ) The complete image of the western blot shown in Figure 5I using PHF6 antibody . The specific band is marked with an arrow ( left ) . The detection antibody recognizes additional non-specific bands . DOI: http://dx . doi . org/10 . 7554/eLife . 04263 . 016 The Phf6 mRNA contains three potential , conserved binding sites for miR-128 ( Figure 5—figure supplement 2 ) . The sensitivity of the mouse 3′UTR to miR-128 was confirmed in a reporter assay upon co-expression of miR-128 , with the response to pre-miR-128-2-RED greater than pre-miR-128-1-RED , as expected ( Figure 5E ) . To determine if miR-128 can regulate endogenous Phf6 , we used two cell lines , HeLa and HEK-293 , that express Phf6 but not miR-128 . In HeLa cells transfection with synthetic miR-128 led to a strong reduction in endogenous PHF6 protein . Transfection of two non-targeting miRNAs , let-7b or miR-125 , had no effect ( Figure 5F ) . Similar results were obtained in HEK-293 cells . As controls , we transfected with synthetic miRNAs for either let-7b , a non-targeting miRNA , or miR-124 , a microRNA with one conserved binding site in the PHF6 3′UTR . Whereas let-7b had no effect , transfection with synthetic miR-128 consistently reduced PHF6 protein levels by an average of approximately 50% ( Figure 5G , quantified in Figure 5H ) . Unlike miR-128 , the reduction in PHF6 in response to miR-124 was not statistically significant , suggesting that the three predicted binding sites for PHF6 act cooperatively to mediate stronger repression than the single site present for miR-124 . To complement the in situ data , we used qRT-PCR to show that Phf6 mRNA levels show a reciprocal temporal relationship to miR-128 , with levels highest in the embryonic cortex and an approximately 50% reduction between E16 . 5 and P3 ( Figure 5I , compare to the miR-128 profile in Figure 1A ) . A similar inverse relationship was observed during maturation of cultured embryonic cortical neurons . Phf6 mRNA was maximally expressed in the first two days of culture and declined with increasing time in culture ( Figure 5K ) . Levels of miR-128 determined in parallel showed an inverse profile with levels increasing over time in culture ( Figure 5L ) . Western blots confirmed the reduction in PHF6 expression at the protein level ( Figure 5J ) . Zhang et al . have shown that shRNA-mediated knockdown of Phf6 in the developing cortex led to a similar effect on radial neuronal migration and morphology as premature miR-128 expression ( Zhang et al . , 2013 ) . To test if miR-128 might be acting via suppression of Phf6 , we generated an expression plasmid containing the open reading frame of Phf6 linked to eGFP via an IRES sequence . As a negative control , we tested a similar construct containing the ORF of Nrp2 , a known regulator of migration but weak miR-128 target ( see Figure 5—figure supplement 1 ) . Co-expression of NRP2 and pre-miR-128-2 after electroporation at E15 . 5 had no effect on the migration of cortical neurons assayed at P7 compared to expression of pre-miR-128-2 alone ( data not shown ) . In contrast , co-expression of PHF6 and pre-miR-128-2 significantly reduced the number of ectopic neurons in the lower layers and promoted their migration into the upper layers ( Figure 6A–C ) . Quantification of neuronal position at P7 confirmed that significantly more PHF6/miR-128 double-positive neurons reached the upper layers than those expressing miR-128 alone ( Figure 6C ) . These results suggest that precise timing of miR-128 expression is required to fine-tune the pro-migratory function of PHF6 . 10 . 7554/eLife . 04263 . 017Figure 6 . PHF6 rescues the migration defect caused by pre-miR-128-2 . ( A and B ) Brain sections of P7 mice electroporated at E15 . 5 with pre-miR-128-2-RED ( A ) or pre-miR-128-2-RED plus PHF6-GFP expression constructs ( B ) . Sections were stained for dsRed and GFP to reveal electroporated cells . The position of bins used to quantify migration is shown on the right . Scale bar represents 50 µm . Cortical layers are labeled on the left , as determined by nuclear staining ( not depicted ) . ( C ) Number of neurons in each bin was determined and expressed as the per cent contained in upper layers ( Bin 1–4 ) vs deeper layers ( Bin 5–10 ) . ( Five mice analyzed per condition . Significance determined by Two-way ANOVA with Bonferroni post-test p < 0 . 01 , error bars represent the Standard deviation ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04263 . 017 The results presented so far indicate that correct temporal control of miR-128 expression is necessary to avoid interference with PHF6-mediated neuronal migration . We therefore wondered if this balance is also important for the maturation of neurons in the cortical plate . For these experiments , electroporations were performed using the same conditions as in the migration experiments but analyzed at P15 . We performed whole-cell patch-clamp recordings in combination with intracellular biocytin labeling of pyramidal cells located in layer II/III and compared control ( Intron-RED ) , miR-128 gain-of-function ( pre-miR-128-2-RED ) and PHF6 rescue ( pre-miR-128-2-RED plus PHF6-GFP expression vectors ) conditions ( Figure 7A ) . 10 . 7554/eLife . 04263 . 018Figure 7 . miR-128 and PHF6 regulate dendritic complexity and intrinsic excitability . ( A ) Cells from electroporations using Intron-RED ( left ) , pre-miR-128-2-RED ( middle ) , or pre-miR-128-2-RED plus PHF6-GFP ( right ) were recorded and filled . Representative reconstructed neurons ( top ) and their voltage responses to a family of current pulses ( bottom ) are shown . Compared to Intron-RED control , AP discharge is increased by pre-miR-128-2-RED and intermediate upon co-expression of pre-miR-128-2-RED and PHF6-GFP . ( B ) Sholl analysis of filled and reconstructed neurons , from Intron-RED ( open circles , n = 7 cells ) , pre-miR-128-2-RED ( gray , n = 9 cells ) , and pre-miR-128-2-RED plus PHF6-GFP ( blue , n = 9 cells ) electroporated neurons . Error bars represent standard error of the mean . ( C , E–H ) Summary bar charts of intrinsic physiological properties: Membrane potential ( VM C ) , Input resistance ( RI E ) , Rheobase ( F ) , Action Potential ( AP ) frequency ( G ) and voltage sag ( H ) . Colors as in ( B ) , bars are overlain by data from individual cells . ( D ) Current–voltage relationship for the three groups of electroporated neurons , color scheme as in ( B ) . Note the steep curve for pre-miR-128-2-RED neurons , and partially recovered RI relationship for PHF6 rescue neurons . Statistics: ns – p > 0 . 05 , p < 0 . 05 , p < 0 . 01 , p < 0 . 001 , Two-way ANOVA for graph in ( B ) and Mann–Whitney non-parametric test for graphs in C , E–H . DOI: http://dx . doi . org/10 . 7554/eLife . 04263 . 01810 . 7554/eLife . 04263 . 019Figure 7—figure supplement 1 . Reconstructed neurons used to perform Sholl analysis at P15 . ( A–C ) Reconstruction of P15 layer II/III neurons expressing Intron-RED ( A ) , pre-miR-128-2-RED ( B ) or pre-miR-128-2-RED plus PHF6-GFP ( C ) . Reconstruction was done on Z-stack images of biocytin-filled cells , here rendered in 2-D . The same neurons were also analyzed for their electrophysiological properties . DOI: http://dx . doi . org/10 . 7554/eLife . 04263 . 01910 . 7554/eLife . 04263 . 020Figure 7—figure supplement 2 . pre-miR-128-2 but not pre-miR-128-1 affects dendritic arbor complexity . ( A–C ) Layer II/III neurons at P21 were reconstructed after antibody staining for dsRed after electroporation at E15 . 5 with Intron-RED ( A ) , pre-miR-128-1-RED ( B ) , or pre-miR-128-2-RED ( C ) . ( D ) Sholl analysis on 3-D reconstructed neurons . Dendritic arbor complexity of Intron-RED ( black ) , pre-miR-128-1-RED ( gray ) and pre-miR-128-2-RED ( light gray ) electroporated neurons is graphed . pre-miR-128-2-RED led to significantly less ramification between 35 µm and 120 µm from the soma compared to either control ( Intron-RED or pre-miR-128-1-RED ) . Significance was tested with Two-way ANOVA , error bars represent Standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 04263 . 020 To determine the effect of miR-128 on neuronal morphology , individual dsRed+ upper layer neurons were reconstructed after staining for biocytin ( Figure 7A and Figure 7—figure supplement 1 ) and their dendritic complexity compared using Sholl analysis ( Figure 7B ) . We observed a significant reduction in the number of dendritic intersections , a measure of dendritic complexity , in cells electroporated with pre-miR-128-2-RED compared to the IntronRED control . The number of dendritic intersections was reduced approximately 37% for the proximal arbors 40–75 µm from the soma ( Figure 7B ) . Co-electroporation of pre-miR-128-2-RED and PHF6-GFP largely counteracted this effect of miR-128 . Compared to cells electroporated with pre-miR-128-2-RED alone , a statistically significant increase in intersection numbers was observed between 40 and 75 µm from the cell body . There was no significant difference in this parameter at any distance from the soma between control cells and cells co-expressing miR-128-2 and PHF6 ( Figure 7B ) . To confirm these results , we also performed Sholl analysis on layer II/III neurons at P21 , comparing control ( Intron-RED ) and miR-128 gain-of-function ( pre-miR-128-2-RED ) conditions . In these experiments individual neurons were reconstructed after staining for dsRed to amplify the fluorescent signal ( Figure 7—figure supplement 2A–C ) . Cells prematurely expressing miR-128 displayed a statistically significant decrease in proximal dendritic complexity throughout the area approximately 35–120 µm from the cell body compared to control ( Figure 7—figure supplement 2D ) . Neither the length nor the orientation of the apical dendrites was noticeably affected . As an additional control , we also tested the less active pre-miR-128-1-RED expression construct . As expected , Sholl analysis of the resulting neurons yielded an intermediate phenotype that was not statistically different than control ( Figure 7—figure supplement 2B , D ) . This result indicates that the reduction in dendritic complexity associated with premature miR-128 expression persists after P15 and is therefore more likely due to interference with , as opposed to a delay in , dendritic outgrowth . In addition to morphological changes , whole-cell patch clamp recordings revealed differences in the intrinsic physiological properties of layer II/III pyramidal cells in response to miR-128 gain-of-function . After electroporation of pre-miR-128-2-RED , the affected neurons had a significantly more depolarized resting membrane potential ( VM ) than cells electroporated with the Intron-RED control ( VM = −64 . 6 ± 1 . 3 mV vs −73 . 0 ± 1 . 4 mV , Figure 7C ) . Furthermore , neurons prematurely expressing miR-128 showed a steeper current–voltage relationship across a range of hyper- and depolarizing current pulses compared to control cells , an effect that can be primarily accounted for by their higher input resistance ( 203 ± 18 MΩ for pre-miR-128-2-RED vs 161 ± 18 MΩ for Intron-RED , 26% change , Figure 7D , E ) . Because we found no difference in the membrane time constant between pre-miR-128-2 expressing cells and control cells ( data not shown ) , the increased input resistance is most likely a consequence of the observed reduction in dendritic complexity ( Figure 7B ) . However , in combination with the depolarized membrane potential it may also indicate a reduction in basal membrane conductance mediated by potassium leak channels . In either case , in their sum , these changes should lead to an increase in excitability . Indeed , we observed a reduction in the current required to trigger action potential discharge ( rheobase ) in pre-miR-128-2 expressing cells compared to control cells ( 79 . 3 ± 1 . 0 pA vs 124 . 5 ± 12 . 8 pA , Figure 7F ) . Furthermore , pre-miR-128-2 expressing cells fired trains of action potentials ( APs ) at substantially higher frequencies ( 42 ± 4 Hz ) in response to large depolarizing current pulses ( 250 pA , 500 ms ) , a 63% increase compared to control cells ( 26 ± 1 Hz , Figure 7G ) . Interestingly , in response to large hyperpolarizing pulses the miR-128 gain-of-function neurons responded with an approximately twofold larger voltage sag than control neurons ( 5 . 6 ± 0 . 5 mV vs 2 . 6 ± 0 . 2 mV , measured for −250 pA current pulses , Figure 7H ) . This suggests that during hyperpolarization an increase in HCN-mediated Ih currents may partially compensate the higher input resistance seen in the miR-128 gain-of-function neurons . Using cells obtained from co-electroporations of pre-miR-128-2-RED and PHF6-GFP , we found that the effects of premature miR-128 expression on the electrophysiological properties of layer II/III neurons are for the most part mediated by PHF6 . Neurons co-expressing PHF6 and pre-miR-128-2 had a VM of −70 . 5 ± 1 . 4 mV and an input resistance of 180 ± 16 MΩ , both comparable to that of control Intron-RED cells ( Figure 7C , E ) . The current–voltage relationship , rheobase and the maximum AP discharge were also partially rescued by PHF6 co-expression ( 35 ± 2 Hz at 250 pA , 500 ms , Figure 7D , F , G ) . The increase in the hyperpolarization-induced voltage sag was also partially reversed by PHF6 ( 4 . 3 ± 0 . 8 mV ) , although it remained higher than in control neurons ( Figure 7H ) . In summary , miR-128 misexpression during corticogenesis results in substantive changes in both the morphological and physiological properties of upper layer neurons . With the exception of the voltage sag and rheobase , which were partially compensated , the observed reductions in dendritic complexity and changes in intrinsic excitability were restored to control levels by co-transfection with PHF6 . By carefully analyzing the expression pattern of miR-128 during cortical development , we present evidence that miR-128 might be part of a regulatory switch required for the transition from migration to outgrowth , thereby promoting functional neuronal maturation . Based on the disparate temporal control of pre-miR-128-2 and miR-128 , post-transcriptional mechanisms appear to contribute to the timing of miR-128 activity . Post-transcriptional regulation of miRNA biogenesis is believed to facilitate dynamic control over miRNA activity that may be required for cells to rapidly change their gene expression in response to developmental or environmental signals ( Krol et al . , 2010 ) . Another possible advantage of post-transcriptional control is that it would allow the timing of miR-128 expression to be partly uncoupled from the regulation of Arpp21 transcription , the host mRNA for miR-128-2 . One example of this in the nervous system is the ability of miR-26 to suppress its host gene Ctdsp2 and allow differentiation of neural stem cells ( Dill et al . , 2012 ) . There is evidence for a similar feedback relationship between miR-128 and Arpp21 in the adult brain during the suppression of fear-evoked memories ( Lin et al . , 2011 ) . However , mice deficient in Arpp21 are viable and without a known defect in cortical development ( Rakhilin et al . , 2004; Davis et al . , 2012 ) . The disparity we observe between pre-miR-128-2 expression and miR-128 accumulation suggests that a delay in cytoplasmic DICER processing of the precursor contributes to the temporal control of miR-128 . For several miRNAs , DICER cleavage is known to be inhibited by precursor-specific RNA binding proteins such as LIN28 in the case of let-7 and miR-9 or DHX36 for miR-134 ( Rybak et al . , 2008; Bicker et al . , 2013; Nowak et al . , 2014 ) . A different mechanism , sequestration by the circular RNA sponge CDR1 , is thought to control miR-7 ( Hansen et al . , 2013; Memczak et al . , 2013 ) . The mechanism or mechanisms responsible for post-transcriptional control of miR-128 remain to be determined , however , it appears to help restrict miR-128 accumulation to the cortical plate after neurogenesis and radial migration have occurred . These observations prompted us to test the effects of premature miR-128 expression on radial migration . Neuronal migration is a complex process necessary for correct cortical lamination and the formation of functional neuronal networks . Previously , three brain-enriched miRNAs ( miR-9 , miR-132 , and miR-137 ) have been implicated in the regulation of neuronal migration ( reviewed in Evsyukova et al . 2013 ) . miR-9 and miR-132 cooperate as positive regulators of migration by preventing the expression of the transcription factor FOXP2 ( Clovis et al . , 2012 ) . Similarly , in utero electroporation of miR-137 leads to increased migration of progenitors into the cortical plate due to the ability of miR-137 to stimulate neuronal differentiation ( Sun et al . , 2011 ) . By contrast , we show that miR-128 is a negative regulator of migration and that the onset of miR-128 activity coincides with the termination of upper neuron migration . Manipulating the timing of miR-128 expression interferes with migration and cortical lamination , at least in part through regulation of the transcriptional repressor PHF6 . Like miR-128 , the Phf6 gene is restricted to vertebrates ( Lower et al . , 2002 ) . Based on cross-species comparisons of predicted miR-128 binding sites available at the TargetScan website ( Friedman et al . , 2008 ) , targeting of the Phf6 mRNA by miR-128 appears to be enhanced in mammals ( 3 sites ) , opossum ( 3 sites ) , and platypus ( 2 sites ) compared to chicken or frog ( no conserved sites ) . Within the nervous system , mutations in PHF6 have been detected in the developmental disorders Börjeson-Forssmann-Lehmann ( BFLS; OMIM 301900 ) and Coffin–Siris ( CSS; OMIM 135900 ) syndromes ( Lower et al . , 2002; Tsurusaki et al . , 2012; Wieczorek et al . , 2013 ) . BFLS is an X-linked recessive intellectual disability disorder associated with epilepsy and other developmental abnormalities . The phenotypic spectrum of CSS phenotypes overlaps BFLS and includes variable intellectual disability and developmental delay . CSS was recently shown to be associated with mutations in several components of SWI/SNF chromatin remodeling complexes in addition to PHF6 , strongly suggesting a role for PHF6 in epigenetic regulation ( Santen et al . , 2012; Tsurusaki et al . , 2012; Wieczorek et al . , 2013 ) . Furthermore , biochemical evidence has linked PHF6 to several chromatin modifying complexes , including the nucleosome remodeling and deacetylation complex ( NuRD ) ( Todd and Picketts , 2012 ) and the Polymerase associated factor 1 complex ( Paf1C ) ( Zhang et al . , 2013 ) . Paf1C has several known functions including histone modification , transcription initiation , and termination ( Jaehning , 2010 ) . PHF6 and Paf1C have been implicated in the control of neuronal migration in the mouse . Knockdown of PHF6 during embryonic corticogenesis resulted in impaired upper layer neuron migration characterized by excessive branching of the leading process . Knockdown of PAF1 led to quantitatively similar effects on migration , suggesting that PHF6 acts in the context of Paf1C to facilitate migration ( Zhang et al . , 2013 ) . The reciprocal expression patterns we observe comparing miR-128 and PHF6 during cortical development and neuronal growth in vitro suggest that miR-128 is a significant regulator of PHF6 . We also show that the effect of miR-128 on the morphology and final distribution of migrating upper layer progenitors is similar to that reported after PHF6 knockdown ( Zhang et al . , 2013 ) . Moreover , co-expression of PHF6 and miR-128 alleviated this phenotype , indicating that miR-128 is a physiological regulator of PHF6 during corticogenesis . The regulation of SWI/SNF-complex subunit composition by miR-124 provides a precedent for temporal control of epigenetic modifiers by miRNAs during neurogenesis ( Ronan , et al . , 2013 ) . By regulating PHF6 , miR-128 may play a similar role for Paf1C or the NuRD complex later in neuronal differentiation . Because premature miR-128 expression inhibited and miR-128 inhibition exaggerated radial migration , the miR-128/PHF6 circuit may play a role in how migrating neurons interpret their position , whether in response to an internal clock , external cues , or cell–cell interactions . Our results suggest that regulation of PHF6 by miR-128 is important for two interdependent aspects of upper layer neuron maturation in the cortical plate . We show for the first time that miR-128 and PHF6 cooperate in the regulation of dendritic arborization of upper layer neurons . Electrophysiological recordings also show that the balance between miR-128 and PHF6 influences cell autonomous excitability . PHF6 knockdown has previously been shown to increase the excitability of heterotopic neurons that were retained in the white matter due to impaired migration ( Zhang et al . , 2013 ) . Although this finding offers a potential explanation for the cognitive deficits and seizure activity observed in BFLS and CSS , the underlying mechanisms are not yet understood . Comparing the intrinsic properties of neurons expressing either ectopic miR-128 alone or miR-128 together with PHF6 , we found that much , but not all , of the difference in intrinsic electrophysiological properties may be directly related to the effects on structural complexity . Layer II/III neurons expressing miR-128 prematurely had reduced complexity of their dendritic arbor , with the most apparent differences observed in their proximal dendrites . This reduction in dendritic complexity was rectified by co-expression of PHF6 and miR-128 . Electrophysiological recordings further showed that the input resistance of recorded neurons was increased following miR-128 expression , as would be expected from a reduction in dendritic complexity . Interestingly , premature miR-128 expression also led to a more depolarized resting membrane potential than control cells . This is unlikely to be a direct effect of the morphological changes , and may reflect a reduction in hyperpolarizing leak currents . The net effect of the physiological changes induced by miR-128 was an increase in excitability , reflected by a reduced rheobase and increased firing frequency in response to depolarizing currents . In addition , exogenous PHF6 dampened the effects of miR-128 for all parameters tested . Thus , neuronal excitability is highly sensitive to the precise timing of miR-128 expression and subsequent repression of PHF6 during network formation in vivo . The lack of complete rescue of some parameters by PHF6 , however , indicates that additional regulatory targets of miR-128 may contribute to some of the physiological effects we see in post-migratory neurons . It is interesting to compare our gain-of-function results in cortical neurons to the phenotype observed upon targeted deletion of miR-128 in dopamine responsive neurons of the striatum ( D1 neurons ) ( Tan et al . , 2013 ) . Loss of miR-128 resulted in heightened excitability that was attributed to the upregulation of ion channels and signal transduction pathways that occurred in the absence of miR-128 . In contrast to D1 neurons , there were no significant differences in either the amplitude or the frequency of postsynaptic IPSCs or EPSCs in the cortical neurons we analyzed . Therefore , the regulatory impact of miR-128 may depend on the region and the developmental time point under investigation . We identify a regulatory interaction between miR-128 and PHF6 that is critical for the proper migration and dendritic outgrowth of upper layer neurons in the developing mouse cortex . These results may have significant relevance for the understanding of cognitive deficits and seizure susceptibility in human patients with mutations in PHF6 and highlight the importance of correct temporal regulation of miR-128 for the establishment of the cortical architecture . FMR1 mice were obtained from Charles River ( Cologne , Germany ) , C57Bl/6 mice from the Forschungseinrichtungen für Experimentelle Medizin , Berlin . Animals were handled according to the rules and regulations of the Berlin authorities and the animal welfare committee of the Charité Berlin , Germany . The expression constructs pre-miR-128-1-RED and pre-miR-128-2-RED contain the respective mouse pre-miRNA sequences together with ≈300 bp upstream and downstream flanking sequences inserted into Intron-RED , the plasmid pEM-157 containing an engineered intron in dsRed ( Makeyev et al . , 2007 ) . The PHF6 sensor construct contains the entire 3′UTR present in NM_032458 cloned downstream of eGFP in a modified peGFP-C1 vector ( Rybak et al . , 2008 ) . The miRNA sensor assay has been described in detail previously ( Rybak et al . , 2008 ) . The PHF6 expression construct contains the PHF6 cDNA cloned into the XhoI and EcoRI restriction sites present upstream of an IRES-GFP cassette in the vector pRS003 . PHF6 expression is documented in Figure 5—figure supplement 3 . Sponge design and cloning strategy are described in Rybak et al . ( 2008 ) . Sixteen high-affinity binding sites were inserted between the SalI and XhoI ones sites in a modified 3′UTR of peGFP-N1 . The repeated sequence is shown in Supplementary file 1 , as are primer sequences used for all plasmid constructs . RNA was isolated from dissected forebrain/cortex of the embryonic and post-natal stages and adult brain , from cultured cortical neurons or from transfected HEK-293 cells ( Lipofectamine 2000 ) using TRIzol ( Life Technologies , Carlsbad , CA ) according to manufacturer's instruction . For qRT-PCR of mRNA , cDNA was synthetized using RevertAid Premium Reverse Transcriptase ( Thermo Scientific , Valencia , CA ) followed by amplification using RT2 SYBR Green ( Sabio Sciences/Qiagen , Venlo , Netherlands ) according to manufacturer's instructions . GAPDH was used for normalization of primary cortical neuron samples and Oaz1 for brain samples . Quantification of miRNA expression made use of miRNA TaqMan Assays for miR-128 normalized against sno135 ( Probe Set ID:000589 and ID:1230 , Life Technologies ) . Western blotting followed standard procedures using HeLa , HEK-293 or primary cortical lysates prepared in 1% NP-40 , 20 mM Hepes pH 7 . 9 , 350 mM NaCl , 1 mM MgCl2 , 0 . 5 mM EDTA , 0 . 5 mM EGTA , 50 mM NaFl , 1 mM DTT with the addition of protease inhibitor cocktail set I ( Calbiochem/EMD Millipore , Schwalbach , Germany ) . An ImageQuant LSA 4000mini ( GE Healthcare , Little Chalfont , United Kingdom ) was used for detection , quantification by normalization to loading controls was done using Fiji software . Electrophoresis and blotting are described in Rybak et al . ( 2009 ) ; Smirnova et al . ( 2005 ) . For hybridization 20 µM LNA probe ( Exiqon A/S , Vedbaek , Denmark ) was radioactively labeled using 60 µCi [gamma32-P] ATP and T4 Polynucleotide Kinase ( Fermentas/Thermo Scientific ) . The labeled probes were diluted in 5 ml hybridization buffer ( 250 mM Na2HPO4 ( pH 7 . 2 ) , 7% SDS , 1 mM EDTA , 1% BSA ) . The membrane was incubated in a rotating hybridization oven at 46°C and then washed twice in 2× SSPE , 0 . 1% SDS and twice in 0 . 5× SSPE , 0 . 1% SDS . The signal was detected by autoradiography . In situ hybridization was performed using 5′ and 3′ digoxygenin labeled LNA probes ( Exiqon A/S ) essentially as described in Silahtaroglu et al . ( 2007 ) . Embryonic and early postnatal brain tissue was collected at the appropriate stage and fixed overnight in 4% PFA , adult brain tissue was collected after perfusion . The tissue was hybridized with double digoxigenin labeled LNA probes ( Exiqon A/S ) at the suggested hybridization temperature . Anti-digoxigen antibodies and any primary antibodies to detect proteins of interest were incubated simultaneously overnight at 4°C . Protein detection was performed first with appropriate labeled secondary antibodies followed by the enzymatic reaction to detect the miRNA . NBT/BCIP ( Roche tablets ) or Fast red ( Roche tablets ) were used , according to manufacturer's instructions ( Hoffmann-La Roche , Basel , Switzerlad ) , to detect miRNAs for bright field or fluorescence microscopy , respectively . For Phf6 mRNA detection the tissue was hybridized at the suggested temperature using a custom LNA probe ( Exiqon , see Table 1 ) with 5′ biotin and 3′ biotin-TEG labels . Anti-streptavidin-HRP antibody ( 1:500 ) was incubated overnight at 4°C . Then the Tyramide Signal Amplification ( TSA ) -Cyanine 3 system ( Perkin Elmer , Waltham , MA ) was used according to manufacturer's instructions: the fluorophore was diluted 1:50 in Amplification buffer and developed in the dark for 7 min . Cryosections were incubated in potassium sulfide solution ( 50% Potassium disulfide dissolved in water ) for 15 min , washed twice in water and incubated in cresyl violet solution ( 1 . 5% cresyl violet dissolved in acetate buffer ) for 30 min . Slices were washed for 1 min in Acetate buffer ( 0 . 01 M Sodium acetate , 0 . 01 M Acetic acid ) , 30 s in Differentiation buffer ( 500 ml water , 700 µl Acetic acid ) , and rinsed once in water . The slides were dehydrated and mounted . After in situ hybridization , using the NBT/BCIP detection method , the sections were imaged using an Olympus BX51 microscope and 40× objective . The colors of the bright field image were inverted in Fiji and the resulting image was used to measure the fluorescent intensity . The area of interest was contoured using the Polygon selection tool . The integrated density , mean fluorescence , and the area were measured . In the same image , an unstained region was contoured and measured for background substraction . The corrected total cell fluorescence ( CTCF ) was calculated using the formula: CTCF = Integrated density − ( area of selected region × mean of background ) . The fluorescence of IZ and VZ/SVZ were normalized to the fluorescence of the CP in each image . The normalized values were used for the analysis . At least three slices per brain and three brains per condition were analyzed . The statistical test used was One-Way ANOVA . Embryonic brain tissue was collected at the appropriate stage and fixed in 2% PFA for 6 hr . Cryosections were not post-fixed but directly incubated in blocking buffer ( 1× PBS , 0 . 25% Triton X , 0 . 1% Tween 20 , 3% BSA ) . The sections were incubated overnight at 4°C with anti-PHF6 antibody ( BETHYL A301-451A 1:100 , Bethyl Laboratories , Montgomery , TX ) . Antibody specificity is documented in Figure 5—figure supplement 3A . For the detection , the tissue was incubated for 1 hr at room temperature with anti-rabbit secondary antibody-HRP conjugate followed by TSA Cyanine 3 system detection according to manufacturer's instructions ( Perkin Elmer ) . The fluorophore was diluted 1:50 in Amplification buffer and developed in the dark for 7 min . In utero electroporation of NMRI mice was performed as described in Saito ( 2006 ) with minor modifications . A 300 ng/µl solution of pre-miR-128-1-RED , pre-miR-128-2-RED or control Intron-RED plasmids and/or IRES-GFP control or PHF6-GFP vector at 150 ng/µl was injected in one lateral ventricle . E15 . 5 embryos were electroporated using 6 pulses of current at 35 mV . The resulting embryos or pups were processed for immunohistochemistry ( migration analysis , marker detection ) or electrophysiology . The migration analysis was assessed at P7 on 50-µm brain slices . The slices were collected from the beginning of the corpus callosum to the middle of the hippocampus . Floating slices were stained for detection of dsRed ( Abcam ab62341 at 1:150 , Cambridge , United Kingdom ) and GFP ( Abcam ab13970 at 1:500 ) . The primary antibodies were dissolved in blocking solution ( 1× PBS , 0 . 25% Triton-X , 0 . 1% Tween-20 , 3% BSA ) . The slices were incubated in primary antibody overnight at room temperature with shaking . Secondary antibodies were incubated 2 hr at room temperature on a shaker . The mounted sections were imaged using a Leica SL confocal microscope with a 10× objective . A grid consisting of 10 bins was applied to the images , positioning the beginning of the first bin at the beginning of layer II and the end of the tenth bin at the end of layer VI , as determined by visual analysis of nuclear staining ( DRAQ5 ) , essentially as described ( Rosário et al . , 2012 ) . When necessary more than one adjacent grid was applied to cover the entire electroporated region . Neurons within each bin were counted using the Cell counter plugin for Fiji . An average of five sections from at least three independent brains per condition was analyzed . The number of neurons in each bin was normalized first for individual brains and then the normalized value was used as n = 1 per condition . The data were analyzed in Prism 5 . 0 using Two-way ANOVA . Intron-RED and pre-miR-128-2-RED electroporated pups from the same litter were analyzed at P0 ( born E20 ) . 10 µm cryosections were stained for dsRed ( Abcam ab62341 1:150 ) , Cux1 ( Santa Cruz Biotechnology , sc-13024 1:150 , Heidelberg , Germany ) , and Ctip2 ( Abcam ab18465 1:500 ) . The slices were imaged using a Leica SL confocal microscope with 40× objective . Using the Cell Counter plugin for Fiji both the total number of electroporated neurons in the cortical plate and the number of electroporated neurons positive for either Cux1 or Ctip2 was counted . The number of neurons positive for the layer marker was normalized to the total number of electroporated neurons . Three independent brains electroporated with pre-miR-128-2-RED and one brain electroporated with Intron-RED were analyzed and for each layer marker at least three slices per brain were counted . The analyzed P0 brains ( born E19 ) from Intron-RED + pRS003 ( n = 3 ) and pre-miR-128-2-RED ( n = 4 ) electroporated animals were from the same litter . 60-µm sections were stained for dsRed ( Abcam ab62341 1:150 ) and GFP ( Abcam ab13970 1:500 ) . Nuclear staining was obtained with DRAQ5 ( Biostatus , Shepshed , United Kingdom ) . Images were taken using a Leica SL confocal microscope . The overview was taken as a single image with 10× objective . Images for reconstruction of migrating neurons were taken with a 40× objective and a 1 µm step Z-stack . The deep layers were defined using nuclear stain and a pool of migrating neurons within the deep layers was reconstructed using the Fiji plugin Simple Neurite tracer . The number of branches and filopodia ( excluding the trailing process ) was counted . To distinguish between branch and filopodium a cut-off of 5 µm was used . Acute brain slices were prepared from P15 mice after electroporation as described in the text . Slice preparation , recordings , visualization of the neurons , and data analysis were performed as described previously ( Booker et al . , 2013 ) . In brief , 300-μm thick coronal slices including the somatosensory cortex were prepared in ice-cold carbogenated sucrose-substituted artificial cerebrospinal fluid ( ACSF; in mM: 87 NaCl , 2 . 5 KCl , 25 NaHCO3 , 1 . 25 NaH2PO4 , 25 glucose , 75 sucrose , 7 MgCl2 , 0 . 5 CaCl2 , 1 Na-Pyruvate , 1 Ascorbic Acid ) , left to recover at 35°C for 30 min , then stored at room temperature . Whole-cell patch clamp recordings were performed in a submerged recording chamber superfused with carbogenated recording ACSF ( in mM: 125 NaCl , 2 . 5 KCl , 25 NaHCO3 , 1 . 25 NaH2PO4 , 25 glucose , 1 MgCl2 , 2 CaCl2 , 1 Na-Pyruvate , 1 Ascorbic Acid ) at 32–34°C , from visually identified GFP-positive neurons within the electroporated region of the somatosensory cortex , using a Multiclamp 700B amplifier ( Molecular Devices , Sunnyvale , CA ) . Patch pipettes were filled with a K-gluconate based intracellular solution ( in mM: 130 K-Gluc , 10 KCl , 2 MgCl2 , 10 EGTA , 10 HEPES , 2 Na2-ATP , 0 . 3 Na2-GTP , 1 Na2-Creatinine and 0 . 1% biotinylated-lysine ( Biocytin , Invitrogen/Life Technologies ) , pH 7 . 3 , 290–310 mOsm ) , resulting in a pipette resistance of 2–5 MΩ . Voltage signals were digitized at 10 kHz ( NI-DAQ , National Instruments , Newbury , UK ) , acquired with WinWCP software ( J Dempster , Strathclyde University ) and analysed offline using Stimfit ( C Schmidt-Hieber; www . stimfit . org ) . The intrinsic physiology of neurons was characterized in current-clamp mode , with a family of hyperpolarizing to depolarizing current pulses ( −250 to 250 pA , 50 pA steps , 500 ms duration ) ; determining the current–voltage relationship , Ih mediated voltage sag and action potential ( AP ) discharge frequency . Small hyperpolarizing current pulses ( −10 pA , 500 ms duration ) were applied to assess the input resistance ( RI ) of the recorded neurons . Membrane potential ( VM ) was calculated as the 50 ms baseline prior to the small hyperpolarizing step . Following intrinsic characterization , outside-out patches were formed and biocytin was allowed to fill the cell for an additional 15 min . Slices were immersion fixed in 4% formaldehyde in 0 . 1 M phosphate buffer ( PB ) overnight at 4°C . Slices were copiously rinsed in PB and the filled cells visualized with Avidin-conjugated Alexa-Fluor-647 ( Invitrogen/Life Technologies; 1:1000 ) , in PB containing 0 . 3% Triton X-100 and 0 . 05% NaN3 , overnight at 4°C . Slices were subsequently rinsed in PB and mounted on glass slides , with a 300 μm agar spacer to prevent compression of the slices after cover-slipping . The slices were imaged using Leica SL confocal ( 1024 × 1024 resolution ) using ×20 objective and 200 Hz speed . The step between stacks was 1 µm . For analysis at P0 60 µm slices were prepared from single litters of electroporated animals and processed for immunostaining with dsRed and eGFP antibodies plus DRAQ5 nuclear stain . Images were taken using a Leica SL confocal microscope , for reconstruction a 40× objective and Z-stack step of 1 µm was used . Nuclear staining was used to identify the deep layers of the cortical plate , individual neurons were reconstructed with the Fiji plugin Simple Neurite tracer . Quantification was essentially as described in Guerrier et al . ( 2009 ) . For analysis at P21 electroporated animals were sacrificed , perfused , and 100-µm slices were prepared . Electroporated neurons were visualized by staining with dsRed antibody . Z-stack images were taken with an inverted epifluorescence microscope ( Olympus IX81 ) with a 1 µm stack and reconstructed as above . For P15 neurons , after recording outside-out patches were formed and cells were filled with biocytin for 15 min . After overnight fixation by immersion in 4% formaldehyde in 0 . 1 M phosphate buffer ( PB ) at 4°C , the filled cells visualized with Avidin-conjugated Alexa-Fluor-647 ( Invitrogen/Life Technologies; 1:1000 ) , in PB containing 0 . 3% Triton X-100 and 0 . 05% NaN3 , overnight at 4°C . After rinsing in PB and mounting on glass slides with a 300 μm agar spacer the slices were imaged using a 20× objective and a Leica SL confocal microscope at 200 Hz . The step between stacks was 1 µm at a resolution of 1024 × 1024 . Neurons were reconstructed using the Simple Neurite Tracer plugin . Sholl analysis was performed on 3-D reconstructions using the Sholl analysis plugin for Fiji . The radius of the first concentric sphere was set at 7 . 5 µm and the increase between radii was 5 µm . The data set for Sholl analysis at P15 and P21 are provided in Supplementary file 2 and Supplementary file 3 , respectively . Statistical analysis was performed using Prism 5 . 0 , when indicated multi-group comparisons were analyzed by Two-way ANOVA with the Bonferroni posttest; when comparing two groups a Student's unpaired t-test was employed as indicated in each legend . Significance is denoted in the Figures as: p < 0 . 001; p < 0 . 01; p < 0 . 05; ns , not significant .	The unique capabilities of the mammalian brain depend on the patterns formed by spatial arrangements and connections between millions ( sometimes billions ) of electrically active cells called neurons , and on the connections between these neurons . During the development of the cortex , the largest part of the brain , neurons are born in stem cell areas that lie deep inside the brain , and these newly made neurons then migrate outwards to their final positions close to the surface of the adult brain . Franzoni et al . have examined how two molecules , a small RNA called miR-128 and a protein called PHF6 , control when and how neurons migrate through the cortex and then grow to form connections with other neurons as they mature . Mutations that disrupt PHF6 can cause intellectual disabilities , and one possible reason for this is that PHF6 is needed to ensure that the neurons migrate to the correction location . Franzoni et al . now show that miR-128 can reduce the production of PHF6 and is therefore responsible for controlling when and where PHF6 is active . Studying miR-128 in detail , they show that although an inactive precursor form of miR-128 is present in stem cells and migrating neurons , the active form of miR-128 is only found in neurons that have already reached their final position in the cortex . Franzoni et al . used genetic methods to override the switch that controls when miR-128 becomes active . When the amount of miR-128 was artificially reduced , the neurons migrated too far . Artificially increasing the amount of miR-128 had the opposite effect: both the movement of the neurons and , later , their growth were defective . PHF6 was the key to these effects: if PHF6 levels were kept close to normal , miR-128 could no longer interfere with the movement and growth of the neurons . Further work will be required to better understand how miR-128 is turned off and on , and how PHF6 acts to control neuronal movement and growth .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "cell", "biology" ]	2015	miR-128 regulates neuronal migration, outgrowth and intrinsic excitability via the intellectual disability gene Phf6
We describe the physical context of the Dinaledi Chamber within the Rising Star cave , South Africa , which contains the fossils of Homo naledi . Approximately 1550 specimens of hominin remains have been recovered from at least 15 individuals , representing a small portion of the total fossil content . Macro-vertebrate fossils are exclusively H . naledi , and occur within clay-rich sediments derived from in situ weathering , and exogenous clay and silt , which entered the chamber through fractures that prevented passage of coarser-grained material . The chamber was always in the dark zone , and not accessible to non-hominins . Bone taphonomy indicates that hominin individuals reached the chamber complete , with disarticulation occurring during/after deposition . Hominins accumulated over time as older laminated mudstone units and sediment along the cave floor were eroded . Preliminary evidence is consistent with deliberate body disposal in a single location , by a hominin species other than Homo sapiens , at an as-yet unknown date . The Rising Star cave system lies in the Bloubank River valley , 2 . 2 km west of Sterkfontein cave . It comprises an area of 250 × 150 m of mapped passageways situated in the core of a gently west dipping ( 17° ) open fold , and is stratigraphically bound to a 15–20 m thick , stromatolitic dolomite horizon in the lower parts of the Monte Christo Formation ( Eriksson et al . , 2006; Figures 1B and 2A ) . This dolomite horizon is largely chert-free , but contains five thin ( <10 cm ) chert marker horizons that have been used to evaluate the relative position of chambers within the system ( Figure 2B ) . The upper contact is marked by a 1–1 . 3 m-thick , capping chert unit that forms the roof of several large cave chambers . Surface mapping indicates that cave openings and flowstone-filled fractures do not penetrate this capping chert unit , except where a dextral fault truncates the stratigraphy to the south of the system ( Figure 2 ) . The network of cavities is developed along west-northwest , north and northwest trending fractures and joints . 10 . 7554/eLife . 09561 . 004Figure 2 . Geological map and cross-section of the Rising Star cave system . ( A ) Geological Map showing the distribution of chert-free dolomite and fracture systems controlling the cave . Inset shows the location of the Cradle of Humankind in southern Africa; ( B ) Northeast-Southwest , schematic cross section through the cave system , relative to several chert marker horizons; ( C ) Detailed map of the Dinaledi Chamber showing the orientation of the floor and the position of the excavation and sampling sites . DOI: http://dx . doi . org/10 . 7554/eLife . 09561 . 004 The fossil-bearing chamber , named the Dinaledi Chamber ( ‘Chamber of Stars’ in the Sotho language; Figure 2A ) , is ∼30 m below surface and ∼80 m , in a straight line , away from the present , nearest entrance to the cave ( Figure 2B ) . It is situated in the central part of the system and was found during speleological surveys ( see methodology section ) . The only identified access point into the Dinaledi Chamber involves an exposed , ∼15 m climb from the bottom of a large ante-chamber ( the Dragon's Back Chamber ) , up the side of a sharp-edged dolomite block that has dislodged from the roof ( the Dragon's Back; Figure 2B ) . From the top of the Dragon's Back , the Dinaledi Chamber is accessed via a narrow , northeast-oriented vertical fissure , and involves a ∼12 m vertical climb down , with squeezes as tight as ∼20 cm , to reach the floor ( Figure 2B , C ) . The main passage forming the Dinaledi Chamber is ∼25–50 cm wide at its narrowest and ∼10 m long , and expands in width near the intersections with cross-cutting passages ( Figure 2C ) . The roof of both the Dinaledi and Dragon's Back chambers is formed by the capping chert ( Figure 2B ) . The Dragon's Back Chamber can currently be accessed in two ways , both involving steep climbs along narrow fractures and tight passages ( Figure 2A ) : route 1 , along an east-northeast trending passage that follows a fracture for a horizontal distance of ∼50 m past a narrow access point called the ‘postbox’; and route 2 , along a more complicated set of broadly east-trending passages , via a network of southeast , east and north trending fractures for ∼120 m , and past a narrow access point called ‘superman crawl’ . Route 1 is the most direct and contains abundant sediment accumulations once the deeper part of the cave is accessed ( i . e . , ∼20 m into the cave ) ; route 2 has a gentler gradient , but is longer and involves a descent along narrow fissures largely devoid of sediment accumulations . An exhaustive search by a professional caving team and researchers has failed to find any other plausible access points into the Dinaledi Chamber , and there is no evidence to suggest that an older , now sealed , entrance to the chamber ever existed . Furthermore , detailed surface mapping of the landscape overlying the Rising Star cave system ( Figure 2A ) illustrates that no large flowstone-filled fractures occur in the region above the Dinaledi Chamber . The skeletal material recovered from the Rising Star cave was collected during two field expeditions in November 2013 and March 2014 , and includes 1550 identifiable fossil hominin specimens as well as six bird and several rodent specimens . Of the 1550 hominin specimens , ∼300 numbered bone fragments were collected from the surface of the Dinaledi Chamber , and ∼1250 numbered fossil specimens were recovered from a small excavation pit in that chamber . Throughout the Rising Star cave system erosional remnants of fossiliferous sediment , breccia , and flowstone units provide evidence for several cycles of sediment-flowstone fill and removal/dissolution as the level of the water table in the cave changed repeatedly . On approaching the Dinaledi Chamber from the closest current cave entrance , sediment deposits in the cave become progressively finer-grained . The coarse-grained clastic deposits encountered furthest into the cave are in the Dragon's Back Chamber , and include channelized sandstone and quartz/chert pebble conglomerate units that terminate against the Dragon's Back . The fill generally dips gently west ( <5° ) in the down slope direction of the passages , and becomes near horizontal in the Dragon's Back and Dinaledi chambers . These chambers contain no evidence of sediment input from proximal sources that would indicate a nearby vertical shaft connecting these chambers to surface ( Pickering et al . , 2007 ) . Mapping reveals no large cave openings on surface above these chambers , although narrow , flowstone-filled fractures occur ( Figure 2A ) that may have allowed passage of water and some fine-grained sediment into caverns below . The Dinaledi Chamber is exclusively filled by flowstone and fine-grained sediment involving two depositional facies distributed across three stratigraphic units that filled the chamber over time . Stratigraphic units are separated by erosional unconformities , or laterally continuous flowstone intercalations . Erosion remnants of the units occur in a variety of stratigraphic positions , and there is extensive evidence of reworking with older units being re-deposited into younger units . Each of the facies and units , and each of the flowstone phases that are interlayered with the various units , are described below ( Figures 3 , 4 ) . 10 . 7554/eLife . 09561 . 005Figure 3 . Cartoon illustrating the geological and taphonomic context and distribution of fossils , sediments and flowstones within the Dinaledi Chamber . The distribution of the different geological units and flowstones is shown together with the inferred distribution of fossil material . DOI: http://dx . doi . org/10 . 7554/eLife . 09561 . 00510 . 7554/eLife . 09561 . 006Figure 4 . Stratigraphic units and flowstones observed in the Dinaledi Chamber . ( A ) Erosional remnant of horizontally laminated Unit 1 strata ( Facies 1 ) . ( B ) Close-up view of Unit 1 ( Facies 1a ) showing fine laminations and small invertebrate burrows ( note fine sand infilling in burrows ) . ( C ) Overview photo of the Dinaledi Chamber , directly to the east of the entrance point into the chamber . Photo shows distribution of Flowstones 1–3 and stratigraphic Units 2 and 3 . ( D ) Close-up view of Flowstone 1 encasing sediment of Unit 2 . Note that several generations of flowstone ( Flowstones 1a–e ) are coating Unit 2 . The thin , clear lower layer is Flowstone 1a , and the overlying white flowstone is either Flowstone 2 or 3 . ( E ) Close-up view of Unit 2 , consisting of generally poorly-cemented Facies 2 sediment . ( F ) View of the chamber floor near the entry point . On the cave floor , a large erosional remnant of Unit 1 ( orange laminated mudstone of Facies 1a ) , is surrounded by mud-clast breccia of Unit 3 ( main hominin bearing unit ) . Note that Flowstone 2 has been undercut by post-depositional erosion of Unit 3 , which , in this location has resulted in a lowering of the floor by as much as 25 cm . ( G ) Flowstone 2 overlying Unit 3 in one of the chamber's side passages . In this location Unit 3 has also been partly eroded after depositional from underneath the flowstone drape , leaving a hanging remnant , with some indurated sediment of Unit 3 attached to its base . Note the continued deposition of sediment above Flowstone 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 09561 . 006 A number of different flowstone formations can be identified in the Dinaledi Chamber . These formed at separate times , and help demarcate stratigraphic units . Each type of flowstone is described below starting with the oldest . Using the distribution of flowstones , sedimentary facies ( e . g . , the presence of Facies 1 mud clasts inside Facies 2 ) , and erosional contacts , a basic stratigraphy of three separate units has been established in the Dinaledi Chamber . This stratigraphic interpretation is preliminary and based solely on geological reasoning ( i . e . , stratigraphic superposition of individual units , textural variations in Facies 2 sediment and the presence or absence of dissolution features ) , because of an absence of reliable age data . It is important to note that the facies described above are not synonymous with stratigraphic ordering although in general Facies 1 sediments appear to be older than Facies 2 sediments . Variations in texture , composition and degree of lithification of Facies 2 sediments , as well as direct contact relationships , make it possible to define different stratigraphic units composed of Facies 2 . Because we do not yet have a clear understanding of the age relationships , nature of disconformable surfaces or the extent of reworking between the units , we refrain from defining these as discrete allostratigraphic units and instead prefer the use of lithostratigraphic units . The three lithostratigraphic units and their distribution in the Dinaledi Chamber are described below . Mineralogical and chemical studies were carried out on sediment samples from the floors of the Dinaledi and Dragon's Back chambers ( Facies 2; Unit 3 ) to assess the nature of the sediment encasing the Dinaledi fossils and to determine whether these chambers were connected at the time of deposition of Units 2 and 3 ( see Supplementary file 1 for analytical results ) . Four samples of cave sediment were studied using different techniques to obtain data on their size fraction , mineralogy and geochemical characteristics . Samples UW101-SO-31 , UW101-SO-34 and UW101-SO-39 were collected near the excavation pit in the Dinaledi Chamber , from brown matrix-supported mud clast breccia of Unit 3 ( Figures 2C and 7 ) . UW101-SO-31 is light-brown in colour , and contains sand grains and has bone fragments . UW101-SO-34 consists of dark-brown mud and is not gritty to feel . UW101-SO-39 is a medium- to dark-brown mud that is slightly gritty to feel . One additional sample , DB-1 , was collected for analysis from unconsolidated floor sediment in the Dragon's Back chamber ( Figure 2B ) . This is a brown silty mud with a fine-gritty feel , containing sub-mm sized bone fragments . The three samples of Unit 3 from the Dinaledi Chamber ( SO31 , 34 , 39; Figures 2C and 5D , E , 8 ) are dominated by reworked , angular mud clasts in a clay matrix , with some chert fragments , but with little externally derived detrital quartz . The sample from the Dragon's Back Chamber is dominated by detrital quartz , with some detrital muscovite , as well as shale and chert fragments , of which some are altered and impregnated and/or coated with Mn- and Fe-oxides . Mudstone fragments are rare in DB-1 ( Figures 5C and 8 ) . Elevated MnO ( 3 . 9–4 . 2% ) and Fe2O3 ( 10 . 3–11% ) levels in the floor sediments of both chambers are associated with alteration of chert or mudstone prior to their comminution , with Mn- and Fe oxides/hydroxides occurring as replacement and micro-vein infilling . XRD diffractograms identified both quartz and muscovite in all samples with high scores compared to other minerals . Hematite was identified with high scores in all the Dinaledi samples . Goethite ( FeO ( OH ) ) and birnessite ( ( Na , Ca , K ) ( Mn4+ , Mn3+ ) 2O4·1 . 5H2O ) were only identified in UW101-SO-31 , with low scores , but high certainty . Other minerals identified with low scores and high certainties are dolomite ( ( CaMg ) ( CO3 ) 2 ) in UW101-SO-34 and kaolinite ( Al2Si2O5 ( OH ) 4 ) in UW101-SO-39 . Using XRD , only quartz and muscovite were identified in DB-1 . Acid tests showed all samples to be free of calcite or aragonite . All analysed Mn oxi-hydroxides in both chambers contain fluorine , with similar ( atomic ) F/Mn ratios of ≈ 0 . 14 , indicating a similar chemical environment during their formation . Different patterns in K2O vs Al2O3 plots reflect a dominance of mudstone fragments consisting mainly of clay minerals , in the Dinaledi Chamber , and the presence of muscovite in the Dragon's Back Chamber ( Figure 8 ) . The contrasting composition in particulate matter of floor sediments in the two chambers , suggests that the Dinaledi Chamber was an isolated sedimentary environment at the time of deposition of Unit 3 , with no or very limited transfer of sediment between the two chambers . 10 . 7554/eLife . 09561 . 010Figure 8 . Comparison of selected fragments and electron microprobe analytical data of Facies 2 ( Unit 3 , floor ) sediment in the Dinaledi Chamber , and floor sediments in the Dragon's Back Chamber . Analytical spot size is 5 μm diameter , which is generally larger than grain sizes . ( A ) Chert fragment impregnated with Mn oxi-hydroxide from the Dragon's Back Chamber . ( B ) Shale fragment from the Dragon's Back Chamber . ( C ) orange mud clast , typical of Facies 2 sediments from the Dinaledi Chamber , note much finer grain size than seen in ( B ) . ( D–G ) Plots of K2O vs Al2O3 for mud clast fragments in Facies 2 samples from both chambers show an important difference between them . M , muscovite compositional field . The samples from the Dinaledi Chamber ( E–G ) yield some data close to the muscovite field , probably indicating sericite grains slightly smaller than the spot size , and all show a trend with K/Al ratios much lower than muscovite , up to a high Al2O3 content >30% , which indicates either illite , or mixtures of sericite and kaolinite or other K-free clay minerals . In ( E ) the analysis of the fragment shown in ( C ) is indicated . The sample from the Dragon's Back Chamber in ( D ) shows data in the muscovite field ( data point corresponding to [B] is indicated in [D] ) , but otherwise only low K- and Al-concentrations , which are typical of Mn oxi-hydroxide impregnation ( data point corresponding to [a] is indicated in [D] ) . No analytical data in d correspond to mudstone fragments such as shown in ( C ) . ( H–K ) , plots of Fe as FeO vs Mn as MnO show similarity between the chambers with respect to Fe-Mn oxi-hydroxide impregnations and alterations within the fragments: in both cases , domains with high Fe rarely coincide with domains high in Mn . ( L–O ) , Plots of F vs Mn as MnO . Some elevated F concentrations at zero Mn values occur , but most data show a correlation for samples from both chambers: elevated Mn content is invariably associated with elevated F , with an atomic ratio F/Mn ≈ 0 . 14 . No Mn oxi-hydroxide minerals with F have been described , but partial substitution of F− for OH− as in apatite might be suspected . The similarity of the F/Mn ratios in Mn oxi-hydroxide impregnated fragments from Facies 2 sediments in both chambers suggests a uniform geochemical environment during the Mn oxi-hydroxide alteration event . DOI: http://dx . doi . org/10 . 7554/eLife . 09561 . 010 Flowstone samples from the Dinaledi Chamber were analysed for uranium to assess the possibility of U-Pb dating . Although analysed samples mostly contain sufficient U for this ( 0 . 3–0 . 7 ppm ) , a fine dusting of a detrital component derived from associated muds is present in all tested pilot samples . This has confounded preliminary attempts at U-Pb dating , because of the high , and isotopically variable , background of common Pb it carries . A detailed taphonomic study of the hominin remains is ongoing , and initial results have been provided in this study to illustrate the broader context of the fossil assemblage . Analyses of taphonomic processes affecting the Dinaledi hominin fossil assemblage have been conducted to describe decomposition , weathering and fracture patterns , surface modifications as a result of invertebrate–bone interactions ( as well as the lack of vertebrate–bone interactions ) , and spatial context including skeletal distribution patterns ( Behrensmeyer et al . , 1986; Galloway , 1999; Straus and Porada , 2003; Loe , 2009; Symes et al . , 2013; Figures 7 , 9–12 ) . Summary tables of results are presented in Table 1 and Supplementary file 2 . 10 . 7554/eLife . 09561 . 011Figure 9 . Taphonomic spatial patterning within the fossil assemblage exposed in the excavation pit . Taphonomic signatures and spatial orientations suggest that some of the assemblage may be para-authochthonous in nature , rather than primary or in situ . This scenario provides a mechanism for explaining the combination of near- or fully-anatomically articulated skeletal material and elements , which are heavily commingled and in a non-horizontal resting state ( from near-vertical to oblique long-axis orientations ) . ( A ) Example of an articulated ankle region . ( B ) Example of an articulated hand . ( C ) Example of cluster of skeletal elements showing disarticulated elements in a non-horizontal resting state . Note long bone fragment in near-vertical alignment , compared to normal horizontal or near-horizontal alignment of the commingled elements surrounding it . Labels denote specimen numbers . DOI: http://dx . doi . org/10 . 7554/eLife . 09561 . 01110 . 7554/eLife . 09561 . 012Figure 10 . Examples of taphonomic traces recorded on hominin remains . ( A ) UW101–1288 tibial diaphysis showing evidence of mineral staining adhering to the cortex . The fossil shows evidence of dark zone sub-aerial or sub-surface weathering . Specimen shows a central midline crack with sediment infill , which separates conjoined manganese concretions . ( B ) UW101–419 ( Cranium A[1] ) showing iron oxide staining around the external auditory meatus . ( C ) UW101–312 and 1040 conjoined fragments of a tibial shaft , showing stepped transverse fracture ( post-mortem ) of the mid-shaft; note longitudinal crack , and evidence of invertebrate modification . ( D ) UW101–1288 tibial diaphysis showing a weathering pattern typical of Stage 1 evidenced by fine longitudinal cracks , without concomitant flaking , delamination , or the formation of fibrous texture . ( E ) UW101–1074 tibial shaft showing manganese mineral concretions overlying yellow staining across the diaphysis . ( F ) Specimen UW 101–419 Cranium A ( 1 ) displaying tide lines of dark brown , reddish brown and yellow staining , which extends across different vault fragment . ( G ) UW101–498 tibial shaft , showing comminuted post-mortem fracture/crushing preserved by sediment infiltrate . ( H ) UW101–1070 segment of tibial diaphysis displaying differential mineral staining patterns between conjoined fragments . DOI: http://dx . doi . org/10 . 7554/eLife . 09561 . 01210 . 7554/eLife . 09561 . 013Figure 11 . Taphonomy—surface modifications . ( A ) Removal of the bone surface with sets of shallow , evenly spaced , multiple parallel striations on fibula ( UW101–1037 ) , which run longitudinal with the main axis of the bone and are interpreted as gastropod radula damage . ( B ) Fibula ( UW101–1037 ) showing removal of the bone surface with sets of shallow , evenly spaced , multiple parallel striations that follow the collagen fibres together with shallow circular pits ranging from 0 . 1 to 3 mm in diameter , the bases of which may be smooth , cupped , or covered with multiple parallel striations . These features have been attributed to gastropod radula damage . ( C ) Tibia ( UW101–484 ) showing removal of the bone surface with sets of shallow , striations that show a smooth scalloped edge together with circular pits ranging from 0 . 1 to 3 mm in diameter interpreted as the result of gnawing by beetle larvae . ( D ) Tibia ( UW101–484 ) with areas of surface removal that have a straight edge associated with scrape marks interpreted as damage made by a beetle mandible . ( E ) Fibula ( UW101–1037 ) with sets of shallow , evenly spaced , multiple parallel striations orientated transverse to the long axis of the bone interpreted as gastropod radula damage , resulting in an etched surface appearance that exposes underlying structures . ( F ) Tibia ( UW101–484 ) showing clusters of large individual striations that are variably arrow-shaped and often overlap , interpreted as damage made by a beetle mandible . Compare with Figure 12 which shows surface modifications made by modern snails and beetles and their larvae . The scale bar in all samples equals 1 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 09561 . 01310 . 7554/eLife . 09561 . 014Figure 12 . Comparative examples of surface modifications on bone made by modern snails and beetles and their larvae after four months in controlled experiments . Gastropods and beetles were found to produce similar modifications to those observed on the Rising Star hominin remains , and remove the surfaces of fresh , dry and fossil bones to an equal degree ( see Figure 11 ) . ( A ) Dry bovid rib showing surface removal associated with evenly spaced , multiple parallel striations made by the radula of an Achatina ( land snail ) . ( B ) Fresh sheep bone that was originally covered with tissue showing how Helix aspersa ( garden snails ) have removed the outer cortical lamellae to produce an etched appearance and create circular shallow pits with smooth and striated bases . ( C ) Dry bovid rib showing shallow , evenly spaced , multiple parallel striations produced by Achatina . ( D ) Dry bird femur showing large individual striations that are variably arrow-shaped and often overlap , made by Omorgus squalidus ( hide beetles ) . ( E ) A weathered bovid tooth showing surface removal with a scalloped edge produced by Dermestes maculatus larvae , and with a straight edge associated with scrape marks . ( F ) Scrape marks created by a D . maculatus adult beetle mandible on a dry medium-sized bovid long bone flake . The scale bar in all samples equals 1 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 09561 . 01410 . 7554/eLife . 09561 . 015Table 1 . Element distribution patterns recording the Minimum Number of Individual and Indentifiable Elements ( MNIE ) for skeletal parts of the H . naledi assemblage from the Dinaledi ChamberDOI: http://dx . doi . org/10 . 7554/eLife . 09561 . 015ElementMNIE left--MNIE rightMandible7Calvaria6C12C22Other cervical3Thoracic13Lumbar3First rib2Second rib11Sternum1Clavicle32Scapula23Humerus35Radius24Ulna23Scaphoid13Lunate12Capitate12Trapezoid12Trapezium12Triquetral1MC143MC234MC333MC41MC511Proximal manual phalanges35Intermediate manual phalanges27Distal manual phalanges14Ilium45Ischium45Pubis32Sacrum1Coccyx1Femur59Patella4Tibia45Fibula44Talus62Calcaneus22Navicular33Medial cuneiform30Intermediate cuneiform34Lateral cuneiform21Cuboid12MT133MT213MT322MT413MT53Proximal pedal phalanges12Intermediate pedal phalanges14Distal pedal phalanges6 Whereas Unit 1 , is a distinct older stratigraphic unit , Units 2 and 3 appear to have formed in a continual manner involving the interaction of three separate sedimentary processes: ( a ) sediment accumulation below access points into the cave chamber ( i . e . , near the current vertical entrance shaft ) ; ( b ) erosion of the accumulating sedimentary pile , as sediment slumped down-slope , into the chamber towards floor drains; and ( c ) stabilisation by flowstone formation . The accumulation of poorly consolidated Facies 2 sediment in Units 2 and 3 , alternated with periods of flowstone deposition . Each of the Flowstones 1a-e interpreted to have formed as crust on top of a debris pile of Facies 2 sediment , starting with the formation of Flowstone 1a on top of a debris cone forming Unit 2 . Formation of Flowstone 1a was followed by a period of erosion during which the poorly consolidated sediment pile , slumped inwards into the cave , probably as a result of the removal of sediment through floor drains deeper down in the cave chamber . This erosional process would have undercut the sediment pile of Unit 2 , leaving behind partly indurated erosional remnants of Unit 2 covered by remnants of Flowstone 1a . As the sedimentary pile below this erosion remnant stabilized and flowstone formation continued , a new crust formed on top of the pile of Facies 2 sediment , to once more , weakly indurate the top-layer . This was followed by the next phase of erosion , slumping and undercutting of the sediment pile , leaving behind erosional remnants of Flowstone 1b again covering the remains of partly indurated Facies 2 mudstone breccia assigned to Unit 2 . This progressive erosional process repeated itself at least three more times to form Flowstones 1c-e , as sediment of Unit 2 was gradually reworked and spread out across the floor of the Dinaledi Chamber , where it accumulated as Unit 3 ( Figure 4 ) . This process was concomitant with continued weathering , auto-brecciation and erosion of remnants of Unit 1 , which also contributed to formation of Unit 3 . Deposition of Unit 3 was followed by a major phase of flowstone formation that formed cascades , curtains and flowstone crusts atop Unit 3 across the entire chamber ( Flowstone 2 ) . In the lower parts of the chamber and in many of the lower side passages , large areas of the cave floor ( i . e . , Unit 3 ) are still covered by Flowstone 2 ( and to a lesser extent , more recent patches of Flowstone 3 ) , including areas where the flowstone directly covers H . naledi bones . However , in the upper and central parts of the main chamber , Unit 3 has been locally eroded , undercutting and partly removing the cover of Flowstone 2; for example , near the fossil excavation pit , erosion remnants of Flowstone 2 along the chamber wall are positioned several cm's above the present chamber floor surface , indicating erosion of the uppermost part of Unit 3 since deposition of Flowstone 2 ( Figure 4F , G ) . In addition , collapsed sediment containing hominin bones derived from hanging remnants of Unit 3 have locally washed down to be re-deposited atop Flowstone 2 lower down in the chamber , where the floor is level and no erosion has occurred . Thus , the erosional and re-depositional processes that led to the formation of Units 2 and 3 , is on-going today , and the distribution of fossils on the cave floor is , in part at least , the result of erosional winnowing . The remains of H . naledi are concentrated in Unit 3 , but remains also occur in Unit 2 . Four hominin bone fragments were observed in situ within Unit 2 , and the shafts of a radius and ulna derived from a juvenile hominin were found in loose material slumped from hanging remnants of Unit 2 . The large number of hominin bones collected from around the weathered base of Unit 2 also suggests that this unit contained a significant number of the fossils that contributed to the H . naledi assemblage . However , many of the remains of H . naledi were probably not eroded out of Unit 2 to be re-deposited in Unit 3 . Unlike Unit 2 , which is limited to small erosion remnants near the cave entrance , Unit 3 is distributed across the Dinaledi Chamber and preserves well-articulated skeletal elements that have not been reworked from older units . This is illustrated by the presence of an articulated ankle , foot and hand in the excavation pit in Unit 3 ( Berger et al . , 2015this volume; Figure 9 ) , which suggests that these remains were deposited in situ during the accumulation of Unit 3 . It is highly unlikely that these remains weathered out from older sediments of Unit 2 , because Unit 2 accumulated near the entrance to the chamber , that is , ∼11 m away from the pit ( Figure 2C ) , and there are no accumulations of Unit 2 above or near the excavation pit . In addition , if a hand or foot would have weathered out of the unconsolidated sediment of Unit 2 , they would not have remained perfectly articulated as the sediment slumped downslope into the chamber , past floor drains . Also , there have not been any flowstone fragments found within Unit 3 that would suggest that reworked skeletal material from unit 2 may have been transported en bloc deep into the chamber from the entrance area . Furthermore , the orientation of the chamber passages and presence of drain systems likely precludes such a possibility . Thus , the fossils of H . naledi were probably deposited over an extended period of time during deposition and reworking of Units 2 and 3 , and before deposition of Flowstone 2 . Continual reworking of these units and the fossils they contain is consistent with taphonomic evidence ( element distribution; breakage patterns , mineral staining ) . The distribution of Unit 2 below the current entry shaft into the Dinaledi Chamber , and the orientation of Flowstones 1a–e that cover Unit 2 and slope into the chamber , together strongly suggest that the current entry shaft has always been the main entry point for sediment into the chamber . This also indicates that the fossils of H . naledi entered via this route . The Dinaledi collection displays taphonomic characteristics indicative of a depositional history that involved several stages of burial with surface modifications and breakage patterns consistent with repeated reworking of at least part of the assemblage within the confines of the Dinaledi Chamber , involving both biotic and abiotic agents ( Supplementary file 2 ) . The distribution of bone material and skeletal part representation indicative of limited winnowing ( Table 1 ) indicate that the fossils of H . naledi must have found their way into the chamber via a difficult route that precluded any other large vertebrates from finding a way in . The distribution of the fossils within reworked material derived from Unit 2 , as well-articulated remains in Unit 3 suggests that H . naledi fossils entered the chamber over an extended period of time; that is , not all remains were deposited at once . The presence of articulated , functional anatomical units that conventionally disassociate early in the decompositional sequence ( Figure 9B ) also suggests that bodies were fresh or in the early stages of decomposition when they entered the chamber and were encapsulated in the matrix ( Haglund , 1993; Lyman , 1994 ) . The lack of green fractures on any of the elements in the assemblage suggests that the bodies did not enter the chamber due to catastrophic accident such as falling into the chamber or due to flooding , or suffered trauma in any other way shortly before or after death . Limited weathering ( physical and chemical ) indicative of sub-aerial , sub-surface processes in a periodically wet or water-saturated , dark environment ( Figure 10 ) indicate that the bones were never exposed to the earth's surface and elements ( the sun and rain ) outside the cave ( Lyman and Fox , 1989; Backwell et al . , 2012; Junod and Pokines , 2013 ) . The preservation of primary and secondary context patterns of skeletal elements , the fragmented nature of many fossil bones , overprinting tide marks of minerals stains on skull fragments ( Figure 10F ) , invertebrate damage on bones by gastropods and beetles ( Figure 11 ) , and bone breakage patterns indicative of post-mortem ( dry bone ) or mechanically incompetent fractures ( Lyman and Fox , 1989 ) probably due to re-working of some of the fossil-bearing sediments ( Figure 10 ) all indicate that the fossils experienced significant periods of exposure and reworking after deposition ( Lyman , 1994; Manhein et al . , 2006 ) . This multi-staged internment history is best understood in terms of the continual reworking of Units 2 and 3 due to the gradual erosion of the cave floor as it slumps toward floor drains in the chamber ( the geological process explained above ) . The collection is also notable for the absence of large vertebrate remains , and the taphonomic processes that are not evidenced through surface and structural modifications and skeletal part representation most importantly there is no evidence for carnivore or rodent modification; no evidence of cut marks or burns and no ( geological or taphonomic ) traces of fluvial transportation or long-distance movement or traces of green bone fractures ( Haglund , 1992; Lyman , 1994; Fernández-Jalvo and Andrews , 2003; Pickering et al . , 2011a ) . Considering the geological and taphonomc context described above , the Dinaledi hominin site represents a depositional scenario that deviates from all other hominin localities in the region ( Brain , 1981; Partridge et al . , 2003; Dirks et al . , 2010; Pickering and Kramers , 2010; Pickering et al . , 2011a ) in two important ways . First , the Dinaledi Chamber appears to have never had unimpeded access , with fossil material deposited in muddy sediment in a geologically and geochemically isolated environment , and second the large vertebrate assemblage is comprised solely of hominin material ( Berger et al . , 2015 ) . Each of these points will be discussed in more detail below . Unlike other southern African cave sites where hidden shafts and sinkholes acted as death-traps to numerous species ( Brain , 1981; Partridge et al . , 2003; Dirks et al . , 2010; Pickering and Kramers , 2010; Pickering et al . , 2011a ) , there is no indication of a direct vertical passageway to surface into the Dinaledi Chamber . The chert layer forming the roof of the chamber is unbroken and coarse-grained clastic deposits transported by water action are absent , as are externally derived sediments . Instead , the Dinaledi Chamber represents an isolated sedimentary environment with fossils preserved in fine-grained , semi-consolidated mudstone breccia that is texturally and chemically distinct from fill in the upper chambers of the cave and that appears to have largely accumulated below the current entry shaft into the chamber . Flowstone formation continues today ( Flowstone 3 ) , changing the morphology of cave passages . This makes it possible that a more direct access-way or easier passage may have existed when hominins entered . A different entrance into the chamber may also explain the presence of rodent bone concentrations in Facies 1b . However , sedimentation patterns indicate that the accumulation of Unit 2 with fossils occurred below the current entry point into the chamber , and alternate routes did not involve vertical access shafts that connected directly to surface in either the Dinaledi Chamber or nearby Dragon's Back Chamber . The lack of other contemporaneous fauna in the assemblage , and complete lack of surface modifications by vertebrates ( carnivores , scavengers or rodents ) further suggests that the Dinaledi Chamber remained undisturbed by other animals , which could not reach the chamber . Taken together , this means H . naledi would have had to traverse difficult terrain to reach the Dinaledi Chamber , with the chamber and the proximal entrance to the chamber positioned in the dark zone . Mono-specific assemblages have been described from Tertiary and Mesozoic vertebrate fossil sites ( Kidwell et al . , 1986; Rogers , 1990; Behrensmeyer , 1991 ) , linked to catastrophic events ( Turnbull and Martill , 1988 ) . Among deposits of non H . sapiens hominins , where evidence of catastrophic events is lacking , mono-specific assemblages have been associated typically with deliberate cultural deposition or burial ( Arsuaga et al . , 1997; Carbonell and Mosquera , 2006; Gaudzinski , 2006 ) . Every previously known case of cultural deposition has been attributed to species of the genus Homo with cranial capacities near the modern human range . Unlike the Dinaledi assemblage , each of these hominin associated occurrences also contains at least some medium- to large-sized , non-hominin fauna ( Arsuaga et al . , 1997; Carbonell and Mosquera , 2006; Gaudzinski , 2006; Behrensmeyer , 2008 ) . There are a number of sites in Europe that represent accumulations of hominin fossils in caves that can be compared with the Dinaledi assemblage . The fossil hominins from Sima de los Huesos , Atapuerca , Spain , are presently believed to have accumulated within a single event ( Aranburu et al . , 2015 ) . The assemblage also includes carnivore remains , including the remains of ( an extensive number of ) cave bears and smaller carnivores . The Sima de los Huesos bone assemblage is highly fractured , with most fractures consistent with post-depositional breakage , but a small portion ( ∼4% ) consistent with peri-mortem breakage that might be explained as a result of fall-down a 13 m shaft into the cave ( Sala et al . , 2015a ) . Alternatively , there is some evidence that peri-mortem trauma may have resulted from lethal interpersonal violence ( Sala et al . , 2015b ) . Both hominin and carnivore bones in the Sima de los Huesos bear carnivore tooth marks at low frequency ( less than 4% ) , which has been interpreted as the sporadic action of carnivores trapped after falling into the chamber ( Sala et al . , 2014 ) . Cut marks have not been reported on the hominin remains ( Andrews and Jalvo , 1997 ) . Another example is presented by Level TD6-2 of Gran Dolina , Spain , which represents an accumulation of hominin and faunal bones in a ∼30 cm thick layer across an excavated extent of 7 m2 . The remains of at least six hominin individuals are present in this assemblage , with cut marks and evidence of intentional de-fleshing similar to associated fauna , as well as tooth marks from carnivores ( Saladié et al . , 2014 ) . The accumulation of hominin bone is interpreted as the result of cannibalism , with later carnivore scavenging at a smaller scale . A similar instance of cannibalism appears to be evidenced at El Sidron , Spain . Here , an almost exclusively hominin assemblage comprises a minimum of 13 individuals bearing cut marks , percussion pitting , and conchoidal scars typical of intentional processing of carcasses ( Rosas et al . , 2012 ) . The Krapina rock shelter in Croatia represents the earliest-excavated large hominin sample . Here remains of a minimum of 23 individual Hominins were recovered with extensive associated fauna from several stratigraphic layers , with most hominin material coming from a single layer , the ‘Homo Zonus’ . Breakage patterns of the Krapina hominin assemblage are mostly consistent with post-depositional fracturing , including excavation damage and prehistoric fracturing due to sedimentary loading ( Russell , 1987a ) . Many of the hominin bones bear cut marks , which have been variably interpreted as evidence of cannibalism or mortuary practices ( Russell , 1987b; Villa , 1992 ) . Pliocene and Early Pleistocene occurrences of single-species hominin bone deposits in Africa thus far contrast with these European cave sites . The AL 333 assemblage from Hadar , Ethiopia includes the fragmentary remains of some 17 hominin individuals and fauna , which appear to have been redeposited within a shallow streambed after possibly having been subject to predation ( Behrensmeyer , 2008 ) . Malapa , in the Cradle of Humankind , South Africa has been interpreted as a ‘death trap’ , with several partial hominin skeletons and associated fauna ( Dirks et al . , 2010 ) . The Malapa hominin and faunal assemblages lack evidence of carnivore activity , and include a disproportionate fraction of climbing species , suggesting that access to the cave was formerly limited and possibly hazardous for ungulates ( Val et al . , 2015 ) , but a broad array of faunal species have been recovered from the site . The Dinaledi assemblage shows no evidence of accumulation by either cannibalism or carnivore activity ( Supplementary file 2 ) , making it different from El Sidron , the Aurora Stratum of Gran Dolina , AL 333 , and Krapina . The Dinaledi situation shows similarities with the Sima de los Huesos assemblage in several respects . However , the Dinaledi hominins recovered to date are entirely free of cut marks , tooth marks , or peri-mortem fractures while each of these is present at low frequencies in the Sima de los Huesos sample . Few previously recognised scenarios operating in South African caves could have produced the selectivity for hominin remains as observed in the Dinaledi Chamber ( Brain , 1981; de Ruiter and Berger , 2000; Dirks et al . , 2010 ) ; a depositional situation common in no species other than modern humans . Below we will briefly discuss five alternative hypotheses for the accumulation of the hominin material in the Dinaledi Chamber , presented in order of what we consider to be reverse likelihood . The original mapping of the Rising Star cave system was conducted by a group of cavers from the Free cavers and CROSA caving societies in Johannesburg , who produced a base map that was used by our caving team when exploring the cave system . This map did not show the location of the Dinaledi Chamber . The Dinaledi Chamber was discovered as a result of detailed speleological surveys of a series of chambers known to contain macro-fossils including the Dragon's Back Chamber . Once discovered , the Dinaledi Chamber and access routes into the chamber were mapped in more detail using traditional mapping techniques involving tape-measures , compass-clinometers , and laser-inclinometers . Prior to our work there is no evidence that excavations for fossils were ever undertaken in the Rising Star cave . When the Dinaledi Chamber was first entered , it was clear that cavers had been in the chamber before , because they had re-arranged some of the bones ( Figure 6B ) , and had left behind several survey pegs . It is unknown to what degree earlier caving expeditions may have disturbed the original context of the fossils , or damaged some of the bone material , by walking over them , although it was noted that upon first entry much of the rubbly floor of the Dinaledi Chamber had remained undisturbed ( i . e . , not walked on before ) . None of the earlier expeditions into the Dinaledi Chamber have left a record of the chamber itself on any survey maps , or have mentioned the fossils in the chamber . Further investigation has established only one confirmed entry into the chamber in the early 1990s ( responsible for leaving the survey pegs ) . Given the complex nature of the entryway , it is unlikely that many , if any , expeditions would have entered the Dinaledi Chamber before or after that one confirmed entry . Since the discovery of the fossils , the entrances into the Dragon's Back Chamber have been locked off , and entry by recreational cavers is no longer physically possible . Analytical work was carried out at the SPECTRUM central analytical facility of the University of Johannesburg . Grain-size analysis and petrography , were carried out at James Cook University . Bulk chemical analyses of 4 samples were carried out by X-ray fluorescence ( XRF ) using a Philips PANalytical MagiX Pro instrument and standard borate fusion ( Figure 5B ) . The bulk mineralogical composition of four samples was determined by X-ray diffraction ( XRD ) using a Philips PANalytical X'Pert instrument . The High Score Plus software was used to refine the XRD diffractograms to identify the various mineral phases within the sediment samples . The cave sediment samples were mounted in 30 mm epoxy blocks and polished for textural analysis using scanning electron microscopy ( SEM; Figures 5 , 8 ) and electron microprobe analysis . SEM studies were carried out using a Tescan Vega3 scanning electron microscope equipped with an Oxford Instruments X-max 50 mm2 EDS detector . Quantitative spot chemical analyses were carried out using a CAMECA SX100 electron microprobe with 4 wavelength dispersive spectrometers and an EDS detector . The instrument was operated at 15 kV with a 5 µm beam width . As many phases within the fragments constituting the samples are not resolvable even at the μm scale , most of the analytical results represent mixtures , and information on their constituent minerals can be obtained from element correlations . The data are given in Supplementary file 1 . Relevant two-element plots are shown in Figure 8 . For the quartz-dominated sample DB-1 from the Dragon's Back Chamber , the non-quartz components were investigated . Following discovery of the fossil deposits , permission for excavation and recovery of fossils was issued by the South African Heritage Resource Agency , Department of Arts and Culture ( PermitID: 952 ) . Excavations are coordinated through the Evolutionary Studies Institute and the National Centre for Excellence in PalaeoSciences , at the University of the Witwatersrand , where all specimens are curated and stored in the fossil vault , and can be studied and accessed by researchers . Due to the difficult operating conditions , and limited physical access to the excavation chamber by the field crew , a unique system of recording and recovery was developed to assist excavation in the Dinaledi Chamber . The excavation and recording strategy focused on maximizing the retrieval of stratigraphic and spatial information during the recovery process by the excavators , and at the same time allowing reflexive supervision from senior investigators in a physically remote location . The technical strategy included the use of high-resolution 3D non-contact scanning , live digital video streaming of the excavation process to an above-ground supervisory team , as well as more conventional archaeological recording methods to facilitate post-excavation analysis . Although the recovery process utilised conventional archaeological methodologies , an explicitly forensic archaeological approach ( Hunter and Cox , 2005 ) was applied to the resolution of the Dinaledi Chamber . We adopted a multidisciplinary framework , bringing a wide range of expertise in buried environments , to ensure that the most complete range of evidence was collected . The excavation strategy proceeded on a single-context recording and recovery basis ( MOLAS , 1994 ) . The limits of the excavation were defined within the site , and sketch plans produced where appropriate . Excavation was undertaken with non-metallic tools in order to limit the possibility of recovery damage to highly fragile and friable skeletal material , and the exposure and recovery of any sediment surface was limited to the production of one or two hand-brush trays of spoil only . All excavated matrix sediment was double-bagged and recovered for analysis . In situ metric measurements were taken where specimens were highly fragmented or fragile . Taphonomic traces were noted where seen . All bone fragments were lifted from the surrounding sediment , double bagged , wrapped in bubble plastic , taped securely , boxed and lifted to the surface for cleaning , photographic recording , and consolidation or preservation of the remains where necessary . Recording pathways adopted are presented in Table 2 . 10 . 7554/eLife . 09561 . 016Table 2 . Recording activity pathway for Dinaledi Chamber excavationsDOI: http://dx . doi . org/10 . 7554/eLife . 09561 . 016Material or actionAssign and recordFormsIn cave: excavator and recorder Sediment contextContext number and attributesContext formPhotographyPhoto registerSketch plan and/or sectionSection logScanScan registerSampleSample register BoneElement numberSkeletal formSpatial propertiesArea sketchPhysical propertiesTaphonomyPeri-M traumaPost-M traumaMetric formRecoveryRecovery log Excavation scanRecord all observed contexts and elementsScan formAbove ground: recorder Context or elementRecord of assigned numberContext log Excavation scanRecord all observed contexts and elementsScan log Bone elementPreliminary ID , spatial attributes and locationElement log Recovered materialWhat is lifted and boxedRecovery log A photographic record of the excavation was taken , augmented with high-resolution , 3D surface scanning to document the location , orientation ( axial and surface ) of every bone . We used an Artec Eva ( http://www . artec3d . com/hardware/artec-eva/ ) , 3D white light scanner with the capacity to capture surface colour and texture ( surface resolution: 0 . 5 mm; 3D point accuracy: 0 . 1 mm ) in order to record and analyze the spatial position and orientation of bones within Unit 3 matrix material . The post-scan process was managed in Artec Studio 9 . Each scan sweep comprised a number of separate images compiled into a single layer . Each scan was then registered to acquire 3D triangulated points . Once registration was complete , the separate layers were manually aligned , by using a minimum of three reference points . The reference points are fixed survey markers within the Dinaledi Chamber , which were captured in each scan to provide optimal registration . Once alignment was complete , a global registration process allowed for the scan data to be merged accurately . This produced a 3D mesh representation of the scanned area . The 3D scan was then overlain with a photographic texture map , which was captured by the Artec Eva at the time of scanning . The entire H . naledi assemblage has been analysed at macroscopic scale , and a reduced sample of specimens from 11 individuals has been studied by microscope . Specimens were viewed at magnifications between 7 and 50 times using an Olympus SZX16 Zoom Stereo Microscope fitted with a DFC420 digital camera and equipped with Stream software . Modifications observed on the Rising Star material were compared with those recorded on bones in a reference collection held at the University of the Witwatersrand , which comprises bones and teeth modified by 58 known agents , including humans and a wide range of other vertebrates; 17 invertebrate taxa; and geological processes , including different forms of sedimentary abrasion , from trampling by large animals to the overall rounding and polishing caused by moving water . Characteristics and traces evaluated in the taphonomic analysis are listed in Table 3 ( after Pokines and Symes , 2013 ) . 10 . 7554/eLife . 09561 . 017Table 3 . Taphonomic recording criteria ( after Pokines and Symes , 2013 ) DOI: http://dx . doi . org/10 . 7554/eLife . 09561 . 017SignatureCharacters or taphonomic traces for recordingPreservationalGeneral state of remains ( excellent , good , fair , or poor ) Cortical erosion/exposure of cancellous boneCortical exfoliation ( bone loss in thin , spalling layers ) Postmortem breakagePerimortem breakage/fragmentation or traumaRounding ( erosion/tumbling in an abrasive environment ) DecalcifiedPostmortem cracking of desiccated tooth enamelIncidental surface striations/scratchesSoil surface exposureSurface cracking/longitudinal splitting from drying of waterlogged boneWeathering ( bleaching and cracking; sensu Behrensmeyer ) Mineral depositionCopper ( green ) , iron ( red ) , calcium ( white ) , manganese ( black ) , or other mineral oxide stainingVivianite formationConcretionWater staining ( presence of a water line from mineral deposits , colour differential line ) MechanicalExcavation damageMicro-abrasionSoil/burial substrateGeneral soil stainingWarping/flattening of elements ( especially the cranial vault ) Crushing/compaction from overburdenAdhering/infiltrating sedimentsFaunalAdhering faunaCarnivore puncture and gnawingGastric corrosion , winnowing , or windowing of boneRodent gnawingInvertebrate surface modification and damage In undertaking the analyses we have adopted a taphonomic approach derived from forensic practice in relation to the death process and burial environment with a degree of resolution usually reserved for medico-legal casework . The co-option of forensic taphonomy into palaeo-taphonomy provides a framework with which to investigate the decompositional and formational histories of ancient deposits by applying analyses of short-period events into the geological past ( Symes et al . , 2013 ) . The remains of H . naledi were analysed following the six stage classification system of Behrensmeyer ( 1978 ) based on faunal remains that weathered on the landscape in Kenya . Although this classification system has been developed for surface deposits , and is therefore less suitable for South African cave deposits , we have chosen to apply this recording system , because it allows us to address issues of possible bone transport into the Dinaledi Chamber from surface . Behrensmeyer ( 1978 ) described six weathering stages ( Stages 0–5 ) based on the progressive pattern of linear cracking and flaking of the cortical surface , followed by formation of a rough fibrous texture , and eventual collapse of bone integrity and structure ( i . e . , greasy bone with tissue is designated as Stage 0 , and splintered fragile bone falling apart in situ is designated as Stage 5 ) . The progressive weathering stages are broadly indicative of the period of surface exposure in sub-aerial conditions , with bone exposed to the elements , including sunlight .	Modern humans , or Homo sapiens , are now the only living species in their genus . But as recently as 20 , 000 years ago there were other species that belonged to the genus Homo . Together with modern humans , these extinct human species , our immediate ancestors and their close relatives are collectively referred to as ‘hominins’ . Now , Dirks et al . describe an unusual collection of hominin fossils that were found within the Dinaledi Chamber in the Rising Star cave system in South Africa . The fossils all belong to a newly discovered hominin species called Homo naledi , which is described in a related study by Berger et al . The unearthed fossils are the largest collection of hominin fossils from a single species ever to be discovered in Africa , and include the remains of at least 15 individuals and multiple examples of most of the bones in the skeleton . Dirks et al . explain that the assemblage from the Dinaledi Chamber is unusual because of the large number of fossils discovered so close together in a single chamber deep within the cave system . It is also unusual that no other large animal remains were found in the chamber , and that the bodies had not been damaged by scavengers or predators . The fossils were excavated from soft clay-rich sediments that had accumulated in the chamber over time; it also appears that the bodies were intact when they arrived in the chamber , and then started to decompose . Dirks et al . discuss a number of explanations as to how the remains came to rest in the Dinaledi Chamber , which range from whether Homo naledi lived in the caves to whether they were brought in by predators . Most of the evidence obtained so far is largely consistent with these bodies being deliberately disposed of in this single location by the same extinct hominin species . However , a number of other explanations cannot be completely ruled out and further investigation is now needed to uncover the series of events that resulted in this unique collection of hominin fossils .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology" ]	2015	Geological and taphonomic context for the new hominin species Homo naledi from the Dinaledi Chamber, South Africa
Males and females typically pursue divergent reproductive strategies and accordingly require different dietary compositions to maximise their fitness . Here we move from identifying sex-specific optimal diets to understanding the molecular mechanisms that underlie male and female responses to dietary variation in Drosophila melanogaster . We examine male and female gene expression on male-optimal ( carbohydrate-rich ) and female-optimal ( protein-rich ) diets . We find that the sexes share a large core of metabolic genes that are concordantly regulated in response to dietary composition . However , we also observe smaller sets of genes with divergent and opposing regulation , most notably in reproductive genes which are over-expressed on each sex's optimal diet . Our results suggest that nutrient sensing output emanating from a shared metabolic machinery are reversed in males and females , leading to opposing diet-dependent regulation of reproduction in males and females . Further analysis and experiments suggest that this reverse regulation occurs within the IIS/TOR network . Sex differences in life history , behaviour and physiology are pervasive in nature . These differences arise mainly from the divergent reproductive strategies between the sexes that are rooted in anisogamy ( Chapman , 2006 ) . Typically , males produce large numbers of small , cheap gametes and evolve traits that facilitate the acquisition of mates and the increase of fertilisation success . Females , on the other hand , produce fewer , energetically costlier gametes and tend to evolve traits that optimise rates of converting resources into offspring ( Trivers and Campbell , 1972 ) . Given these fundamental differences between male and female reproductive investments , one of the key areas of divergence between the sexes concerns physiology , metabolism and responses to diet ( Jensen et al . , 2015 ) . Studies in insect species ( Jensen et al . , 2015; Reddiex et al . , 2013; Maklakov et al . , 2008; Maklakov et al . , 2009; Simpson and Raubenheimer , 2011 ) have shown that the two sexes require different diets to maximise fitness . Female fitness is typically maximised on a high concentration of protein , which fulfils the demands of producing and provisioning eggs . Males , in contrast , achieve optimal fitness with a diet consisting of more carbohydrate , which can fuel activities such as locating and attracting mates . Work on nutritional choices has shown that individuals tailor their diet in line with their physiological needs . In insects , females overall prefer diets with higher protein content , whereas males chose a more carbohydrate-rich diet ( Lee et al . , 2008; Corrales-Carvajal et al . , 2016 ) . These choices are further adapted to reflect the individual's current condition and reproductive investment ( Corrales-Carvajal et al . , 2016; Ribeiro and Dickson , 2010 ) . For example , Camus et al . ( 2018 ) found that the female preference for protein in fruit flies was significantly higher in mated females ( who require resources to produce eggs ) than virgins , while the preferences of males ( who start producing sperm before reaching sexual maturity ) did not significantly differ between mated and virgin flies . But individuals not only choose diets to suit their needs where possible , they also adapt their physiology and reproductive investment in response to the quality and quantity of nutrition available . This has been studied extensively using experiments that either alter the macronutrients composition ( carbohydrates vs . protein ) of the diet while keeping the overall caloric intent constant , or by manipulating the overall nutrient content of the food—dietary restriction ( DR ) . These studies have shown that a wide range of life history traits respond to changes in both the composition of the food ( Simpson and Raubenheimer , 2011; Moatt et al . , 2019; Solon-Biet et al . , 2014 ) and the quantity of nutrients supplied ( Piper et al . , 2005; Regan et al . , 2016; Piper and Partridge , 2018 ) . For example , DR typically causes an extension of lifespan at the cost of reduced reproduction ( Partridge et al . , 2005 ) , and a similar response can be triggered by a shift from protein to carbohydrates in the diet ( Solon-Biet et al . , 2014 ) . Although most studies manipulating diet have concentrated on females only , those including both sexes suggest that DR responses are broadly similar in males and females—despite their large differences in optimal diet . In fruit flies , DR extends lifespan in both sexes ( Magwere et al . , 2004; Zajitschek et al . , 2014; Zajitschek et al . , 2013 ) , even though the observed increase in longevity appears smaller in males than females and the degree of DR that maximises lifespan can differ between the sexes ( Magwere et al . , 2004 ) . Qualitatively similar results have been obtained when manipulating the macronutrient composition of the diet . Studying field crickets , ( Maklakov et al . , 2008 ) found that shifting the dietary balance away from protein and towards carbohydrates increased lifespan in both sexes , even though the effect of nutrients on reproductive investment differed between the sexes ( Maklakov et al . , 2008 ) . These quantitative sex differences in dietary lifespan effects can at least in part be attributed to sex-biased responses in individual tissues . Thus , Regan and co-workers showed that D . melanogaster males in which the gut had been genetically feminised had DR responses more similar to those of females ( Regan et al . , 2016 ) . The contrast between large differences in optimal diet but similar responses to diet manipulation raises the question of how males and females differ in their diet-dependent regulation of metabolism and reproductive allocation . Due to the predominant focus on female responses to nutrition , we currently know relatively little about the degree to which regulation is shared or differs between the sexes ( Hoedjes et al . , 2017 ) , in particular at the molecular level . Work in females has shown that nutrient-sensing pathways play a key role in the observed DR phenotype ( Clancy et al . , 2002; Slack et al . , 2011; Zandveld et al . , 2017; Emran et al . , 2014; Bjedov et al . , 2010 ) . Specifically , two evolutionarily conserved signalling pathways—insulin/insulin-like growth factor 1 ( IIS ) and Target of Rapamycin ( TOR ) —are thought to regulate longevity in a diet-dependent way ( Hoedjes et al . , 2017; Alic and Partridge , 2011; Gallinetti et al . , 2013 ) . Recent transcriptomic work in female D . melanogaster has further shown that DR and rapamycin treatment ( which inhibits TORC1 activity ) elicit similar changes in gene expression ( Dobson et al . , 2018 ) . Both responses share a significant number of overlapping genes , and are mediated by transcription factors in the GATA family; in line with the involvement of these regulators in amino acid signalling and lifespan modulation across eukaryotes ( Dobson et al . , 2018 ) . While these data are starting to paint an increasingly detailed picture of nutrient-dependent regulation in females , the lack of information on males severely limits our understanding of how diet shapes metabolism and life history decisions . For example , it is not clear to which degree the regulation identified in females reflects their specific dietary requirements and physiology . Further , we cannot tell whether males and females differ in their general metabolism and its nutrient-dependent regulation , or whether diet responses are largely shared , and sex-specific effects limited to the regulation of reproductive investment . Interestingly , perturbing the IIS/TOR network in virgin flies has been shown to elicit sex-specific expression changes in males and females ( Graze et al . , 2018 ) , but the link to nutrition and the effect on reproductive investment remains unclear . Addressing these questions is important because they have implications for the degree to which male and female physiology and its regulation are uncoupled and able to independently evolve . Thus , a shared physiology and diet-dependent regulation of metabolism across the two sexes would constrain the degree to which each sex is able to independently optimise its life-history decisions in response to the current nutritional environment . Here , we are starting to address these fundamental questions by investigating male and female diet responses in gene expression . We study this in the context of shifts of nutritional composition ( amino acid-to-carbohydrate ratio ) between the male and female optima . This manipulation is more subtle than classic dietary restriction , given we are changing the quality of the diet whilst keeping caloric intake the same . This approach allows us to contrast , for each sex , an optimal and a non-optimal condition , as well as , across sexes , a more amino acid- and a more carbohydrate-rich diet . Furthermore , we can compare the female responses to a smaller , more quantitative shift in diet composition to existing data on responses to DR . We use nutritional geometry techniques to establish the male and female optimal diets in an outbred D . melanogaster population and then examine the transcriptomic responses of both sexes to the male-optimal diet ( protein-to-carbohydrate ratio 1:4 ) and the female-optimal diet ( 2:1 ) . We then assess the degree to which expression changes from male- to female-optimal diets are shared or divergent between the sexes , and how this relates to the function and regulation of genes . Our analysis reveals that most of the core metabolic gene network is shared between the sexes , responding to diet changes in a sexually concordant manner . However , we also find smaller sets of genes where male and female responses diverge , either by being restricted to one sex or by males and females showing opposing diet-induced expression changes and observe that sex-limited reproductive genes are generally up-regulated on each sex's optimal diet . These results indicate that while males and females share a common , and concordantly regulated metabolic machinery , the sexes diverge in how nutritional information is translated into reproductive regulation . Further results allow us to link this divergent regulation to the Tor pathway . First , we find that our genes with diet-dependent regulation overlap with genes previously associated with responses to DR , rapamycin treatment and perturbation of the IIS/TOR network and known targets of the TOR pathway . Second , we can show experimentally that inhibiting TORC1 with rapamycin has a disproportionately negative effect on reproductive fitness on each sex's optimal diet . These results are compatible with the shared nutrient-sensing signal being inverted in males and females to produce diametrically opposed Tor-dependent regulation of reproduction in the two sexes . We first examined the effects of diet composition on male and female fitness . We recovered previous results , finding that males and females differ significantly in their dietary requirements to maximise fitness ( parametric bootstrap analysis: PB-stat = 78 . 002 , p<0 . 001 ) . For females , the number of eggs produced differed significantly between diets ( Analysis of Variance , F7 = 41 . 4703 , p<0 . 001 ) and was maximised on the 2:1 ( P:C ) nutritional rail ( Figure 1 and Figure 1—figure supplement 3 ) . Male competitive fertilisation success also differed between diets ( F7 = 3 . 5927 , p<0 . 001 ) , but peaked at the 1:4 ratio ( Figure 1 and Figure 1—figure supplement 4 ) . Dietary choices also differed between the sexes ( F2 = 27 . 826 , p<0 . 001 ) . The choices of both sexes closely matched their previously established optimal composition , with females choosing to consume a more protein-rich diet than males ( Figure 1 ) . We also found that females , on average , tend to consume more liquid food than males but this relationship depends on the diet ( sex ×diet: F7 = 5 . 66 , p<0 . 001 , Figure 1—figure supplement 2 ) . We measured gene expression in males and females maintained on food of either the female-optimal ( 2:1 ) or male-optimal ( 1:4 ) protein-to-carbohydrate ratio . We separately analysed transcriptomic responses in genes that were expressed in both males and females ( hereafter 'shared genes' , N = 8888 ) and those that showed sex-limited expression ( Nmale-limited = 1879 and Nfemale-limited = 165 , see Supplementary file 2 for full gene lists ) . For each shared gene , we tested for the effect of sex , diet and the sex-by-diet interaction on expression level . As expected , we found evidence for sex-differences in expression for a large number of genes ( a total of 8318 genes with significant sex effect ) . In addition , we found large-scale transcriptomic responses to diet ( 806 genes with significant diet effect ) . Despite the large differences between male and female dietary requirements and food choices , the largest part of the transcriptional responses to diet is shared between the sexes ( significant diet effect but no interaction , category 'D' in Table 1 , 639 genes ) . Here , males and females show parallel shifts in expression ( although in most cases from a sexually dimorphic baseline expression ) when reared on high-carbohydrate vs . high-protein food , and fold-changes between the two diets are strongly positively correlated between males and females ( Figure 2; r = 0 . 76 , p<0 . 001 ) . In addition to these sexually concordant responses , however , we also find a significant number of genes where the sexes show different responses to diet shifts ( significant sex-by-diet interaction ) . For some of these genes , male and female expression change in opposing directions ( category 'D × S' in Table 1 , 51 genes ) . Thus , genes that are more highly expressed on a protein-rich diet in one sex are more lowly expressed on that diet in the other sex , resulting in negatively correlated fold-changes in the two sexes ( Figure 2; r = −0 . 75 , p<0 . 001 ) . For another , larger group of genes ( category 'D+D × S' , 116 genes ) , both sexes tend to show expression shifts in the same direction ( significant diet effect ) but differ in the magnitude of their responses ( significant interaction term ) . These genes typically show a large expression response in one sex , but only a small or no response in the other sex , with overall a lower correlation of fold changes across sexes ( r = 0 . 53 , p<0 . 001 ) . For the most part , the dominant expression change occurs in females , but there is a small number of genes where only male expression responds to diet ( Figure 2 ) . We next analysed diet responses in genes with sex-limited expression . Similar to shared genes , we also observed significant expression changes in response to diet ( Table 2 ) . Thus , 56 out of 165 female-limited genes showed significant expression change between carbohydrate- and protein-rich diets . The majority of these ( 50 genes ) had higher expression on the protein-rich diet preferred by females , while only a small number ( six genes ) had higher expression on the less beneficial carbohydrate-rich diet ( Figure 3 ) . In males , 30 out of the 1879 genes with male-limited expression showed significant diet responses . All of these had higher expression in the males' preferred carbohydrate-rich diet , compared to the less beneficial protein-rich media ( Figure 3 ) . Taken together , these results show that both sexes respond to their nutritional environment by upregulating sex-limited genes on their respective optimal diets . We used several approaches to investigate the functions of the genes showing diet responses . First , we performed Gene Ontology ( GO ) enrichment analyses for the shared genes of the three categories ( D , D × S , D+D × S ) defined above . We found distinct and significant enrichment in each class , with a predominance of GO terms relating to neuronal and metabolic biological processes ( Figure 4 ) . Second , we took a more targeted approach and analysed male and female expression changes for genes with specific GO annotations . With this analysis we aimed to assess how metabolic genes responded to diet manipulation , compared to the rest of the genome . For this , we fist created a ‘baseline’ of gene expression by extracting a list of genes that fall under the parent term ‘Biological Process’ ( GO:0008150 ) . From that list , we then removed the genes in the offspring category ‘Metabolic Process’ to create a set of genes performing biological functions unrelated to metabolism . We then compared this baseline to genes that fell within the following GO categories: ‘Metabolic Process’ ( GO: 0008152 ) , ‘Glycolysis’ ( GO:0006096 ) and ‘TCA cycle’ ( GO:0006099 ) . The latter two were chosen as core processes in carbohydrate and protein metabolism . For the sets of genes in each of these categories that showed shared expression across the sexes , we found positive correlations between male and female fold changes between the two diet treatments ( RMP = 0 . 35 , RGLY = 0 . 74 , RTCA = 0 . 6 , Figure 5A ) . These correlations were significantly more positive than the ( also slightly positive ) correlation observed in the non-metabolic baseline gene set , despite the fact that correlations for the small Glycolysis and TCA gene sets have wide confidence intervals ( Figure 5B ) . This indicates that , even though there is a general shared response to diet between males and females , male and female responses are more similar in genes involved in core metabolic processes than the rest of the genome . For the sex-limited differentially expressed genes , we unsurprisingly found an enrichment of GO terms involved in reproduction ( Figure 6 ) . In females , differentially expressed genes were enriched for functions associated with egg production ( chorion-containing eggshell formation ) , but also hormonal control ( ecdysone biosynthetic pathway and hormone synthetic pathway ) . Male differentially expressed genes were enriched for sperm function ( sperm competition ) . Since responses in both sexes consisted predominantly of up-regulation of genes under their respective optimal diets , these results show that for both males and females , the expression of reproductive genes is increased in the condition that maximises the fitness of that sex . In order to infer the regulators that drive the observed expression responses to diet , we searched for enrichment of transcription factor binding motifs upstream of the genes in the three categories . Our analyses revealed significant enrichment of regulatory motifs in each group ( see Supplementary file 3 for a full list ) . Genes that showed only significant diet responses ( concordant response between the sexes , D ) , presented an overrepresentation of binding motifs for the transcription factors CrebB and lola . Genes that showed opposing changes in males and female ( D × S ) were enriched for motifs for vri and GATA transcription factors ( grn , pnr , srp , GATAd , GATAe ) . Finally , genes that showed diet responses largely restricted to one sex ( D+D × S ) were enriched primarily for GATA motifs , irrespectively of whether the response occurred predominantly in females or predominantly in males . Female-specific genes were mostly enriched for the transcription factors Blimp-1 , slbo and Dfd , whereas male-specific genes were enriched for regulation by pan and Sox . We used comparisons to previously published transcriptomic datasets to relate the responses to shifts in diet quality observed here to those triggered by dietary restriction and perturbations of nutrient signalling . First , we compared genes in our three categories of diet-dependent regulation overlapped significantly with sets of genes that change expression in response to dietary restriction and rapamycin in females , analysed separately for brain , thorax , gut , and fat body ( Dobson et al . , 2018 ) . We found significant overlap in the majority of comparisons made ( Table 3A and B ) . Non-significant results were only obtained for some comparisons involving the list of genes in the D × S category , where males and females show opposing responses to diet . While this might reflect biological reality , it has to be noted that the numbers of genes—and hence statistical power to detect overlap—are smallest in the D × S category . Overall , the results of these comparisons demonstrate that transcriptional responses to the more subtle changes in dietary composition that we apply here generally mirror those that have previously been observed under dietary restriction . We then compared our gene categories with a dataset from heads of virgin males and females in which IIS/TOR signalling had been perturbed by expressing a dominant-negative allele of the insulin receptor InRDN ( Graze et al . , 2018 ) . Reanalysing this dataset ( see Materials and methods ) we obtained a list of genes that were altered by an IIS/TOR perturbation across both sexes ( N = 5200 genes ) similar to the results obtained in the original paper . However , subjecting the data to an analysis analogous to that we performed on our own , we further found that IIS/TOR perturbation causes expression changes similar to those observed for our diet treatments . Thus , a large number of genes show concordant responses to altered insulin signalling in males ( significant InR effect ) and females , while a second set shows opposing responses ( InR-by-sex interaction , InR×S ) and a third shows largely sex-specific responses ( InR+InR×S ) ( Figure 2—figure supplement 1 , Supplementary file 4 ) . Furthermore , we detect parallelism in the effects of diet manipulation and InR perturbation on several levels . At the most basic level , the genes that are significantly affected by IIS/TOR perturbation overlap significantly with the genes that are significantly affected by diet quality ( 489 genes observed , 351 expected , 39% excess , Fisher's exact test , p < 0 . 001 ) . Second , genes that show a significant diet effect ( 'D' ) are more likely to also show a significant effect of InR perturbation ( 'InR' ) ( 436 genes with both terms significant , 37% excess , Fisher's exact test , p < 0 . 001 ) and genes with a significant diet-by-sex interaction are more likely to also show a significant InR-by-sex interaction ( 51 genes , 108% excess , Fisher's exact test , p < 0 . 001 ) . Third , a full comparison based on a contingency table containing all possible combinations of classes also showed a significant correspondence ( Chi-squared test , Χ92 = 248 . 53 , p < 0 . 001 ) , with excess overlap in most combinations of classes as well in genes that are classified in neither analyses ( Supplementary file 1 — Table 5 ) . And finally , fold changes in male and female gene expression in response to IIS/TOR perturbation correlate positively with those in response to diet manipulations ( see Figure 2—figure supplement 2 , Supplementary file 4 ) , despite the fact that the two datasets analyse different tissues ( head vs . whole body ) . These results indicate that manipulating diet quality and manipulating IIS/TOR signalling produces parallel and overlapping expression responses . We also investigated the overlap between our diet-responsive genes and genes that have been identified as TORC1-regulated due to their dependence on REPTOR and REPTOR-BP ( Tiebe et al . , 2015 ) . While based on expression in S2 cells only , this to our knowledge is the best characterised set of TOR-responsive genes . In line with the similarity between expression responses to diet and IIS/TOR-manipulation described above , we find significant overlap between our gene categories and genes with REPTOR- or REPTOR-BP–dependent expression , specifically in our category that responds to diet ( 'D' , 28 genes ) and our sex-biased category ( 'D+D × S' , nine genes , Table 3C , Supplementary file 4 ) . The overlap with previously described responses raises the potential for the IIS/TOR network , and specifically TORC1 , mediating the diet-dependent phenotypes that we observe here . This appears plausible for the modulation of female fecundity in response to diet , where the role of TORC1 is well established , but has not been assessed in males . We therefore directly tested the phenotypic effect of varying doses of rapamycin and its interaction with diet , on our proxies for male and female fitness . Our experiment showed that , across the two sexes , rapamycin leads to a reduction in reproductive output ( rapamycin effect: p<0 . 001 , Figure 7 and Figure 7—figure supplement 1 , Supplementary file 1 - Table 4 ) . More importantly , however , we also found a significant interaction between diet and rapamycin treatment that was shared across males and females , where rapamycin lead to a larger reduction in reproductive output on each sex's optimal diet ( sex ×rapamycin: p=0 . 001 ) . Finally , our experiment revealed possible quantitative differences between the sexes in the effect of rapamycin on reproduction ( sex ×rapamycin × diet: p=0 . 068 ) ; while the effect of the treatment in females correlated roughly with the dose administered , males showed a threshold response where all rapamycin levels in the optimal diet resulted in a reduction in reproductive output to the level observed on the non-optimal diet . Our analyses demonstrated the existence of a core metabolic transcriptome that shows sexually concordant regulation in response to diet . Overall , expression fold changes from carbohydrate- to protein-rich food among metabolic genes are positively correlated between the sexes , and significantly more so than for the transcriptomic background . This indicates that gene expression in males and females responds generally similarly to changes in dietary composition . In line with this interpretation , the large majority of genes with diet-dependent expression show significant changes only in response to diet , independently of sex ( 639 out of 806 genes , 79% ) . Functionally , genes in this core metabolic transcriptome are enriched for carboxylic acid metabolism and neurological biological processes . Carboxylic acid metabolism is an integral part of both protein and carbohydrate processing—for instance , part of the components of amino acids are carboxylic acid sidechains . The prominence of neurological biological processes , on the other hand , supports the notion of a neural gut-brain connection that is conserved evolutionarily ( Kaelberer and Bohórquez , 2018 ) and shared between the sexes . Specifically , the sensory mechanisms in the gastrointestinal tract convey information about the nutritional status to regulate satiety ( and thereby feeding behaviour ) , metabolism , and digestion ( Kaelberer et al . , 2018 ) in a way that is similar between males and females . We were also able to infer key regulators of sexually concordant , diet-dependent gene expression , using motif enrichment tools . Upstream regions of genes with sexually concordant diet responses were enriched for motifs of two main transcription factors CreB and lola transcription factors . CrebB is involved in diurnal rhythms and memory formation ( Bittinger et al . , 2004; Kogan et al . , 1997 ) , but also in energy homeostasis and starvation resistance , mediated by insulin signalling ( Wang et al . , 2008 ) . The lola transcription factor , on the other hand , is mainly involved in axon guidance in Drosophila ( Horiuchi et al . , 2003; Goeke et al . , 2003 ) . But interestingly , some protein isoforms have also been associated with octopamine synthesis pathways which are essential for nutrient sensing ( Dinges et al . , 2017 ) . Besides the large , shared core metabolic transcriptome , we also identified smaller groups of genes with sex-specific expression responses to diet . A first group showed opposing diet responses in males and female ( D × S , 51 out of 806 genes , 6 . 3% ) . These genes are enriched for transport functions and synapse assembly/organisation . One of our candidate antagonistic genes is fit ( female-specific independent of transformer ) . Known to be sexually dimorphic in expression , fit has been found to be rapidly upregulated in male heads during the process of male courtship and mating , along with another antagonistic candidate Odorant binding protein 99b , Obp99B ( Ellis and Carney , 2010; Carney , 2007 ) . Interestingly , fit has also been implicated in protein satiety in a sex-specific manner ( Sun et al . , 2017 ) . Following the ingestion of protein-rich food , fit expression increases in both sexes ( although more so in females than males ) , but only supresses protein appetite in females ( Sun et al . , 2017 ) . Both fit and Obp99B were found to be significantly altered in a sex-specific way when flies were starved , further cementing their role in nutrient response ( Fujikawa et al . , 2009 ) . Together with previous work , our results therefore cement the tight link between nutritional sensory mechanisms and reproduction , however this response is sex-specific . Another group of genes showed mostly responses in one sex ( D+D × S , 116 genes , 14 . 4% ) . Most of the genes observed in this category show expression changes in females ( with little change in male expression levels ) and are mainly involved in carbohydrate metabolism and female receptivity . One notable gene in this category is the transcription factor doublesex , which plays a key role in sexual differentiation and the regulation of sex-specific behavioural traits ( Shirangi et al . , 2006 ) . Expression levels of this gene are higher in females that are fed a high-protein diet ( unless the difference in dsx mRNA levels is due to growth in a sexually dimorphic , and hence dsx-expressing , tissue type ) . Of interest among the few genes with male-limited diet response ( Figure 2 ) is Adenosylhomosysteinase ( Ahcy ) , which we find males to express at lower levels on the carbohydrate-rich ( optimal ) diet . Ahcy is involved in methionine metabolism and has been linked to male lifespan regulation . Ahcy knock-outs were shorter lived , while knock-outs for two putative Ahcy-repressors extended male life- and health-span ( Parkhitko et al . , 2016 ) . These effects are in line with the under-expression we observe on high carbohydrate , under the assumption that greater investment in current reproduction is associated with decreased lifespan ( which may not generally hold in the context of nutrient manipulation; Jensen et al . , 2015 ) . Both the genes with opposing ( D × S ) and those with sex-limited diet-dependent regulation ( D+D × S ) show significant enrichment for GATA transcription factors . This class of transcription factors has been previously implicated in female nutritional and reproductive control . For example , the ovary-specific dGATAb binds upstream of both yolk protein genes Yp1 and Yp2 ( Lossky and Wensink , 1995 ) . GATA-related motifs have also previously been shown to be enriched in genes showing differential expression in response to DR and rapamycin treatment in female flies ( Dobson et al . , 2018 ) . The shared regulation is further supported by the fact that the diet-responsive genes we identify here also overlap significantly with those previously inferred to respond to DR- and rapamycin-treatment . These results suggest that changing the quality of the diet elicits a similar response as changing the quantity via protein dilution . This may not be surprising , if DR is considered a response mainly to the quantity of protein ingested ( Lee et al . , 2008; Grandison et al . , 2009 ) , and fits with previous work that found the ratio of macronutrients—not caloric intake—to be the main determinant of healthy ageing in mice ( Solon-Biet et al . , 2014 ) . However , the overlap highlights that DR-phenotypes are not an all-or-nothing response but instead are part of a continuum of life history adjustments in response to how suitable the dietary environment is for current reproduction . We also found diet responses in reproductive genes that are exclusively expressed in either males or females . Regulation largely reflects diet-dependent reproductive investment , with most genes being more highly expressed on a sex's optimal diet with lower expression on the suboptimal diet . In females , a significant number of these genes are involved in egg production and thus linked to diet-dependent reproductive investment ( Trivers and Campbell , 1972 ) . Also among the genes is insulin-like peptide-7 ( dILP-7 ) , one of a family of peptides known to having the functional as hormones and neuropeptides ( Sisodia and Singh , 2012 ) involved in nutrient foraging control ( Shim et al . , 2013 ) . More specifically , dILP-7 is expressed in neurons that play an active role in female fertility . These neurons have been linked with the egg-laying decision process ( Yang et al . , 2008; Lihoreau et al . , 2016 ) and dILP-7 is among a number of genes show sexually dimorphic expression in these neuronal cells ( Castellanos et al . , 2013 ) . Interestingly , IIS/TOR perturbation also results in sex-specific changes in dILP peptides ( dILP2 , 3 , 5 and 6 ) in the head ( Graze et al . , 2018 ) ( where dILP7 is not expressed; Nässel and Broeck , 2016 ) . Mirroring expression responses in females , we also find higher expression of reproductive genes on the optimal diet in males . This is surprising—based on the view that male fitness is limited by the acquisition of mates and the supposedly low investment required for sperm production ( Trivers and Campbell , 1972 ) , one could expect that males do not modulate their reproductive investment in response to the nutritional environment but remain primed to maximally exploit any mating opportunity . Assuming that expression of these genes reflects reproductive investment , the fact that they do respond to the nutritional environment suggests that male reproductive strategies are maybe more subtle , and their investment more costly , than previously appreciated . This is plausible , as work on other insects has shown that the production of high quality sperm is costly ( Bunning et al . , 2015 ) ( but courtship activity does not appear to carry a significant cost , at least in fruit flies; ( Flintham et al . , 2018 ) . Superficially , it may seem obvious that male and female reproductive genes are upregulated on each sex's respective optimal diet . In the presence of a largely shared and concordantly regulated metabolic machinery , however , this pattern implies that the output of nutrient sensing pathways is used in different , and potentially inversed ways in males and females . While our analyses do not allow us to identify the exact point of reversal within the regulatory hierarchy , our data provide some interesting insights . First , it is noteworthy that GATA transcription factors are inferred to be regulating genes that show a wide range of expression patterns , being enriched among genes with opposing expression changes in males and females ( the D × S set ) , as well as those that show largely sex-limited responses ( D+D × S ) . This could imply that the main role played by these factors is to convey information about the metabolic and nutritional state of the animal ( similar to homeotic genes in development ) , which is then incorporated combinatorially with additional factors to give rise to the sex- and diet-specific expression patterns that we observe . Second , several lines of bioinformatic evidence suggest that the expression changes that we describe here are at least in part regulated by IIS/TOR signalling . Thus , the genes that we find to respond to diet manipulation significantly overlap with genes affected by manipulation of IIS/TOR signalling as described by Graze et al . ( 2018 ) , a dataset that our reanalysis reveals to show a similar structure of genes with sexually concordant , sexually opposing and sex-biased expression changes . This pattern and the overlap with our data is all the more noteworthy as Graze et al . assessed the effect of IIS/TOR perturbation in virgin flies , where males and females have more similar dietary requirements , and hence presumably more similar physiological states , than in mated flies ( Camus et al . , 2018 ) . In addition to showing parallels with IIS/TOR-dependent expression , our diet-dependent genes also significantly overlap with the arguably best-defined set of TORC1-dependent genes currently available ( Tiebe et al . , 2015 ) . These results suggest that diet-dependent expression responses , and their sex-specific differences , are mediated by IIS and the TOR pathway . This conclusion is corroborated by the results of our experiment combining diet manipulation with rapamycin treatments , which are consistent with TORC1-dependent upregulation of reproduction on optimal diets in both sexes . Here we find that while rapamycin generally lowers reproductive output , this effect is more pronounced on the respective optimal diet of each sex . This is expected in females , where a large body of work implicates the IIS/TOR network in life-history shifts between reproduction and longevity ( Wullschleger et al . , 2006 ) . Accordingly , a nutritionally favourable environment should lead to increased TORC1 activity and elevated reproductive output . What our data show , however , is that a parallel effect of increased reproduction on the optimal diet is detectable in males , even though the composition of that diet is the one that is unfavourable in females , leading to low TORC1 and reduced reproduction . Across the sexes , TORC1 activity would thus not reflect a specific dietary composition but a measure of nutritional optimality and regulate reproductive investment accordingly . We note that , while tantalising , these inferences will require further careful validation . Due to the focus on females , diet-dependent regulation of male reproduction has been little explored . Knock-down of Tor and raptor in males has been found to result in an accumulation of germline stem cells , combined with deficient differentiation ( Liu et al . , 2016 ) . Future work will need to assess the effect of these changes on male reproductive output and , more importantly , whether and how the signal of the nutrient sensing mechanisms that feed into the Tor pathway are modulated in a sex-specific way . Independently of how the regulatory reversal is achieved mechanistically , our data also suggest that the relationship between the composition of the diet consumed and reproductive output does not merely reflect the passive effect of metabolic conversion rates from nutritional components to gametes and energy but is at least in part the result if an active regulation of immediate reproductive investment . This has important implications for our interpretation of variation in diet-specific reproductive success , which has been documented in the population studied here ( Camus et al . , 2017 ) . Thus , variation between genotypes in the dietary composition that maximises , for example , male reproductive fitness is therefore probably at least partly caused by genetic variation in how nutrients are sensed or how this sensory output is used to regulate reproductive investment . Studying this variation in more detail will provide a fruitful avenue to better understand the regulatory mechanisms involved , as well as the selective forces that shape variation in its components . We used the D . melanogaster laboratory population LHM for our experiments . This has been sustained as a large outbred population for over 400 non-overlapping generations ( Chippindale et al . , 2001; Rice , 1996 ) , maintained on a strict 14 day regime , with constant densities at larval ( ~175 larvae per vial ) and adult ( 56 vials of 16 male and 16 females ) stages . All LHM flies were reared at 25°C , under a 12 hr:12 hr light:dark regime , on cornmeal-molasses-yeast-agar food medium . We used a modified liquid version of the synthetic diet described in Piper et al . ( 2014 ) , that is prepared entirely from purified components to enable precise control over nutritional value ( see Supplementary file 1 Tables 1-3 ) . Previous studies have used diets based on natural components , typically sugar as the carbon source and live or killed yeast as the protein source ( Piper and Partridge , 2007 ) . Such diets offer only approximate control over their composition , because the yeast-based protein component also contains carbohydrates and is required to provide other essential elements ( vitamins , minerals , cholesterol , etc . ) that vary in relative abundance . As a consequence , phenotypic responses to such diets cannot be straightforwardly interpreted in a carbohydrate-to-protein framework as they are confounded by responses to other dietary components . Our use of a holidic diet completely eliminates these problems without causing any apparent stress in the flies ( Piper et al . , 2014 ) . Eight isocaloric artificial liquid diets were made that varied in the ratio of protein ( P , incorporated as individual amino acids ) and carbohydrate ( C , supplied as sucrose ) , while all other nutritional components were provided in fixed concentrations . Nutrient ratios used were [P:C] – 4:1 , 2:1 , 1:1 , 1:2 , 1:4 , 1:8 , 1:16 and 1:32 , with the final concentration of each diet ( sum of sugar and amino acids ) being 32 . 5 g/L . These ratios span the P:C ratio of the molasses medium on which the LHM population is maintained . Based on the media recipe used in our laboratory and the approximate protein and carbohydrate content of the ingredients , we estimate our standard food to have a P:C ratio of about 1:8 . The diets in our experiments on the edges of our nutritional space , with the highest carbohydrate- or protein-bias , can thus considered to be ‘extreme’ in comparison to our standard laboratory media—even taking into account the fact that ratios in synthetic and organic diets may not be directly comparable , as nutrients in synthetic food appear to be more readily accessible ( Piper et al . , 2014 ) . For diet preference assay we used two diets; protein and carbohydrate . Each diet contained all nutritional components ( vitamins , minerals , lipids ) at equal concentration , with the protein diet containing amino acids and the carbohydrate diet containing sucrose . Preliminary experiments established that flies would not eat purified amino acids with the vitamin/mineral/lipid buffer , so we diluted our protein solution with 20% of a suspension of dried yeast extract , made at the same protein concentration as the synthetic solution ( 16 . 25 g/L ) . Given that yeast extract also contains sugars , the final protein diet then included 4% carbohydrate . Alongside the dietary setup used for measuring diet-dependant fitness , we tested what flies preferred to eat , given the choice . For this , flies were supplied with two 5 µl microcapillary tubes ( ringcaps , Hirschmann ) ; one containing the protein solution and the other the carbohydrate solution . Capillary tubes were replaced daily , and food consumption for each fly trio was recorded for a period of three days . As a control , the rate of evaporation for all diet treatments was measured in six vials that contained the two solution-bearing capillary tubes but no flies and placed randomly in the controlled temperature room . Their average evaporation per day was used to correct diet consumption for evaporation . Male and female flies were assayed for fitness in the same way as previously described for Experiment 2 . However , rather than just feeding either a protein-rich or a carbohydrate-rich diet , we combined each of the two dietary treatments with one of four different concentrations of the drug rapamycin ( 0 µM , 5 µM , 10 µM , 50 µM ) . Rapamycin is a drug that very specifically inhibits TORC1 , and hence TOR-signalling , with this function being highly conserved from S . cerevisiae to humans ( Crespo and Hall , 2002 ) . Nutritional compositions and rapamycin levels were combined in a full factorial design resulting in a total of eight different diets ( two nutritional compositions times four rapamycin levels , eight diets in total ) for each sex . We had approximately 20 vials for each experimental unit . We performed a joint analysis on a dataset combining male and female fitness data . Before statistical analyses , male and female fitness measures were transformed to obtain normally distributed residual values . Female egg numbers were log-transformed , whereas male competitive fertility data was arcsine-transformed . Moreover , to be able to compare across sexes , male and female fitness measures were further centred and scaled ( separately for each sex ) using Z-transformations . We fitted a linear fixed effects model to the transformed fitness values with sex , diet and rapamycin concentrations ( coded as a categorical factor to accommodate possible non-linearity in the effect ) and their interactions . For the main analysis we categorised diet as optimal/non-optimal ( where the nutritional composition of the 'optimal' diet category is carbohydrate-rich for males and protein-rich for females ) . This encoding makes it more straightforward to assess how rapamycin treatment interacts with diet-quality in each sex . We also ran analysis where diet composition was encoded as 'carbohydrate-rich' and 'protein-rich' .	"You are what you eat" is a popular saying that can often make scientific sense . Everything an animal eats gets broken down into smaller molecules that fuel the many biological processes required to survive , move and reproduce . However , the food that the sexes need to maximize their fertility may not be exactly the same , as males make lots of small , mobile sperm cells while females create a small number of large eggs . In fruit flies for example , females benefit most from foods that contain lots of protein , while males are more fertile when they eat foods that are rich in carbohydrates . However , it remained unclear how these differences have evolved . Here , Camus et al . examine the genes that are active in male and female fruit flies which eat a diet rich in either carbohydrates or in proteins . Their experiments showed that both sexes share a large collection of genes which respond to the two diets in the same way . However , the type of food had opposite effects on the activity of certain genes involved in male and female reproduction . When the fruit flies had a protein-rich diet , for example , genes that promoted reproduction got turned on in females , but switched off in males . The opposite pattern was observed when the insects were exposed to carbohydrate-rich diets . Further analyses suggested that these different responses might be linked to a molecular network called IIS/TOR , which is a specific cascade of reactions that responds to nutrient availability . The findings of Camus et al . suggest that male and female flies produce different signals in reaction to food , which helps them to reproduce when they are able to meet their particular nutritional needs . Armed with a better understanding of the fundamental differences between the sexes , it may be possible to improve research into human health and animal keeping .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology", "genetics", "and", "genomics" ]	2019	Sex-specific transcriptomic responses to changes in the nutritional environment
In mouse embryo gastrulation , epiblast cells delaminate at the primitive streak to form mesoderm and definitive endoderm , through an epithelial-mesenchymal transition . Mosaic expression of a membrane reporter in nascent mesoderm enabled recording cell shape and trajectory through live imaging . Upon leaving the streak , cells changed shape and extended protrusions of distinct size and abundance depending on the neighboring germ layer , as well as the region of the embryo . Embryonic trajectories were meandrous but directional , while extra-embryonic mesoderm cells showed little net displacement . Embryonic and extra-embryonic mesoderm transcriptomes highlighted distinct guidance , cytoskeleton , adhesion , and extracellular matrix signatures . Specifically , intermediate filaments were highly expressed in extra-embryonic mesoderm , while live imaging for F-actin showed abundance of actin filaments in embryonic mesoderm only . Accordingly , Rhoa or Rac1 conditional deletion in mesoderm inhibited embryonic , but not extra-embryonic mesoderm migration . Overall , this indicates separate cytoskeleton regulation coordinating the morphology and migration of mesoderm subpopulations . In mice , a first separation of embryonic and extra-embryonic lineages begins in the blastocyst at embryonic day ( E ) 3 . 5 when the trophectoderm is set aside from the inner cell mass . A second step is the segregation of the inner cell mass into the epiblast , the precursor of most fetal cell lineages , and the extra-embryonic primitive endoderm ( Chazaud and Yamanaka , 2016 ) . At E6 , the embryo is cup-shaped and its anterior-posterior axis is defined . It comprises three cell types , arranged in two layers: the inner layer is formed by epiblast , distally , and extra-embryonic ectoderm , proximally; the outer layer , visceral endoderm , covers the entire embryo surface . The primitive streak , site of gastrulation , is formed at E6 . 25 in the posterior epiblast , at the junction between embryonic and extra-embryonic regions , and subsequently elongates to the distal tip of the embryo . The primitive streak is the region of the embryo where epiblast cells delaminate through epithelial-mesenchymal transition to generate a new population of mesenchymal cells that form the mesoderm and definitive endoderm layers . All mesoderm , including the extra-embryonic mesoderm , is of embryonic epiblast origin . At the onset of gastrulation , emerging mesoderm migrates either anteriorly as so-called embryonic mesodermal wings , or proximally as extra-embryonic mesoderm ( Arnold and Robertson , 2009; Sutherland , 2016 ) . Cell lineages studies showed that there is little correlation between the position of mesoderm progenitors in the epiblast and the final localization of mesoderm descendants ( Lawson et al . , 1991 ) . Rather , the distribution of mesoderm subpopulations depends on the temporal order and anterior-posterior location of cell recruitment through the primitive streak ( Kinder et al . , 1999 ) . Posterior primitive streak cells are the major source of extra-embryonic mesoderm , while cells from middle and anterior primitive streak are mostly destined to the embryo proper . However , there is overlap of fates between cells delaminating at different sites and timings ( Kinder et al . , 1999; Kinder et al . , 2001 ) . Extra-embryonic mesoderm contributes to the amnion , allantois , chorion , and visceral yolk sac . It has important functions in maternal-fetal protection and communication , as well as in primitive erythropoiesis ( Watson and Cross , 2005 ) . Embryonic mesoderm separates into lateral plate , intermediate , paraxial and axial mesoderm , and ultimately gives rise to cranial and cardiac mesenchyme , blood vessels and hematopoietic progenitors , urogenital system , muscles and bones , among others . Endoderm precursors co-migrate with mesoderm progenitors in the wings and undergo a mesenchymal-epithelial transition to intercalate into the visceral endoderm ( Viotti et al . , 2014 ) . Mesoderm migration mechanisms have mostly been studied in fly , fish , frog and chicken embryos . During fly gastrulation , mesodermal cells migrate as a collective ( Bae et al . , 2012 ) . In the fish Fundulus heteroclitus , deep cells of the dorsal germ ring move as loose clusters with meandering trajectories ( Trinkaus et al . , 1992 ) . At mid-gastrulation , zebrafish lateral mesoderm cells are not elongated and migrate as individuals along indirect paths , while by late gastrulation , cells are more polarized and their trajectories are straighter , resulting in higher speed ( Jessen et al . , 2002 ) . In zebrafish prechordal plate , all cells have similar migration properties but they require contact between each other for directional migration ( Dumortier et al . , 2012 ) . In chick , cells migrate in a very directional manner at high density . Cells are continually in close proximity , even though they frequently make and break contacts with their neighbors ( Chuai et al . , 2012 ) . Relatively little is known about mesoderm migration in mice because most mutant phenotypes with mesodermal defects result from anomalies in primitive streak formation , mesoderm specification , or epithelial-mesenchymal transition ( Arnold and Robertson , 2009 ) , precluding further insight into cell migration mechanisms . We previously identified a role for the Rho GTPase Rac1 , a mediator of cytoskeletal reorganization , in mesoderm migration and adhesion ( Migeotte et al . , 2011 ) . Recent advances in mouse embryo culture and live imaging have overcome the challenge of maintaining adequate embryo growth and morphology while performing high-resolution imaging . It facilitated the uncovering of the precise spatial and temporal regulation of cellular processes and disclosed that inaccurate conclusions had sometimes been drawn from static analyses ( Viotti et al . , 2014 ) . Live imaging of mouse embryos bearing a reporter for nuclei has pointed towards individual rather than collective migration in the mesodermal wings ( Ichikawa et al . , 2013 ) . Very recently , a spectacular adaptive light sheet imaging approach allowed reconstructing fate maps at the single cell level from gastrulation to early organogenesis ( McDole et al . , 2018 ) . However , little is known about how mesoderm populations regulate their shape and migration mechanisms as they travel across distinct embryo regions to fulfill their respective fates . Here , high-resolution live imaging of nascent mesoderm expressing membrane-bound GFP was used to define the dynamics of mesoderm cell morphology and its trajectories . Mesoderm cells exhibited a variety of cell shape changes determined by their spatial localization in the embryo , and the germ layer they were in contact with . The embryonic mesoderm migration path was meandrous but directional , and depended on the Rho GTPases Rhoa and Rac1 . Extra-embryonic mesoderm movement was , strikingly , GTPases independent . Transcriptomes of different mesoderm populations uncovered specific sets of guidance , adhesion , cytoskeleton and matrix components , which may underlie the remarkable differences in cell behavior between mesoderm subtypes . The T box transcription factor Brachyury is expressed in posterior epiblast cells that form the primitive streak , maintained in cells that delaminate through the streak , then down-regulated once cells progress anteriorly in the mesodermal wings ( Wilkinson et al . , 1990 ) . In order to visualize nascent mesoderm , Brachyury-Cre ( hereafter referred to as T-Cre ) transgenic animals , in which a construct encoding Cre cDNA fused to the regulatory elements of the Brachyury gene directing gene expression in the primitive streak was randomly inserted ( Feller et al . , 2008; Stott et al . , 1993 ) , were crossed to a membrane reporter line: Rosa26::membrane dtTomato/membrane GFP ( Muzumdar et al . , 2007 ) ( referred to as mTmG ) ( Figure 1 ) . In T-Cre; mTmG embryos , primitive streak and mesoderm-derived cells have green membranes ( mG ) , whereas all other cells have red membranes ( mT ) . Embryos dissected at E6 . 75 or E7 . 25 were staged according to Downs and Davies ( 1993 ) ( Figure 1—figure supplement 1a ) and examined in different orientations by confocal or two-photon excitation live imaging for 8 to 12 hr ( Figure 1 , Figure 1—figure supplement 1c and d , Videos 1 and 2 ) . Conversion of mT to mG was first observed at Early/Mid Streak ( E/MS ) stage , and was initially mosaic , which facilitated the tracking of individual migrating cells with high cell shape resolution . From Mid/Late Streak ( M/LS ) onwards , most primitive streak cells underwent red to green conversion ( Figure 1—figure supplement 1e , f ) . The shape of mesoderm cells and their tracks were recorded through imaging of embryos from different perspectives between ES and Early Bud ( EB ) stages of development , in order to obtain images of optimal quality for each embryo region ( Figure 1 and Videos 1 and 2 ) . Posterior views ( Figure 1a ) showed proximal to distal primitive streak extension and basal rounding of bottle-shaped cells exiting the streak , as previously described ( Williams et al . , 2012 ) . Lateral views ( Figure 1b ) allowed comparing cells as they migrated laterally in mesodermal wings , or proximally in extra-embryonic region . Anterior views ( Figure 1c ) showed cell movement towards the midline . The imaging time frame did not allow following individual cells from their exit at the primitive streak to their final destination . However , the trajectories we acquired ( Videos 3 and 4 ) fitted with the fate maps built using cellular labeling or transplantation ( Kinder et al . , 1999; Kinder et al . , 2001 ) , or adaptive light sheet microscopy ( McDole et al . , 2018 ) . The first converted ( GFP positive ) cells in ES embryos dissected around E6 . 75 usually left the posterior site of the primitive streak to migrate towards the extra-embryonic compartment . Embryonic migration started almost simultaneously , and migration towards both regions proceeded continuously . Strikingly , migration behavior ( Figure 2a , Videos 3 and 4 ) and cell shape ( Figure 2b ) were different depending on the region cells migrated into . In the embryonic region , mesoderm cells had a global posterior to anterior path , even though they zigzagged in all directions ( proximal-distal , left-right , and even anterior-posterior ) . Cells did not migrate continuously , but showed alternations of tumbling behavior with straighter displacement , as described for zebrafish mesendoderm progenitors ( Diz-Muñoz et al . , 2016 ) . Embryonic mesoderm cells from ES/MS embryos tracked for 2 . 5 hr moved at a mean speed of 0 . 65 µm/min to cover approximately 90 µm and travel a net distance of 40 µm ( Figure 2a' , Table 1 ) . Straightness ( the ratio of net over travel displacement , so that a value of 1 represents a linear path ) was approximately 0 . 5 ( Figure 2a' , Table 1 ) . Cells in the extra-embryonic region moved slightly slower ( 0 . 45 µm/min ) to do approximately 70 µm , but their net displacement ( 20 µm ) and straightness ( 0 . 3 ) were significantly smaller , reflecting trajectories with no obvious directionality ( Figure 2a' , Table 1 , and Video 5 ) . Extra-embryonic cells were twice larger in volume , and more elongated ( Figure 2b , b' , Table 2 ) . The increased size can be attributed to a lower frequency of division ( Figure 1—figure supplement 1b ) . They had few large protrusions , and filopodia were scarce and short ( Figure 2b , b' , Table 2 ) . Cells passing through the primitive streak were , as reported ( Ramkumar et al . , 2016; Williams et al . , 2012 ) , bottle shaped with a basal round cell body and an apical thin projection ( Figure 3a ) . Mesoderm cells in contact with the epiblast and visceral endoderm sent thin protrusions towards their respective basal membranes ( Figure 3b , b' , and Video 6 ) . Interestingly , the density of thin protrusions was much higher in cells in contact with the visceral endoderm . As this phenotype was observed as early as ES in the posterior region , it could reflect the putative signaling role of the proximal posterior visceral endoderm , which remains coherent along gastrulation ( Kwon et al . , 2008 ) . Intercalation of prospective definitive endoderm cells in the visceral endoderm layer , which occurs from LS stage onwards ( Viotti et al . , 2014 ) , was very rarely observed , either because Brachyury does not label endoderm progenitors ( Burtscher and Lickert , 2009 ) , or because the T-Cre transgenic line does not bear the regulatory elements driving T expression in prospective endoderm . Cells in tight clusters surrounded by other mesoderm cells in the wings had a smoother contour , with the caveat that protrusions couldn’t be visualized between cells of similar membrane color . Reconstruction of cells in the anterior part of the wings , where recombination was incomplete , showed thin protrusions between mesoderm cells ( Figure 1c ) . Cells also extended long broad projections , which spanned several cell diameters and were sent in multiple directions before translocation of the cell body , in what seemed a trial and error process ( Figure 3c and Videos 1 , 2 and 7 ) . The presence of potential leader cells could not be assessed , as the first cells converted to green are not the most anterior ones ( Figure 1—figure supplement 1e ) . Nonetheless , cells with scanning behavior were observed at all times , which suggests that all cells are capable to explore their surroundings . To address the collectiveness of mesoderm migration , we assessed the impact of cell proximity on behavior by comparing the trajectories of daughter cells after mitosis , pairs of unrelated cells that were in immediate proximity at the beginning of observation , and cells colliding along the way . As expected from fate mapping experiments , daughter cells resulting from mitosis within the mesoderm layer followed close and parallel trajectories ( Figure 3d and Figure 3—source data 2 ) . They travelled a similar net distance over 204 min ( net displacement ratio: 0 . 91 ± 0 . 01 , n=12 pairs from four embryos at E/MS stage ) , in the same direction ( angle: 7 ± 1 . 13° ) , with one daughter cell displaying a higher straightness ( travel displacement ratio: 0 . 61 ± 0 . 07 ) . They remained close to one another ( mean distance between daughter cells: 15 . 6 ± 2 . 35 µm ) , but not directly apposed . Interestingly , they stayed linked by thin projections for hours , even when separated by other cells ( Figure 3d ) . Pairs of cells that were in immediate proximity at time zero traveled a similar distance over 152 min ( net displacement ratio: 0 . 88 ± 0 . 03 , n=13 pairs from four embryos ) , in the same direction ( angle: 9 . 9° ± 2 . 77 ) ; they remained close to one another ( travel displacement ratio and final distance were respectively 0 . 75 ± 0 . 04 and 15 . 2 µm ± 3 . 84 µm ) ( Figure 3—source data 3 ) . Contact between cell protrusions could be spotted in most pairs . Mesoderm cell migration is often compared to neural crest migration , as both cell types arise through epithelial-mesenchymal transition ( Roycroft and Mayor , 2016 ) . An important feature of neural crest migration is contact inhibition of locomotion , where cells that collide tend to move in opposite directions . In contrast , most mesoderm cells stayed in contact upon collision ( Figure 3—source data 4 ) : 16 out of 24 cell pairs from 5 ES to Zero Bud ( 0B ) embryos remained attached for 2 . 5 hr ( one briefly lost contact before re-joining ) , 3/24 stayed joined for around 1 hr , and 5/24 pairs separated instantly . We segmented 8 pairs for 166 min , and observed a mean distance at the end of tracking of 52 . 5 ± 32 . 5 µm; they followed parallel trajectories ( angle: 8 . 25 ± 1 . 7° , n=8 pairs ) , for a similar net distance ( net displacement ratio: 0 . 85 ± 0 . 07; travel displacement ratio: 0 . 64 ± 0 . 11 ) . Thin projections could occasionally be observed between them after contact . Those data suggest that cells coming in close proximity tend to have a similar behavior . The presence of thin projections between them may reflect cell-to-cell communication . Embryonic and extra-embryonic mesoderm cells were isolated through fluorescence ( GFP ) -assisted cell sorting from E7 . 5 MS and LS T-Cre; mT/mG embryos in order to generate transcriptomes . Biological replicates ( 4 MS , 2 LS ) from both stages resulted in grouping of samples according to the embryonic region ( Figure 4a , b ) . Non-supervised clustering based on embryonic and extra-embryonic gene signatures identified by single cell sequencing in Scialdone et al . ( 2016 ) showed that samples segregated as expected ( not shown ) . We performed pairwise comparison of embryonic and extra-embryonic samples obtained at both stages and selected the genes that were consistently differentially expressed with a fold change >2 . Gene ontology analysis of genes clusters enriched either in embryonic or extra-embryonic mesoderm highlighted expected developmental ( angiogenesis and hematopoiesis in extra-embryonic , somitogenesis in embryonic ) , and signaling ( BMP and VEGF in extra-embryonic , Wnt and Notch in embryonic ) biological processes ( Figure 4—figure supplement 1a , a' ) . Interestingly , differences were also seen in gene clusters involved in migration , adhesion , cytoskeleton , and extracellular matrix organization . Genes with known expression pattern in gastrulation embryos found enriched in embryonic mesoderm included well-described transcription factors , as well as FGF , Wnt , Notch , TGFβ and Retinoic Acid pathways effectors ( Figure 4—figure supplement 1b ) . Genes expected to be more expressed in extra-embryonic mesoderm included the transcription factors Ets1 and Tbx20 , and several members of the TGFβ pathway ( Inman and Downs , 2007; Pereira et al . , 2011 ) ( Figure 4—figure supplement 1c ) . Primitive hematopoiesis , the initial wave of blood cell production which gives rise to primitive erythrocytes , macrophages , and megakaryocytes , takes place around E7 . 25 in hemogenic angioblasts of the blood islands ( Lacaud and Kouskoff , 2017 ) . Expression of genes involved in hemangioblast development , endothelium differentiation , and hematopoiesis increased from MS to LS ( Figure 4f and Figure 4—figure supplement 1c ) . In addition , we confirmed two extra-embryonic genes identified through subtractive hybridization at E7 . 5 ( Kingsley et al . , 2001 ) : Ahnak ( Figure 4—figure supplement 1c ) , see also Downs et al . ( 2002 ) and the imprinted gene H19 ( Figure 4—figure supplement 1c ) . Of particular interest among the genes with higher expression in embryonic mesoderm for which no expression data was available at the stage of development were genes related to matrix ( Lama1 , Galnt16 , Egflam ) , adhesion ( Itga1 , Itga8 , Pcdh8 , Pcdh19 , Pmaip1 ) , and guidance ( Epha1 and 4 , Robo3 , Sema6d , Ntn1 ) ( Figure 4c ) . Epha4 expression in the mouse embryo has been described in the trunk mesoderm and developing hindbrain at Neural Plate ( NP ) stage ( Nieto et al . , 1992 ) . In LS embryos , Epha4 expression was higher in the primitive streak and embryonic mesoderm ( Figure 4d and Figure 4—figure supplement 1e ) . Dynamic Epha1 , Efna1 and Efna3 expression patterns have been shown during gastrulation ( Duffy et al . , 2006 ) . In LS/0B embryos , Epha1 mRNA was present in the primitive streak , mostly in its distal part . Its ligand Efna1 was in the primitive streak with an inverse gradient , and was mainly expressed in the extra-embryonic region , notably in amnion and in chorion . Efna3 was very abundant in the chorion ( Figure 4d and Figure 4—figure supplement 1e ) . In parallel , in extra-embryonic mesoderm , we found higher expression of distinct sets of genes with putative roles in guidance ( Unc5c , Dlk1 , Scube2 , Fzd4 ) , matrix composition ( Hapln1 , Col1a1 , Lama4 ) , adhesion ( Itga3 , Pkp2 , Podxl , Ahnak , Adgra2 , Pard6b ) , Rho GTPase regulation ( Rasip1 , Stard8 , Rhoj ) , and cytoskeleton ( Myo1c , Vim , Krt8 and Krt18 ) ( Figure 4e and not shown ) . Interestingly , Podocalyxin ( Podxl ) was abundant in extra-embryonic mesoderm ( Figure 4f and Figure 4—figure supplement 1d ) , which fits with data from embryo and embryoid body single cell sequencing showing that Podxl is a marker for early extra-embryonic mesoderm and primitive erythroid progenitors of the yolk sac ( Zhang et al . , 2014 ) . In view of the differences in cell shape and migration , we focused on the cytoskeleton , in particular actin , and intermediate filaments proteins ( Vimentin and Keratins ) . Vimentin was found in all mesoderm cells as expected , but more abundant in extra-embryonic mesoderm ( Figure 5a ) . Remarkably , within the mesoderm layer , Keratin 8 was selectively expressed in extra-embryonic mesoderm cells ( amniochorionic fold , amnion , chorion , and developing allantois ) ( Figure 5b–d ) . In contrast , the filamentous actin ( F-actin ) network , visualized by Phalloidin staining , seemed denser in embryonic mesoderm ( Figure 5a ) . To visualize F-actin only in mesoderm , we took advantage of a conditional mouse model expressing Lifeact-GFP , a peptide that binds specifically to F-actin with low affinity , and thus reports actin dynamics without disrupting them ( Schachtner et al . , 2012 ) . Live imaging of T-Cre; LifeAct-GFP embryos at MS and LS stage confirmed that while LifeAct-GFP positive filaments could be visualized clearly in embryonic mesoderm cells , GFP was weaker and diffuse in extra-embryonic mesoderm ( Figure 5e and Video 8 ) . Rho GTPases are molecular switches that relay signals from cell surface receptors to intracellular effectors , leading to a change in cell behavior ( Hodge and Ridley , 2016 ) . They are major regulators of cytoskeletal rearrangements ( Hall , 1998 ) , and the spatiotemporal fine regulation of Rho GTPases activities determines cytoskeletal dynamics at the subcellular level ( Spiering and Hodgson , 2011 ) . Therefore , inactivation of a given Rho GTPase may result in variable consequences depending on cell type and context . We previously established that Sox2-Cre mediated deletion of Rac1 in the epiblast before onset of gastrulation causes impaired migration of embryonic mesoderm while extra-embryonic mesoderm migration is less severely affected ( Migeotte et al . , 2011 ) . We thus hypothesized that Rho GTPases might be differentially regulated in cells invading both regions , resulting in some of the observed distinctions in cytoskeletal dynamics , cell shape and displacement mode . Mutations were induced in cells transiting the primitive streak by crossing heterozygous wild-type/null Rhoa ( Jackson et al . , 2011 ) or Rac1 ( Walmsley et al . , 2003 ) animals bearing the T-Cre transgene with animals homozygous for their respective conditional alleles bearing the mTmG reporter ( mutant embryos are referred to as RhoaΔmesoderm and Rac1Δmesoderm ) . The phenotypes of RhoaΔmesoderm and Rac1Δmesoderm embryos were less severe than that of RhoaΔepi and Rac1Δepi embryos ( our unpublished data and Migeotte et al . , 2011 ) ( Figure 6—figure supplement 1 ) . Mutants were morphologically indistinguishable at E7 . 5 . At E8 . 5 , RhoaΔmesoderm embryos were identified , though with incomplete penetrance , as being slightly smaller than their wild-type littermates ( 11/12 mutants had a subtle phenotype , including five with reduced numbers of somites , and six with abnormal heart morphology ) ( Figure 6—figure supplement 1a ) . By E9 . 5 , all RhoaΔmesoderm embryos had an obvious phenotype ( 12/12 mutants had a small heart , 9/12 had a reduced number of somites , 2/12 had an open neural tube , 2/12 had a non-fused allantois ) ( Figure 6—figure supplement 1b ) . Rac1Δmesoderm embryos also had subtle phenotypes at E8 . 5 ( 15/16 embryos were slightly smaller than wild-type littermates , 4/16 had a small heart ) ( Figure 6—figure supplement 1d ) . In situ hybridization for Brachyury showed weaker staining in the tail region in 5/10 mutant embryos , indicative of reduced presomitic mesoderm ( Figure 6—figure supplement 1c ) . By E9 . 5 , all mutants had abnormal heart morphology and reduced body length , and 3 embryos out of 9 were severely delayed ( Figure 6—figure supplement 1e ) . At E10 . 5 , penetrance was complete; 7/7 embryos had a short dysmorphic body and pericardial edema ( not shown ) . The phenotypic variability at early time points likely reflects mosaicism of T-Cre mediated recombination . Embryonic mesoderm explants from E7 . 5 MS/LS mTmG; Rac1Δmesoderm or RhoaΔmesoderm embryos were plated on fibronectin . In wild-type explants , cells showed a radial outgrowth from the explants , displaying large lamellipodia in the direction of migration ( Video 9 ) . After cell-cell contact , they remained connected through long thin filaments . Rhoa deficient explants showed less release of individual cells ( Figure 6—figure supplement 2a ) . Rhoa mutant cells appeared more cohesive , and were rounder than wild-type cells ( Figure 6—figure supplement 2c ) . Remarkably , live imaging of explants from Rhoa mutant embryos showed a phase of compaction preceding cell migration ( Video 10; 2/4 RhoaΔmesoderm mutant explants displayed compaction ) . In Rac1 explants ( Figure 6—figure supplement 2b ) , GFP-expressing cells remained within the domain of the dissected explant and displayed pycnotic nuclei , while wild-type non-GFP cells could migrate . Live imaging could not be performed as 4 out of 5 mutant explants detached from the plate . This is similar to explants from Rac1 epiblast-specific mutants ( Migeotte et al . , 2011 ) , and is attributed to lack of adhesion-dependent survival signals . Live imaging of mTmG; RhoaΔmesoderm or Rac1Δmesoderm embryos dissected at E6 . 75 or 7 . 25 ( Figure 6 ) showed that the majority of Rhoa and Rac1 mesoderm-specific mutants ( 4/8 for Rhoa , 6/9 for Rac1 ) displayed an accumulation of cells at the primitive streak , which formed a clump on the posterior side between epiblast and visceral endoderm ( Figure 6a–d and Video 11 ) , indicating a mesoderm migration defect . Interestingly , although embryonic mesoderm migration was impaired , with only a handful of cells visible on the anterior side by E7 . 5 , extra-embryonic mesoderm migration was maintained ( Figure 6e , f and Videos 12–15 ) . There was a significant decrease in embryonic , but not extra-embryonic mesoderm cells speed for both Rhoa and Rac1 mutants , compared to wild-type embryos ( Figure 6—figure supplement 1f ) . Similarly , in embryos deleted for Rac1 in epiblast and epiblast-derived cells upon Sox2-Cre ( Hayashi et al . , 2002 ) recombination , GFP positive mesoderm cells were dispersed in the extra-embryonic region , while embryonic mesoderm cells were confined in a bulge adjacent to the primitive streak ( Figure 6—figure supplement 2d ) . Accordingly , staining for mesoderm-derived vascular structures ( Pecam-1 ) in the yolk sac at E8 . 5 showed no difference between mutant and wild-type embryos ( Figure 6—figure supplement 2e , f ) . Those findings suggest that extra-embryonic mesoderm cells either do not rely on Rac1 and Rhoa for movement , or are able to compensate for loss of Rac1 or Rhoa , which is consistent with their paucity in actin-rich protrusions . Mesoderm cell delamination from the epiblast requires basal membrane disruption , apical constriction , loss of apicobasal polarity , changes in intercellular adhesion , and acquisition of motility ( Nieto et al . , 2016 ) . The transcriptional network and signaling pathways involved in epithelial-mesenchymal transition are conserved ( Ramkumar and Anderson , 2011 ) . However , pre-gastrulation embryo geometry varies widely between species , which has important consequences on interactions between germ layers and mechanical constrains on nascent mesoderm cells ( Williams and Solnica-Krezel , 2017 ) . Live imaging of mouse embryo has allowed recording posterior epiblast rearrangements , as well as cell passage through the primitive streak ( Ramkumar et al . , 2016; Williams et al . , 2012 ) . Contrary to the chick embryo , there is no global epiblast movement towards the primitive streak in the mouse . However , cell shape changes , including apical constriction and basal rounding , are similar . Morphological data on mouse mesoderm cells acquired through scanning electron microscopy of whole mount samples ( Migeotte et al . , 2011 ) , and transmission electron microscopy of embryo sections ( Spiegelman and Bennett , 1974 ) revealed an array of stellate individual cells linked by filopodia containing a lattice of microfilaments . We took advantage of mosaic labeling of nascent mesoderm to define the dynamics of cell shape changes associated with mesoderm migration . Cells just outside the streak retracted the long apical protrusion and adopted a round shape with numerous filopodia making contacts with adjacent , but also more distant mesoderm cells . In mesodermal wings , cells close to the epiblast were more loosely apposed , and extended fewer filopodia towards its basal membrane , compared to cells adjacent to the visceral endoderm , which were tightly packed and displayed numerous filopodia pointing to the visceral endoderm basal membrane . Cells travelling in a posterior to anterior direction towards the midline displayed long protrusions , up to twice the cell body size , which extended , retracted , occasionally bifurcated , several times before the cell body itself initiated movement , suggesting an explorative behavior . Remarkably , extension of long protrusions was not limited to the first row of cells . Migration was irregular in time and space , as cells often stopped and tumbled , and displayed meandrous trajectories . After division , cells remained attached by thin protrusions . Contrary to neural crest cells , mesoderm cells did not show contact inhibition of locomotion . Cells in close proximity tended to follow parallel paths . Extra-embryonic mesoderm first accumulates between extra-embryonic ectoderm and visceral endoderm at the posterior side of the embryo , leading to formation of the amniochorionic fold that bulges into the proamniotic cavity ( Pereira et al . , 2011 ) . This fold expands , and lateral extensions converge at the midline . Accumulation and coalescence of lacunae between extra-embryonic mesoderm cells of the fold generate a large cavity closed distally by the amnion , and proximally by the chorion . At LS stage , extra-embryonic mesoderm forms the allantoic bud , precursor to the umbilical cord , in continuity with the primitive streak ( Inman and Downs , 2007 ) . Extra-embryonic mesoderm cells had striking differences in morphology and migration mode , compared to embryonic mesoderm cells . They were larger and more elongated , displayed fewer filopodia , and almost no large protrusions . They migrated at a similar speed , but in a much more tortuous fashion , resulting in little net displacement . Direction cues could come from cell-matrix contact , homotypic or heterotypic ( with epiblast or visceral endoderm ) cell-cell interaction , diffuse gradients of morphogens , and/or mechanical constraints . Transcriptome data were compatible with roles for guidance molecules such as Netrin1 and Eph receptors in directing mesoderm migration . Epha4 was strongly expressed in the PS and mesoderm , particularly in the embryonic region . In Xenopus , interaction of Epha4 in mesoderm and Efnb3 in ectoderm allows separating germ layers during gastrulation ( Rohani et al . , 2014 ) . Epha1 and its ligands Efna1 and 3 had partially overlapping , but essentially reciprocal compartmentalized expression patterns during gastrulation ( Duffy et al . , 2006 ) . In addition , we found abundant Epha1 expression in somites and presomitic mesoderm at E8 . 5 ( not shown ) . Interestingly , Epha1 KO mice present a kinked tail ( Duffy et al . , 2008 ) . The specific and dynamic expression patterns of Epha4 , Epha1 , and their respective ligands during gastrulation are compatible with roles in germ layers separation , including nascent mesoderm specification and migration . Identification of those potential guidance cues will help design strategies to better understand how mesoderm subpopulations reach their respective destinations . Visualization and modification of Rho GTPases activity through FRET sensors and photoactivable variants has shed light on their fundamental role during cell migration in in vivo contexts , such as migration of fish primordial germ cells ( Kardash et al . , 2010 ) or Drosophila border cells ( Wang et al . , 2010 ) . Study of a epiblast-specific mutant showed that Rac1 acts upstream of the WAVE complex to promote branching of actin filaments , lamellipodia formation , and migration of nascent mesoderm ( Migeotte et al . , 2011 ) . Remarkably , extra-embryonic mesoderm cells did not display leading edge protrusions , and Rac1 and Rhoa mesoderm-specific mutants were deficient for embryonic , but not extra-embryonic mesoderm migration . Interestingly , in embryos mutants for Fgfr1 ( Yamaguchi et al . , 1994 ) or Fgf8 ( Sun et al . , 1999 ) , extra-embryonic mesoderm populations are almost normal , while embryonic mesoderm derivatives are severely affected . Measurement of Rho GTPases activity in those mutants would allow exploring the possibility that Rac1 and Rhoa act downstream of the FGF pathway to promote mesoderm migration , as proposed in Drosophila ( van Impel et al . , 2009 ) . In addition , F-actin filaments were more abundant in embryonic , compared to extra-embryonic , mesoderm , reinforcing the hypothesis that they might rely on distinct cytoskeletal rearrangements . Intermediate filaments are major effectors of cell stiffness , cell-matrix and cell-cell adhesion , as well as individual and collective migration ( Pan et al . , 2013 ) . Members of type I and II keratin families form obligate heterodimers , which assemble into filaments ( Loschke et al . , 2015 ) . Type II Keratins 7 and 8 , and type I Keratins 18 and 19 are the first to be expressed during embryogenesis . Combined Keratins 8/19 and 18/19 deletions cause lethality at E10 attributed to fragility of giant trophoblast cells ( Hesse et al . , 2000 ) . Deletion of the entire type II Keratins cluster results in growth retardation starting at E8 . 5 ( Vijayaraj et al . , 2009 ) . Recently , knockdown of Keratin 8 in frog mesendoderm highlighted a role for intermediate filaments in coordinating collectively migrating cells . Keratin-depleted cells were more contractile , displayed misdirected protrusions and large focal adhesions , and exerted higher traction stress ( Sonavane et al . , 2017 ) . Transcripts for Keratins 8 and 18 , as well as Vimentin , were enriched in extra-embryonic , compared to embryonic mesoderm . While Vimentin was present in all mesoderm , Keratin 8 was only detectable in extra-embryonic mesoderm . An antagonistic relationship between Vimentin intermediate filaments and Rac1-mediated lamellipodia formation has been described ( Helfand et al . , 2011 ) , and a similar opposition may exist between Rac1 activity and keratin intermediate filaments ( Weber et al . , 2012 ) . Extra-embryonic mesoderm cells' elongated morphology , paucity in lamellipodia , and lack of directional migration may thus result from their high content in intermediate filaments , and low Rho GTPase activity ( Figure 7 ) . The recent development of a K8-YFP reporter mouse strain for intermediate filaments ( Schwarz et al . , 2015 ) , and the availability of reliable Rho GTPases FRET sensors ( Spiering and Hodgson , 2011 ) , will be instrumental in dissecting their relationship in mesoderm . The mesoderm germ layer has the particularity to invade both embryonic and extra-embryonic parts of the conceptus , and its migration is important for both fetal morphogenesis and development of extra-embryonic tissues including the placenta . We found that embryonic and extra-embryonic mesoderm populations , both arising by epithelial-mesenchymal transition at the primitive streak , display distinct shape dynamics , migration modes , Rho GTPase dependency , cytoskeletal composition , as well as expression of different sets of guidance , adhesion , and matrix molecules . Landmark experiments in the 1990 s showed that the fate of a mesoderm cell depends on the time and place at which it emerges from the primitive streak . We have unveiled morphological and behavioral specificities of mesoderm populations through whole embryo live imaging , and provided a molecular framework to understand how cells with distinct fates adapt to , and probably modify , their tridimensional environment . The T-Cre line was obtained from Achim Gossler ( Feller et al . , 2008 ) , the Rac1 line from Victor Tybulewicz ( Walmsley et al . , 2003 ) , the Rhoa line from Cord Brakebusch ( Jackson et al . , 2011 ) , the mTmG ( Muzumdar et al . , 2007 ) and Sox2-Cre ( Hayashi et al . , 2002 ) lines from The Jackson Laboratory , and the conditional LifeAct-GFP line from Laura Machesky ( Schachtner et al . , 2012 ) . Mice were kept on a CD1 background . Mice colonies were maintained in a certified animal facility in accordance with European guidelines . Experiments were approved by the local ethics committee ( CEBEA ) . Mouse genomic DNA was isolated from ear biopsies following overnight digestion at 55°C with 1 . 5% Proteinase K ( Quiagen ) diluted in Lysis reagent ( DirectPCR , Viagen ) , followed by heat inactivation . Embryos were dissected in Dulbecco’s modified Eagles medium ( DMEM ) F-12 ( Gibco ) supplemented with 10% FBS and 1% Penicillin-Streptomycin and L-glutamine and 15 mM HEPES . They were then cultured in 50% DMEM-F12 with L-glutamine without phenol red , 50% rat serum ( Janvier ) , at 37°C and 5% CO2 . Embryos were observed in suspension in individual conical wells ( Ibidi ) to limit drift , under a Zeiss LSM 780 microscope equipped with C Achroplan 32x/0 . 85 and LD C Apochromat 40x/1 . 1 objectives . Stacks were acquired every 20 min with 3 μM Z intervals for up to 10 hr . Embryos were cultured for an additional 6 to 12 hr after imaging to check for fitness . Antibodies were goat anti-Pecam-1 1:500 ( R and D systems ) ; rabbit anti-Podocalyxin 1:200 ( EMD Millipore ) ; rat anti-Keratin 8 1:100 ( Developmental Studies Hybridoma Bank ) ; rabbit anti-Vimentin 1:200 ( abcam ) . F-actin was visualized using 1 . 5 U/ml TRITC-Phalloidin ( Invitrogen ) , and nuclei using DAPI ( Sigma ) . Secondary antibodies were anti rabbit Alexa Fluor 488 and 647 , anti rat Alexa Fluor 647 ( all from Life technologies ) , and anti goat Alexa Fluor 647 ( Jackson ) . Whole-mount in situ hybridization was carried out as described in Eggenschwiler and Anderson ( 2000 ) . For in situ hybridization on sections , embryos were dissected in PBS and fixed for 30 min at 4°C in 4% PFA . They were washed in PBS , embedded directly in OCT ( Tissue-Tek ) , and cryosectioned at 7–10 microns . Slides were re-fixed for 15 min on ice in 4% PFA . RNA probes were obtained from ACDBio , and hybridization was performed using the RNAscope 2 . 5 HD Reagent Kit-RED ( ACDBio ) according to manufacturer’s instructions . Slides were counterstained with 50% Gill’s Hematoxylin . For immunofluorescence , embryos were fixed in PBS containing 4% paraformaldehyde ( PFA ) for 2 hr at 4°C , cryopreserved in 30% sucrose , embedded in OCT and cryosectioned at 7–10 microns . Staining was performed in PBS containing 0 . 5% Triton X-100% and 1% heat-inactivated horse serum . Sections and whole-mount embryos were imaged on a Zeiss LSM 780 microscope . Primary explant cultures of nascent mesoderm were generated as described in Burdsal et al . ( 1993 ) . Explants were cultured for 24–48 hr in DMEM F-12 supplemented with 10% FBS and 1% Penicillin-Streptomycin and L-glutamine on fibronectin ( Sigma ) coated glass bottom microwell 35 mm dishes with 1 . 5 cover glass ( MatTek ) . They were fixed for 30 min in PBS containing 4% PFA prior to staining . For live imaging , explants were let to adhere for 4–6 hr , and then imaged every 15 min for up to 12 hr . Images were processed using Arivis Vision4D v2 . 12 . 3 ( Arivis , Germany ) . Embryo contours were segmented manually on each Z-slice and time point , and then registered using the drift correction tool of Arivis Vision4D . Embryo rotation was adjusted manually if necessary . We chose embryos where successful registration could be achieved , so that the embryo's residual slight movements were much smaller than cell displacement . Similarly , we found embryo growth to be negligible compared to cell displacement ( data not shown ) . Cells were then manually segmented on each Z-slice and time point by highlighting cellular membranes using Wacom’s Cintiq 13HD . Net displacement , path length , speed and angle between two cells were based on the centroid coordinates of segmented cells from Arivis , and calculated by a homemade Python script ( Python Software Foundation , https://www . python . org ) . To extract speed behavior , we interpolated the path length curve and derivated it . The path length over time was closely linear , so we extracted the mean of the speed values . Surface , volume , long/short axis ratio of 2D inner ellipse , and straightness were calculated by Arivis . 2D Z projections of late embryos were used to quantify the filopodia length and density . Filopodia size and density were measured on Icy ( de Chaumont et al . , 2012 ) and analyzed using a homemade Python script . Videos were generated using the StackReg ImageJ plugin ( Thévenaz et al . , 1998 ) . All data are presented as Mean ± SEM . Depending on whether data had a Gaussian distribution or not , we used either the Mann-Whitney-Wilcoxon or the t-test . A p value < 0 . 05 was considered statistically significant . T-Cre; mTmG embryos were collected from different mice , and those at the appropriate stage were pooled . Embryonic and extra-embryonic portions were separated by manually cutting the embryo with finely sharpened forceps . The embryos were digested using 2X Trypsin , and pure GFP + populations were sorted through flow cytometry ( FACSARIA III , BD ) , directly in extraction buffer . RNA was extracted using the PicoPure kit ( ThermoFisher Scientific ) . RNA quality was checked using a Bioanalyzer 2100 ( Agilent technologies ) . Indexed cDNA libraries were obtained using the Ovation Solo RNA-Seq System ( NuGen ) following manufacturer recommendation . The multiplexed libraries ( 18 pM ) were loaded on flow cells and sequences were produced using a HiSeq PE Cluster Kit v4 and TruSeq SBS Kit v3-HS from a Hiseq 1500 ( Illumina ) . Paired-end reads were mapped against the mouse reference genome ( GRCm38 . p4/mm10 ) using STAR software to generate read alignments for each sample . Annotations Mus_musculus . GRCm38 . 87 . gtf were obtained from ftp . Ensembl . org . For transcript quantification , all the Reference Sequence ( RefSeq ) transcript annotations were retrieved from the UCSC genome browser database ( mm10 ) . Transcripts were quantified using the featureCounts ( Liao et al . , 2014 ) software tool using the UCSC RefSeq gene annotations ( exons only , gene as meta features ) . Normalized expression levels were estimated using the EdgeR rpm function and converted to log2 FPKM ( fragments per kilobase of exon per million mapped reads ) after resetting low FPKMs to one to remove background effect . Differential analysis was performed using the edgeR method ( quasi-likelihood tests ) ( McCarthy et al . , 2012 ) . The edgeR model was constructed using a double pairwise comparison between embryonic mesoderm versus extra-embryonic mesoderm at two different time points ( MS and LS ) . First , the count data were fitted to a quasi-likelihood negative binomial generalized log-linear model using the R glmQLFit method . To identify differentially expressed genes , null hypothesis EM_E7 . 0==EEM_E7 . 0 and EM_E7 . 25==EEM_E7 . 25 were tested using the empirical Bayes quasi-likelihood F-tests ( glmQLFTest method ) applied to the fitted data . The F-test P-values were then corrected for multi-testing using the Benjamini-Hochberg p-value adjustment method . Transcripts with a greater than background level of expression ( mean log2 count per million >0 ) , an absolute fold change >2 , and a low false discovery rate ( FDR <0 . 05 ) were considered as differentially expressed . The sample visualization map was produced by applying the t-Distributed Stochastic Neighbor Embedding ( tSNE ) dimensionality reduction method ( Van Der Maaten and Hinton , 2008 ) to log2 FPKM expression levels ( all transcripts ) . The R tSNE method from 'Rtsne' library was applied without performing the initial PCA reduction and by setting the perplexity parameter to 2 . The heatmap was produced using the R heatmap . 2 methods using the brewer . pal color palette . GO analysis was performed using the DAVID software ( Huang et al . , 2009 ) .	As an embryo develops , its cells divide and specialize to form different tissues and organs . Early in development the cells arrange into so-called germ layers , which each produce particular types of tissue . One of these layers , called the mesoderm , develops into the muscles , bones and circulatory system of the embryo . It also contributes to the support structures that feed and protect the embryo , such as the placenta , umbilical cord and yolk sac . If these ‘extra-embryonic’ structures do not develop correctly , the embryo may not grow properly . Much of what we know about how the cells of the mesoderm move around to form different tissues comes from studies of species that lay eggs; for example , chicks , frogs and fish . The initial steps of embryo development in these animals are similar to how mammals develop , but bigger differences emerge as the extra-embryonic tissues start to form . Recent methodological advances are now making it possible to dynamically study this later stage of development in live mouse embryos . Saykali et al . studied mouse embryos whose mesoderm cells contained a ‘reporter’ that allowed them to be identified when viewed using a microscopy technique known as two-photon live imaging . This approach allows cells to be tracked as they move through living tissue . Saykali et al . found that the mesoderm cells change shape depending on which region of the embryo they are in , and on which germ layer they are next to . The cells that become extra-embryonic are larger and longer , and develop small protrusions . Instead of moving directly to their destinations , they tend to zigzag . Further experiments revealed that embryonic and extra-embryonic mesoderm cells produce different amounts of several proteins , including the distinct types of filaments that act as the cell’s internal skeleton . Mesoderm cells that are destined to become extra-embryonic depend less on signaling proteins called Rho GTPases to move around . Knowing how mesoderm cells form extra-embryonic structures will help researchers to understand how problems with these structures can affect how embryos grow . The techniques used by Saykali et al . will also help to design new ways to cultivate mesoderm cells in the laboratory for future experiments . These could , for example , investigate whether human mesoderm cells develop in the same way as mice mesoderm cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2019	Distinct mesoderm migration phenotypes in extra-embryonic and embryonic regions of the early mouse embryo
To understand the neural origins of rhythmic behavior one must characterize the central pattern generator circuit and quantify the population size needed to sustain functionality . Breathing-related interneurons of the brainstem pre-Bötzinger complex ( preBötC ) that putatively comprise the core respiratory rhythm generator in mammals are derived from Dbx1-expressing precursors . Here , we show that selective photonic destruction of Dbx1 preBötC neurons in neonatal mouse slices impairs respiratory rhythm but surprisingly also the magnitude of motor output; respiratory hypoglossal nerve discharge decreased and its frequency steadily diminished until rhythm stopped irreversibly after 85±20 ( mean ± SEM ) cellular ablations , which corresponds to ∼15% of the estimated population . These results demonstrate that a single canonical interneuron class generates respiratory rhythm and contributes in a premotor capacity , whereas these functions are normally attributed to discrete populations . We also establish quantitative cellular parameters that govern network viability , which may have ramifications for respiratory pathology in disease states . Central pattern generator ( CPG ) circuits give rise to common behaviors such as swimming , walking , and breathing ( Grillner , 2006; Grillner and Jessell , 2009; Kiehn , 2011 ) . To understand the cellular origins of these behaviors , key problems are to identify the rhythmogenic and premotor populations , and then quantify the requisite number of neurons to sustain functionality . We address these issues in the mammalian breathing CPG by cumulatively ablating a genetically identified interneuron population hypothesized to form the rhythmogenic core while monitoring effects on network output in real time . The brainstem pre-Bötzinger complex ( preBötC ) putatively drives inspiratory breathing rhythms ( Smith et al . , 1991; Feldman et al . , 2013; Moore et al . , 2013 ) . These rhythms persist in reduced slice preparations that retain the preBötC and can be monitored via respiratory hypoglossal ( XII ) nerve discharge , providing a powerful in vitro model of breathing behavior ( Lieske et al . , 2000; Koizumi et al . , 2008; Funk and Greer , 2013 ) . Rhythmogenic preBötC interneurons are distinguished by glutamatergic transmitter phenotype or the expression of neuropeptides and peptide receptors , or their intersection ( Funk et al . , 1993; Gray et al . , 1999 , 2001; Guyenet et al . , 2002; Stornetta et al . , 2003; Wallén-Mackenzie et al . , 2006; Tan et al . , 2008 ) . Glutamatergic , peptidergic , and peptide receptor-expressing preBötC interneurons develop from a common set of precursors that express the homeobox gene Dbx1 during embryonic development . Dbx1-derived interneurons ( hereafter Dbx1 neurons ) in perinatal mice are inspiratory modulated , and Dbx1-null mice die at birth without ever breathing ( Bouvier et al . , 2010; Gray et al . , 2010; Picardo et al . , 2013 ) . Therefore , we—and others—proposed that Dbx1 preBötC neurons comprise the core inspiratory rhythm generator , i . e . , the Dbx1 core hypothesis . Previously , we estimated the cellular mass critical for respiratory rhythm generation by laser-ablating preBötC inspiratory interneurons identified by Ca2+ imaging . The destruction of ∼120 rhythmic neurons irreversibly stopped respiratory rhythmogenesis ( Hayes et al . , 2012 ) . However , inspiratory-modulated preBötC neurons may be excitatory or inhibitory . Ca2+ fluorescence changes cannot differentiate rhythmogenic glutamatergic neurons ( Funk et al . , 1993; Wallén-Mackenzie et al . , 2006 ) from GABA- or glycinergic neurons ( Kuwana et al . , 2006; Winter et al . , 2009 ) , which influence sensory integration and coordinated patterns of muscle contraction for respiratory behaviors , but are non-rhythmogenic because inspiratory rhythms in vivo and in vitro do not require synaptic inhibition ( Feldman and Smith , 1989; Brockhaus and Ballanyi , 1998; Kuwana et al . , 2003; Ren and Greer , 2006; Fujii et al . , 2007; Janczewski et al . , 2013 ) . Therefore , we predicted that the selective destruction of Dbx1 preBötC neurons , which are predominantly glutamatergic ( Bouvier et al . , 2010; Gray et al . , 2010 ) —unlike locomotor Dbx1 neurons in lumbar spinal cord of which ∼70% express inhibitory transmitters ( Lanuza et al . , 2004; Talpalar et al . , 2013 ) —would impair rhythmogenesis with a cell-ablation tally much lower than 120 ( Hayes et al . , 2012 ) . To test this prediction of the Dbx1 core hypothesis , we used photonics to map the positions of Dbx1 preBötC neurons , and then laser ablated them—one cell at a time—while continuously monitoring respiratory motor output ( Wang et al . , 2013 ) . As predicted , cumulative destruction of Dbx1 preBötC neurons progressively decreased respiratory frequency until rhythm ceased after ablation of ∼85 neurons . Surprisingly , cumulative Dbx1 cellular ablations also diminished the amplitude of respiratory XII nerve discharge , suggesting that Dbx1 preBötC neurons also influence motor output . In simulations that assign only rhythm-generating function to preBötC neurons , cumulative ablations decreased frequency and stopped rhythmogenesis but at much lower tallies and without perturbing the output amplitude . These ablation results and model-experiment discrepancies , combined with antidromic activation from the XII nucleus and axonal projection patterns , ascribe rhythm-generating and premotor roles to Dbx1 preBötC neurons . Thus , we demonstrate that one cardinal class of hindbrain interneurons serves two distinct roles in a key mammalian CPG; the data further establish quantitative cellular parameters that minimally ensure network functionality . Dbx1 neurons were detected and mapped within the preBötC , and then laser ablated individually , in sequence , while monitoring respiratory network functionality via XII motor nerve output . Experiments began with an ‘initialization phase’ that defined the domain for detection and ablation , which was bilateral . We used a Dbx1 Cre-driver line ( Dbx1CreERT2 ) coupled with floxed reporter mice ( Gt ( ROSA ) 26Sorflox-stop-tdTomato ) to locate Dbx1 neurons via fluorescence . Viewed in the transverse plane of slices that expose the preBötC at their surface ( i . e . , preBötC-surface slices ) , Dbx1 neurons form a bilaterally symmetrical V-shape starting dorsally at the border of the XII motor nuclei and continuing ventrolaterally to the preBötC ( Figure 1A and Figure 1—figure supplement 1A ) . Dorsally , the preBötC adjoins the semi-compact division of the nucleus ambiguus ( scNA ) ; the ventral border of the preBötC is orthogonal to the dorsal boundary of the principal sub-nucleus of the inferior olive ( IOPloop ) ( Ruangkittisakul et al . , 2011 , 2014 ) . These spatial relationships visible in bright field or epifluorescence allow us to pinpoint the preBötC ( Figure 1A ) . At the cellular level , identifying putative rhythmogenic neurons on the basis of fluorescent protein expression alone is acceptable because the overwhelming majority of Dbx1 preBötC neurons are inspiratory ( e . g . , Figure 1B ) ( Picardo et al . , 2013 ) . 10 . 7554/eLife . 03427 . 003Figure 1 . Dbx1 preBötC neurons . ( A ) Bright field ( left ) and fluorescence ( right ) images of the right half of a preBötC-surface slice preparation . Anatomical landmarks are illustrated including: XII , the hypoglossal motor nucleus; scNA , semi-compact nucleus ambiguus; IOPloop , the dorsal loop of the principal inferior olive; and the ventral border of the preBötC , which is orthogonal to the IOPloop . Scale bar is 300 µm . At right , the larger white box shows the detection and ablation domain . ( B ) Expansion of smaller white box in A , showing tdTomato expression in Dbx1 neurons and intracellular dialysis via patch pipette with Alexa 488 from the recorded neuron whose robust inspiratory discharge is illustrated at right ( scale bar is 10 µm ) . Respiratory motor output from the XII nerve is shown in raw and RMS-smoothed form . Voltage and time calibration bars represent 20 mV and 2 s . Baseline membrane potential in the recorded neuron was −60 mV . ( C ) Mask of targets showing validated Dbx1 interneuron targets ( red ) and regions of fluorescence that do not pass muster and were rejected as targets ( blue ) for focal planes at depths z = ( 30–60 µm ) . The region shown in each case maps to the 412 × 412 µm2 square shown by the larger white box in A ( right ) . Only a subset of the masks are shown for economy of display . ( D ) 3D reconstruction of detected targets for all focal planes z = ( 10–80 µm ) from the left and right preBötC . Each Dbx1 neuron is represented by a single red point centered on its soma . DOI: http://dx . doi . org/10 . 7554/eLife . 03427 . 00310 . 7554/eLife . 03427 . 004Figure 1—figure supplement 1 . Detection of Dbx1 preBötC neurons . ( A ) Fluorescent image of a transverse slice from a Dbx1+/CreERT2; Rosa26tdTomato mouse pup . Anatomical landmarks are illustrated including: XII , the hypoglossal motor nucleus; scNA , semi-compact nucleus ambiguus; and IOP , the principal inferior olive . The domain for detection and ablation is indicated by the white boxes , bilaterally . Scale bar is 500 µm . ( B ) Mask of targets showing validated Dbx1 ( red ) and invalidated ( blue ) cells for all focal planes to a depth of −80 µm . Each image is 412 × 412 µm2 ( as in Figure 1C ) . Image processing routines for detecting and validating Dbx1 neuron targets are detailed in ‘Materials and methods’ , Figure 1—figure supplement 2 , and a methodological paper ( Wang et al . , 2013 ) . Note that the highest fraction of validated Dbx1 target cells is found at deeper focal planes , e . g . , −80 µm due to the ‘priority rule’ , which applies to overlapping ROIs in adjacent focal planes . According to the priority rule , the ROI from the deeper focal is accepted as a ‘bona fide’ target and the redundant ROI at the superficial level is rejected . Also see Figure 1—figure supplement 2C , D . DOI: http://dx . doi . org/10 . 7554/eLife . 03427 . 00410 . 7554/eLife . 03427 . 005Figure 1—figure supplement 2 . Detection of Dbx1 neuron targets via fluorescence and image processing . ( A1 , B1 , C1 , D1 ) Images from the preBötC of Dbx1+/CreERT2; Rosa26tdTomato mice showing tdTomato in neurons derived from Dbx1-expressing precursors ( i . e . , Dbx1 neurons ) . Scale bar in A1 is 20 µm and applies to all panels . C1 and D1 show the same field of view at two different depths ( −20 and −10 µm , respectively ) . ( A2 , B2 , C2 , D2 ) Masks of ROIs obtained by analyzing the corresponding images above . Red ROIs are deemed valid targets by the circularity test , which evaluates somatic shape; blue ROIs that fail the circularity test are rejected . Circularity analyses distinguish somata from auto-fluorescent detritus ( A1 , A2 ) as well as contiguous soma-dendrite images ( B1 , B2 ) and isolated segments ( shafts ) of dendrites ( C1 , C2 , D1 , D2 ) . Non-somatic auto-fluorescence is rejected because it does not accurately indicate underlying neurons . Dendritic segments are not valid targets because they are difficult to target in the ablation phase of the experiments and their cell bodies are detectable in adjacent focal planes . Often , a cell rejected by the circularity test in one focal plane ( e . g . , C2 , graygray double arrowhead ) is validated in the adjacent plane ( D2 , graygray double arrowhead ) . When ROIs that pass the circularity test are detected in more than one focal plane , they are validated or rejected according to the priority rule . ROIs from a deeper focal plane ( −20 µm ) are validated by circularity and thus colored red ( C2 , circled ROIs ) . Subsequent detection of overlaying ROIs at the superficial focal plane ( −10 µm ) , which also pass the circularity test , are nonetheless rejected by the priority rule and thus colored blue ( D2 , circled ROIs ) . These criteria for target detection are more fully described in ‘Materials and methods’ and Wang et al . ( 2013 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03427 . 00510 . 7554/eLife . 03427 . 006Figure 1—figure supplement 3 . Average number of Dbx1 neurons detected at each acquisition depth from z = 0 ( surface ) to z = −80 µm in preBötC-surface slices and control slices with the ventral respiratory column ( VRC ) exposed at the slice surface . The number of Dbx1 neurons detected per focal plane per side ( in 10-µm increments of the focal plane ) is shown individually for each individual experiment ( graygray unfilled circles ) along with the mean ±SD for all experiments ( black unfilled squares with black lines showing SD ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03427 . 006 In the subsequent ‘detection phase’ , a visible wavelength laser scanned the domain and an iterative threshold-crossing algorithm analyzed the image to then draw regions of interest ( ROIs ) for putative cell targets based on fluorescence brightness . Potential targets were evaluated on the basis of shape to differentiate cell bodies from auto-fluorescent debris , and to reject the fluorescence from dendrites and neuropil whose somata were detectable in adjacent focal planes ( Wang et al . , 2013 ) ( Figure 1—figure supplement 2 ) . The map of ROIs for validated cell targets was retained at each focal plane ( Figure 1C—figure supplements 1B and 2 , red ROIs ) . Potential targets that did not meet these criteria were discarded ( but displayed for demonstration purposes in Figure 1C and Figure 1—figure supplements 1B and 2 , blue ROIs ) . Target detection was repeated at 10-µm increments through the z-axis and the final three-dimensional map of targets was stored in memory ( Figure 1D ) . Typically , we detected 26–50 Dbx1 neurons per focal plane per side ( Figure 1—figure supplement 3 ) for a total average number of 705 targets in the preBötC ( SD 119 , SEM 59 , range: 548 to 802 , n = 8 slices ) . During the ‘ablation phase’ of the experiments , Dbx1 neurons in preBötC-surface slices were randomly selected for photonic lesioning . Each target in the domain was individually spot scanned with a Ti:sapphire laser using maximum intensity 800-nm pulses until target destruction was confirmed by three forms of optical criteria ( Wang et al . , 2013 ) or was deemed a failure . Generally >90% of lesion attempts are successful ( Figure 2—figure supplement 1 ) ( Hayes et al . , 2012; Wang et al . , 2013 ) . Only confirmed lesions add to the running tally . The frequency and amplitude of inspiratory motor output diminished at the onset of the ablation phase ( Figures 2A and 3A ) . XII amplitude decreased steeply with the tally of ablated cells , and then stabilized at 44% of its pre-lesion value ( SD 4% , SEM 1% , suction electrode recordings are reported in normalized arbitrary units ) . Frequency , however , continued to decrease ( i . e . , cycle period increased ) throughout the ablation phase . Initially , within the first dozen ablations , the average decrease in respiratory frequency was nearly twofold , and it continued to fall steeply until rhythm cessation ( range of frequencies: 0 . 22–0 . 007 Hz , Figure 3B [inset shows bi-exponential increase in cycle period] , n = 5 slices ) . Furthermore , the rhythm destabilized during the ablation phase . We defined regularity score ( RS ) as the ratio of the present cycle period with respect to the mean period over 10 prior cycles ( see ‘Materials and methods’ for RS formula ) . Cycle-to-cycle variations in RS indicate irregularity; the system is trending slower when RS exceeds unity . The RS of preBötC-surface slices measured 1–9 during the ablation phase ( Figure 3C ) . Respiratory rhythm ceased altogether after an average of 85 confirmed Dbx1 neuron ablations in preBötC-surface slices ( SD 44 , SEM 20 , range 42–137 , n = 5 slices ) , well before exhausting the average list of 705 targets per slice . These ablations were bilateral and the tally reflects the sum of both sides ( Figure 2—figure supplement 2 ) . The representative experiment in Figure 2A shows rhythm cessation after 62 confirmed ablations , corresponding to 9% of the total 677 detected targets . 10 . 7554/eLife . 03427 . 007Figure 2 . Cumulative serial ablation of Dbx1 neurons in preBötC-surface slices ( A ) and control slices whose surface exposes the ventral respiratory column , not preBötC ( B ) . ( A and B ) The x-axis is a timeline . The y-axis plots XII amplitude ( normalized units , top ) and respiratory period ( bottom ) . The respiratory period axis is continuous ( 0–200 s ) but plotted with two scales . Major ticks are separated by 10 s from 0 to 20 s ( with unlabeled minor ticks at 5 s increments ) , and thereafter major ticks are plotted in 100 s divisions from 21 to 200 s ( with unlabeled minor ticks at 50 s increments ) . The discontinuity in the y-axis stops at 20 s ( lower portion ) and starts at 21 s ( upper portion ) . There is one data point for every individual respiratory period measured . The recording in A is no longer displayed after 6 min of XII quiescence . Substance P ( SP ) injection in A is displayed at higher sweep speed in Figure 5C . The recording in B is no longer displayed after 90 min of continuous stable XII output following the end of the ablation phase . Time calibrations in A and B are shown separately . DOI: http://dx . doi . org/10 . 7554/eLife . 03427 . 00710 . 7554/eLife . 03427 . 008Figure 2—figure supplement 1 . Cellular laser ablation and confirmation . ( A ) The image acquired during maximum-intensity Ti:sapphire laser scanning of the target cell with a 560–615 nm band-pass filter , which indicates cell destruction . This image was acquired with higher digital magnification compared to panels B–D; scale bar is 2 µm . ( B1–2 ) Images of native tdTomato expression in Dbx1 preBötC neurons before ( B1 ) and after ( B2 ) a single cell laser ablation . The target cell ( arrowhead ) is visible pre-lesion but not in the post-lesion image . Neighboring ( unlesioned ) neurons are unaffected . Scale bar of 10 µm applies to all images in B–D . ( C ) Bright field images of the target cell ( arrowhead ) prior to laser lesion . ( D1–5 ) Images of the target cell post-lesion ( arrowhead ) at 5-µm increments in the z plane . The focal plane in C was normalized to z = 0 µm for relative comparison with panels D1–5 . DOI: http://dx . doi . org/10 . 7554/eLife . 03427 . 00810 . 7554/eLife . 03427 . 009Figure 2—figure supplement 2 . Cumulative tally of laser ablations for preBötC-surface slices ( magenta ) and control slices whose surface exposes the ventral respiratory column , not preBötC ( cyan ) . The total tally and the individual side tallies are shown for each preparation . Black bars show the mean . For preBötC-surface slices , the tally was always lower on the side that was being lesioned when the rhythm stopped because rhythm cessation halted the ablation sequence . DOI: http://dx . doi . org/10 . 7554/eLife . 03427 . 00910 . 7554/eLife . 03427 . 010Figure 3 . Ablation effects on respiratory frequency and the amplitude of XII motor output . ( A–D ) Measurements are displayed in light grey and red for preBötC-surface slices and dark grey and blue for control slices that expose the ventral respiratory column ( VRC ) . ( A ) XII amplitude and ( B ) respiratory frequency for preBötC-surface and control slices are plotted vs cumulative percent of total lesions during the ablation phase ( bars show SD ) . Inset in B shows respiratory period in lieu of frequency ( bars show SD ) for preBötC-surface slices . ( C and D ) The regularity score ( RS ) is plotted vs cumulative percent of total lesions for preBötC-surface ( C ) and control slices ( D ) . B , C , and D are plotted on semi-log axes . B and C are labeled with subordinate ticks at 2 , 4 , and 6 . Tick labels are omitted from D because they match C exactly . B ( inset ) has linear axes . DOI: http://dx . doi . org/10 . 7554/eLife . 03427 . 010 Detection and ablation were similarly performed bilaterally in control slices whose rostral surface exposed the ventral respiratory column , which occupies a comparable domain for detection and ablation in the transverse plane , but this domain is ∼100 µm rostral to preBötC . The ventral respiratory column contains inspiratory and expiratory-modulated neurons that are not associated with rhythmogenesis ( Smith et al . , 1990; Feldman et al . , 2013 ) . During the initialization phase in control slices , the domain was centered on the highest density of fluorescent Dbx1 neurons bounded by the compact division of the nucleus ambiguus ( cNA ) dorsally and the ventral margin of the slice ( Figure 4 ) . We acquired 38–60 Dbx1 neuron targets per focal plane per side ( Figure 1—figure supplement 3 ) for a total average of 906 targets per slice ( SD 97 , SEM 34 , range: 722–1004 , n = 8 slices ) . 10 . 7554/eLife . 03427 . 011Figure 4 . Dbx1 neurons in the ventral respiratory column . ( A ) Bright field ( left ) and fluorescence ( right ) images of the right half of a control slice preparation . Anatomical landmarks are illustrated including: XII , the hypoglossal motor nucleus; cNA , the compact division of the nucleus ambiguus; IOPloop , the ventral portion ( loop ) of principal sub-nucleus of the inferior olive; and VRC , the ventral border of the ventral respiratory column . Scale bar is 300 µm . At right , the larger white box shows the detection and ablation domain . ( B ) Expansion of smaller white box in A , showing tdTomato expression in Dbx1 ventral respiratory column neurons ( scale bar is 10 µm ) , one of which was recorded . Intracellular dialysis via patch pipette with Alexa 488 is visible in the recorded neuron whose inspiratory depolarization and discharge pattern are illustrated at right . Respiratory motor output from the XII nerve is shown in raw and RMS-smoothed form . Voltage and time calibration bars represent 20 mV and 1 s . ( C ) Masks of targets showing validated Dbx1 interneuron targets ( red ) and regions of fluorescence that do not pass muster and were rejected as targets ( blue ) for focal planes at depths z = ( 40–70 µm ) . ( D ) 3D reconstruction of detected targets for all focal planes z = ( 0–80 µm ) in the ventral respiratory column ( VRC ) of the left and right side . A single red point centered on its soma represents each Dbx1 neuron . The highest fraction of accepted Dbx1 target cells is found at deeper focal planes ( see Figure 1—figure supplement 2 and ‘priority rule’ explained in ‘Materials and methods’ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03427 . 011 Dbx1 ventral respiratory column neurons were lesioned in random sequence during the ablation phase of control experiments . The amplitude of XII motor output decreased to 77% of control ( SD 2% , SEM 1% , normalized arbitrary units ) over the course of the ablation phase ( Figures 2B and 3A ) . Frequency did not change . It measured 0 . 36 Hz ( SD 0 . 2 Hz , SEM 0 . 04 Hz ) during the detection phase compared to 0 . 39 Hz ( SD 0 . 3 Hz , SEM 0 . 01 Hz ) during the ablation phase , which was not significant ( p=0 . 49 , Mann–Whitney U-test , Figure 3B ) . RS did not deviate from ∼1 throughout the ablation phase ( Figure 3D , n = 8 slices ) . These data indicate that cumulative sequential ablation of Dbx1 ventral respiratory column neurons has no effect on the stability or the period of the respiratory cycle . The ablation protocol exhausted the entire set of targets in every control experiment without stopping the rhythm ( e . g . , 923 confirmed ablations in Figure 2B ) . Ablations in control slices were also performed bilaterally , where the tally reflects the sum of both sides ( Figure 2—figure supplement 2 ) . Cumulative deletion of Dbx1 preBötC neurons appeared to degrade respiratory oscillator function . Nonetheless , an alternative explanation could involve the loss of excitatory drive ( rather than destruction of CPG core circuitry ) . Dbx1 preBötC neurons express neurokinin-1 peptide receptors ( NK1Rs ) ( Bouvier et al . , 2010; Gray et al . , 2010 ) that stimulate respiratory rhythmogenesis ( Gray et al . , 1999; Pagliardini et al . , 2005; Ballanyi and Ruangkittisakul , 2009 ) . Monoaminergic and peptidergic raphé neurons project to , and receive feedback projections from , the preBötC to elevate excitability in the respiratory network ( Ptak et al . , 2009 ) . Therefore , laser ablation in the preBötC could break the link with the raphé and thus diminish excitatory drive . To test that idea , we exposed the lesioned preBötC to a bolus of neuropeptide substance P ( SP , 1 mM ) after the respiratory cycle period exceeded 120 s , which we previously determined was a reliable benchmark of a slice that would cease rhythmic function within 5–10 min without pharmacological intervention ( Hayes et al . , 2012 ) . First , as a control , we applied SP to unlesioned preBötC-surface slices , which transiently increased respiratory frequency , i . e . , lowered the cycle period from 4 . 7 s ( SD 0 . 8 s , SEM 0 . 1 s ) to 3 . 3 s ( SD 0 . 5 s , SEM 0 . 1 s , average period computed for 25 cycles ) , and then equilibrated in 21 min ( SD 4 min , SEM 2 min , n = 4 slices ) ( Figure 5A ) . The regularity score was ∼1 throughout the bout , which is consistent with the stable rhythm expected in unlesioned slices ( Figure 5B ) . Then , SP was injected into five preBötC-surface slices wherein the cumulative laser ablation of Dbx1 neurons caused 120 s of quiescence . SP transiently revived respiratory rhythm; the average cycle period was 1 . 7 s ( SD 0 . 4 s , SEM 0 . 2 s , computed for 10 cycles after SP bolus injection ) but the cycle period slowed down rapidly , surpassing the control period previously measured during the detection phase ( 4 . 7 s ) within 3 min ( SD 2 min , SEM 1 min ) . Cycle period continuously lengthened and fluctuated from cycle to cycle , and then the rhythm stopped altogether ( Figure 5C ) . Judged on the basis of equilibration time , the transient effects of SP were significantly briefer in lesioned slices ( p=0 . 02 , Mann–Whitney U-test ) . Furthermore , lesioned slices ultimately fell inexorably silent ( Figures 2A and 5C ) , whereas unlesioned slices maintained rhythmicity for 4–6 hr ( Figure 5A ) . More importantly , the SP-evoked activity in lesioned preBötC-surface slices was irregular: RS measured 2–10 ( Figure 5D , compare to unlesioned slice in Figure 5B ) . These data indicate that NK1R-expressing Dbx1 neurons can evoke transient cycles of respiratory activity , but the loss of ∼85 Dbx1 preBötC neurons slows the respiratory oscillator frequency and renders XII motor output nonfunctional . 10 . 7554/eLife . 03427 . 012Figure 5 . Substance-P ( SP ) injections in preBötC-surface slices . ( A ) SP bolus injected in an un-lesioned preBötC-surface slice . XII output magnitude is plotted with cycle period as a time series . ( B ) Semi-log plot of regularity score ( RS ) for 30 min after SP injection from the slice preparation in A . RS axis is continuous but plotted with two scales . ( C ) preBötC-surface slice shown in the acquisition phase ( left ) and during the ablation phase ( right ) , which were separated by a time gap of 3 hr . After 120 s of quiescence ( data point circled in red ) , SP injection revived the rhythm transiently . ( D ) Semi-log plot of RS for 15 min after SP injection from the slice preparation in C . Data in C and D were from the same preparation as in Figure 2A . DOI: http://dx . doi . org/10 . 7554/eLife . 03427 . 012 We used graph theory and simulations to investigate how Dbx1 neuron ablations affect preBötC structure and function . The Rubin–Hayes preBötC neuron model ( Rubin et al . , 2009 ) was assembled in Erdős-Rényi G ( n , p ) graphs ( Newman et al . , 2006 ) with population sizes n from 200–400 and connection probabilities p from 0 . 1 to 0 . 2 . These parameter ranges encompass n = 325 , an empirical estimate of the number of excitatory neurons in the perinatal mouse preBötC ( Hayes et al . , 2012 ) as well as p=0 . 13 , the only experimentally determined connection probability among putatively rhythmogenic preBötC neurons in acute mouse slices ( Rekling et al . , 2000 ) . Networks within the above n–p parameter range that generated respiratory-like cycle periods of ∼4 s are shown with asterisks in Figure 6A and Figure 6—figure supplement 1A . This set of model networks also generated network-wide bursts within 200–300 ms following brief focal glutamatergic stimulation of five or more constituent neurons ( Figure 6B ) in agreement with focal glutamate un-caging experiments in neonatal mouse slices , which showed that simultaneous stimulation of 4–9 preBötC neurons can trigger inspiratory bursts with similar latency ( Kam et al . , 2013b ) . These results substantiate that the model networks well represent the neonatal mouse preBötC in vitro . 10 . 7554/eLife . 03427 . 013Figure 6 . Numerical simulations . ( A ) Networks of Dbx1 preBötC neurons with population size ( n ) and synaptic connection probability ( p ) . Blocks show the mean cycle period according to the colorimetric scale ( right ) for 10 ( or more ) realizations of the network for each ( n , p ) pair . Asterisks denote networks that generated respiratory-like ( ∼4 s ) cycle periods in ≥80% of individual realizations of the network . ( B ) Focal glutamatergic stimulation of constituent neurons in a model network ( n , p ) = ( 330 , 0 . 125 ) . Network-wide bursts can be evoked when five or more individual cells are stimulated . These simulations mimic holographic laser-mediated glutamate un-caging experiments ( Kam et al . , 2013b ) and are included because they bolster confidence that our model networks accurately capture features and behaviors of the preBötC in newborn mice . Raster plots show spike activity in six constituent neurons randomly selected from the network and focally stimulated ( see ‘Materials and methods’ for numerical simulation of glutamate un-caging protocol ) . If focal stimulation evoked EPSPs ( not spikes ) then the raster reports ‘EPSPs’; spikes are indicated by short vertical lines . From left to right , the number of stimulated units increments by one; five ( or more ) units evoked an inspiratory-like burst . A running-time histogram of network activity is shown at the bottom . Calibration bars represent 100 spikes/10-ms bin ( vertical ) and 0 . 5 s ( horizontal ) . ( C ) Running-time histogram for one simulation of sequential ablation in a network ( n , p ) = ( 330 , 0 . 125 ) . Cell ablation tally is shown ( top ) . Time calibration is 30 s . Spikes-per-bin calibration bar is the same as the inset ( lower ) , 100 spikes/10-ms bin . Insets show a raster plot of spike activity in the entire network with a running-time histogram . The numerical y-axis reports cell index for each neuron model in the network . Left inset shows the first ablation ( magenta arrow ) . Right inset shows all cumulative 36 ablations ( magenta arrows ) . Time calibration for both insets is 1 s ( at right ) . ( D ) Cycle period and ( E ) spikes-per bin ( i . e . , a measure of the magnitude of simulated network output as in C ) are plotted vs cumulative percent of total ablations for 10 networks with ( n , p ) = ( 330 , 0 . 125 ) . D plotted in semi-log axes , E in linear axes . Magenta shows data from individual networks , cyan plots the mean response . DOI: http://dx . doi . org/10 . 7554/eLife . 03427 . 01310 . 7554/eLife . 03427 . 014Figure 6—figure supplement 1 . Numerical simulations of Dbx1 neuron laser ablation experiments . Networks of Dbx1 preBötC neurons parameterized by population size ( n ) and synaptic connection probability ( p ) . Erdős-Rényi random directed graphs G ( n , p ) ( Newman et al . , 2006 ) determined the underlying connectivity structure . Each node in G ( n , p ) was populated by a Rubin-Hayes preBötC neuron model ( Rubin et al . , 2009 ) with dynamic excitatory synapses for links . Each block in the panels reports a measure of network performance . ( A ) The grey scale reports the percent of model networks that generated spontaneous rhythmic activity . Asterisks denote networks that generated respiratory-like cycle periods in ≥80% of individual realizations , which were then subjected to simulated laser ablation experiments ( results in B and C ) . ( B ) The colorimetric scale reports the mean cycle period for 10 ( or more ) realizations of the network for each ( n , p ) pair ( same as Figure 6A and panel C ) . Networks with asterisks ( from Figure 6A and this figure's panel A ) were subject to laser ablations in random sequence; the numbers in the blocks report the average final cycle period ( in s ) prior to rhythm cessation in the lesioned network at each ( n , p ) pair . ( C ) The numbers in the blocks report the average cell ablation tally at the point of rhythm cessation for five or more laser ablation simulations . DOI: http://dx . doi . org/10 . 7554/eLife . 03427 . 014 Sequentially deleting neurons in the model networks decreased frequency until the rhythm stopped altogether ( Figure 6C , D , and Figure 6—figure supplement 1B; Supplementary file 1 ) . These simulations generally agreed with the experimental results except for two discrepancies . First , the amplitude of network output did not diminish ( compare Figure 6C , E to Figures 2A and 3A ) . Second , rhythm cessation required the average deletion of 41 constituent model neurons ( SD 15 , SEM 6 , range 19–67 , Figure 6—figure supplement 1B , C; Supplementary file 1 ) as opposed to the experimental cell ablation tally of ∼85 . First , we examine the loss of rhythmic function , and then address these discrepancies . To assess whether a collapse of network structure could explain the breakdown in rhythmicity , we computed canonical local and global measures of topology for the graph G ( n , p ) underlying each network simulation . These measures were computed after each cellular deletion and thus tracked continuously in parallel with the simulated networks . The table in Supplementary file 2 reports the value of each topological measure prior to any deletions and after the final deletion associated with rhythm cessation ( see ‘Materials and methods’ for definitions and computational methods ) . From the start to finish , the cumulative ablation sequence caused no major change in local metrics including cluster coefficient , closeness centrality , and betweenness centrality . Global connectivity metrics such as the K-core , which has been applied to analyze rhythmic neural systems including the preBötC ( Schwab et al . , 2010 ) , showed only modest changes that were incommensurate with the large changes in frequency observed in experiments and simulations . The number of strongly connected components in the model networks did not depart from unity , thus the underlying graph was not fractured and every constituent interneuron could be reached via a finite number of synaptic links from every other interneuron , even after the rhythm stopped . These calculations show that these relatively low numbers of cumulative cellular ablations do not disconnect or disintegrate the core CPG , which suggests that a breakdown in network structure cannot explain the impairment and cessation of rhythmic function . The alternative is that neurons and synapses confer non-linear functional properties to the underlying rhythmic system that are not captured by the graph connectivity alone ( see ‘Discussion’ ) . Lower cell ablation tallies perturbed and stopped the rhythm in simulations , and the aggregate burst magnitude did not decline ( Figure 6C , E , and Figure 6—figure supplement 1C , Supplementary file 1 ) . Both disparities could be explained if a subset of the experimentally lesioned population consists of premotor—rather than rhythmogenic—interneurons . Thus , we tested whether Dbx1 preBötC neurons project to the XII motor nucleus . Of the eight Dbx1 neurons with inspiratory modulation ( Figure 7A–D ) , two could be antidromically activated by stimulation within the XII nucleus . Figure 7 shows representative data from such a Dbx1 neuron whose XII-evoked antidromic spike was extinguished by collision with an orthodromic spike triggered by a somatic current pulse ( Figure 7E ) . Most Dbx1 preBötC neurons are inspiratory and show commissural axons that cross the midline and innervate the contralateral preBötC ( Figure 8A–C ) , as shown previously ( Bouvier et al . , 2010; Picardo et al . , 2013 ) . Here , we identify Dbx1 preBötC neurons that are also inspiratory modulated but send axons ipsilaterally toward the XII nucleus ( Figure 8D–F and Figure 8—figure supplement 1 ) , consistent with a role related to premotor transmission of inspiratory drive from preBötC to XII motoneurons . 10 . 7554/eLife . 03427 . 015Figure 7 . Dbx1 preBötC neurons with premotor function . ( A ) Fluorescence and ( B ) bright field images of a slice preparation . Anatomical landmarks are illustrated including: XII , the hypoglossal motor nucleus; scNA , semi-compact division of the nucleus ambiguus; IOPloop , the ventral loop of the principal inferior olive , and the ventral surface of the preBötC . Scale bar is 100 µm and applies to A and B . A patch-recording pipette is visible , marking the inspiratory-modulated neuron detailed in C–E . A dotted circle indicates the tip of the pipette and cell body . ( C and D ) tdTomato expression , intracellular dialysis of Alexa 488 , and merged image ( C ) from the inspiratory neuron shown with XII nerve output ( D ) . Voltage and time calibration bars represent 20 mV and 1 s . Baseline membrane potential in the recorded neuron was −60 mV . ( E ) Antidromic activation of the Dbx1 inspiratory neuron from C and D . Action potentials were evoked by XII stimulation ( left ) and intracellular 5-ms supra-threshold current pulses ( middle ) . When the antidromic XII stimulus was preceded immediately by a supra-threshold intracellular current pulse , the antidromic spike was occluded ( collision test , right ) . Several sweeps , all from a −62 mV baseline membrane potential , are superimposed with vertical offset in each case . Voltage calibration is the same as panel D . Applied current ( Iapp ) calibration is shown . Time calibration bar for E is 25 ms . DOI: http://dx . doi . org/10 . 7554/eLife . 03427 . 01510 . 7554/eLife . 03427 . 016Figure 8 . Commissural and premotor projections of inspiratory Dbx1 preBötC neurons . ( A ) Biocytin-filled and reconstructed Dbx1 preBötC neuron with commissural axon projection . The axon , which meanders in depth in this confocal image stack , was digitally traced ( yellow ) and superimposed in one plane for display . Axon trajectory crosses the midline of the slice and enters the preBötC contralaterally . Scale bar is 25 µm . ( B ) Mosaic image of the entire slice . The biocytin-filled soma ( green ) of neuron in A is shown at lower right ( white arrow ) . Scale bar is 200 µm . Panels A and B have exactly the same orientation ( dorsal up , ventral down ) . ( C ) Inspiratory discharge from the neuron in A and B . Top trace is membrane potential of the recorded Dbx1 preBötC neuron . Lower trace is XII output . Scale bars are 10 mV and 0 . 5 s . ( D ) Biocytin-filled and reconstructed Dbx1 preBötC neuron that projects toward the XII motor nucleus . Scale bar is 25 µm . The axon remained largely coplanar and thus is readily visible , except near its distal tip . In Figure 8—figure supplement 1 , this same neuron is shown with a digitally traced ( yellow ) axon superimposed on the confocal image . ( E ) Mosaic image of the entire slice . Neuron in D is shown at lower left ( white arrow ) . Scale bar is 200 µm . Panels D and E have exactly the same orientation ( dorsal up , ventral down ) . ( F ) Inspiratory discharge from the neuron in D and E . Top trace is membrane potential of the recorded Dbx1 preBötC neuron . Lower trace is XII output . Scale bars are 10 mV and 0 . 5 s . DOI: http://dx . doi . org/10 . 7554/eLife . 03427 . 01610 . 7554/eLife . 03427 . 017Figure 8—figure supplement 1 . Magnified view of the Dbx1 preBötC neuron from Figure 8D-F in which the axon has been digitally traced in the confocal stack and superimposed over the image to better illustrate the axon projection toward the XII motor nucleus . Scale bar is 25 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03427 . 017 Two-photon lasers can destroy cells of a well-defined class with minimal damage to surrounding tissues ( Eklöf-Ljunggren et al . , 2012; Wang et al . , 2013 ) . Here , we use this technique to study the contribution of Dbx1 neurons in slices that capture essential components of the breathing CPG and generate measurable motor nerve output . Target detection relies on native fluorescent protein expression . An overwhelming majority of Dbx1 neurons in the ventral medulla have a glutamatergic transmitter phenotype and inspiratory modulated firing patterns ( Bouvier et al . , 2010; Gray et al . , 2010; Picardo et al . , 2013 ) , so the Cre/lox Dbx1 reporter system is a reliable means to identify neurons with inspiratory function and target them for laser ablation . Dbx1 is also expressed in rostral parts of the ventral respiratory column between the caudal pole of the facial nucleus and the preBötC along the anterior–posterior axis ( Feldman et al . , 2013; Gray , 2013 ) . The ventral respiratory column contains auxiliary inspiratory neurons ( Figure 4 ) ( Smith et al . , 1990; Ballanyi et al . , 1999; Barnes et al . , 2007 ) , which served as a control population . Laser ablating these neurons that do not have significant rhythmogenic function facilitates a comparative analysis of Dbx1 neuron ablations at the level of the preBötC . Laser ablation of Dbx1 neurons in the ventral respiratory column had a minor effect on XII motor output amplitude and negligible effects on frequency and regularity . These negative results show that laser–tissue interactions are not generally deleterious for respiratory function in vitro ( Eklöf-Ljunggren et al . , 2012; Wang et al . , 2013 ) . Periodicity is the hallmark feature of an oscillator . Here , sequential laser ablation of Dbx1 preBötC neurons steadily diminished the inspiratory burst frequency , caused cycle period fluctuations , and ultimately the cessation of rhythmic motor output . We conclude that the oscillator was continuously degraded until it could no longer sustain spontaneous function . These data strengthen the proposal that Dbx1 neurons comprise the core inspiratory rhythm generator , which was originally based on Dbx1 knockout mice that fail to breathe at birth ( Pierani et al . , 2001 ) , and an array of neuroanatomical and physiological criteria including glutamatergic transmitter phenotype , the expression of peptides and peptide receptors , strong inspiratory rhythmic phenotype , and the ability to synchronize the preBötC bilaterally ( Bouvier et al . , 2010; Gray et al . , 2010; Picardo et al . , 2013 ) . We previously laser-ablated rhythmic preBötC neurons identified by Ca2+ imaging . In that study , deleting all the detected targets ( 120 on average ) slowed , destabilized , and then stopped the rhythm ( Hayes et al . , 2012 ) . The interpretability of these experiments suffered two caveats: the rhythm stopped after a delay of ∼30 min following the final target ablation , and furthermore , the transmitter phenotype of the ablated targets was unknown . In this study , use of the Dbx1 Cre-driver line ensured that the target neurons were glutamatergic , a requisite characteristic for respiratory rhythmogenic function ( Greer et al . , 1991; Funk et al . , 1993; Ge and Feldman , 1998; Shao et al . , 2003; Wallén-Mackenzie et al . , 2006 ) . And here , destroying an average of 85 Dbx1 neurons stopped the XII motor rhythm in the midst of the ablation phase , before exhausting the target list , which suggests a more direct impact on the core rhythmogenic circuit . We cannot rule out the possibility that subsets of preBötC neurons generate ‘burstlets’ observable in local field recordings ( Kam et al . , 2013a ) . However , there is no collective inspiratory motor output after sequential laser ablation of Dbx1 preBötC interneurons , which indicates that the CPG is nonfunctional . Here , the debilitating effects on respiratory rhythm generation at a much lower ablation tally suggest that the preBötC core in vitro is very sensitive to the loss of just a few constituent interneurons ( i . e . , <100 ) . This sensitivity to neuron loss may be accentuated in reduced slice preparations lacking excitatory and neuromodulatory drive from the rostral medulla and pons , as well as peripheral chemosensory and mechanosensory feedback via the vagus nerve . Extrinsic sources of drive raise preBötC network excitability and enhance respiratory rhythm . An acute sensitivity to neuron loss , such as we report for slices , may not apply to the preBötC network in vivo , but this remains to be tested via quantitative cellular ablation experiments with physiological monitoring in intact animal models . We detected an average of 705 Dbx1 target cells in preBötC-surface slices , but we conclude that a significant number were non-rhythmogenic . Some fraction of the detected targets can be discounted as Dbx1-derived non-rhythmogenic glia ( Gray et al . , 2010 ) . However , more significantly , some fraction manifests premotor function . The present evidence for premotor function in Dbx1 preBötC neurons with verified inspiratory discharge patterns ( e . g . , Figures 7 and 8D–F ) is consistent with large-scale pressure ejections of biocytin in the preBötC region of another strain of Dbx1-reporter mice ( Dbx1LacZ knock-in ) , which labeled many midline-crossing axons as well as axons projecting to the XII nucleus ( Bouvier et al . , 2010 ) . Even 20 years ago it was recognized that a fraction of the excitatory neurons in the preBötC , and immediately dorsal to preBötC , had premotor functionality ( Funk et al . , 1993 ) . Because laser ablations in preBötC-surface slices decreased XII magnitude ( e . g . , Figures 2A and 3A ) , we propose that a non-negligible fraction of the ablated Dbx1 neurons were inspiratory modulated but non-rhythmogenic , and most likely constitute XII premotor neurons ( Peever et al . , 2002; Chamberlin et al . , 2007; Koizumi et al . , 2008; Volgin et al . , 2008 ) . This scenario explains why there was a decline in XII motor output in experiments ( deletion of Dbx1 premotor neurons causes motor output to decline ) that was not mimicked by model simulations of a pure rhythmogenic circuit . It also explains why sequential ablations in simulations perturbed and stopped the rhythm at much lower cell ablation tallies; a significant fraction of the photonically ablated Dbx1 neurons were unrelated to rhythmogenesis per se . Although we lack quantitative certainty , if we assume that each of the two caveats above ( i . e . , the existence of Dbx1 glia and premotor neurons ) explains ∼10% of the detected targets , then the size of the essential preBötC core would be N = 705 – 2 [0 . 1 ( 705 ) ] = 564 , which is remarkably close to the estimate of ∼600 from adult rat studies that enumerated the population size based on NK1R expression in the preBötC ( Gray et al . , 2001 , 1999 ) . In our previous laser ablation study , we estimated population size to be ∼325 ( Hayes et al . , 2012 ) , which probably underestimates the population size because incomplete fluorescent Ca2+ dye loading in slices precludes the detection of a significant fraction of the rhythmogenic preBötC network . Our data suggest that the preBötC contains rhythmogenic and premotor interneurons that both develop from Dbx1-expressing precursors . It is surprising that Dbx1 neurons play these two roles in respiration when the role of Dbx1 neurons in spinal locomotor systems seems to be coordinating left–right limb alternation at any speed ( Lanuza et al . , 2004; Talpalar et al . , 2013 ) rather than rhythm generation or premotor transmission . In that regard , Dbx1 preBötC neurons appear to have more in common with excitatory Shox2 interneurons of the lumbar spinal cord ( a subset of V2 interneurons ) , which contribute to locomotor rhythm generation and premotor circuits downstream of the rhythm-generating core ( Dougherty et al . , 2013 ) . The present measurements imply that destroying on average X¯=85 of N = 564 Dbx1 preBötC neurons ( 15% ) precludes spontaneous respiratory motor rhythm in vitro . The mean and its 95% confidence intervals are expressed as follows: X¯± ( Zα/2σn ) , where Zα/2 is the cutoff value for a two-tailed normal distribution with probability α=0 . 05 , and σn is standard error . Thus , we conclude the ability of Dbx1 preBötC neurons to spontaneously generate rhythmic respiratory motor output in slice preparations is sensitive to deletion of 85 ± 1 . 96 ( 20 ) Dbx1 interneurons , i . e . , 8–22% of its core . Although it cannot function spontaneously post-lesion , the rhythmogenic preBötC core does not appear to be destroyed by piecewise lesioning . Peptide injections evoked irregular transient bursts in lesioned preBötC-surface slices . Also , measures of local and global connectivity in lesioned network models remained undiminished after cumulative cell deletions stopped and precluded rhythmic function . Therefore , we propose that preBötC function depends on non-trivial properties that emerge from non-linear synaptic and intrinsic membrane properties . Although such properties remain to be definitively determined , we advocate—and have explicitly modeled—a ‘group pacemaker’ rhythmogenic mechanism . In a group pacemaker , each constituent neuron forms recurrent connections with other constituent neurons in a network of finite size and amplifies excitatory drive via synaptically triggered inward currents ( Rekling et al . , 1996; Rekling and Feldman , 1998; Rubin et al . , 2009 ) . If that is a viable explanation for rhythmogenesis , then it could account for the loss of spontaneous function in the laser ablation context . Far before the cell ablation tally destroys the underlying network and its connectivity , the removal of each constituent neuron that contributes to rhythmic burst generation through its ability to amplify synaptic drive has a profound and deleterious effect on network functionality . The likelihood that arrhythmic respiratory networks retain considerable numbers of constituent neurons and interconnectivity suggests that unraveling the cellular and synaptic mechanisms of rhythmogenesis and motor output could be exploited to restore functionality in lesioned slices and , to the extent that our observations apply in vivo , to develop clinical therapies that bolster respiratory function in pathological conditions of animal models or human patients . Dbx1 respiratory neurons in the medulla represent excellent potential targets for pharmacological intervention or gene therapy to treat respiratory pathologies . Potentially enhancing premotor functionality in Dbx1-derived neurons could ameliorate obstructive sleep apnea . Boosting the function of rhythmogenic Dbx1 neurons may mitigate central apneas of prematurity as well as opiate respiratory depression . Treatment strategies aimed at rhythmogenic Dbx1 neurons may help overcome the effects of a reduced quantity or efficacy of neurons within the preBötC due to neurodegenerative diseases or aging ( Benarroch , 2003; Benarroch et al . , 2003; Tsuboi et al . , 2008 ) . The Institutional Animal Care and Use Committee at The College of William & Mary , which ensures compliance with United States federal regulations concerning care and use of vertebrate animals in research , approved the following protocols . The anesthesia and surgery protocols are consistent with the 2011 guidelines of the Animal Research Advisory Committee , which is part of the Office of Animal Care and Use of the National Institutes of Health ( Bethesda , MD ) . We used transgenic mice that express Cre recombinase fused to the tamoxifen-sensitive estrogen receptor ( CreERT2 ) in cells that express the Dbx1 gene ( Dbx1+/CreERT2 ) ( Hirata et al . , 2009; Gray et al . , 2010; Picardo et al . , 2013 ) . Dbx1+/CreERT2 mice were coupled to floxed reporter mice whose Rosa26 locus was modified by targeted insertion of a loxP-flanked STOP cassette followed by tandem dimer ( td ) Tomato ( Gt ( ROSA ) 26Sorflox-stop-tdTomato , i . e . , Rosa26tdTomato , Jax No . 007905 ) ( Madisen et al . , 2010 ) . Tamoxifen administration to pregnant females on the tenth day after the plug date produces bright native fluorescence in Dbx1-derived neurons ( i . e . , Dbx1 neurons ) in ∼50% of the offspring: Dbx1+/CreERT2; Rosa26tdTomato . Dbx1 neurons can be visualized via native fluorescence in the preBötC and contiguous regions of the medulla . The Dbx1+/CreERT2 heterozygous line has a CD-1 background . The Rosa26tdTomato line is homozygous with C57BL/6J background . We verified animal genotype via real-time PCR using primers specific for Cre and tandem dimer red fluorescent protein . Neonatal pups aged postnatal days 0–5 ( P0–5 ) were anesthetized for at least 4 min of immersion in crushed ice in order to render the animals insentient to the same degree as would occur with inhalation anesthetics ( Danneman and Mandrell , 1997; Fox et al . , 2007 ) . Anesthesia via hypothermia facilitates the rapid isolation of the intact brainstem and spinal cord , which would otherwise be damaged by cervical dislocation . The brainstem and spinal cord were removed within 90 s and then dissected in a dish filled with artificial cerebrospinal fluid containing ( in mM ) : 124 NaCl , 3 KCl , 1 . 5 CaCl2 , 1 MgSO4 , 25 NaHCO3 , 0 . 5 NaH2PO4 , and 30 d-glucose , equilibrated with 95% O2 and 5% CO2 ( pH = 7 . 4 ) . After removing the meninges and isolating the XII nerve rootlets , the brainstem and contiguous upper cervical spinal cord were fixed in position on a paraffin-coated paddle , or glued to an agar block , with rostral side up . The paddle or block was mounted to the vise of a vibrating microtome . The advancing blade approached the ventral surface of the tissue for sectioning in the transverse plane . XII nerve rootlets remained visible during the sectioning sequence . We cut a single slice of thickness 400–450 µm , which invariably retained the preBötC , XII premotor neurons and XII motoneurons that modulate and control airway resistance during breathing . We employed two discrete slice-cutting strategies to differentially expose respiratory nuclei at the slice surface , as previously described ( Hayes et al . , 2012 ) . The first slice type exposed the preBötC at the rostral face , and thus is called a preBötC-surface slice . The second slice type exposes the ventral respiratory column ∼100 µm rostral to the preBötC at the rostral slice surface and served as a control slice for laser ablations . Histology atlases for newborn mice were used to calibrate slices online during sectioning ( Ruangkittisakul et al . , 2011 , 2014 ) . For premotor recording experiments ( Figure 7 ) , we modified the preBötC-surface slice for the whole cell recordings such that the preBötC was exposed on the caudal surface . Slices were perfused with 27°C ACSF at 4 ml/min in a recording chamber on a fixed stage upright microscope . The external K+ concentration was raised to 9 mM and inspiratory motor output was recorded from XII nerve roots using a suction electrode and an AC-coupled differential amplifier . The amplified electrical signal and a root-mean-squared ( smoothed ) version of the signal were recorded by a 16-bit analog-to-digital converter and stored on a digital computer . Because the composition of neural circuits at the rostral surface of the slice is critical for data interpretation , we fixed and stained each slice used for ablations at the end of the experiment to more precisely benchmark the neuroanatomical boundaries of respiratory-related nuclei according to the respiratory brainstem mouse atlases referred to above ( Ruangkittisakul et al . , 2011 , 2014 ) . Fixation solution contained 4% paraformaldehyde in phosphate buffer ( 33 mM NaH2PO4 and 67 mM Na2HPO4 , pH = 7 . 2 ) . After 1-hr in fixation solution , slices were rinsed in phosphate buffer for 2 min , and then submerged for 60–75 s in staining solution containing 1% thionin acetate , 0 . 1 M sodium acetate trihydrate , and 0 . 1 M acetic acid . After washing in a series of ethanol solutions , slices were mounted in a well slide , obliquely illuminated , and digitally imaged via stereomicroscope . Control slices were characterized by the compact division of the nucleus ambiguus ( cNA ) , a thick dorsal inferior olive ( IOD ) , and a minimally developed principal loop of the inferior olive ( IOPloop ) . The preBötC-surface slices were characterized by very little ( if any ) visible portion of the cNA yet a clear semi-compact nucleus ambiguus ( scNA ) , a fully developed IOPloop , as well as medial inferior olive ( IOM ) . We performed whole cell recordings using a Dagan ( Minneapolis , MN ) IX2-700 current-clamp amplifier . Patch pipettes were fabricated from borosilicate glass ( OD: 1 . 5 mm , ID: 0 . 87 mm , 4–6 MΩ in the bath ) and filled with solution containing ( in mM ) : 140 K-gluconate , 10 HEPES , 5 NaCl , 1 MgCl2 , 0 . 1 EGTA , 2 Mg-ATP , 0 . 3 Na-GTP , 50-µM Alexa 488 hydrazide , and 2-mg/ml biocytin . Empirical measurement of the liquid junction potential was 1 mV and thus not corrected . Access ( series ) resistance was ∼10–15 MΩ , which was countered by bridge balance . Conventional current-clamp analog recordings were digitized at 4 kHz with a 16-bit A/D converter after 1 kHz low-pass filtering ( PowerLab , AD Instruments , Colorado Springs , CO ) . Neurons were selected for recording based on native tdTomato fluorescence in neurons preferentially in the dorsal preBötC . After identifying an inspiratory Dbx1 preBötC neuron , we tested for antidromic activation using a concentric bipolar electrode ( FHC Inc . , Bowdoin , ME ) placed at the surface of the XII nucleus . Stimuli were triggered by a pulse generator ( Tenma TGP110 10 MHz Pulse Generator , Aim-TTi USA , Fairport , NY ) and amplitude and polarity were controlled by a stimulus isolation unit ( Iso-Flex , AMPI , Jerusalem , Israel ) . We applied cathodic stimuli at increasing intensities , to a maximum of 0 . 4 mA , to elicit short latency antidromic action potentials . Then , brief ( 1 ms ) current pulses , at magnitudes at or exceeding rheobase , were applied before the antidromic stimulation such that both ortho- and antidromic spikes were evoked . The delay between stimuli was progressively decreased until a collision was observed , i . e . , the antidromic spike was occluded . Biocytin-loaded neurons were fixed in 4% paraformaldehyde in 0 . 1 M Na-phosphate buffer for at least 16 hr at 4°C . Then , the slices were treated with Scale solution containing 4 M urea , 10% ( mass/volume ) glycerol and 0 . 1% ( mass/volume ) Triton X-100 , for 10 days to clear the tissue and remove opaque background staining ( Hama et al . , 2011 ) . Slices were then washed in phosphate buffered saline ( PBS ) for 1 hr , followed by a 15-min cycle with PBS containing 10% heat-inactivated fetal bovine sera ( F4135; Sigma-Aldrich ) . Next , slices were incubated in PBS containing fetal bovine sera with additional 1% Triton X-100 . Finally , the slices were incubated in FITC ( i . e . , fluorescein-isothiocyanate ) -conjugated ExtrAvidin ( E2761; Sigma-Aldrich ) overnight at 4°C , and then rinsed twice with PBS , followed by six 20-min washes in PBS , and then cover-slipped in Vectashield ( H-1400 Hard Set , Vector Laboratories , Burlingame , CA ) . We visualized recorded neurons using a laser-scanning confocal microscope ( Zeiss LSM 510 , Thornwood , NY ) or a spinning-disk confocal microscope ( Olympus BX51 , Center Valley , PA ) . Images were contrast enhanced and pseudo-colored using the free ImageJ software ( National Institutes of Health , Bethesda , MD ) , and then digitally reconstructed using the free Neuromantic software for morphological reconstruction ( Myatt et al . , 2012 ) . Dbx1 neurons were detected and mapped within three-dimensional ( 3D ) volumes of the preBötC or ventral respiratory column , and then subsequently laser ablated while monitoring respiratory network functionality . The instrument incorporated a Zeiss LSM 510 laser scanning head and fixed-stage microscope body with a 20×/1 . 0 numerical aperture water-immersion objective , an adjustable wavelength 1 . 5 W Ti:sapphire tunable laser ( Spectra Physics , Irvine , CA ) , and a robotic xy translation stage ( Siskiyou Design , Grants Pass OR ) . The methodology has been described in a technical report ( Wang et al . , 2013 ) and in an original research report ( Hayes et al . , 2012 ) . We wrote custom software dubbed Ablator that automated a three-step routine . The first step ( initialization phase ) defines the domain for target detection and ablation . The domain can be bilaterally distributed , like the preBötC and ventral respiratory column . The maximum size of any part of the domain in the transverse ( xy ) plane must fit within an area of maximum dimensions 412 square micrometers . The z domain ( depth ) is a function of tissue opacity , laser power ( Ti:sapphire ) , and the emission properties of the fluorescent reporter . For neonatal mouse brainstem tissue ( P0–5 ) , using 800-nm pulses emitted at ∼1 W , which measured 36 mW at the specimen plane , the z domain generally measured less than 100 µm . The second step ( detection phase ) acquires high-resolution images via confocal microscopy with a visible-wavelength laser ( HeNe 543 nm for tdTomato ) . Dbx1 neurons were identified by native fluorescent protein expression using a threshold-crossing target detection algorithm in Ablator software , which is open-source and available for free download at the sourceforge . net archive , i . e . , http://sourceforge . net/projects/ablator/ . Additional image processing routines differentiate Dbx1 somata from auto-fluorescent debris and neuropil ( Figure 1—figure supplement 2 ) . The final map of Dbx1 neuron targets reflects the position of the center of each cell body in the 3D volume of the domain ( see Figure 1D ) . Ablator chooses Dbx1 neuron targets in random order and advances until all the targets are exhausted or the respiratory rhythm ceases for longer than 120 s . The Ti:sapphire laser scans a 10 square micrometer spot centered on each target with 800-nm pulses at maximum intensity . The ablation is confirmed if fluorescence is detected in the band 560–615 nm , which reflects presumed water vapor in the cell cavity and excludes infrared reflections of the long-wavelength laser ( Figure 2—figure supplement 1A ) ( Wang et al . , 2013 ) . In addition , lesioned targets disappear from the fluorescence image ( Figure 2—figure supplement 1B ) , and their pre-lesion bright field image ( Figure 2—figure supplement 1C ) is replaced post-lesion by a pock mark ( Figure 2—figure supplement 1D ) . Confirmed lesions add to a running tally . If lesion confirmation cannot be obtained , then the target selection algorithm does not advance and subsequent attempts are made to lesion the ROI . With each subsequent iteration , the scanning speed is decreased to improve the likelihood of lesioning the target . This loop repeats a total of five times . If confirmation of lesion cannot be ascertained after the fifth attempt , then it is deemed a failed lesion . Failed lesions do not contribute to the tally and their ROIs are removed from the list of targets to avoid reselection for the remainder of the experiment . A log file documents lesions by index number and time of confirmation . The XII rhythm is monitored and recorded continuously so its state can be directly correlated with the lesion tally in real time . Cell targets are destroyed in successful lesions so their effects are cumulative . The laser lesions are performed bilaterally in the preBötC . After a batch of lesions on one side , the robotic xy translation stage translates to the contralateral side and performs another batch , and then switches sides again , and so on until the targets are exhausted or the XII rhythm ceases . We measured XII burst magnitude ( amplitude and area ) and computed cycle period ( the interval between consecutive XII bursts ) using LabChart software ( ADInstruments , Colorado Springs , CO ) . The regularity score ( RS ) was defined as the quotient of period of the present cycle Tn with respect to the mean cycle period for ten previous cycles:RS=Tn1j∑i=1jTn−i , where j = 10 . We defined the control epoch as 30 min of continuous recording from the end of the detection phase to the beginning of the ablation phase . Data sets were tested for normality using a Shapiro–Wilk test . We rejected the null hypothesis that the data are drawn from a normal distribution if the p-value of the test statistic was less than α=0 . 05 . Data that could be considered normally distributed were compared using two-tailed paired t-tests , whereas data that did not conform to the normal distribution were compared using non-directional ( two-tailed ) Mann–Whitney U-tests . XII burst amplitude and frequency/cycle period were reported with standard deviation ( SD ) and standard error of the mean ( SEM ) . Discrete cell counts that pertain to the number of neurons detected or the number of neurons lesioned are reported with SD , SEM and min–max range . We wrote a Matlab ( MathWorks Inc . , Natick , MA ) script to generate Erdős-Rényi G ( n , p ) -directed random graphs ( Newman et al . , 2006 ) with key parameters of population size ( n ) and connection probability ( p ) . Vertices ( a . k . a . , nodes ) of G ( n , p ) were populated by Rubin–Hayes preBötC neuron models and the directed edges ( a . k . a . , links ) between vertices were modeled by excitatory glutamatergic synapses ( Rubin et al . , 2009 ) . We simulated the network models on the SciClone computing complex at The College of William & Mary , which features 193 nodes with a total of 943 CPU ( central processing unit ) cores , 5 . 9 terabytes of physical memory , 220 terabytes of disk capacity , and peak performance of 21 . 2 teraflops . We used a Runge–Kutta fourth-order numerical integration routine with fixed time step of 0 . 25 ms . Network models were subject to 100 random deletions , one deletion every 25 s . Neuron deletions were achieved by setting the synaptic state variable and its corresponding differential equation to zero , which essentially removes the cell from the network . Deleted neurons no longer contributed to running-time histograms of network activity and were removed from raster plots ( e . g . , Figure 6C ) . Transient glutamatergic stimulation of constituent model neurons mimicked the experimental glutamate un-caging protocol by Kam et al . which evoked respiratory bursts in the preBötC ( Kam et al . , 2013b ) . Focal stimulation was achieved by setting the synaptic state variable to 0 . 9 for 200 ms , without modifying the differential equation , so the glutamatergic excitation was indeed transient . Focal stimulation was applied to rhythmically active networks several seconds following an endogenous burst ( Figure 6B ) . Since there is uncertainty regarding the exact network size , we conducted a series of simulations for a range of ( n , p ) with the aim of finding a reasonable parameter range to produce respiratory-like rhythms ( 3-4 s cycle period prior to ablations ) . We varied n from 200 to 400 with a step size of 10 and p from 0 . 1 to 0 . 2 with a step size of 0 . 0125 . For each parameter set , 10 simulations without deletion were conducted for 25 s to assess network rhythmicity ( Figure 6—figure supplement 1 ) . For the parameter sets whose initial period fell between 3 and 4 s , we performed 5–6 simulations with deletions ( for n = 320 , 330 , 340 we performed 16 simulations in each case ) and then calculated the longest period , the ablation tally , and discrete network metrics pertaining to G ( n , p ) . The results are documented in Figure 6 and Figure 6—figure supplement 1 , as well as the table in Supplementary file 1 . During the simulations , raster graphs were simultaneously generated to detect the spiking for each individual neuron ( Figure 6C ) . The running time histogram is based on the raster graph for each simulation , from which we computed the cycle period and amplitude ( number of spikes per time bin , Figure 6D , E ) . A network with n vertices can be represented by its adjacency matrix A ( n×n ) in a manner that if there is a connection from vertex i to vertex j then Aij=1 , otherwise Aij=0 . The adjacency matrices are asymmetric for neuronal networks , which are directed ( i . e . , the chemical synapses are unidirectional ) . In discrete simulations , the lesion of neurons is modeled by removing vertices from the adjacency matrix along with their edges , i . e . , connections ( in and out ) . We computed three global metrics ( K-core , number of strongly connected components , average in and out degree ) for the initial network and the remaining network after a sequence of 100 random deletions . Also for each deleted vertex , we computed three local network metrics ( local cluster coefficient , closeness centrality , betweenness centrality ) to indicate the importance of the vertex within the previous network . The metrics are defined below and reported in the table of Supplementary file 2 .	Our first breath , moments after we are born , is the result of a pattern of activity in our brain that started in the embryo and will continue almost effortlessly until we die . Like other rhythmic activities , such as walking and swimming , breathing originates from circuits of neurons in the brain that generate patterns . These circuits pass messages to other cells that translate them into the physical movements required to take a breath . Interrupting these patterns by injury or illness can lead to breathing disorders or cause death . Previous studies have identified a class of neuron , which all express a specific gene , that is necessary for breathing . Mice born without this class of cell failed to ever take a breath and died at birth . These neurons are found in part of the brainstem and can continue to generate rhythm even when this section of the brainstem is removed from newborn mice and cut into very thin slices . However , it is unclear how many of these neurons are needed to maintain a breathing rhythm . Wang et al . used a laser to destroy the breathing rhythm-generating neurons in these slices one at a time and found that the rhythm of breathing in ( i . e . , inspiration ) stopped after ∼15% of the neurons were destroyed . This suggests that a high percentage of these neurons must be maintained for breathing to continue normally . Wang et al . also discovered that destroying the rhythm-generating neurons reduced the strength of the signals sent from the brainstem to trigger the movements that cause breathing in . This suggests that the same class of neurons also sends messages to the muscles involved in breathing; it was previously thought that a separate class of cell in the same part of the brain sent these messages . Studies involving live animals are now needed to confirm the results . If confirmed , the findings may be used to develop new treatments for a number of breathing disorders . Medications that boost the signals sent to the muscles by these neurons might be useful for treating sleep apnea . Wang et al . also suggest that medications that boost rhythm generation might be useful for premature infants with breathing difficulties and people with drug-induced breathing problems . Moreover , finding ways to maintain breathing rhythms with fewer of these neurons may help those with neurodegenerative disorders , which cause cells in the brain to be lost .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2014	Laser ablation of Dbx1 neurons in the pre-Bötzinger complex stops inspiratory rhythm and impairs output in neonatal mice
We recently reported that the C2AB portion of Synaptotagmin 1 ( Syt1 ) could self-assemble into Ca2+-sensitive ring-like oligomers on membranes , which could potentially regulate neurotransmitter release . Here we report that analogous ring-like oligomers assemble from the C2AB domains of other Syt isoforms ( Syt2 , Syt7 , Syt9 ) as well as related C2 domain containing protein , Doc2B and extended Synaptotagmins ( E-Syts ) . Evidently , circular oligomerization is a general and conserved structural aspect of many C2 domain proteins , including Synaptotagmins . Further , using electron microscopy combined with targeted mutations , we show that under physiologically relevant conditions , both the Syt1 ring assembly and its rapid disruption by Ca2+ involve the well-established functional surfaces on the C2B domain that are important for synaptic transmission . Our data suggests that ring formation may be triggered at an early step in synaptic vesicle docking and positions Syt1 to synchronize neurotransmitter release to Ca2+ influx . Synchronized rapid release of neurotransmitters at the synapse is a highly orchestrated cellular process . This involves maintaining a pool of synaptic vesicles ( SV ) containing neurotransmitters docked at the pre-synaptic membrane , ready to fuse and release their contents upon the influx of calcium ions ( Ca2+ ) following an action potential , while also preventing the spontaneous fusion of SVs in absence of the appropriate cue ( Südhof and Rothman , 2009; Jahn and Fasshauer , 2012; Südhof , 2013; Rizo and Xu , 2015 ) . The core machinery required for the Ca2+ triggered neurotransmitter release are the SNARE proteins ( VAMP2 , Syntaxin , and SNAP25 ) as well as Munc13 , Munc18 , Complexin and Synaptotagmin ( Südhof and Rothman , 2009; Jahn and Fasshauer , 2012; Südhof , 2013; Rizo and Xu , 2015 ) . A combination of biochemical , genetic and physiological results have pinpointed Synaptotagmin as a central component involved in every step of this coordinated process ( Wang et al . , 2011; Jahn and Fasshauer , 2012; Südhof , 2013; Rizo and Xu , 2015 ) . The principal neuronal isoform , Synaptotagmin 1 ( Syt1 ) , is a SV-associated protein , with a cytosolic domain consisting of tandem Ca2+-binding C2 domains ( C2A and C2B ) attached to the membrane via a juxtamembrane ‘linker’ domain ( Brose et al . , 1992; Takamori et al . , 2006 ) . Accordingly , Syt1 acts as the immediate and principal Ca2+ sensor that triggers the rapid and synchronous release of neurotransmitters following an action potential ( Brose et al . , 1992; Geppert et al . , 1994; Fernández-Chacón et al . , 2001 ) . Upon Ca2+ binding , the adjacent aliphatic surface loops on each of the C2 domains partially insert into the membrane and this enables the SNAREs to complete membrane fusion by mechanisms that are still uncertain ( Tucker et al . , 2004; Rhee et al . , 2005; Hui et al . , 2006; Paddock et al . , 2011 ) . Syt1 is also needed for the initial stage of close docking of SVs to the plasma membrane ( PM ) , requiring in particular the interaction of the polybasic region on C2B domain with the anionic lipid , phosphatidylinositol 4 , 5-bisphosphate ( PIP2 ) at the PM ( Bai et al . , 2004; Wang et al . , 2011; Parisotto et al . , 2012; Park et al . , 2012; Honigmann et al . , 2013; Lai et al . , 2015 ) . The C2B domain also binds to the neuronal t-SNAREs ( Syntaxin/ SNAP25 ) on the PM , which positions the Syt1 on the pre-fusion SNARE complexes and contributes to the docking of the SV but is by itself insufficient ( de Wit et al . , 2009; Parisotto et al . , 2012; Mohrmann et al . , 2013; Kedar et al . , 2015; Park et al . , 2015; Zhou et al . , 2015 ) . Despite a wealth of information on Syt1 function and underlying molecular mechanism , critical questions remain . Deletion ( or mutations ) of Syt1 eliminates fast synchronous release and increases the normally small rate of asynchronous/spontaneous release ( Geppert et al . , 1994; Littleton et al . , 1994; Bacaj et al . , 2013 ) . Reciprocally , removing Complexin increases the spontaneous release amount and the remaining Syt1 is only capable of mounting asynchronous release , though this release is still Ca2+-dependent ( Huntwork and Littleton , 2007; Hobson et al . , 2011; Jorquera et al . , 2012; Cho et al . , 2014; Trimbuch and Rosenmund , 2016 ) . This suggests that Syt1 , acting in concert with Complexin , also functions as a clamp to both restrain and energize membrane fusion to permit rapid and synchronous release ( Giraudo et al . , 2006; Krishnakumar et al . , 2011; Kümmel et al . , 2011 ) . How this clamping is accomplished still remains a mystery . In addition , fast neurotransmitter release exhibits a steep cooperative dependency on Ca2+ concentration , which implies that several Ca2+ ions need to be bound to one or more Syt1 molecules to trigger release ( Schneggenburger and Neher , 2000 , 2005; Matveev et al . , 2011 ) . Further , reduced Ca2+ binding affinity does not change this Ca2+ cooperativity ( Striegel et al . , 2012 ) , suggesting multiple copies of Syt1 molecules might be involved in gating release . However , the exact mechanism of the cooperative triggering of SV fusion is unclear . We have recently shown that Syt1 C2AB domains can form Ca2+-sensitive ring-like oligomers on phosphatidylcholine ( PC ) /phosphatidylserine ( PS ) lipid surfaces ( Wang et al . , 2014 ) . This finding suggests a simple and elegant mechanism: If these Syt1 ring-like oligomers were to form at the interface between SVs and the plasma membrane , they could act sterically to prevent fusion , until this barrier is removed when Ca2+ enters and triggers ring disassembly i . e . the Syt1 ring would synchronize fusion to Ca2+ influx . In addition , the oligomeric nature of Syt1 could explain the observed Ca2+ cooperativity of neurotransmitter release . Here we show that the ring-like oligomer is a common structural feature of the C2 domain containing protein and describe the physiological correlates of the Syt1 ring oligomer which argues for a functional role for the Syt1 ring in orchestrating the synchronous neurotransmitter release . We had previously described the formation of Ca2+-sensitive ring-like oligomers on lipid monolayers with the C2AB domain of Syt1 ( Wang et al . , 2014 ) . To explore this further , we analyzed the organization of membrane bound C2AB domains of other neuronal isoforms of Synaptotagmin ( Syt2 , Syt7 and Syt9 ) on lipid surface under Ca2+-free conditions by negative stain electron microscopy ( EM ) . Syt2 and Syt9 act as Ca2+ sensors for synchronous SV exocytosis but are expressed in only a subset of neurons ( Xu et al . , 2007 ) , while Syt7 has been posited to mediate the Ca2+-dependent asynchronous neurotransmitter release ( Bacaj et al . , 2013 ) . EM analysis on lipid monolayer was carried out as described previously ( Wang et al . , 2014 ) . Briefly , the lipid monolayer formed at the air/water interface was recovered on a carbon-coated EM grid and protein solution was added to the lipid monolayer under Ca2+-free conditions ( 1 mM EDTA ) and incubated for 1 min at 37°C . Negative-stain analysis revealed the presence of ring-like oligomers for all the Syt isoforms tested ( Figure 1 ) . Despite the variability in the number of ring-like structures between different isoforms , the size of the ring oligomers were remarkably similar , with an average outer diameter of ~30 nm ( Figure 1 ) . In all cases , each ring was composed of an outer protein band of a width of ~55Å , which is consistent with the dimensions of a single C2AB domain ( Fuson et al . , 2007 ) . This data shows that the ability to form the circular oligomers is not unique to Syt1 , but conserved among the Syt isoforms and further suggests that it might be an intrinsic property of the C2 domains . 10 . 7554/eLife . 17262 . 003Figure 1 . Ring-like oligomers are a common structural feature of C2 domain proteins . EM analysis showing the C2AB domains of neuronal isoforms of Syt , namely Syt1 , Syt2 , Syt7 , and Syt9 form ring like oligomers on monolayers under Ca2+-free conditions . Similar ring-like structures were observed for other related C2 domain proteins , like Doc2B and E-Syt 1 & 2 . The number of ring-oligomers observed on the monolayers varied , but the dimensions of the rings were remarkably consistent ( ~30 nm ) . All EM analyses were carried out using 5 µM protein on monolayer containing 40% PS and buffer containing 15 mM KCl and 1 mM free Mg2+ . Representative micrographs and average values , along with standard error of the means ( SEM ) from a minimum of three independent trials are included . DOI: http://dx . doi . org/10 . 7554/eLife . 17262 . 00310 . 7554/eLife . 17262 . 004Figure 1—figure supplement 1 . Ring assembly is not a conserved property of all C2 domains . Negative stain EM analysis showing the Syt1C2B domain alone can form ring-like oligomers , but the Syt1C2A domain cannot . Syt1C2B rings are relatively un-stable and are smaller ( ~22 nm ) compared to the Syt1C2AB ring oligomers ( ~30 nm ) . EM analysis were carried out using 5 µM protein on monolayer containing 40% PS and buffer containing 15 mM KCl and 1 mM free Mg2+ . Representative micrographs from 3 independent trials are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 17262 . 00410 . 7554/eLife . 17262 . 005Figure 1—figure supplement 2 . Effect of Ca2+ addition on pre-formed ring-like oligomers of Syt isoforms and other C2 domain proteins . Pre-formed ring oligomers were washed briefly ( 10 s ) with buffer containing calcium ( final concentration of 1 mM free ) and analyzed via negative stain EM . All Syt isoforms and Doc2B were sensitive to the Ca2+ treatment and number of rings observed reduced drastically with Ca2+ treatment . *Note: With Syt2C2AB , we observed irregular patches of protein arrays together with circular structures following Ca2+ treatment . These circular structures were quite different from the Ca2+-free rings as they exhibit a more uniform protein density , indicating these are patches of protein arrays , arranged into round shape probably due to the local buckling of monolayer by insertion of Syt2 calcium loops . Over longer incubation time , the number of 'circular structure' decrease and the size of irregular patches increase and begin to pack orderly , suggesting the circular structure is the intermediate building block of 2D-array formation . Ca2+ addition had divergent effect on the E-Syt isoforms , while the E-Syt2 was largely un-affected , the E-Syt1 rings were stabilized ( discussed below ) . All EM analyses were carried out using 5 µM protein on monolayer containing 40% PS and buffer containing 15 mM KCl and 1 mM free Mg2+ . Representative micrographs and averages and SEM from 3–4 independent trials are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 17262 . 005 Therefore , we next tested the C2AB domains of Doc2B , C2ABCDE domains of extended Synaptotagmin 1 ( E-Syt1 ) and the C2ABC domains of E-Syt2 . Doc2B is a C2 domain protein expressed in the pre-synaptic terminals and a putative Ca2+ sensor that regulates both spontaneous ( Groffen et al . , 2010 ) and asynchronous release ( Yao et al . , 2011 ) . E-Syts are endoplasmic reticulum ( ER ) resident proteins , which contain multiple C2 domains and have been implicated in ER-PM tethering , the formation of membrane contact sites , and in lipid transport and Ca2+ signaling ( Giordano et al . , 2013; Reinisch and De Camilli , 2016; Fernandez-Busnadiego , 2016; Herdman and Moss , 2016 ) . Doc2B and E-Syt2 formed circular oligomeric structures on lipid monolayers analogous to those seen with Syt isoforms ( Figure 1 ) . However , we observed very few and un-stable ring-like oligomers with E-Syt1 ( Figure 1 ) . The lack of ring-like oligomers for E-Syt1 might be due to the insufficient concentration of this protein on the membrane surface as E-Syt1 has very weak affinity to the membrane under Ca2+-free conditions ( Idevall-Hagren et al . , 2015 ) . The uniform dimensions of the ring oligomers of the multi-C2 domain proteins suggested that the ring is formed by a single C2 domain , with the other C2 domain ( s ) projecting away radially ( Figure 1 ) . This implies that the ring oligomerization is not a general property of all C2 domains , but only a select few . Consistent with this , we find that the Syt1C2B domain alone can form the ring-like oligomers albeit a bit smaller in size , but the Syt1C2A cannot ( Figure 1—figure supplement 1 ) . Brief treatment of the pre-formed ring oligomers with 1 mM Ca2+ ( Figure 1—figure supplement 2 ) revealed that all of the Syt isoforms ( Syt1 , Syt2 , Syt7 , and Syt9 ) and Doc2B were sensitive to Ca2+ and are rapidly disrupted , but E-Syt were either un-affected ( E-Syt2 ) or even stabilized ( E-Syt1 ) . Altogether , our data suggests that ring-like oligomers are a common structural feature of C2 domain containing proteins , but their sensitivity to Ca2+ is divergent ( discussed below in detail ) . To assess the functional relevance of the Syt1 ring oligomers , we sought to understand the molecular aspects of the oligomer assembly and the Ca2+ susceptibility under physiologically-relevant conditions . The ring oligomers assembled with the minimal C2AB domain of Syt1 were highly sensitive to the ionic strength of the buffer and the anionic lipid content on the monolayer . A minimum of 35% PS in the monolayer and buffers containing <50 mM KCl were required to obtain stable ring structures ( Wang et al . , 2014 ) . We reasoned that the inclusion of conserved N-terminal juxtamembrane region ( ~60 residues ) that connects the C2AB domains to the membrane anchor , might help stabilize the ring oligomers . The juxtamembrane linker domain has been shown to be vital for Syt1 role in activating synchronous release and in clamping the spontaneous release ( Caccin et al . , 2015; Lee and Littleton , 2015 ) . It also has the ability to interact with the membrane and has been shown to self-oligomerize ( Fukuda et al . , 2001; Lai et al . , 2013; Lu et al . , 2014 ) . We purified the entire cytoplasmic domain of Syt1 ( Syt1CD , residues 83–421 ) using a stringent purification protocol ( Seven et al . , 2013; Wang et al . , 2014 ) to remove all polyacidic contaminants , which could promote non-specific aggregation of the protein ( Seven et al . , 2013 ) and this is confirmed by a single peak in the size-exclusion chromatography ( Figure 2—figure supplement 1A ) . As expected , lipid binding analysis showed that the juxtamembrane domain enhances and stabilizes the membrane interaction of Syt1 under physiologically-relevant experimental conditions ( Figure 2—figure supplement 1B ) . To visualize the organization of the Syt1CD on lipid monolayers under Ca2+-free conditions , we adapted the conditions used previously to obtain Syt1C2AB rings ( Wang et al . , 2014 ) . Negative stain EM analysis showed that Syt1CD can form stable ring-like oligomers ( Figure 2A ) on monolayers under physiologically-relevant lipid ( PC/PS at 3:1 molar ratio ) and buffer ( 100 mM KCl , 1 mM free magnesium , Mg2+ ) composition . The outer diameter of these Syt1CD rings ranged from 19–42 nm , with an average size of 30 ± 4 . 5 nm ( Figure 2B ) , analogous to the Syt1C2AB rings ( Wang et al . , 2014 ) . Based on the helical indexing of the Syt1C2AB tubes ( Wang et al . , 2014 ) , we estimate that this corresponds to 12–25 copies of Syt1 molcule , with average ~17 copies of Syt1 . The Syt1CD rings were robust as we did not observe many collapsed ring structures , like the ‘clams’ or ‘volcanos’ , routinely seen with C2AB rings ( Wang et al . , 2014 ) and were stable under a wide-range of the ionic strengths and anionic lipid content ( Figure 2C ) . Therefore , we used the Syt1CD to delineate the mechanistic details of the Syt1 ring oligomer assembly and its Ca2+-sensitivity in a physiologically relevant environment . 10 . 7554/eLife . 17262 . 006Figure 2 . The entire cytoplasmic domain of Syt1 ( Syt1CD ) forms ring-like oligomers under physiologically relevant conditions . ( A ) Negative stain EM analysis shows ring-like oligomers of Syt1CD on PC/PS ( 3:1 molar ratio ) lipid monolayers in buffer containing 100 mM KCl and 1 mM MgCl2 . ( B ) The size distribution of the Syt1CD rings as measured from the outer diameter ( n = ~400 ) under these experimental conditions using ImageJ software . ( C ) The Syt1CD ring-oligomers were observed under a wide-ranging conditions . Under all conditions tested , the dimension of these ring oligomers were very consistent ( ~30 nm ) , but the number of rings observed depended on amount of the anionic lipid in the monolayer and the salt ( KCl ) concentration of the buffer ( D ) EM analysis showing that the polylysine ( K326/K327 ) motif of C2B domain is critical to the ring formation , but the other conserved polybasic regions of Syt1 , namely K190/K191 on C2A and R398/R399 on C2B are not involved in ring formation . All EM analyses were carried out using 5 µM protein on monolayers containing 25% PS and in buffer containing 100 mM KCl and 1 mM free Mg2+ . Representative micrographs and averages/SEM from three independent trials are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 17262 . 00610 . 7554/eLife . 17262 . 007Figure 2—figure supplement 1 . Purification and Characterization of Syt1CD . ( A ) Size Exclusion Chromatography profile on a Superdex75 10/300 GL column of Syt1CD . Syt1CD was purified using a stringent purification protocol , including benzonase treatment , high salt wash and ion exchange chromatography . The Syt1CD sample shows a single peak confirming that is free of polyacidic impurities , which could trigger aggregation of the protein . Inset: SDS-PAGE Coomaisse analysis of the peak shows a single band consistent with to the size ( ~41 kDa ) of the Syt1CD protein . ( B ) Lipid binding analysis shows that the inclusion of juxtamembrane domain enhances the Syt1 membrane interaction . To measure binding , 10 µM of Syt1C2AB or Syt1CD were mixed with 1 mM small unilamellar vesicles ( SUV ) containing 25% PS+3% PIP2 ( remainder was PC ) and incubated for 1 hr at RT with in buffer containing 100 mM KCl and 1 mM free Mg2+ . The SUVs were isolated using discontinuous density gradient and analyzed on SDS-PAGE/ Coomaisse analysis , after adjusting for the amount of lipid recovered . The amount of protein bound was estimated using density measurement using ImageJ software . DOI: http://dx . doi . org/10 . 7554/eLife . 17262 . 00710 . 7554/eLife . 17262 . 008Figure 2—figure supplement 2 . Presence of anionic lipid is required to assemble the Syt1CD ring oligomers . Negative stain EM analysis was carried out using 5 µM protein on monolayer with DOPC alone in buffer containing 100 mM KCl and 1 mM free Mg2+ . Representative micrographs from 3–4 independent trials are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 17262 . 008 The assembly of the Syt1CD ring oligomers strictly required the presence of anionic lipid ( PS ) in the monolayer ( Figure 2—figure supplement 2 ) and the amount of the negative charge in the monolayer and the ionic strength of the buffer affected the number and integrity of the Syt1CD rings ( Figure 2C ) . Therefore , to identify which parts of Syt1 are involved in positioning the Syt1 on the membrane to promote the ring assembly , we focused on the conserved polybasic regions of Syt1 . Disrupting the polylysine motif on the C2A ( K190A , K191A ) or the arginine cluster on the C2B ( R398A , R399A ) did not affect the ring formation ( Figure 2D ) , but mutations of key lysine residues ( K326A , K327A ) within the polybasic patch on the C2B drastically reduced ( ~90% ) the number of the Syt1CD rings , even when 25% PS was included in the monolayer ( Figure 2D ) . This suggests that the electrostatic interaction between the polylysine motif on C2B and the anionic lipids on the membrane surface is required for the ring formation . Consequently , we tested the effect of PIP2 on the ring assembly as the polylysine motif on C2B has been shown to preferentially bind PIP2 with high affinity ( Bai et al . , 2004; Parisotto et al . , 2012; Park et al . , 2012; Honigmann et al . , 2013; Krishnakumar et al . , 2013; Lai et al . , 2015 ) . Syt1CD ring formation did not require PIP2 , but inclusion of PIP2 in the lipid monolayer ( 25% PS , 3% PIP2 , 72% PC ) improved the number and the integrity of the Syt1CD rings ( Figure 3A and E ) . However , PIP2 was essential to obtain stable Syt1CD ring oligomers when ATP at physiological concentrations ( 1 mM Mg-ATP ) was included ( Figure 3B , C and E ) . ATP is a critical co-factor , which modulates Syt1 function as it reverses the inactivating cis- interaction of Syt1 with its own membrane while preserving the functional trans- association to the plasma membrane ( Park et al . , 2012; Vennekate et al . , 2012 ) . This is because ATP effectively screens the interaction of Syt1 with weakly anionic PS , but not with the strong negative charges on the PIP2 head group found exclusively on the PM ( Park et al . , 2012 , 2015 ) . Correspondingly , lipid binding assays showed that the ATP blocks the binding of Syt1CD to PS-containing vesicles , but not to PS/PIP2 membranes ( Figure 3—figure supplement 1 ) . Corroborating this , 6% PIP2 as the sole anionic lipid ( 6% PIP2 , 94% PC ) in the lipid monolayer was found to be sufficient to form ring oligomers , even in the presence of 1 mM ATP ( Figure 3D and E ) . Taken together , our data shows that under physiological ionic conditions , the Ca2+-independent interaction of the C2B domain with PIP2 on the PM , which has been implicated in the vesicle docking both in vitro and in vivo ( Wang et al . , 2011; Parisotto et al . , 2012; Park et al . , 2012; Honigmann et al . , 2013; Lai et al . , 2015 ) , is key to assembling the Syt1 ring-like oligomers . 10 . 7554/eLife . 17262 . 009Figure 3 . Syt1-PIP2 interaction is key to ring-formation under physiologically relevant conditions . ( A ) Inclusion of 3% PIP2 ( in addition to 25% PS ) in the monolayer stabilized the ring structures and increased the number of rings observed . ( B , C ) Addition of ATP drastically reduced the number of rings observed in monolayers containing 25% PS only , but not when supplemented with 3% PIP2 . ( D ) PIP2 ( 6% ) as the only anionic lipid on the bilayer was sufficient to assemble ring-like oligomers , even in the presence of 1 mM ATP . All EM analyses were carried out using 5 µM protein in buffer containing 100 mM KCl and 1 mM free Mg2+ . Representative micrographs and average values/SEM from a minimum of three independent trials are shown in ( E ) . The rings observed under all conditions shown in ( E ) were similarly ( ~30 nm ) sizedDOI: http://dx . doi . org/10 . 7554/eLife . 17262 . 00910 . 7554/eLife . 17262 . 010Figure 3—figure supplement 1 . Lipid binding analysis shows that ATP effectively screens the interaction of Syt1CD to PS-only membrane , but not membrane containing 3% PIP2 . To assess Syt1CD-membrane binding , 10 µM of Syt1CD were mixed with 1 mM small unilamellar vesicles ( SUV ) containing either 25% PS or 25% PS+3% PIP2 ( remainder was PC ) and incubated for 1 hr at RT with in in buffer containing 100 mM KCl and 1 mM free Mg2+ supplemented with 1 mM Mg2+ or 1 mM Mg-ATP . The SUVs were isolated using discontinuous density gradient and analyzed on SDS-PAGE/ Coomaisse analysis after adjusting for the amount of lipid recovered . The amount of protein bound in each case was estimated using density measurement using ImageJ software . DOI: http://dx . doi . org/10 . 7554/eLife . 17262 . 010 Similar to Syt1C2AB , Syt1CD rings were sensitive to Ca2+ and brief treatment ( ~10 s ) with Ca2+ drastically disrupted the integrity of the preformed Syt1CD ring oligomers ( Figure 4A ) . Calcium ions at concentrations in the range measured in intra-terminal region during synaptic transmission ( Schneggenburger and Neher , 2000 , 2005; Neher and Sakaba , 2008 ) fragmented and disassembled the rings in a Ca2+ concentration-dependent fashion ( Figure 4A ) . PIP2 had little or no effect on the Ca2+ sensitivity of the Syt1CD as we observed very similar reduction in Syt1CD rings with or without 3% PIP2 across all Ca2+ concentration tested ( Figure 4—figure supplement 1 ) . To verify that the Ca2+ sensitivity of the Syt1CD rings is indeed due to specific Ca2+ binding to Syt1 and to map this sensitivity , we generated and tested Syt1CD mutants that disrupt Ca2+ binding to the C2A and C2B domains respectively ( Shao et al . , 1996 ) . As shown in Figure 4B , disrupting Ca2+ binding to C2B ( Syt1CD D309A , D363A , D365A; C2B3A ) rendered the ring oligomers insensitive to calcium ions , while blocking Ca2+ binding to the C2A domain ( Syt1CD D178A , D230A , D232A; C2A3A ) did not alter the effect of Ca2+ on the Syt1CD rings ( Figure 4—figure supplement 2 ) . Likewise , mutations of aliphatic loop residues in the C2B domain ( Syt1CD V304N , Y364N , I367N; C2B3N ) , which insert into the membrane following Ca2+ binding , made the Syt1CD ring oligomers insensitive to Ca2+ wash , but corresponding mutations in the C2A calcium loops ( Syt1CD F231N , F234N , S235N; C2A3N ) had no effect ( Figure 4C , Figure 4—figure supplement 3 ) . The mutation analysis shows that the rapid disruption of the Syt1 rings requires Ca2+ binding to the C2B and the subsequent reorientation of the C2B domain into the membrane . In other words , the dissociation of the Syt1 ring oligomers is coupled to the conformational changes in C2B domain , which is involved in Ca2+ activation and is physiologically required for triggering synaptic transmission . 10 . 7554/eLife . 17262 . 011Figure 4 . Ca2+ binding and subsequent re-orientation of the C2B domain into the membrane are needed to disassemble the Syt1 ring oligomer . ( A ) Syt1CD ring oligomers were sensitive to Ca2+ and brief treatment ( 10 s ) of the pre-formed rings with physiological levels of Ca2+ greatly reduced the number of rings observed . ( B ) Ca2+sensitivity of the Syt1CD rings maps to the C2B domain as disrupting Ca2+ binding to C2B ( Syt1CD D309A , D363A , D365A; C2B3A ) but not C2A ( Syt1CD D178A , D230A , D232A; C2A3A ) rendered the rings Ca2+ insensitive . ( C ) Ca2+-induced insertion of just the C2B domain is necessary to disrupt the ring oligomers as hydrophilic mutation that blocks its insertion of the C2B loop ( Syt1CD V304N , Y364N , I367N; C2B3N ) but not the C2A loop ( Syt1CD F231N , F234N , S235N; C2A3N ) makes the rings insensitive to Ca2+ . All EM analyses were carried out using 5 µM protein on monolayers containing 25% PS and in buffer containing 100 mM KCl and 1 mM free Mg2+ . Effect of addition of 1mM Ca2+ ( final concentration ) is shown in ( B ) & ( C ) . Representative micrographs and average values and deviations ( SEM ) from 3–4 independent trials are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 17262 . 01110 . 7554/eLife . 17262 . 012Figure 4—figure supplement 1 . Inclusion of PIP2 does not change the Ca2+ sensitivity of the Syt1CD ring oligomers . The pre-formed rings were sensitive to Ca2+ and brief treatment ( 10 s ) of the pre-formed rings with physiological levels of Ca2+ drastically reduced the number of rings observed , even when 3% PIP2 ( in addition to 25% PS ) was included in the monlayer . The reduction in the number of ring oligomers observed was comparable to no PIP2 condition . ( Figure 4A ) . Representative micrographs and averages and SEM from 3–4 independent trials are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 17262 . 01210 . 7554/eLife . 17262 . 013Figure 4—figure supplement 2 . Disrupting the calcium binding to C2B ( Syt1CD D309A , D363A , D365A; C2B3A ) , but not C2A ( Syt1CD D178A , D230A , D232A; C2A3A ) renders the Syt1CD rings insensitive to calcium . All EM analyses were carried out using 5 µM protein on monolayers containing 25% PS and in buffer containing 100 mM KCl and 1 mM free Mg2+ . Pre-assembled ring oligomers were briefly ( 10 s ) washed with buffer containing 1 mM free Ca2+ to assess the calcium sensitivity . Representative micrographs from 3–4 independent trials are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 17262 . 01310 . 7554/eLife . 17262 . 014Figure 4—figure supplement 3 . Disrupting the Ca2+-induced membrane insertion of C2B loop ( Syt1CD V304N , Y364N , I367N; C2B3N ) but not the C2A loop ( Syt1CD F231N , F234N , S235N; C2A3N ) makes the rings insensitive to calcium . All EM analyses were carried out using 5 µM protein on monolayers containing 25% PS and in buffer containing 100 mM KCl and 1 mM free Mg2+ . Pre-assembled ring oligomers were briefly ( 10 s ) washed with buffer containing 1 mM free Ca2+ to assess the calcium sensitivity . Representative micrographs from 3–4 independent trials are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 17262 . 014 In support of a functional role for the Syt1 ring-oligomers , we find that the molecular basis of the Syt1 ring oligomer assembly and its reversal are coupled to well-established mechanisms of Syt1 action . The interaction of the conserved lysine residues in the polybasic region of the C2B domain with PIP2 on the inner leaflet of the pre-synaptic plasma membrane is a key determinant in both ring assembly and in synaptic vesicle docking in vivo ( Martin , 2012; Honigmann et al . , 2013 ) , suggesting these processes are mechanistically linked . In addition , Syntaxin clusters PIP2 ( by binding via its basic juxtamembrane region ) and it has been suggested that it is these clusters that recruit the SVs ( Honigmann et al . , 2013 ) . Given the high local concentration of both PIP2 ( estimated to be up to ~80 mol% in such micro-domains [Honigmann et al . , 2013] ) and Syt1 ( anchored in the synaptic vesicles ) , it is easy to imagine how the ring-like oligomers could form at the docking site in between the synaptic vesicle and the PM . There are ~16–22 copies of Syt1 on a synaptic vesicle ( Takamori et al . , 2006; Wilhelm et al . , 2014 ) , enough to form a ring oligomer of ~27–37 nm in diameter , assuming no contribution from the plasma membrane pool of Syt1 . This is consistent with the Syt1 ring diameters observed on the lipid monolayers ( Figure 2B ) . Several studies have shown that the Syt1-PIP2 docking interaction precedes the engagement of the v- with t-SNAREs ( van den Bogaart et al . , 2011; Parisotto et al . , 2012 ) . The prior formation of a Syt1 ring would thus position it to ideally prevent the complete zippering of the SNAREs , in addition to acting as a washer ( or spacer ) to separate the two membranes . The height of the ring , ~4 nm ( Wang et al . , 2014 ) would allow for the N-terminal domain of the SNARE complex to assemble , but such a gap would impede complete zippering . In effect , the Syt1 rings would block SNARE-mediated fusion and hold the SNARE in a pre-fusion half-zippered state ( Figure 5 ) . This is consistent with the earlier observation that docked vesicles appear to be 3–4 nm away from plasma membrane ( Fernandez-Busnadiego et al . , 2011 ) . 10 . 7554/eLife . 17262 . 015Figure 5 . ‘Washer’ model for the regulation of neurotransmitter release by Syt1 . ( A ) The SV docking interaction of the Syt1 polylysine motif ( blue dots ) with the PIP2 ( yellow dots ) on the plasma membrane positions the Syt1 on the membrane to promote the ring-oligomer formation . The ring assembly might precede the engagement of the SNARE proteins . ( B ) Syt1 ring-oligomers assembled at the SV-PM interface act as a spacer or ‘washer’ to separate the two membranes . The height of the ring ( ~4 nm ) would allow the partial assembly of the SNARE complex , but prevent complete zippering and thus , block fusion . NOTE: The positioning and occupancy of SNAREs on Syt1 ring is not known and are shown for illustrative purposes only . ( C ) Upon binding calcium ions ( red dots ) , the Ca2+ loops that locates to the oligomeric interface , re-orients and inserts into the membrane , thus disrupting the ring oligomer to trigger fusion and release neurotransmitters . Thereby , the Syt1 ring oligomers will synchronize the release neurotransmitters to the influx of calcium ions . DOI: http://dx . doi . org/10 . 7554/eLife . 17262 . 01510 . 7554/eLife . 17262 . 016Figure 5—figure supplement 1 . Organization of Syt1 C2 domains in the ring oligomers . ( A ) X-ray structure of Syt1C2AB fitted into the EM density map of the Syt1C2AB coated monolayer tubes [Adapted from ( Wang et al . , 2014 ) ] shows that C2B domain ( gray ) mediates both the membrane ( yellow ) association and the ring assembly and C2A domain ( cyan ) projects away from the ring . Surface representation of the tube is sectioned to the thickness of a single strand of the 4-start helix to reveal the shape of individual asymmetric unit . ( B ) Organization of the C2B domain in the Syt1 ring oligomer shows that the PIP2 binding polylysine motif ( K326 , K327 , blue ) are packed against the membrane surface , which holds back the Ca2+ binding site/loop ( red ) of C2B from the membrane and it localizes to the protein-protein interface involved in ring formation . This arrangement explains the Ca2+-sensitivity of the Syt1 rings as Ca2+ binding would include reorientation of the C2B domain from the ring geometry . Further , in this model , the recently identified ( Zhou et al . , 2015 ) primary SNARE binding interface on the C2B ( R281 , E295 , Y338 , R398 & R399 , green ) is accessible and free to interact with the SNAREs . However , the occupancy and orientation of the SNAREs on the Syt1 ring is not known . DOI: http://dx . doi . org/10 . 7554/eLife . 17262 . 016 Besides positioning the Syt1 to promote the ring assembly , the binding of the polybasic region to the PIP2 clusters on the PM would also hold back the Ca2+ binding loops from the membrane ( Figure 5 ) . In fact , modeling of the C2AB domain onto the EM density map of the tubular structures of the Syt1C2AB suggests that the C2B calcium loops locates at the interface of the Syt1 oligomer ( Figure 5—figure supplement 1 ) . Such an arrangement would explain how the Syt1 ring could synchronize SV fusion to Ca2+ influx . Ca2+ binding to the C2B domain and subsequent conformational change , which incidentally is required to trigger neurotransmitter release ( Fernández-Chacón et al . , 2001; Rhee et al . , 2005; Paddock et al . , 2011 ) , would induce reorientation of the C2B domain from the ring geometry and thus , break the ring oligomers . As such , this would remove the steric barrier and permit the stalled SNAREpins to complete zippering and trigger SV fusion to release neurotransmitters ( Figure 5 ) . This is congruent with the recent report ( Bai et al . , 2016 ) , showing that the switch between the functional states ( clamped vs . activated ) of Syt1 involves large conformational change in the C2 domains . Besides membranes , Syt1 also binds to t-SNAREs and this interaction is functionally relevant for fast neurotransmitter release ( de Wit et al . , 2009; Mohrmann et al . , 2013; Zhou et al . , 2015; Wang et al . , 2016 ) . Recent reports have mapped the key t-SNARE binding interface to the C2B domain ( Zhou et al . , 2015 ) , which is believed to form before the influx of Ca2+ and is maintained during Ca2+ activation process ( Krishnakumar et al . , 2013; Zhou et al . , 2015; Wang et al . , 2016 ) . We note that in our Syt1 ring oligomer model , this binding interface on the C2B ( Figure 5—figure supplement 1 ) is accessible and free to interact with the SNAREs . However , the occupancy and positioning of the SNARE complexes on the Syt1 ring oligomer is not known and as such , is the focus of our ongoing research . Nevertheless , it is easy to imagine that such an interaction would allow the Syt1 ring to act as a primer to organize the core components of the fusion machinery to allow for a rapid and synchronous neurotransmitter release . Further , the oligomeric structure could provide a mechanistic basis for the observed Ca2+-cooperativity in triggering SV fusion . Obviously , the ‘washer’ model is speculative and functional and physiological studies are required to ascertain its relevance . Based on our data , the key principles of the ring oligomer assembly and its Ca2+ sensitivity can be summarized as follows: The ring-oligomer formation is mediated by a single C2 domain ( within multi-C2 domain protein ) , which binds the anionic lipids on membrane surface via the polybasic motif ( Figures 1 and 2 ) and Ca2+ induced re-orientation of the same C2 domain away from the ring geometry disrupts the ring oligomers ( Figure 4 ) . In other words , the Ca2+ sensitivity of the ring oligomers requires the same C2 domain to have the capacity to bind both anionic lipids and Ca2+ . This is true for the C2AB domains of the Syt isoforms and Doc2B and hence , these ring oligomers are Ca2+ sensitive ( Figure 1—figure supplement 2 ) . However , in the case of the E-Syts , the C-terminal C2 domains ( C2E for E-Syt1 and C2C in E-Syt1 ) that are involved in anionic lipid dependent membrane tethering ( thereby the ring formation ) lack the putative Ca2+ binding loops , with the N-terminal C2 domains mediating the Ca2+-dependent membrane interaction ( Giordano et al . , 2013; Reinisch and De Camilli , 2016 ) . Hence , the E-Syt rings are insensitive to Ca2+ ( Figure 1—figure supplement 2 ) . Further , E-Syt1 exhibits very weak membrane binding under Ca2+ free conditions , which is enhanced upon Ca2+ addition ( Idevall-Hagren et al . , 2015 ) . The increased surface concentration of the E-Syt1 in the presence of Ca2+ could explain the improvement in the number of E-Syt1 rings observed under these conditions ( Figure 1—figure supplement 2 ) . In summary , we find that ring-like oligomers are a common structural feature of C2 domain containing proteins , not all of which are regulators of exocytosis . Particularly interesting are the E-Syts , which function to enable the ER and plasma membrane to come into intimate contact – close enough for lipids to be transferred . Our results suggest this might be achieved by bridging two membranes with an intervening structure , most probably based on ring oligomers . Such an organization could stabilize the contact sites and also enhance the lipid transfer function of E-Syts . However , more research is required to understand this better . Interestingly , yeast cells have both E-Syts ( for membrane adhesion ) and SNAREs ( for membrane fusion ) but do not contain vesicle-associated Syt protein and do not carry out calcium-regulated exocytosis . Perhaps this set the stage for exocytosis to evolve when the C2 domains combined with a vesicle-associated protein to form ring-like oligomer i . e . washers that reversibly impeded SNAREpins . The Syt1CD wild-type and mutant proteins were expressed and purified as a His6-tagged protein using a pET28 vector , while SytC2AB isoforms and Doc2B were expressed and purified as a GST-construct . The proteins were purified as described previously ( Seven et al . , 2013; Wang et al . , 2014 ) , with few modifications . Briefly , Escherichia coli BL21 ( DE3 ) expressing Sytconstructs were grown to an OD600 ~0 . 7–0 . 8 , induced with 0 . 5 mM isopropyl β-D-1-thiogalactopyranoside ( IPTG ) . The cells were harvested after 3 hr at 37°C and suspended in lysis buffer ( 25 mM HEPES , pH 7 . 4 , 400 mM KCl , 1 mM MgCl2 , 0 . 5 mM TCEP , 4% Triton X-100 , protease inhibitors ) . The samples were lysed using cell disrupter , and the lysate was supplemented with 0 . 1% polyethylimine before being clarified by centrifugation ( 100 , 000 ×g for 30 min ) . The supernatant was loaded onto Ni-NTA ( Qiagen , Valencia , CA ) , or Glutathione-Sepharose ( Thermo Fisher Scientific , Grant Island , NY ) beads ( 3 hr or overnight at 4°C ) and the beads was washed with 20 ml of lysis buffer , followed by 20 ml of 25 mM HEPES , 400 mM KCl buffer containing with 2 mM ATP , 10 mM MgSO4 , 0 . 5 mM TCEP . Subsequently , the beads were resuspended in 5 ml of lysis buffer supplemented with 10 μg/mL DNaseI , 10 μg/mL RNaseA , and 10 μl of benzonase ( 2000 units ) and incubated at room temperature for 1 hr , followed by quick rinse with 10 ml of high salt buffer ( 25 mM HEPES , 1 . 1 M KCl , 0 . 5 mM TCEP ) to remove the nucleotide contamination . The beads were then washed with 20 ml of HEPES , 400 mM KCl buffer containing 0 . 5 mM EGTA to remove any trace calcium ions . The proteins were eluted off the affinity beads in 25 mM HEPES , 100 mM KCl , 0 . 5 mM TCEP buffer , either with 250 mM Imidazole ( His-tag proteins ) or using Precission protease for GST-tagged constructs and further purified by anionic exchange ( Mono-S ) chromatography . Size-exclusion chromatography ( Superdex75 10/300 GL ) showed a single elution peak ( ~12 mL ) consistent with a pure protein , devoid of any contaminants . Coding sequences of C2A-E domains from human E-Syt1 was cloned into pCMV6-AN-His vector ( OriGene ) . The plasmid was transfected into Expi293 cells ( Thermo Fisher Scientific , Grant Island , NY ) for protein expression . After three days of transfection , cells were collected and lysed by three cycles of freeze and thaw ( liquid N2 and 37°C water bath ) . His-tagged E-ESyt1C2ABCDE was then purified by His60 Nickel Resin ( Clontech , Mountain View , CA ) , with Imidazole elution . For E-Syt2ABC production , the coding sequence was cloned into a modified pCDFDuet-1 vector ( Novagen , Danvers , MA ) , which has an N-terminal GST tag and a Prescission protease cleavage site and transformed into BL21 ( DE3 ) . The cells were grown at 37°C to an OD600 of ∼0 . 6–0 . 8 , then were shifted to 22°C before induction with 0 . 5 mM IPTG . Cells were harvested 18 hr after induction . The proteins were purified by Glutathione Sepharose 4B chromatography . GST tags were removed by treatment with Prescission protease . Both E-Syt proteins were further purified by gel filtration on a Superdex200 column . The gel filtration buffer contained 20 mM HEPES at pH 8 . 0 , 150 mM NaCl , and 0 . 5 mM TCEP . All chromatrography was carried out using AKTA system ( GE Healthcare , Marlborough , MA ) In all cases , the protein concentration was estimated using Bradford assay with BSA as standard and the nucleotide contamination was tracked using the 260 nm/280 nm ratios . The protein was flash frozen and stored at −80°C with 10% glycerol ( 20% glycerol for Syt1CD ) without significant loss of ring-forming activity . To form the lipid monolayer , degased ultrapure H2O was injected through a side port to fill up wells ( 4 mm diameter , 0 . 3 mm depth ) in a Teflon block . The surface of the droplet was coated with 0 . 5 μl of phospholipid mixture ( 0 . 5 mM total lipids ) . The lipid mixtures , DOPC/DOPS & DOPPC/DOPS/PIP2 were pre-mixed as required , dried under N2 gas and then re-suspended in chloroform to the requisite concentration before adding to the water droplet . The Teflon block then was sealed in a humidity chamber for 1 hr at room temperature to allow the chloroform to evaporate . Continuous carbon-coated EM grids ( 400 mesh; Ted Pella Inc . , Redding , CA ) were baked at 70°C for 1 hr and washed with hexane to improve hydrophobicity . Lipid monolayers formed at the air/water interface were then recovered by placing the pre-treated EM grid carbon side down on top of each water droplet for 1 min . The grid was raised above the surface of the Teflon block by injecting ultrapure H2O into the side port and then was lifted off the droplet immediately . Proteins were rapidly diluted to 5 μM in 20 mM MOPS , pH 7 . 5 , 5 mM KCl , 1 mM EDTA , 2 mM MgAC2 , 1 mM DTT , 5% ( wt/vol ) trehalose buffer and then added to the lipid monolayer on the grid and incubated in a 37°C humidity chamber for 1 min . The final KCl concentration in the buffer were adjusted to 100 mM or 140 mM as required . To facilitate structural analysis of the rings , we further optimized the incubation conditions by using an annealing procedure: Rings were nucleated at 37°C for 1 min followed by a 30-min annealing step at 4°C . The grids were rinsed briefly ( ∼10 s ) with incubation buffer alone or with buffer supplemented with CaCl2 ( 0 . 1 , 0 . 5 and 1 mM free ) for Ca2+ treatment studies . The free [Ca2+] was calculated by Maxchelator ( maxchelator . stanford . edu ) . Subsequently , the grids were blotted with Whatman#1 filter paper ( Sigma-Aldrich , St . Louis , MO ) , negatively stained with uranyl acetate solution ( 1% wt/vol ) , and air dried . The negatively stained specimens were examined on a FEI Tecani T12 operated at 120 kV . The defocus range used for our data was 0 . 6–2 . 0 μm . Images were recorded under low-dose conditions ( ∼20 e−/Å2 ) on a 4K × 4K CCD camera ( UltraScan 4000; Gatan , Inc . , Pleasanton , CA ) , at a nominal magnification of 42 , 000× . Micrographs were binned by a factor of 2 at a final sampling of 5 . 6 Å per pixel on the object scale . The image analysis , including size distribution measurements was carried out using ImageJ software .	Reliable communication between neurons is essential for the brain to work properly . This is accomplished by tightly controlling how chemical messengers , called neurotransmitters , move between neurons . Neurotransmitters are typically packaged into bubble-like structures called synaptic vesicles and are released only when the neuron receives an input electrical signal . A set of proteins orchestrates the release of the neurotransmitters from the neuron , which happens after the synaptic vesicles fuse with the cell membrane . Synaptotagmin , a protein found on the surface of the synaptic vesicle , plays many roles in neurotransmitter release . It helps to attach the synaptic vesicle to the cell membrane and also prevents the vesicles from fusing to the membrane in the absence of an appropriate input signal . Most importantly , it detects when the electrical signal arrives at the neuron by binding to calcium ions that flood the cell following the input signal . This triggers the rapid fusion of the vesicles to the cell membrane . It is not clear how Synaptotagmin is able to carry out its different roles and in particular , control how neurotransmitters are released as calcium ions enter the cell . Zanetti et al . have now used a technique called negative stain electron microscopy to investigate how Synaptotagmin molecules taken from mammals arrange themselves on the surface of a membrane . In this technique , individual Synaptotagmin proteins on the surface of a synthetic membrane are chemically marked and their structure is imaged using an electron beam . Using this approach under conditions resembling those in cells , Zanetti et al . found that 15–20 copies of Synaptotagmin came together and formed ring-like structures on the membrane surface . These ring structures were rapidly broken apart when calcium ions were added to them . Further investigations suggest that the ring structures form when synaptic vesicles first attach to a membrane . Overall , it appears that the Synaptotagmin rings act as washers or spacers to prevent the vesicle from fusing to the cell membrane until the rings are disrupted by the arrival of calcium ions . Future studies are now needed to investigate whether the ring structures form inside cells and whether they act together with other proteins involved in neurotransmitter release .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics", "neuroscience" ]	2016	Ring-like oligomers of Synaptotagmins and related C2 domain proteins
COPI coated vesicles carry material between Golgi compartments , but the role of COPI in the secretory pathway has been ambiguous . Previous studies of thermosensitive yeast COPI mutants yielded the surprising conclusion that COPI was dispensable both for the secretion of certain proteins and for Golgi cisternal maturation . To revisit these issues , we optimized the anchor-away method , which allows peripheral membrane proteins such as COPI to be sequestered rapidly by adding rapamycin . Video fluorescence microscopy revealed that COPI inactivation causes an early Golgi protein to remain in place while late Golgi proteins undergo cycles of arrival and departure . These dynamics generate partially functional hybrid Golgi structures that contain both early and late Golgi proteins , explaining how secretion can persist when COPI has been inactivated . Our findings suggest that cisternal maturation involves a COPI-dependent pathway that recycles early Golgi proteins , followed by multiple COPI-independent pathways that recycle late Golgi proteins . The COPI coat was first visualized nearly 30 years ago on vesicles budding from Golgi cisternae ( Orci et al . , 1986 ) . This coat was shown to consist of a soluble heptameric complex that is recruited to Golgi membranes by the small GTPase Arf1 ( Serafini et al . , 1991; Waters et al . , 1991; Yu et al . , 2012 ) . During vesicle formation , COPI polymerizes to form a curved lattice that captures specific cargoes , including p24 family proteins and certain SNAREs ( Beck et al . , 2009 ) . COPI can be divided into the B subcomplex , which consists of α- , β’- and ε-COP , and the F subcomplex , which consists of β- , δ- , γ- , and ζ-COP ( Lowe and Kreis , 1995; Gaynor et al . , 1998; Lee and Goldberg , 2010; Jackson , 2014 ) . In mammalian and plant cells , COPI vesicles bud from cisternae throughout the Golgi stack except for cisternae of the trans-Golgi network ( TGN ) , which produces clathrin-coated vesicles ( Ladinsky et al . , 1999; Staehelin and Kang , 2008; Klumperman , 2011 ) . Yet despite this wealth of biochemical , morphological , and structural information , the functions of COPI have been hard to elucidate . The strongest data implicate COPI in retrograde transport to the ER . Transmembrane ER resident proteins occasionally escape from the ER , and are then retrieved from the Golgi or ER-Golgi intermediate compartment ( ERGIC ) in retrograde COPI vesicles ( Szul and Sztul , 2011; Barlowe and Miller , 2013 ) . Some transmembrane ER proteins contain cytosolically-oriented C-terminal KKxx-type signals , which are recognized by COPI for retrieval to the ER ( Cosson and Letourneur , 1997 ) . COPI also retrieves transmembrane ER proteins that associate with the Rer1 recycling factor , as well as transmembrane ER proteins that contain arginine-based sorting signals ( Sato et al . , 2001; Michelsen et al . , 2007 ) . Finally , COPI plays a role in retrieving the KDEL receptor ( Orci et al . , 1997 ) , which binds escaped luminal ER proteins . Retrograde COPI vesicles are captured at the ER by the Dsl1 tethering complex in a process that involves recognition of the coat ( Ren et al . , 2009; Zink et al . , 2009 ) . COPI also plays a role in intra-Golgi traffic , but the evidence is open to multiple interpretations . Initially the proposal was that COPI vesicles carry secretory cargoes forward from one Golgi cisterna to the next in a “vesicle shuttle” ( Malhotra et al . , 1989; Orci et al . , 1989 ) . After the discovery of COPI-mediated Golgi-to-ER recycling , the vesicle shuttle model was extended by proposing that COPI vesicles act as bidirectional carriers in the ER-Golgi system ( Pelham , 1994; Rothman , 1996 ) . However , the idea that COPI vesicles carry secretory cargoes from one cisterna to the next faced the problem that some secretory cargoes are much larger than COPI vesicles . Examples include the cell-surface scales that are secreted by certain algae , and procollagen bundles in mammalian fibroblasts ( Leblond , 1989; Becker et al . , 1995 ) . This problem was addressed by the cisternal maturation model , which states that cisternae form at the cis face of the Golgi , then move through the stack to the trans face , then finally peel off to become secretory vesicles ( Glick et al . , 1997; Mironov et al . , 1997; Bonfanti et al . , 1998; Glick and Malhotra , 1998; Pelham , 1998 ) . Thus , entire cisternae could act as forward carriers for secretory cargoes . COPI vesicles have been proposed to recycle resident Golgi proteins within the organelle ( Rabouille and Klumperman , 2005 ) . Consistent with this idea , resident Golgi proteins have been detected in mammalian COPI vesicles ( Martínez-Menárguez et al . , 2001; Malsam et al . , 2005; Gilchrist et al . , 2006; Pellett et al . , 2013; Eckert et al . , 2014 ) —although conflicting results have been reported in other studies ( Orci et al . , 2000a; Cosson et al . , 2002; Kweon et al . , 2004 ) —and the localization of some yeast and plant Golgi proteins has been shown to involve COPI ( Todorow et al . , 2000; Tu et al . , 2008; Woo et al . , 2015 ) . Early versions of the cisternal maturation model postulated that COPI vesicles move in a directed fashion from older to younger cisternae ( Glick et al . , 1997; Glick and Malhotra , 1998 ) . However , no mechanism for such directed movement has yet emerged , suggesting instead that COPI vesicles “percolate” bidirectionally between different cisternae ( Orci et al . , 2000b; Day et al . , 2013 ) . Regardless of the specific traffic pattern of COPI vesicles within the Golgi , the result is thought to be a net retrograde movement of resident Golgi proteins as the cisternae mature ( Glick and Malhotra , 1998; Day et al . , 2013 ) . The cisternal maturation model does not rule out additional roles for COPI in the traffic of secretory cargoes . For example , some secretory cargoes could move forward through the Golgi on a “fast track” involving anterograde COPI vesicles ( Pelham and Rothman , 2000 ) . This concept is supported by evidence that both resident Golgi proteins and secretory cargoes can be incorporated into COPI vesicles ( Orci et al . , 1997; Malsam et al . , 2005; Pellett et al . , 2013 ) . Furthermore , COPI-dependent tubules that connect heterologous cisternae have been implicated in anterograde traffic through the mammalian Golgi ( Yang et al . , 2011; Park et al . , 2015 ) . A prediction of current models is that COPI should be required for secretion . According to the vesicle shuttle model , COPI carries secretory cargoes through the Golgi . According to the cisternal maturation model , COPI drives the maturation process , thereby continually regenerating the Golgi cisternae that serve as anterograde carriers for secretory cargoes . A second prediction is that COPI should be required for Golgi maturation . Presumably , as a Golgi cisterna matures into a trans-Golgi network ( TGN ) compartment , COPI vesicles bud to remove resident Golgi proteins , thereby helping to drive the Golgi-to-TGN biochemical conversion ( Papanikou and Glick , 2014 ) . Both of these predictions about the role of COPI have been tested using the yeast Saccharomyces cerevisiae . Golgi stacking was lost during the evolution of S . cerevisiae ( Mowbrey and Dacks , 2009 ) , but this yeast retains a compartmentalized Golgi that functionally resembles the stacked organelle seen in other organisms , with a late Golgi compartment that corresponds to the mammalian TGN ( Papanikou and Glick , 2009; Myers and Payne , 2013 ) . Surprisingly , when COPI function was disrupted in S . cerevisiae using thermosensitive mutant COPI subunits , secretion was reportedly inhibited for some proteins but not others ( Gaynor and Emr , 1997 ) . Equally surprising results were obtained when a strain with a thermosensitive mutant COPI subunit was examined by video fluorescence microscopy . Golgi maturation can be readily observed in wild-type S . cerevisiae cells ( Losev et al . , 2006; Matsuura-Tokita et al . , 2006 ) , and when the COPI mutant strain was imaged at the nonpermissive temperature , maturation was slowed but not blocked ( Matsuura-Tokita et al . , 2006 ) . The mildness of these phenotypes has added to the uncertainty about how COPI acts in the secretory pathway . Here , we have reexamined the functions of yeast COPI using a new approach . Yeast COPI was rapidly inactivated using the anchor-away method ( Haruki et al . , 2008; Bharucha et al . , 2013 ) , in which FK506-rapamycin binding protein ( FKBP ) was fused to an “anchor” protein while FKBP-rapamycin binding domain ( FRB ) was fused to a COPI subunit . Addition of rapamycin caused the FRB-tagged protein to be tethered at the anchor site , thereby preventing COPI from carrying out its cellular activities . Fluorescence microscopy revealed that COPI inactivation generated hybrid Golgi structures that contained both early and late Golgi proteins . These structures showed unusual dynamics , with an early Golgi protein often persisting for many minutes while late Golgi proteins underwent relatively normal cycles of arrival and departure . The implication is that COPI selectively drives recycling of early but not of late Golgi proteins . After COPI inactivation , the Golgi remained partially functional , indicating that recycling of early Golgi proteins is dispensable for secretion . This analysis helps to integrate the yeast data into a broader understanding of COPI function . COPI has been assumed to operate in vivo as a stable heptameric complex ( Hara-Kuge et al . , 1994; Sahlmüller et al . , 2011; Yip and Walz , 2011 ) . Although COPI can be experimentally separated into the coat-like B subcomplex and the adaptor-like F subcomplex ( Beck et al . , 2009; Jackson , 2014 ) , recent structural data indicate that the two subcomplexes assemble together to generate the coat ( Dodonova et al . , 2015 ) . However , an earlier genetic study hinted that different yeast COPI subunits might act in distinct cellular pathways ( Gabriely et al . , 2007 ) . To examine the in vivo distribution of COPI , we used fluorescence microscopy to visualize several COPI subunits in S . cerevisiae ( Gaynor et al . , 1998 ) . The tagged subunits were Sec26 ( β-COP ) , which is part of the F subcomplex , together with either the Sec21 ( γ-COP ) subunit of the F subcomplex or the Ret1 ( α-COP ) subunit of the B subcomplex . Gene replacement was used to tag Sec26 with GFP , and then to tag Sec21 or Ret1 with mCherry . In both cases the colocalization at punctate structures was essentially complete ( Figure 1A ) . This result suggests that the two subcomplexes of COPI are associated in vivo , at least when COPI is bound to membranes . 10 . 7554/eLife . 13232 . 003Figure 1 . Localization of COPI subunits in yeast . ( A ) COPI subunits were tagged by gene replacement to generate Sec26-GFP , Sec21-mCherry , or Ret1-mCherry . Cells grown in NSD medium at 23°C were imaged by confocal microscopy , and two-color Z-stacks were average projected . Projections from the red and green channels were overlaid to generate merged images . Representative cells are shown . Scale bar , 2 μm . ( B ) Ret1-mCherry was expressed together with the early Golgi marker GFP-Vrg4 . Imaging conditions were as in ( A ) . Scale bar , 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 13232 . 003 The COPI fluorescence pattern was somewhat variable depending on the cell being examined and the imaging conditions . Virtually all of the cells showed large COPI puncta , but a subset of the cells also showed small COPI puncta that may represent intermediates in Golgi assembly ( Figure 1A ) ( Arai et al . , 2008 ) . COPI colocalized strongly with GFP-Vrg4 , which is a GDP-mannose transporter that marks the early Golgi ( Figure 1B ) ( Dean et al . , 1997; Abe et al . , 2004; Losev et al . , 2006 ) . These observations fit with electron tomography studies of mammalian and plant cells , where COPI buds were detected on Golgi but not TGN cisternae ( Ladinsky et al . , 1999; Mogelsvang et al . , 2004; Staehelin and Kang , 2008 ) . Thus , yeast COPI is likely to act at the early Golgi . To test the role of COPI in yeast , we turned to the anchor-away method ( Haruki et al . , 2008 ) . The parental S . cerevisiae strain carried two mutations . First , the cells were rendered resistant to the growth-inhibiting effects of rapamycin by introducing the dominant TOR1-1 point mutation ( Helliwell et al . , 1994 ) . Second , the FPR1 gene , which encodes the yeast homolog of FKBP , was deleted to ensure that FRB-tagged proteins would be trapped at the anchor site rather than binding to cytosolic Fpr1 ( Lorenz and Heitman , 1995; Haruki et al . , 2008 ) . The TOR1-1 fpr1△ strain grew nearly as fast as the parental strain and was fully resistant to rapamycin . This rapamycin-resistant strain was engineered to express the originally described ribosomal Rpl13A-FKBPx2 anchor ( Haruki et al . , 2008 ) , and to tag yeast COPI with FRB . Some commonly used FRB variants have the T2098L mutation , which allows binding of rapamycin derivatives ( Bayle et al . , 2006 ) . The T2098L mutation has also been reported to strengthen the affinity of FRB for FKBP-rapamycin ( Grünberg et al . , 2010 ) . Therefore , gene replacement was used to tag Sec21 , which is essential for viability ( Hosobuchi et al . , 1992 ) , with either wild-type FRB or FRB ( T2098L ) . The Sec21-FRB strain grew as well as the strain carrying untagged Sec21 , but the Sec21-FRB ( T2098L ) strain showed a pronounced growth defect ( Figure 2A ) . T2098L is known to be a destabilizing mutation ( Stankunas et al . , 2007 ) , so we suspect that FRB ( T2098L ) caused degradation of tagged Sec21 . Interestingly , the growth defect could be rescued by appending maltose-binding protein ( MPB ) ( Figure 2A ) or GFP ( data not shown ) to the C-terminus of FRB ( T2098L ) . After appending MBP or GFP , the destabilized FRB ( T2098L ) domain was no longer at an end of the polypeptide chain , and this change may have slowed degradation ( Prakash et al . , 2004; Fishbain et al . , 2011 ) . Regardless of the specific mechanisms at work , these results indicate that FRB ( T2098L ) is unsuitable as a tag for the anchor-away method , and that wild-type FRB should be used instead . 10 . 7554/eLife . 13232 . 004Figure 2 . Evaluation of tag-anchor combinations by growth curve analysis . TOR1-1 fpr1△ strains were grown with shaking in YPD medium at 30°C to mid-log phase , then diluted in fresh medium to an optical density at 600 nm ( OD600 ) of 0 . 15 . Where indicated , rapamycin ( “Rap” ) was then added to 1 μg/mL . After further incubation at 30°C , aliquots were taken at the indicated time points to measure the OD600 . ( A ) The function of Sec21 is preserved after tagging with wild-type FRB but is compromised after tagging with mutant FRB ( T2098L ) . Gene replacement was used to extend Sec21 at the C-terminus with either wild-type FRB , or FRB ( T2098L ) , or an FRB-MBP dual tag , or an FRB ( T2098L ) -MBP dual tag . A control strain expressed wild-type Sec21 . ( B ) COPI can be inactivated by extending Sec21 with a single FRB tag followed by anchoring to ribosomes using the Rpl13A-FKBPx2 anchor , or by extending Sec21 with a double FRB tag followed by anchoring to mitochondria using the OM45-FKBPx4 anchor . ( C ) COPII can be inactivated by extending Sec31 with a single or double FRB tag followed by anchoring to either ribosomes or mitochondria . DOI: http://dx . doi . org/10 . 7554/eLife . 13232 . 00410 . 7554/eLife . 13232 . 005Figure 2—figure supplement 1 . Effect of rapamycin concentration on anchor- mediated growth inhibition . As in Figure 2 , liquid cultures from a strain expressing Sec21-FRB and the Rpl13A-FKBPx2 anchor were treated with rapamycin , and growth was tracked by measuring the optical density at 600 nm for up to 24 hr . Rapamycin concentrations are indicated in μg/mL . DOI: http://dx . doi . org/10 . 7554/eLife . 13232 . 00510 . 7554/eLife . 13232 . 006Figure 2—figure supplement 2 . Localization of FKBP-tagged OM45 to mitochondria . Cells containing mitochondria labeled with matrix-targeted mCherry ( red ) were engineered to express OM45-FKBPx4 , which includes a single C-terminal HA epitope tag . The cells were processed for immunofluorescence microscopy using a polyclonal rabbit anti-HA antibody followed by Alexa Fluor 488-conjugated goat anti-rabbit IgG ( green ) . Samples were then visualized by widefield microscopy . Merged images show colocalization of the green and red signals . Scale bar , 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 13232 . 006 The next step was to test whether growth was inhibited when Sec21 was anchored away . Rpl13A-FKBPx2 served as an anchor on ribosomes . When this anchor was present in a rapamycin-resistant strain encoding wild-type Sec21 , rapamycin addition had no effect on growth , as expected ( Figure 2B ) . By contrast , when Sec21 was tagged with FRB , rapamycin completely blocked growth ( Figure 2B ) . This experiment employed rapamycin at 1 μg/mL , a concentration that was sufficient for maximal growth inhibition ( Figure 2—figure supplement 1 ) . A similar analysis was performed using the mitochondrial outer membrane protein OM45 ( Yaffe et al . , 1989 ) as an anchor . As previously observed ( Cerveny et al . , 2001 ) , tagged and overexpressed OM45 localized to mitochondria ( Figure 2—figure supplement 2 ) . With a strain expressing OM45-FKBPx2 and Sec21-FRB , growth was only partially inhibited by rapamycin ( Figure 2B ) . A likely explanation is that binding of FRB to FKBP-rapamycin is readily reversible ( Banaszynski et al . , 2005 ) . If so , then why is the ribosomal anchor more effective than the mitochondrial anchor ? We believe the answer is that ribosomes are present throughout the cytoplasm whereas mitochondria have a more restricted location . After dissociating from an Rpl13A-FKBPx2 anchor , Sec21-FRB will quickly be captured again by encountering another ribosome , whereas after dissociating from an OM45-FKBPx2 anchor , Sec21-FRB may have time to function at the Golgi before encountering another mitochondrion . To test this hypothesis , we added a double FRB tag to Sec21 while increasing the number of FKBP domains on OM45 to four . These changes were predicted to increase the functional affinity of the anchoring interaction at mitochondria . Indeed , with a strain expressing OM45-FKBPx4 and Sec21-FRBx2 , rapamycin completely blocked growth ( Figure 2B ) . We also tried constructing a strain that expressed Rpl13A-FKBPx4 and Sec21-FRBx2 , but growth was slow , perhaps because the functional affinity of the interaction was high enough to compromise COPI even in the absence of rapamycin ( data not shown ) . In conclusion , growth assays indicate that effective anchoring can be achieved using either Sec21-FRB with a ribosomal Rpl13A-FKBPx2 anchor , or Sec21-FRBx2 with a mitochondrial OM45-FKBPx4 anchor . As a control , we verified that anchoring of COPII also caused growth inhibition , as previously observed with Pichia pastoris ( Bharucha et al . , 2013 ) . For this purpose , we added a single- or double-FRB tag to the essential COPII coat subunit Sec31 ( Salama et al . , 1997 ) . The anchors were Rpl13A or OM45 , linked to either two or four copies of FKBP . As shown in Figure 2C , all of the strains grew well , even the one with the Rpl13A-FKBPx4/Sec31-FRBx2 combination , and rapamycin blocked growth of all of the strains , even the one with the OM45-FKBPx2/Sec31-FRB combination . Thus , the anchor-away method is effective for inactivating components of the secretory pathway , but different components vary in their sensitivities to specific tag-anchor combinations . For further analysis , we relied mainly on the mitochondrial anchor because rapamycin-driven redistribution of labeled proteins to mitochondria can be easily visualized . The first step was to monitor the time course of the anchor-away process using Sec21-FRBx2-GFP with OM45-FKBPx4 . The mitochondria were also labeled with matrix-localized mCherry . Prior to rapamycin addition , Sec21-FRBx2-GFP was present on punctate early Golgi structures that showed no consistent association with mitochondria ( Figure 3A ) . At 10 min after addition of rapamycin at 23°C , Sec21-FRBx2-GFP showed substantial association with mitochondria ( Figure 3A ) . Quantitation of the overlap indicated that this redistribution was maximal by 10–15 min after rapamycin addition , reaching a level of 75–80% colocalization of Sec21-FRBx2-GFP with the mitochondrial matrix marker ( Figure 3B ) . We then examined strains in which Sec21-FRBx2 was nonfluorescent , with the GFP tag fused instead to other COPI subunits . Rapamycin-driven anchoring of Sec21-FRBx2 caused redistribution of Ret1-GFP or Sec26-GFP to mitochondria ( Figure 3—figure supplement 1 ) . Thus , the anchor-away method redistributes the entire COPI complex . 10 . 7554/eLife . 13232 . 007Figure 3 . Visualization of COPI anchoring to mitochondria . ( A ) Sec21 tagged with FRBx2-GFP can be anchored to mitochondria containing OM45-FKBPx4 . The mitochondrial matrix was labeled with mCherry . Representative cells are shown before drug treatment , and after treatment for 10 min at 23°C with 1 μg/mL rapamycin ( “Rap” ) . Cells were imaged by confocal microscopy followed by deconvolution . Scale bar , 2 µm . ( B ) Maximal anchoring to mitochondria occurs within 10–15 min . Anchoring of Sec21-FRBx2-GFP was quantified by measuring the fraction of the GFP signal that colocalized with the mCherry signal ( Levi et al . , 2010 ) at different times after rapamycin addition at 23°C . Approximately 20–30 cells were analyzed at each time point . Error bars show s . e . m . ( C ) Anchoring COPI also anchors early Golgi cisternae . Sec21 was tagged with FRBx2 in a strain with the OM45-FKBPx4 mitochondrial anchor . The strain also expressed GFP-Vrg4 to label early Golgi cisternae , Ret1-mCherry to label COPI , and mito-TagBFP to label the mitochondrial matrix . Cells were imaged before or after rapamycin addition as in ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13232 . 00710 . 7554/eLife . 13232 . 008Figure 3—figure supplement 1 . Simultaneous anchoring of multiple COPI subunits to mitochondria . Quantitation of time-dependent anchoring to mitochondria was performed as in Figure 3B , except that Sec21 was tagged with FRBx2 and the GFP tag was appended to either Ret1 or Sec26 . DOI: http://dx . doi . org/10 . 7554/eLife . 13232 . 008 An interesting question was whether COPI became anchored to mitochondria as an isolated complex , or whether it remained bound to Golgi membranes that became tethered to mitochondria . The anchored COPI was often punctate rather than being evenly distributed over the mitochondria ( Figure 3A ) , suggesting that entire Golgi compartments were being tethered . We analyzed cells that contained GFP-Vrg4 as an early Golgi marker and Ret1-mCherry as a COPI marker , plus a blue fluorescent marker for the mitochondrial matrix ( Murley et al . , 2013 ) . Rapamycin-driven anchoring of COPI to mitochondria using Sec21-FRBx2 also displaced GFP-Vrg4 ( Figure 3C ) , indicating that at least some of the anchored COPI retained its association with Golgi membranes . Although anchored COPI can still bind to the Golgi , the simultaneous anchoring to mitochondria or ribosomes evidently prevents COPI from performing its essential functions . If anchored COPI is functionally inactive as suggested by the growth curve analysis , then recycling of proteins from the Golgi to the ER should be inhibited . To test this prediction , we used a construct encoding the transmembrane domain ( TMD ) of the ER protein Sec71 fused to GFP ( Sato et al . , 2003 ) . Sec71TMD-GFP localizes to the ER with the aid of the retrieval receptor Rer1 ( Sato et al . , 2003 ) , which recycles from the Golgi to the ER in a COPI-dependent manner ( Boehm et al . , 1997; Sato et al . , 2001 ) . Control experiments confirmed that Sec71TMD-GFP was in the ER as indicated by nuclear envelope fluorescence plus additional fluorescence from peripheral ER structures ( Sato et al . , 2003 ) . When Sec71TMD-GFP was expressed using an integrating vector in a wild-type strain , a bright ER signal was seen in essentially all of the cells ( Figure 4A ) . By contrast , in an rer1△ mutant strain , the Sec71TMD-GFP signal in the ER was faint ( Figure 4A ) . The rer1△ strain showed a vacuolar lumen signal as previously reported ( Sato et al . , 2003 ) , although the strength of this signal varied between cells , probably because older cells had degraded more mislocalized Sec71TMD-GFP molecules and had therefore accumulated more GFP in the vacuole . These results validate the use of ER localization of Sec71TMD-GFP as a readout for Golgi-to-ER recycling . 10 . 7554/eLife . 13232 . 009Figure 4 . Depletion of Sec71TMD-GFP from the ER after COPI inactivation . ( A ) Sec71TMD-GFP requires Rer1 for normal ER localization . An expression vector for Sec71TMD-GFP was integrated into the wild-type JK9-3da strain or into an isogenic rer1△ strain . Cells grown to log phase in NSD at 30°C were compressed beneath a coverslip and visualized in a single focal plane by widefield microscopy . Representative images taken at the same exposure are shown . Scale bar , 2 μm . ( B ) Anchoring of COPI causes loss of Sec71TMD-GFP from the ER . Both of the strains shown here contained integrating vectors for expressing the mitochondrial anchor OM45-FKBPx4 as well as Sec71TMD-GFP , and one strain also expressed Sec21-FRBx2 to anchor COPI . Cells growing at 30°C were incubated for 10 min with 0 . 0002% Hoechst 33342 to stain nuclei , followed by an additional treatment for up to 60 min with 1 μg/mL rapamycin ( “Rap” ) . At the indicated time points after rapamycin addition , cells were compressed beneath a coverslip and visualized in a single focal plane by widefield microscopy . Representative images of GFP fluorescence are shown . Scale bar , 2 μm . ( C ) The data from ( B ) were quantified by measuring the GFP fluorescence signals from nuclei as a proxy for the ER signals . For this purpose , the signal in the blue channel was used to select nuclei in ImageJ using the Magic Wand tool . These selections were expanded by 12 pixels to include the nuclear envelope , and the GFP fluorescence from the selected areas in the green channel was quantified . Plotted are average fluorescence signals per unit area . Each data point was derived from approximately 40–80 cells , and was adjusted by subtracting an average background signal from nuclei in the parental strain lacking GFP . The dashed line represents the average background-corrected nuclear fluorescence signal from rer1Δ cells that exhibited low to moderate vacuolar fluorescence . DOI: http://dx . doi . org/10 . 7554/eLife . 13232 . 00910 . 7554/eLife . 13232 . 010Figure 4—figure supplement 1 . Labeling of nuclei to quantify Sec71TMD-GFP fluorescence in the nuclear envelope . Representative cells are shown from the analysis in Figure 4B , C at 60 min after rapamycin addition . The Sec71TMD-GFP fluorescence signal is shown in green , and cellular DNA stained with Hoechst dye is shown in blue . Hoechst-labeled nuclei were used to identify the nuclear envelope component of the Sec71TMD-GFP signal . DOI: http://dx . doi . org/10 . 7554/eLife . 13232 . 010 We predicted that inactivation of COPI would prevent Rer1-dependent Golgi-to-ER recycling while allowing ER export , and would therefore reduce Sec71TMD-GFP levels in the ER . This reduction was expected to be gradual because Sec71TMD-GFP lacks a known ER export signal and should be incorporated at a low efficiency into COPII vesicles ( Dancourt and Barlowe , 2010 ) . In a control strain that contained the mitochondrial OM45-FKBPx4 anchor , addition of rapamycin had no effect on the amount of Sec71TMD-GFP in the ER , as indicated by the fluorescence signal from the nuclear envelope ( Figure 4B , C and Figure 4—figure supplement 1 ) . By contrast , when the OM45-FKBPx4 strain also contained Sec21-FRBx2 for anchoring COPI , addition of rapamycin caused progressive depletion of Sec71TMD-GFP from the ER ( Figure 4B , C and Figure 4—figure supplement 1 ) . After 60 min of rapamycin treatment , the level of Sec71TMD-GFP in the ER was comparable to that seen in the rer1△ strain ( dashed line in Figure 4C ) . This observation provides further evidence that anchoring to mitochondria inactivates COPI . To determine how COPI inactivation affects forward traffic through the secretory pathway , we examined general secretion . Cells were incubated at 30°C with 35S-labeled amino acids during a 10-min pulse and then chased for 30 min , followed by analysis of the radiolabeled proteins secreted into the culture medium ( Gaynor and Emr , 1997 ) . This procedure was performed either with untreated cells , or with cells that had received rapamycin 10 min before the pulse . In a control strain that contained the mitochondrial OM45-FKBPx4 anchor , a series of radioactive protein bands was detected in the absence of rapamycin , and this pattern was unchanged by addition of rapamycin ( Figure 5A ) . However , when the OM45-FKBPx4 strain also contained Sec31-FRBx2 for anchoring the outer layer of the COPII coat , general secretion was completely blocked by rapamycin addition ( Figure 5A ) . As a control , rapamycin had no effect on cellular protein synthesis ( Figure 5—figure supplement 1 ) . The inhibitory effect of anchoring Sec31-FRBx2 was expected because proteins following the conventional secretory pathway all require COPII to leave the ER ( Barlowe and Miller , 2013 ) . 10 . 7554/eLife . 13232 . 011Figure 5 . Effects of anchoring COPI or COPII on general secretion . ( A ) Anchoring COPII inhibits secretion more strongly than anchoring COPI . The yeast strains used here all expressed the mitochondrial anchor OM45-FKBPx4 . Where indicated , a strain also expressed either Sec31-FRBx2 to anchor COPII , or Sec21-FRBx2 to anchor COPI . Cells growing at 30°C were either left untreated or treated for 10 min with 1 μg/mL rapamycin ( “Rap” ) , then pulsed for 10 min with 35S amino acids , then chased for 30 min . The culture medium was analyzed by SDS-PAGE and autoradiography to detect secreted proteins that had been labeled during the pulse . Numbers represent the molecular weight in kDa of marker proteins . After anchoring COPI , the arrow marks a band of secreted protein that was only partially diminished , and the arrowhead marks a band of secreted protein that was severely diminished . ( B ) Efficient anchoring of Sec21 requires two copies of FRB with a mitochondrial anchor but only one copy of FRB with a ribosomal anchor . General secretion was visualized by a pulse-chase procedure as in ( A ) , except that the control strain expressed neither an anchor nor an FRB-tagged protein . Where indicated , a strain expressed either a ribosomal Rpl13A-FKBPx2 anchor or a mitochondrial OM45-FKBPx2 or OM45-FKBPx4 anchor , plus either Sec21-FRB or Sec21-FRBx2 . DOI: http://dx . doi . org/10 . 7554/eLife . 13232 . 01110 . 7554/eLife . 13232 . 012Figure 5—figure supplement 1 . Insensitivity of cellular protein synthesis to rapamycin . The experimental procedure was identical to the one in Figure 5A , except that cellular proteins were analyzed rather than proteins secreted into the culture medium . A control ( data not shown ) confirmed that as in Figure 5A , rapamycin inhibited general secretion completely when COPII was anchored away , but only partially when COPI was anchored away . Numbers represent the molecular weight in kDa of marker proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 13232 . 01210 . 7554/eLife . 13232 . 013Figure 5—figure supplement 2 . Time course of the rapamycin effect on general secretion . The yeast strain expressed the mitochondrial anchor OM45-FKBPx4 , and also expressed Sec21-FRBx2 to anchor COPI . Cells growing at 30°C were either left untreated or treated for the indicated times with 2 μg/mL rapamycin ( “Rap” ) , then pulsed for 10 min with 35S amino acids , then chased for 30 min . The culture medium was analyzed by SDS-PAGE and autoradiography to detect secreted proteins that had been labeled during the pulse . Numbers represent the molecular weight in kDa of marker proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 13232 . 01310 . 7554/eLife . 13232 . 014Figure 5—figure supplement 3 . Partial inhibition of secretion after anchoring different COPI subunits . ( A ) Growth was monitored with or without rapamycin addition as in Figure 2 . The parental strain carried the OM45-FKBPx4 anchor . Where indicated , the strains also carried Sec21-FRBx2 or Ret1-FRBx2 or both . ( B ) A strain carrying the OM45-FKBPx2 anchor was modified to express Ret1-FRBx2-GFP and Sec21-mCherry . Confocal microscopy revealed that addition of 1 μg/mL rapamycin ( “Rap” ) for 10 min at 23°C anchored both tagged COPI subunits to mitochondria , which were labeled with mito-TagBFP . Scale bar , 2 μm . ( C ) The strains in ( A ) were subjected to pulse-chase analysis to detect general secretion as in Figure 5A . DOI: http://dx . doi . org/10 . 7554/eLife . 13232 . 014 When the OM45-FKBPx4 strain also contained Sec21-FRBx2 for anchoring COPI , general secretion was reduced but not abolished ( Figure 5A ) . We extended this test by comparing three different anchoring configurations for COPI ( Figure 5B ) . In a strain that contained the ribosomal Rpl13A-FKBPx2 anchor and Sec21-FRB , rapamycin reduced secretion to a similar extent as in the strain that contained the mitochondrial OM45-FKBPx4 anchor and Sec21-FRBx2 . However , in a strain that contained the mitochondrial OM45-FKBPx2 anchor and Sec21-FRB , rapamycin had only a marginal effect on secretion . These results fit with the growth tests indicating that a single copy of FRB is sufficient to inactivate COPI using the ribosomal anchor , but that two copies of FRB are needed using the mitochondrial anchor ( see above ) . Figure 5A shows that after anchoring COPI , the secretion of some proteins was only weakly inhibited as marked by the arrow , whereas the secretion of other proteins was strongly inhibited as marked by the arrowhead . Many proteins fell between these two extremes , showing extensive but not complete inhibition . Similar results were obtained when the rapamycin concentration was doubled to 2 µg/mL and the drug pretreatment was varied between 5 and 30 min ( Figure 5—figure supplement 2 ) . Moreover , when the FRBx2 tag was placed on Ret1 instead of Sec21 , rapamycin inhibited growth and anchored COPI but only partially inhibited secretion ( Figure 5—figure supplement 3 ) , mimicking the effects seen with Sec21-FRBx2 . Even when both Sec21 and Ret1 were tagged with FRBx2 , rapamycin did not fully inhibit secretion ( Figure 5—figure supplement 3 ) . Based on these data , we conclude that COPI inactivation allows general secretion to continue at a reduced level , with the magnitude of the reduction varying for different proteins . Because the effects of anchoring COPI are similar to those previously seen after thermal inactivation of COPI in a sec21-3 mutant ( Gaynor and Emr , 1997 ) , we performed a direct comparison of the two approaches . Sequencing of genomic DNA from a sec21-3 mutant strain revealed that this allele carries a single F720S point mutation . We introduced this mutation into our parental strain by gene replacement . As anticipated , the sec21-3 mutant cells grew at 23°C but not at 37°C ( Figure 6—figure supplement 1 ) . The sec21-3 mutant was shifted to 37°C for 30 min prior to pulse-chase analysis . In parallel , anchoring to mitochondria was performed at 37°C . As shown in Figure 6 , the sec21-3 mutation had the same effect on general secretion as anchoring Sec21-FRBx2 . In both cases , secretion was inhibited to varying extents for different proteins . Anchoring Sec31-FRBx2 at 37°C completely blocked general secretion ( Figure 6 ) , confirming that secretion has a more stringent requirement for COPII than for COPI . 10 . 7554/eLife . 13232 . 015Figure 6 . Comparison of methods for inactivating Sec21 . General secretion was visualized by a pulse-chase analysis as in Figure 5A , except that cells were grown at 23°C and then shifted to 37°C for 30 min before the procedure . A control strain expressed the mitochondrial OM45-FKBPx4 anchor . Where indicated , a strain expressed OM45-FKBPx4 plus either Sec31-FRBx2 to anchor COPII or Sec21-FRBx2 to anchor COPI . The sec21-3 mutation was in the parental strain background . “Rap” indicates a 10-min treatment with 1 μg/mL rapamycin prior to the pulse . Numbers represent the molecular weight in kDa of marker proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 13232 . 01510 . 7554/eLife . 13232 . 016Figure 6—figure supplement 1 . Thermosensitivity of the sec21-3 strain . Ten μL of 10-fold serial dilutions of wild-type or isogenic sec21-3 mutant cells were spotted in parallel on two YPD plates . The plates were then incubated for two days at either 23°C or 37°C . DOI: http://dx . doi . org/10 . 7554/eLife . 13232 . 01610 . 7554/eLife . 13232 . 017Figure 6—figure supplement 2 . Perturbed Golgi structure in sec21-3 cells at the permissive temperature . A wild-type strain or an isogenic sec21-3 strain were transformed to express the early Golgi marker GFP-Vrg4 and the late Golgi marker Sec7-DsRed . Cells were imaged at 23°C by widefield microscopy . Representative cells are shown . Scale bar , 2 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 13232 . 017 These results suggested that the sec21-3 mutant strain might be useful for a fluorescence microscopy analysis of how COPI inactivation affects the Golgi . Unfortunately , when the sec21-3 strain was transformed to express the early Golgi marker GFP-Vrg4 and the late Golgi marker Sec7-DsRed ( Losev et al . , 2006 ) , Golgi organization was perturbed even at the permissive temperature . Compared to a wild-type control strain , the size of the labeled puncta was more variable , and a greater fraction of the GFP-Vrg4 marker was found in non-punctate structures ( Figure 6—figure supplement 2 ) . Thus , the anchor-away approach is preferable for microscopy because Golgi organization should be normal prior to rapamycin addition . We asked whether Golgi organization was altered by inactivating COPI . The first test employed fluorescence microscopy with GFP-Vrg4 and Sec7-DsRed . As previously observed ( Losev et al . , 2006 ) , early and late cisternae appeared as distinct puncta in the absence of rapamycin ( Figure 7A ) . In a strain carrying the OM45-FKBPx4 anchor and wild-type Sec21 , addition of rapamycin had no effect on this pattern ( Figure 7A ) . However , in a strain carrying OM45-FKBPx4 plus Sec21-FRBx2 , addition of rapamycin for 10 min to anchor COPI caused a dramatic change: the labeled structures were fewer in number and often larger , and many of them showed colocalization of the early and late Golgi markers ( Figure 7A ) . The green and red fluorescence patterns were typically not identical , suggesting that the early and late Golgi markers were associated but partially segregated . This altered distribution of Golgi markers remained largely unchanged after prolonged treatment with rapamycin , except that after 30 min , some of the GFP-Vrg4 began to appear at the vacuolar membrane and plasma membrane ( data not shown ) . A similar conversion of separate to partially overlapping localizations of early and late Golgi markers was seen when a ribosomal anchor was used instead of a mitochondrial anchor ( Figure 7—figure supplement 1 ) , or when the early Golgi marker was the Cog1 subunit of the COG vesicle tethering complex ( Figure 7—figure supplement 2 ) ( Willett et al . , 2013 ) . We infer that COPI inactivation generates hybrid Golgi structures containing both early and late Golgi markers . 10 . 7554/eLife . 13232 . 018Figure 7 . Effects of anchoring COPI or COPII on Golgi organization . ( A ) Anchoring COPI leads to association of early and late Golgi markers . A strain carrying the OM45-FKBPx4 anchor was transformed to express the early Golgi marker GFP-Vrg4 and the late Golgi marker Sec7-DsRed . A derivative strain also expressed Sec21-FRBx2 to anchor COPI . Cells were grown and imaged as in Figure 1 , except that the confocal images were deconvolved . “+ Rap” indicates a 10-min treatment with 1 μg/mL rapamycin prior to imaging . Scale bar , 2 μm . ( B ) Anchoring COPII does not lead to association of early and late Golgi markers . The experiment was performed with rapamycin addition as in ( A ) , except that the strain expressed Sec31-FRBx2 to anchor COPII , and deconvolution was omitted . Similar results were seen when COPII was anchored to mitochondria by incubating with rapamycin for 10 min as in the figure , or for 20 min ( data not shown ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13232 . 01810 . 7554/eLife . 13232 . 019Figure 7—figure supplement 1 . Formation of hybrid Golgi structures with a ribosomal anchor . The analysis was performed as in the lower half of Figure 7A , except that the strain carried the Rpl13A-FKBPx2 ribosomal anchor plus Sec21-FRB to anchor COPI . Scale bar , 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 13232 . 01910 . 7554/eLife . 13232 . 020Figure 7—figure supplement 2 . Visualization of hybrid Golgi structures with an alternative early Golgi marker . The analysis was performed as in the lower half of Figure 7A , except that the early Golgi was marked with Cog1-GFP . Scale bar , 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 13232 . 020 A concern was that formation of hybrid Golgi structures might be a nonspecific consequence of perturbing membrane traffic . To exclude this possibility , we visualized Golgi markers after inactivating COPII . When a strain carrying OM45-FKBPx4 plus Sec31-FRBx2 was treated with rapamycin to anchor COPII , most of the GFP-Vrg4 and Sec7-DsRed were dispersed in the cytoplasm , with the remaining punctate structures showing little colocalization of the two markers ( Figure 7B ) . This result is consistent with an earlier report that inactivation of yeast COPII disrupts the Golgi ( Morin-Ganet et al . , 2000 ) . Thus , the association of early and late Golgi markers after anchoring COPI is a specific effect . To visualize the hybrid Golgi structures at higher resolution , we built on our previous experience with imaging GFP-tagged yeast compartments by correlative fluorescence microscopy and electron tomography ( Bhave et al . , 2014 ) . This method was extended to dual-color imaging by visualizing both GFP-Vrg4 and Sec7-DsRed in plastic-embedded samples ( Figure 8 , insets ) , and was further enhanced by examining thicker sections with scanning transmission electron microscopy ( STEM ) tomography ( Aoyama et al . , 2008; Hohmann-Marriott et al . , 2009; Sousa et al . , 2011 ) . Figure 8A shows Golgi cisternae in untreated cells . Individual early and late Golgi cisternae are identifiable in the tomograms by correlation with the fluorescence data . The tomographic reconstruction ( Video 1 ) indicates that the Golgi cisternae were relatively simple curved structures ( Preuss et al . , 1992; Bhave et al . , 2014 ) , and that the early and late cisternae were not visibly connected . Figure 8B shows Golgi structures after COPI inactivation . The tomographic reconstruction ( Video 2 ) indicates that the early and late Golgi markers were present in irregularly shaped membrane compartments of various sizes . As predicted from the fluorescence analysis ( see above ) , some of the Golgi membranes were closely apposed to mitochondria . We conclude that COPI inactivation converts the yeast Golgi into complex membrane structures that can contain both early and late Golgi proteins . 10 . 7554/eLife . 13232 . 021Figure 8 . Correlative fluorescence microscopy and electron tomography of the yeast Golgi . The yeast strain expressed GFP-Vrg4 and Sec7-DsRed , as well as OM45-FKBPx4 as a mitochondrial anchor plus Sec21-FRBx2 to anchor COPI . Cells were frozen , freeze substituted , and embedded in plastic in a manner that preserved fluorescence . Plastic sections were then examined by fluorescence microscopy followed by STEM tomography . The figures show either ( A ) a section ~0 . 5 μm thick from a representative untreated cell , or ( B ) a section ~1 . 0 μm thick from a representative cell after treatment for 10 min with 1 μg/mL rapamycin . In each case , the left panel shows the fluorescence signals overlaid on a projection of five 3- to 4-nm tomographic slices , the inset shows fluorescence and differential interference contrast analysis of the same cell , and the right panel shows a tomographic model of the relevant labeled structures . Full tomographic reconstructions for the untreated and rapamycin-treated cells are animated in Video 1 and Video 2 , respectively . Green indicates early Golgi cisternae , red indicates late Golgi cisternae , yellow indicates hybrid Golgi structures , and blue indicates mitochondria . Scale bars , 1 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 13232 . 02110 . 7554/eLife . 13232 . 022Video 1 . Tomographic reconstruction of Golgi cisternae in an untreated cell . Yeast cells were analyzed by STEM tomography as described in Figure 8 . The first third of the movie shows every tenth tomographic slice . The second third of the movie shows the same tomographic slices after contours were traced to label early Golgi membranes green and late Golgi membranes red . The final third of the movie shows a rotation of the tomographic model . See also Figure 8A . DOI: http://dx . doi . org/10 . 7554/eLife . 13232 . 02210 . 7554/eLife . 13232 . 023Video 2 . Tomographic reconstruction of hybrid Golgi structures in a cell that was treated with rapamycin to inactivate COPI . The procedure was the same as for Video 1 , except that COPI was inactivated prior to freezing the cells . Putative Golgi membranes are labeled yellow and mitochondrial outer membranes are labeled blue . See also Figure 8B . DOI: http://dx . doi . org/10 . 7554/eLife . 13232 . 023 How does COPI inactivation generate hybrid Golgi structures ? This question was addressed using 4D confocal microscopy . We previously showed that in unperturbed cells expressing GFP-Vrg4 and Sec7-DsRed , a cisterna labels for ~1–3 min with the green early Golgi marker , then loses green fluorescence while acquiring red fluorescence , then labels for ~1–3 min with the red late Golgi marker , then loses red fluorescence ( Losev et al . , 2006 ) . The same maturation dynamics for GFP-Vrg4 and Sec7-DsRed were seen when a strain expressing the OM45-FKBPx4 anchor plus Sec21-FRBx2 was imaged in the absence of rapamycin ( data not shown ) . By contrast , after the same strain had been incubated with rapamycin for 10 min to inactivate COPI , Golgi dynamics were markedly altered as described below . Rapamycin-driven tethering to mitochondria probably facilitated this analysis by making the Golgi structures less mobile and easier to track . After COPI inactivation , two behaviors were evident from the 4D movies ( Video 3 ) . First , when the early and late Golgi markers were both present in a hybrid structure , the two markers remained closely associated even though the shape of the structure fluctuated rapidly . Second , in many of the hybrid Golgi structures , the Sec7-DsRed marker disappeared and then subsequently returned while the GFP-Vrg4 marker persisted the entire time . An example from Video 3 is the Golgi structure marked by an arrowhead in Figure 9—figure supplement 1 . The dynamics of this structure are illustrated in Video 4 , which was generated by cropping the movie frames and then manually erasing the fluorescence of nearby structures at each time point . Figure 9A shows representative frames from Video 4 together with a quantitation of the green and red fluorescence intensities for the entire 15-min movie . The red signal lasted for ~4 . 5 min , then completely disappeared , then returned for ~7 min , then briefly disappeared before beginning to increase once again . A similar pattern was seen with the structure marked by the arrow in Figure 9—figure supplement 1 , although this structure could be tracked reliably for only ~8 . 5 min , as illustrated in Video 5 . Representative frames from Video 5 are shown in Figure 9B . In this case , the red signal disappeared for ~2 min before returning even stronger than before . More generally , although the kinetics varied for different Golgi structures , many of them showed persistent GFP-Vrg4 labeling with alternating high and low levels of Sec7-DsRed labeling . Thus , Sec7-DsRed continued to recycle within the Golgi after COPI inactivation . 10 . 7554/eLife . 13232 . 024Video 3 . Dynamics of Vrg4 and Sec7 after anchoring COPI . Cells in which Sec21 had been anchored to mitochondria were imaged by dual-color 4D confocal microscopy to visualize the dynamics of the early Golgi marker GFP-Vrg4 and the late Golgi marker Sec7-DsRed . Scattered light images were recorded in the blue channel . Complete Z-stacks were collected every 2 s for 15 min , and the data were deconvolved , bleach corrected , and average projected . See also Figure 9 . DOI: http://dx . doi . org/10 . 7554/eLife . 13232 . 02410 . 7554/eLife . 13232 . 025Video 4 . Edited movie showing Golgi Structure 1 from Video 3 . Video 3 was cropped to include only a single budded cell . At each time point , a montage of the Z-stack was generated , and fluorescence signals were erased for all structures except the one marked with the arrowhead in Figure 9—figure supplement 1 . This structure was designated Golgi Structure 1 . An edited version of the cropped movie was then generated . See also Figure 9A . DOI: http://dx . doi . org/10 . 7554/eLife . 13232 . 02510 . 7554/eLife . 13232 . 026Figure 9 . Effects of anchoring COPI on Golgi maturation dynamics . A strain expressing the OM45-FKBPx4 anchor as well as Sec21-FRBx2 was transformed to express the early Golgi marker GFP-Vrg4 and the late Golgi marker Sec7-DsRed . Logarithmically growing cells were attached to a coverglass-bottom dish , and were then treated with 1 μg/mL rapamycin for 10 min , followed by dual-color 4D confocal imaging to generate Video 3 . Edited data sets were generated to analyze the two hybrid Golgi structures marked in Figure 9—figure supplement 1 , yielding Video 4 for Golgi Structure 1 and Video 5 for Golgi Structure 2 . Times are indicated in min:sec format . Scale bars , 2 μm . ( A ) Representative frames are shown from Video 4 , and the green and red fluorescence intensities from Golgi Structure 1 are plotted versus time . ( B ) Representative frames are shown from Video 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 13232 . 02610 . 7554/eLife . 13232 . 027Figure 9—figure supplement 1 . Sample frame from Video 3 . This frame from Video 3 shows the two cisternae that were analyzed in Video 4 , Video 5 , and Figure 9 . The arrowhead indicates Golgi Structure 1 and the arrow indicates Golgi Structure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 13232 . 02710 . 7554/eLife . 13232 . 028Video 5 . Edited movie showing Golgi Structure 2 from Video 3 . Video 3 was cropped to include only a single cell . The procedure was the same as for Video 4 , except that this edited movie displays Golgi Structure 2 , which is marked with an arrow in Figure 9—figure supplement 1 . See also Figure 9B . DOI: http://dx . doi . org/10 . 7554/eLife . 13232 . 028 Because Sec7 is a late Golgi peripheral membrane protein , we wondered whether late Golgi transmembrane proteins would also continue to recycle after COPI inactivation . A late Golgi transmembrane protein that may recycle within the Golgi is the processing protease Kex2 ( Fuller et al . , 1988 ) . It has been proposed that Kex2 traffics between the Golgi and prevacuolar endosomes ( Sipos et al . , 2004; De et al . , 2013 ) , but Kex2 has no known role at prevacuolar endosomes , and the following two results suggest that the major recycling pathway for Kex2 is actually within the Golgi . First , we looked for colocalization of Kex2 with markers of the late Golgi or prevacuolar endosomes . The marker for the late Golgi was Sec7 . The marker for prevacuolar endosomes was Vps8 , a subunit of the CORVET tether ( Arlt et al . , 2015 ) . Late Golgi compartments and prevacuolar endosomes are clearly distinct because in projected confocal images , Sec7-mCherry and Vps8-GFP showed no colocalization apart from a background level that likely represents chance overlap after projection ( Figure 10 ) ( Arlt et al . , 2015 ) . As a control , we examined Vps10 , a vacuolar hydrolase receptor that cycles between the late Golgi and prevacuolar endosomes and localizes to both compartments ( Marcusson et al . , 1994; Cooper and Stevens , 1996; Chi et al . , 2014 ) . Vps10-GFP was present in a subset of the late Golgi structures labeled with Sec7-mCherry and in almost all of the prevacuolar endosomes labeled with Vps8-mCherry ( Figure 10 ) . By contrast , Kex2-GFP colocalized strongly with Sec7-mCherry but showed very little colocalization with Vps8-mCherry ( Figure 10 ) . In some cells , prevacuolar endosomes did contain detectable Kex2-GFP , but this green signal was weak ( Figure 10—figure supplement 1 ) . These results suggest that Kex2 resides in the Golgi , and that cycling through prevacuolar endosomes represents at most a minor pathway for Kex2 trafficking . 10 . 7554/eLife . 13232 . 029Figure 10 . Distinct localization patterns of tagged Vps10 and Kex2 . ( A ) Vps10-GFP localizes to both the Golgi and prevacuolar endosomes whereas Kex2-GFP is largely restricted to the Golgi . Late Golgi compartments were tagged with Sec7-mCherry , or else prevacuolar endosome compartments were tagged with Vps8-mCherry . The localizations of the Vps10-GFP and Kex2-GFP proteins were then examined with reference to these two compartments . As a control , Vps8-GFP was expressed together with Sec7-mCherry to confirm that the two compartments were separate . Representative projected confocal images are shown . Scale bar , 2 μm . ( B ) Quantitation of the data from ( A ) . To analyze an image , a mask was created from the punctate compartment signal , and the percentage of the protein signal visible through the mask was then measured ( Levi et al . , 2010 ) . For each strain , 40 images with ~2–4 cells per image were quantified . Bars represent mean percentage values with s . e . m . Based on the analysis of Vps8 and Sec , the background signal due to chance overlap in this assay was approximately 8–16% . DOI: http://dx . doi . org/10 . 7554/eLife . 13232 . 02910 . 7554/eLife . 13232 . 030Figure 10—figure supplement 1 . Example of an unusual cell with some Kex2-GFP visible in prevacuolar endosomes . In the experiment of Figure 10A , a fraction of the cells had weak but detectable Kex2-GFP signal overlapping with the Vps8-mCherry signal from prevacuolar endosomes , as indicated by the arrowheads in this example . Scale bar , 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 13232 . 030 Second , we used 4D confocal microscopy to compare the dynamics of Kex2-GFP and Sec7-mCherry , and to determine why the colocalization of these two markers is substantial but incomplete ( Franzusoff et al . , 1991; Redding et al . , 1991 ) . It was recently reported that Kex2 departs from Golgi cisternae somewhat earlier than Sec7 ( McDonold and Fromme , 2014 ) . We confirmed and extended that result by showing that Kex2-GFP both arrived and departed slightly before Sec7-mCherry , with the two kinetic traces typically offset by 5–20 s ( Figure 11A and Video 6 ) . Thus , a green spot that lacks red fluorescence could represent a Golgi cisterna that has acquired Kex2-GFP and will soon acquire Sec7-mCherry . Alternatively , a green spot that lacks red fluorescence could represent a non-Golgi compartment that contains Kex2-GFP and will never contain Sec7-mCherry . To determine the relative abundance of these two classes of Kex2-GFP-labeled structures , we analyzed Video 6 and identified the green spots that lacked red fluorescence and that could be reliably tracked for at least 30 s after their initial appearance . A total of 26 such structures were detected . As shown in Video 7 , all 26 structures subsequently acquired Sec7-mCherry . Additional Kex2-GFP-labeled structures that could not be tracked for the full 30 s also acquired Sec7-mCherry ( Figure 11—figure supplement 1 ) . Therefore , most or all of the structures that label strongly with Kex2-GFP are late Golgi cisternae . The straightforward interpretation of these data is that Kex2 recycles within the Golgi , with kinetics slightly offset from those of Sec7 . 10 . 7554/eLife . 13232 . 031Figure 11 . Analysis of Kex2 maturation dynamics with functional or inactivated COPI . ( A ) Kex2 maturation dynamics are slightly offset from those of Sec7 . Cells expressing Kex2-GFP and Sec7-mCherry were attached to a coverglass-bottom dish and imaged by dual-color 4D confocal microscopy to generate Video 6 , in which the top panel is the unedited movie and the bottom panel was generated from edited data sets used to quantify the fluorescence intensities from two cisternae . Representative frames from Video 6 are shown together with the quantitation . Plotted in the bottom panel is fluorescence from the cisterna in the cell at the upper left , and plotted in the top panel is fluorescence from the cisterna in the cell at the lower right . Scale bar , 2 μm . ( B ) GFP-Vrg4 departs as Kex2-mCherry arrives during cisternal maturation . Imaging was performed as in ( A ) , and representative frames from Video 8 are shown together with the quantitation . Plotted in the bottom panel is fluorescence from the cisterna in the cell at the left , and plotted in the top panel is fluorescence from the cisterna in the cell at the right . Scale bar , 2 μm . ( C ) After COPI inactivation , GFP-Vrg4 persists in Golgi structures while Kex2-mCherry cycles . Imaging was performed as in ( A ) , and representative frames from Video 9 are shown together with the quantitation . Plotted in the bottom panel is fluorescence from the Golgi structure in the cell at the left , and plotted in the top panel is fluorescence from the Golgi structure in the cell at the right . These two Golgi structures were tracked for as long as they could be resolved from other fluorescent structures . Scale bar , 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 13232 . 03110 . 7554/eLife . 13232 . 032Figure 11—figure supplement 1 . Additional examples showing the appearance of Sec7 in Kex2-containing structures . In Video 7 , Kex2-GFP-containing structures from Video 6 were analyzed only if they could be tracked for at least 30 s after becoming visible , but additional structures that could not be tracked for the full 30 s showed initial labeling with Kex2-GFP and subsequent appearance of Sec7-mCherry . Three examples are illustrated here . See Video 7 for further details . DOI: http://dx . doi . org/10 . 7554/eLife . 13232 . 03210 . 7554/eLife . 13232 . 033Video 6 . Combined original and edited movie comparing the dynamics of Kex2 and Sec7 . Cells expressing Kex2-GFP and Sec7-mCherry were imaged by dual-color 4D confocal microscopy . Scattered light images were recorded in the blue channel . Complete Z-stacks were collected every 2 s for 5 min , and the data were deconvolved , bleach corrected , and average projected . Edited movies tracking two representative cisternae were generated , merged , and appended below the original movie . See also Figure 11A . DOI: http://dx . doi . org/10 . 7554/eLife . 13232 . 03310 . 7554/eLife . 13232 . 034Video 7 . Frame pairs from Video 6 showing the consistent appearance of Sec7 in Kex2-containing structures . Video 6 was analyzed to identify structures that labeled for Kex2-GFP but not Sec7-mCherry , and that could be followed for at least 30 s after becoming visible . Each frame pair highlights a single Kex2-GFP-containing structure at two closely spaced time points . In the left or right half of the frame pair , the merged green and red fluorescence signals are at the top and the red fluorescence signal is at the bottom . The arrowheads in the left half of the frame pair indicate the structure prior to the appearance of Sec7-mCherry , and the arrowheads in the right half of the frame pair indicate the same structure after the appearance of Sec7-mCherry . For convenience , the 26 frame pairs corresponding to distinct Kex2-GFP-containing structures are displayed as frames in a single movie . See also Figure 11A and Figure 11—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 13232 . 034 Given that Sec7 arrives at a cisterna as Vrg4 is departing ( Losev et al . , 2006 ) , we predicted that GFP-Vrg4 and Kex2-mCherry would show green-to-red maturation with a brief period of overlap . Although Kex2-mCherry gave a comparatively weak signal that made the analysis challenging , maturation from GFP-Vrg4 to Kex2-mCherry was indeed observed ( Figure 11B and Video 8 ) . 10 . 7554/eLife . 13232 . 035Video 8 . Combined original and edited movie showing the dynamics of GFP-Vrg4 and Kex2-mCherry . Cells expressing GFP-Vrg4 and Kex2-mCherry were analyzed as in Video 6 , except that the duration of the movie was 7 min . See also Figure 11B . DOI: http://dx . doi . org/10 . 7554/eLife . 13232 . 035 These control experiments set the stage for testing the effects of anchoring COPI . When COPI was inactivated , the dynamics of Kex2-mCherry were similar to those of Sec7-DsRed , with the levels of Kex2-mCherry alternately rising and falling in Golgi structures that were marked continuously by GFP-Vrg4 ( Figure 11C and Video 9 ) . We conclude that not only peripheral membrane proteins , but also transmembrane proteins of the late Golgi can recycle independently of COPI . 10 . 7554/eLife . 13232 . 036Video 9 . Combined original and edited movie showing the dynamics of GFP-Vrg4 and Kex2-mCherry after anchoring COPI . Cells in which Sec21 had been anchored to mitochondria were imaged as in Video 7 to visualize the dynamics of GFP-Vrg4 and Kex2-mCherry , except that the duration of the movie was 10 . 5 min and the interval between Z-stacks was 3 s . See also Figure 11C . DOI: http://dx . doi . org/10 . 7554/eLife . 13232 . 036 The combined data suggest that COPI inactivation selectively blocks recycling of early Golgi proteins . Thus , instead of an early Golgi cisterna maturing into a late cisterna , an early cisterna matures into a hybrid structure , which eventually loses its late Golgi proteins and then begins the process anew . Thermosensitive yeast mutants are versatile tools for studying the secretory pathway ( Duden and Schekman , 1997 ) , but such a mutant has disadvantages . The molecular basis of the thermosensitivity is typically unknown , creating uncertainty about whether the mutant protein has been completely inactivated by the temperature shift . Moreover , a thermosensitive mutation may have detrimental effects even at the permissive temperature , in which case cellular functions will be compromised before the experiment begins . Both of these issues have been encountered with thermosensitive COPI mutants . For example , some of the commonly used COPI mutants show variable and relatively weak phenotypes ( Gaynor and Emr , 1997 ) . That issue was addressed by isolating the strong sec21-3 mutation ( Gaynor and Emr , 1997 ) , but in our hands , Golgi morphology was perturbed in sec21-3 cells even at the permissive temperature . Thus , thermosensitive mutants have not been ideal for examining the role of COPI in yeast . We addressed this problem by using the anchor-away method . Growth tests indicated that wild-type FRB is suitable as a tag for gene replacement while the destabilized FRB ( T2098L ) mutant is not . With regard to anchors , our initial trials employed an FKBPx2-tagged version of the plasma membrane protein Pma1 ( Haruki et al . , 2008 ) , but strains expressing Pma1-FKBPx2 had growth defects and were genetically unstable . Better results were obtained with the ribosomal Rpl13A-FKBPx2 anchor ( Haruki et al . , 2008 ) . We also generated a mitochondrial OM45-FKBPx4 anchor , which is effective in combination with an FRBx2 tag . Control experiments indicated that these versions of the anchor-away system allow COPI to be inactivated quickly and reliably , and can therefore serve to complement thermosensitive mutants for studying COPI function . COPII can also be inactivated with the anchor-away system , although the response is slightly different than for COPI , indicating that this system needs to be tested for each component that is being inactivated . It should be noted that “anchor-away” is something of a misnomer here because the anchored COPI remained associated with Golgi membranes . Thus , when OM45-FKBPx4 was used as the anchor , entire Golgi compartments apparently became tethered to mitochondria . Such an effect is not unexpected because Golgi cisternae are mobile in the yeast cytoplasm ( Wooding and Pelham , 1998; Losev et al . , 2006 ) . Despite retaining its association with Golgi membranes , the anchored COPI was no longer functional as judged by inhibition of cell growth and of Golgi-to-ER recycling , implying that this method is suitable for studying the roles of COPI . Our data confirm the earlier conclusion that general secretion is arrested by inactivating COPII but is only partially inhibited by inactivating COPI ( Gaynor and Emr , 1997 ) . However , instead of classifying secretory proteins in a binary fashion as being either COPI-dependent or -independent , we propose that COPI inactivation has a spectrum of effects that range from mild to severe depending on the protein . Among the proteins whose traffic was reported to be severely reduced in sec21-3 cells at 37°C were the α-factor and carboxypeptidase Y precursors ( Gaynor and Emr , 1997 ) , both of which rely on the ER export receptor Erv29 , which recycles from the Golgi to the ER ( Dancourt and Barlowe , 2010 ) . A plausible interpretation is that for certain secretory proteins , preventing COPI-dependent recycling of the cognate ER export receptors strongly inhibits traffic ( Gaynor and Emr , 1997 ) . This line of reasoning raises a question: how can the secretory pathway operate at all after a block in COPI-dependent recycling , given that ER-to-Golgi traffic relies on SNARE proteins that are retrieved to the ER by COPI ( Barlowe and Miller , 2013 ) ? We propose that when COPI is inactivated , the cell replenishes ER-to-Golgi SNAREs through new protein synthesis . Other components such as ER export receptors will also be replenished , but at varying rates depending on their synthesis kinetics . According to this model , COPI inactivation will have the following effects on ER export . ( a ) COPII vesicle production will be slowed but not halted . ( b ) For a given secretory protein , the rate of ER export will depend on how rapidly the cognate ER export receptor is replenished and/or how efficiently the protein is packaged into COPII vesicles in the absence of an ER export receptor . After a secretory protein leaves the ER , COPI is thought to help drive traffic through the Golgi , yet when yeast COPI is inactivated , proteins can still be secreted . Insight into this puzzle came from fluorescence microscopy . Soon after COPI is inactivated , early and late Golgi proteins change from marking separate compartments to associating with one another in hybrid Golgi structures . These hybrid structures are very dynamic , and we suspect that the compartments containing early and late Golgi markers exchange material , although our analysis cannot determine whether this exchange involves transient fusion events or other types of transport intermediates . In any case , the hybrid Golgi structures have early Golgi character , so secretory proteins can presumably reach these structures in COPII vesicles , and the hybrid Golgi structures show relatively normal cycling of late Golgi components , so secretory proteins can presumably depart to the cell surface in transport carriers . Compared to the unperturbed Golgi , the hybrid Golgi structures may yield altered glycosylation ( Gaynor and Emr , 1997 ) but they are functional for membrane traffic . This analysis plausibly explains how yeast cells can continue to secrete after COPI inactivation . While it is interesting to explore the effects of COPI inactivation , the larger goal is to understand the normal role of COPI in Golgi traffic . We have proposed that Golgi maturation occurs in discrete stages , and that the transition from the “carbohydrate synthesis” stage to the “carrier formation” stage involves the COPI-dependent recycling of resident Golgi proteins to younger cisternae ( Day et al . , 2013; Papanikou and Glick , 2014 ) . In yeast , Vrg4 is a marker of the carbohydrate synthesis stage while Sec7 is a marker of the carrier formation stage . As a cisterna acquires Sec7 , it loses Vrg4 . Loss of Vrg4 probably occurs by COPI-mediated transport , based on genetic , biochemical , and electron microscopic evidence that Vrg4 is recycled in COPI vesicles ( Abe et al . , 2004; Mari et al . , 2014 ) . The implication is that if COPI is inactivated , a cisterna could acquire Sec7 while failing to lose Vrg4 , thereby generating a hybrid compartment ( Figure 12A ) . 10 . 7554/eLife . 13232 . 037Figure 12 . Working hypotheses for Golgi protein recycling by COPI-dependent and COPI-independent pathways . ( A ) The existence of multiple intra-Golgi recycling pathways can explain why inactivating COPI generates hybrid structures that repeatedly gain and lose late Golgi proteins . Green represents early Golgi proteins that recycle in COPI vesicles , and red represents late Golgi proteins that recycle by COPI-independent pathways . Under normal conditions , early Golgi proteins depart while late Golgi proteins arrive , and then late Golgi proteins depart in turn . When COPI is inactivated , the recycling of early Golgi proteins is inhibited , so when late Golgi proteins arrive , a hybrid structure is generated . This hybrid structure can subsequently lose late Golgi proteins by COPI-independent pathways , and then the process begins again . ( B ) During maturation of the late Golgi , proteins are likely to recycle by several pathways . The thick arrow represents cisternal maturation that occurs during and after conversion to a late Golgi compartment ( Daboussi et al . , 2012; Day et al . , 2013 ) . Peripheral membrane proteins such as Sec7 are recruited to late Golgi cisternae by activated GTPases , and are subsequently released from the membrane by GTP hydrolysis . Some transmembrane proteins such as Vps10 travel to the prevacuolar endosome in clathrin-coated vesicles with the aid of the Gga1 and Gga2 adaptors , and then recycle to newly formed late Golgi cisternae with the aid of cargo scaffolds such as retromer . Other transmembrane proteins such as Kex2 are postulated to recycle from older to younger late Golgi cisternae in clathrin-coated vesicles with the aid of the AP-1 adaptor . DOI: http://dx . doi . org/10 . 7554/eLife . 13232 . 037 We tested this idea by capturing 4D movies after COPI had been inactivated . As predicted , after COPI inactivation , Golgi structures acquired Sec7-DsRed while failing to lose GFP-Vrg4 . This result provides the first direct evidence that COPI plays a role in cisternal maturation . Specifically , we conclude that COPI helps to drive cisternal maturation by recycling early Golgi proteins such as Vrg4 from maturing cisternae . Vrg4 could recycle to younger cisternae either by intra-Golgi traffic , or by Golgi-to-ER traffic followed by delivery to newly forming cisternae . To evaluate these possibilities , we note that when COPII was inactivated by anchoring to mitochondria , GFP-Vrg4 showed no ER accumulation of the type that would be expected if this protein frequently returned to the ER . The data therefore suggest that COPI recycles Vrg4 within the Golgi . After a hybrid Golgi structure was generated by COPI inactivation , Sec7-DsRed frequently disappeared from the structure and subsequently reappeared . Thus , the recycling pathway for Sec7 remained functional after COPI inactivation . A likely mechanism for COPI-independent recycling of Sec7 is dissociation from the membrane of one cisterna followed by reassociation with the membrane of a younger cisterna ( Figure 12B ) . Consistent with this model , Sec7 recruitment to the membrane is known to require activated GTPases , implying that GTP hydrolysis releases Sec7 into the cytosol for another round of recruitment ( McDonold and Fromme , 2014 ) . Can late Golgi transmembrane proteins also recycle independently of COPI ? Some late Golgi transmembrane proteins traffic to prevacuolar endosomes and back . An example is Vps10 , which is delivered to prevacuolar endosomes in clathrin-coated vesicles with the aid of Gga adaptors ( Costaguta et al . , 2001; Deloche et al . , 2001; Abazeed and Fuller , 2008 ) . Recycling of Vps10 from prevacuolar endosomes to the Golgi is mediated by the retromer complex ( Seaman et al . , 1997; Seaman , 2005 ) . This loop through the prevacuolar endosome is presumably independent of COPI ( Figure 12B ) . Other late Golgi transmembrane proteins may recycle within the Golgi itself ( Wong and Munro , 2014 ) . Candidates for such an intra-Golgi recycling pathway include trafficking machinery proteins such as Tlg1 ( Valdivia et al . , 2002 ) , phospholipid translocases such as Drs2 ( Liu et al . , 2008 ) , proteins such as Chs3 and Pin2 that reside in the late Golgi before undergoing regulated export to the plasma membrane ( Valdivia et al . , 2002; Ritz et al . , 2014; Spang , 2015 ) , and processing proteases such as Kex2 that act on secretory cargoes ( Fuller et al . , 1988 ) . These proteins are often assumed to recycle from early endosomes to the late Golgi , but we found that at least some of the yeast compartments described as early endosomes are identical to the late Golgi ( Bhave et al . , 2014 ) , suggesting that the proposed early endosome-to-Golgi recycling pathway could actually be an intra-Golgi recycling pathway . This intra-Golgi recycling pathway might or might not involve COPI . To explore this issue , we focused on Kex2 , which was described together with Sec7 as one of the first known markers of the yeast Golgi ( Franzusoff et al . , 1991; Redding et al . , 1991 ) . By fluorescence microscopy , Kex2 and Sec7 show substantial but incomplete overlap . The existence of structures that label with either Kex2 alone or Sec7 alone can now be explained , by our work plus a recent study ( McDonold and Fromme , 2014 ) , as being due to an offset in the kinetic behaviors of these two proteins during cisternal maturation . Kex2 arrives at a Golgi cisterna ~5–20 s before Sec7 and then departs ~5–20 s before Sec7 . Thus , Kex2 apparently recycles within the Golgi somewhat ahead of Sec7 . Our interpretation argues against the view that Kex2 resembles Vps10 in cycling between the Golgi and prevacuolar endosomes ( Abazeed et al . , 2005; De et al . , 2013 ) . Although we occasionally see Kex2-GFP in prevacuolar endosomes , this pool is very small , probably because Kex2 trafficking to prevacuolar endosomes reflects either a secondary recycling pathway or a degradation pathway . The proposed recycling of Kex2 within the Golgi merits further exploration . Meanwhile , we found that COPI inactivation did not prevent recycling of Kex2 . This result , together with the finding that COPI is concentrated at the early Golgi , suggests that intra-Golgi recycling of late Golgi transmembrane proteins is independent of COPI . A candidate for the carriers that recycle late Golgi transmembrane proteins is clathrin-coated vesicles containing the AP-1 adaptor ( Figure 12B ) . Yeast AP-1 has been implicated in the late Golgi localization of multiple proteins including Tlg1 , Drs2 , Chs3 , and Pin2 ( Valdivia et al . , 2002; Foote and Nothwehr , 2006; Liu et al . , 2008; Barfield et al . , 2009; Myers and Payne , 2013; Ritz et al . , 2014 ) . One study provided evidence that AP-1 also plays a role in Kex2 localization ( Abazeed and Fuller , 2008 ) . Those data have been thought to reflect a recycling pathway from early endosomes , but yeast AP-1 localizes to the late Golgi ( Daboussi et al . , 2012 ) , supporting the alternative view that AP-1 mediates recycling from older to younger cisternae at a late stage in Golgi maturation ( Valdivia et al . , 2002; Liu et al . , 2008 ) . We speculate that Kex2 recycling involves AP-1 , which might act in conjunction with the co-adaptor Ent5 ( Costaguta et al . , 2006; Copic et al . , 2007; Daboussi et al . , 2012 ) . In mammalian cells , the functions attributed to AP-1 include retrograde transport from recycling endosomes and TGN-derived transport carriers ( Hirst et al . , 2012; Bonifacino , 2014; Matsudaira et al . , 2015 ) , and those pathways are potentially analogous to recycling within the late Golgi of yeast . AP-1-containing retrograde vesicles may be captured at the yeast Golgi by effectors of Ypt6 , a Rab GTPase that is recruited prior to Sec7 ( Suda et al . , 2013 ) . One Ypt6 effector is the GARP complex , which has been implicated in endosome-to-Golgi recycling ( Bonifacino and Hierro , 2011 ) but could also play a similar role in intra-Golgi recycling . Another Ypt6 effector is the Sgm1 tether ( Siniossoglou and Pelham , 2001 ) , and intriguingly , the homologous TMF tether in mammalian cells captures intra-Golgi transport carriers carrying late Golgi proteins ( Wong and Munro , 2014 ) . The roles of Ypt6 effectors and AP-1 in late Golgi recycling can be tested in yeast using approaches like the ones described here . Our working hypothesis is that yeast Golgi maturation involves distinct pathways that act in sequence . At the early Golgi , transmembrane proteins are recycled by intra-Golgi COPI vesicles . At the late Golgi , proteins are recycled by multiple mechanisms that involve either transit through the cytosol , or traffic to prevacuolar endosomes and back , or retrograde transport from older to younger cisternae ( Figure 12B ) . An open question is how these various pathways are coordinated to maintain organellar homeostasis . When COPI is inactivated , Golgi compartmentation is lost , and when Golgi compartmentation is lost , COPI is no longer needed for secretion . We interpret these findings to mean that Golgi compartmentation and COPI-driven cisternal maturation are aspects of the same phenomenon . Although Golgi compartmentation is broadly conserved , and is thought to provide fine control of glycan assembly while keeping secretory cargoes in the organelle long enough for complete processing ( Stanley , 2011; Ruiz-May et al . , 2012; Day et al . , 2013 ) , some organisms such as microsporidia forgo Golgi compartmentation and rely instead on a fused Golgi network ( Beznoussenko et al . , 2007; Takvorian et al . , 2013 ) . Yet COPI components are present in microsporidia ( Beznoussenko et al . , 2007; Mowbrey and Dacks , 2009 ) , perhaps because an efficient secretory pathway always requires Golgi-to-ER recycling , even with a non-compartmentalized Golgi . In this view , Golgi-to-ER recycling is the core function of COPI , and many eukaryotes have adapted COPI for the additional purpose of maintaining separate Golgi compartments through cisternal maturation . Experiments were done with derivatives of the haploid S . cerevisiae strain JK9-3da , which carries the mutations leu2-3 , 112 ura3-52 rme1 trp1 his4 ( Kunz et al . , 1993 ) . Yeast cells were grown in rich glucose medium ( YPD ) , or in minimal glucose dropout medium ( SD ) ( Sherman , 1991 ) or nonfluorescent minimal glucose dropout medium ( NSD ) ( Bevis et al . , 2002 ) , with shaking at 200 rpm in baffled flasks . Growth media were obtained from Difco Laboratories ( Detroit , MI , USA ) . The TOR1-1 mutation was introduced using the pop-in/pop-out method for gene replacement ( Rothstein , 1991; Rossanese et al . , 1999 ) . To delete the FPR1 gene , the kanMX cassette was amplified with the following primers to append sequences flanking the FPR1 open reading frame: AACTCGAGTATAAGCAAAAAATCAATCAAAACAAGTAATACGTACGCTGCAGGTCGAC and TAAAAAGCAGAAAGGCGGCTCAATTGATAGTACTTTGCTTATCGATGAATTCGAGCTCG . The resulting fragment was transformed into cells , which were plated on YPD containing 250 μg/mL G418 ( Sigma-Aldrich , St . Louis , MO , USA ) to select for double-crossover replacement of FPR1 . A similar method was used to delete the RER1 gene by replacing it with a hygromycin resistance cassette from pAG32 ( Goldstein and McCusker , 1999 ) . The mitochondrial matrix was labeled by transforming cells with a centromeric plasmid that drove expression from the constitutive ADH1 promoter of a mitochondrially targeted fluorescent protein , either mCherry in the case of pHS12-mCherry ( Sesaki and Jensen , 1999; Bevis and Glick , 2002 ) , or TagBFP in the case of p416-ADH::mito-TagBFP ( Murley et al . , 2013 ) , which was obtained from Laura Lackner . To create a mitochondrial anchor , the gene encoding OM45 ( Yaffe et al . , 1989 ) was inserted between the strong constitutive TPI1 promoter and the CYC1 terminator in a vector that was integrated at the TRP1 locus . The Sec71TMD-EGFP construct was subcloned from a plasmid provided by Ken Sato into the integrating vector YIplac128 ( Gietz and Sugino , 1988 ) . For video microscopy , an integrating vector was used to overexpress Sec7-DsRed . M1x6 ( Losev et al . , 2006 ) . All other tags were introduced by using pop-in/pop-out gene replacement to express proteins at endogenous levels , except that to boost the signal for Kex2-mCherry , a second copy of this construct was expressed from a centromeric plasmid . For GFP tagging , the variants used for gene replacement were either the monomeric mEGFP ( Zacharias et al . , 2002 ) or the monomeric superfolder msGFP ( Fitzgerald and Glick , 2014 ) . The FKBP and FRB genes were obtained from Ariad Pharmaceuticals ( Cambridge , MA ) , and the MBP gene was obtained from New England Biolabs ( Ipswich , MA ) . The mCherry gene ( Shaner et al . , 2004 ) was obtained from Roger Tsien ( University of California at San Diego ) , and was modified at the N- and C-termini to create the mCherry2B variant used here . DNA manipulations were simulated and recorded using SnapGene software ( GSL Biotech , Chicago , IL ) . Annotated sequence files for 34 of the plasmids used in this study are included as a zip archive ( Supplementary file 1 ) , and can be opened with the free SnapGene Viewer ( http://www . snapgene . com/products/snapgene_viewer/ ) . To minimize the background signal for fluorescence microscopy , yeast cultures were grown in SD or NSD medium . Static images were captured with living cells that were compressed beneath a coverslip without fixation and then immediately viewed , and 4D data sets were acquired with cells attached to a concanavalin A-coated coverglass-bottom dish containing NSD medium ( Losev et al . , 2006 ) . To capture static images by widefield microscopy , we used an Axioplan2 epifluorescence microscope ( Zeiss , Thornwood , NY ) equipped with a 1 . 4-NA 100x Plan Apo objective and a digital camera ( Hamamatsu , Skokie , IL ) . To capture static images by confocal microscopy , we used either an SP5 ( Leica , Buffalo Grove , IL ) or an LSM 710 ( Zeiss ) scanning confocal microscope to collect Z-stacks , with pixel sizes of 50-60 nm and Z-step intervals of ~0 . 3 μm . To capture 4D confocal movies with two fluorescence channels ( red and green ) and a scattered light channel ( blue ) , cells were imaged at room temperature essentially as previously described ( Losev et al . , 2006 ) , except that we used an SP5 microscope with pixel sizes of 50–60 nm and Z-step intervals of 0 . 29 μm to collect ~20–24 optical sections every 2 s . For the strain expressing GFP-Vrg4 and Kex2-mCherry , the pixel size was increased to 80 nm , the Z-step interval was increased to 0 . 34 μm , and the interval between Z-stacks for rapamycin-treated cells was increased to 3 s . Some of the static confocal images and all of the 4D confocal data were deconvolved using Huygens software ( Scientific Volume Imaging , Hilversum , The Netherlands ) . For the strain expressing GFP-Vrg4 and Kex2-mCherry , a single pass with a 2D hybrid median filter ( Hammond and Glick , 2000 ) was performed to smooth the data before deconvolution . Adobe Photoshop and ImageJ ( http://rsbweb . nih . gov/ij/ ) were used to colorize and merge the images , adjust brightness , and create average projections . Correction for exponential photobleaching was performed with an ImageJ plugin ( http://cmci . embl . de/downloads/bleach_corrector ) . Editing and quantitation of 4D data sets was performed using custom plugins for ImageJ . These plugins allowed a hyperstack to be hybrid median filtered , converted to a montage time series , edited to remove extraneous fluorescence signals , converted back to a hyperstack , and quantified to measure red and green fluorescence intensities . A zip archive Supplementary file 2 provides detailed instructions for capturing and processing 4D movies , together with our custom ImageJ plugins . Cells were grown to log phase ( OD600 = 0 . 5–0 . 8 ) in SD medium . Prior to labeling , the cells were collected on a bottle-top filter , washed with SD lacking methionine ( SD – Met ) , then resuspended in SD – Met at a concentration of 5 OD600 units/mL . The concentrated cells were incubated for 30 min . For pulse labeling , 25 μCi of TRAN35S-LABEL ( MP Biomedicals , Santa Ana , CA ) was added per OD600 unit of cells , and the cells were incubated for 10 min . A 30-min chase was initiated by adding a 10x solution to give a final concentration of 5 mM unlabeled methionine plus 2 mM unlabeled cysteine . All of these manipulations were carried out with constant aeration at 30°C , except in the case of Figure 6 , for which cells were grown at room temperature and then shifted to 37°C for the SD – Met preincubation , pulse , and chase . After the chase , cells were separated from medium by centrifugation at 5000 rpm ( 2300xg ) in a microcentrifuge . To analyze secreted proteins , the medium was adjusted to a final concentration of 10% trichloroacetic acid ( TCA ) . This mixture was incubated for 5 min at 60°C . TCA-precipitated material was collected by centrifugation for 5 min at full speed ( 16 , 000xg ) in a microcentrifuge , then solubilized by vigorous vortexing in 50 μL of SDS-PAGE sample buffer supplemented with 0 . 1 M dithiothreitol and 50 mM Na+ PIPES , pH 7 . 5 . The sample was incubated for 30 min at 37°C , followed by a 3-min spin at full speed in a microcentrifuge to remove insoluble material . To analyze cellular proteins , the cell pellet was resuspended in the original volume of medium , and then TCA was added to a final concentration of 10% . This mixture was incubated for 5 min at 50°C followed by 5 min on ice . TCA-precipated material was collected by centrifugation for 5 min at 3000 rpm ( 1000xg ) in a microcentrifuge , then resuspended in SDS-PAGE sample buffer supplemented with 0 . 1 M dithiothreitol and 50 mM Na+ PIPES , pH 7 . 5 . The sample was incubated for 30 min at 60°C , followed by a 3-min spin at full speed in a microcentrifuge to remove insoluble material . Each gel lane was loaded with a sample of secreted or cellular proteins corresponding to 0 . 25 OD600 units of cells . SDS-PAGE was performed with Mini-PROTEAN TGX Tris/glycine 4–20% gradient gels using the Precision Plus Protein Dual Color Standards molecular weight markers ( Bio-Rad , Hercules , CA ) . Gels were dried , and radioactive signals were detected using a Storm 860 molcular Imager ( Molecular Dynamics , Sunnyvale , CA ) . A 100-mL culture of untreated or rapamycin-treated yeast cells was grown in SD to mid-log phase at 30°C with shaking . Cells were then concentrated by vacuum filtration using a 0 . 22-μm bottle-top filter ( EMD Millipore , Billerica , MA ) . The cell paste was transferred to planchettes ( Ted Pella , Redding , CA ) , cryo-fixed using a Bal-Tec HPM 010 high-pressure freezing machine ( RMC , Tucson , AZ ) , and placed immediately into cryo-tubes containing a frozen cocktail of 0 . 1% uranyl acetate in anhydrous acetone . Samples were freeze substituted at -80°C for 48–60 hr in an EM AFS2 freeze substitution device ( Leica ) . The temperature was then raised to -50°C and the samples were washed three times with acetone , followed by successive increasing overnight infiltrations with Lowicryl K4M resin ( 25 , 50 , 75 , and 100% ) , followed by three incubations of 1 hr each with 100% resin . Infiltrated samples were placed in molds and polymerized with ultraviolet light at -50°C for 13 hr . To preserve fluorescence , the plastic blocks were stored at -20°C prior to sectioning . Sections of 300–1500 nm were cut with an EM UC6 ultramicrotome ( Leica ) and placed on 200 mesh carbon-formvar coated London-Finder copper grids ( Electron Microscopy Sciences , Hatfield , PA ) . For fluorescence microscopy , a grid was placed on a glass slide with the resin side up , and a 22x22 mm No . 1 . 5 glass coverslip with a 10-μl droplet of 500 mM Na+-HEPES , pH 7 . 5 was inverted onto the grid . The coverslip was immediately sealed with wax . Imaging was performed with an LSM 710 confocal microscope . Dual-color Z-stacks were captured with a 1 . 4-NA 100x Plan-Apo oil objective using a pinhole of 1 . 2 Airy units and with voxels ranging in each dimension from 0 . 30 to 0 . 43 µm . The grid was then retrieved for analysis by electron microscopy . For transmission electron microscopy ( TEM ) as well as TEM tomography and scanning transmission electron microscopy ( STEM ) tomography , images were collected on a Tecnai G2 F30 electron microscope ( FEI , Hillsboro , OR ) with a Schottky field-emission gun operating at 300 kV . Grids were prepared for either TEM or STEM tomography by floating each side of the grid for 10 min on a 10-μl drop of 15 nm colloidal gold bead solution ( British BioCell International , obtained from Ted Pella ) . The samples were then stained for 8–15 min with 2% uranyl acetate , and placed into a Model 2040 dual-axis tomography holder ( Fischione Instruments , Export , PA ) . For TEM tomography , 300–400 nm sections were analyzed using Serial EM ( Mastronarde , 2005 ) to collect digital images at 15 , 000x magnification with a 4K UltraScan camera ( Gatan , Pleasanton , CA ) as a dual-axis tilt series over a range of -60° to +60° at tilt angle increments of 1° . For STEM tomography , the imaging conditions were as follows: extraction voltage = 4250 V , gun lens = 3 , condenser aperture = 50 mm , and camera length range = 200–500 mm . Images were collected using a Model 3000 annular dark field detector ( Fischione ) placed above the viewing screen , and a Model 805 bright- and dark-field detector ( Gatan ) below the viewing screen . Images were collected as a dual-axis tilt series over a range of -60° to +60° at tilt angle increments of 1° using the FEI STEM tomography software . All tomograms were reconstructed and analyzed using IMOD software ( Kremer et al . , 1996 ) . The estimated resolution of the STEM tomograms is 8–12 nm ( Radermacher , 1992 ) .	Proteins play many important roles for cells , and these roles often require the proteins to be in particular locations in or around the cells . A set of cell compartments called the Golgi packages certain proteins into bubble-like structures called vesicles to enable the proteins to be used elsewhere in the cell or released to the outside of the cell , in a process called the secretory pathway . The operation of the secretory pathway requires the Golgi compartments to be continually remodeled . Proteins and other materials can be ferried between the compartments of the Golgi by another type of vesicle . These vesicles are coated with a group , or complex , of proteins called COPI , which forms a curved lattice around the vesicles and helps them to capture the materials they will transport . However , it is not clear whether COPI is also involved in remodeling of the Golgi compartments . Papanikou , Day et al . addressed this question using a technique called the “anchor-away method” combined with microscopy to study COPI in yeast cells . The yeast were genetically engineered so that COPI activity was effectively shut down in the presence of a drug called rapamycin . The experiments show that COPI is involved in the early stages of remodeling the Golgi compartments , but not the later stages . This finding supports the emerging view of the Golgi as a self-organizing cellular machine , and it provides a framework for uncovering the engineering principles that underlie the secretory pathway .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2015	COPI selectively drives maturation of the early Golgi
Ethylene plays critical roles in plant development and biotic stress response , but the mechanism of ethylene in host antiviral response remains unclear . Here , we report that Rice dwarf virus ( RDV ) triggers ethylene production by stimulating the activity of S-adenosyl-L-methionine synthetase ( SAMS ) , a key component of the ethylene synthesis pathway , resulting in elevated susceptibility to RDV . RDV-encoded Pns11 protein specifically interacted with OsSAMS1 to enhance its enzymatic activity , leading to higher ethylene levels in both RDV-infected and Pns11-overexpressing rice . Consistent with a counter-defense role for ethylene , Pns11-overexpressing rice , as well as those overexpressing OsSAMS1 , were substantially more susceptible to RDV infection , and a similar effect was observed in rice plants treated with an ethylene precursor . Conversely , OsSAMS1-knockout mutants , as well as an osein2 mutant defective in ethylene signaling , resisted RDV infection more robustly . Our findings uncover a novel mechanism which RDV manipulates ethylene biosynthesis in the host plants to achieve efficient infection . Rice is a staple food crop in many regions , a model monocot plant for research , and a host to many viruses ( Wu et al . , 2015; 2017 ) . Viral infection causes substantial losses in yield and quality in rice crops and current knowledge on the antiviral responses of monocotyledonous crops is very limited ( Soosaar et al . , 2005; Mandadi and Scholthof , 2013; Wang , 2015 ) . Rice dwarf virus ( RDV ) , a member of the genus Phytoreovirus in the family Reoviridae , is one of the most widespread and devastating viruses that infect rice ( Wu et al . , 2015; Jin et al . , 2016 ) . RDV is transovarially transmitted by the green rice leafhopper ( Nephotettix cincticeps ) in a persistent-propagative manner ( Cao et al . , 2005; Zhou et al . , 2007; Wei and Li , 2016 ) . RDV infection greatly inhibits rice growth and causes severe symptoms including dwarfism , increased tillering , and white chlorotic specks and dark-green discoloration on the leaves . RDV has a double-stranded RNA genome consisting of 12 segments ( S1 to S12 ) . Seven segments , S1 , S2 , S3 , S5 , S7 , S8 , and S9 , encode structural proteins P1 , P2 , P3 , P5 , P7 , P8 , and P9 , respectively , which form double-layered virions; the remaining segments , S4 , S6 , S10 , S11 , and S12 , encode the nonstructural proteins Pns4 , Pns6 , Pns10 , Pns11 , and Pns12 ( Cao et al . , 2005; Zhou et al . , 2007; Liu et al . , 2014; Jin et al . , 2016; Wei and Li , 2016 ) . To survive under the continuous threat of viral infection , plants have evolved multiple defense mechanisms that are activated via different signal transduction pathways . The intensively studied pathways , based on dicot model plants , include nucleotide-binding-site leucine-rich-repeat dominant resistance genes ( R-genes ) , recessive resistance genes ( eukaryotic initiation factors ( eIFs ) ) , RNA interference ( RNAi ) antiviral immunity , and phytohormone-mediated resistance pathways; these pathways interact synergistically or antagonistically and result in a highly complex three-dimensional defense signaling network ( Kasschau et al . , 2003; Baulcombe , 2004; Soosaar et al . , 2005; Ding , 2010; Endres et al . , 2010; Mandadi and Scholthof , 2013; Nicaise , 2014; Carbonell and Carrington , 2015; Collum and Culver , 2016; Wang , 2015; Wu et al . , 2017 ) . As a counter-defense , plant viruses often manipulate plant responses for their own benefit . For example , viruses have evolved strategies to target hormone pathways , often exploiting the antagonistic interactions mediated by phytohormones such as salicylic acid ( SA ) , jasmonic acid ( JA ) , and ethylene ( Soosaar et al . , 2005; Broekaert et al . , 2006; Pieterse et al . , 2009; Santner et al . , 2009; Denancé et al . , 2013; Mandadi and Scholthof , 2013; Alazem and Lin , 2015 ) . The gaseous phytohormone ethylene functions in seed germination and organ senescence , as well as in the response of plants to abiotic and biotic stresses ( Broekaert et al . , 2006; van Loon et al . , 2006; Alazem and Lin , 2015; Kazan , 2015 ) . However , the functions of ethylene in the plant response to viral infection remain poorly understood . Previous studies found that ethylene could modulate host defense in both positive and negative manners ( Knoester et al . , 2001; Love et al . , 2007; Pieterse et al . , 2009; Santner et al . , 2009; Chen et al . , 2013b; Zhu et al . , 2014; Casteel et al . , 2015 ) . The P6 protein encoded by Cauliflower mosaic virus ( CaMV ) was found to interact with components of the ethylene-signaling pathway , and transgenic Arabidopsis expressing P6 became less responsive to ethylene treatment and more resistant to CaMV infection ( Geri et al . , 2004 ) . Another study used ein2 ( ethylene insensitive 2 ) and etr1 ( ethylene response 1 ) mutants and found that the ethylene-signaling pathway is required for Turnip mosaic virus ( TuMV ) -mediated suppression of resistance to the green aphid , Myzus persicae , in Arabidopsis , and that TuMV may modulate ethylene responses to increase plant susceptibility to viral infection ( Casteel et al . , 2015 ) . Chen et al . ( 2013b ) reported that Arabidopsis plants with mutations of the ethylene biosynthesis pathway , such as acs1 ( 1-aminocyclopropane-1-carboxylate synthase ) , erf106 ( ethylene responsive transcription factor 106 ) , and ein2 , were resistant to Tobacco mosaic virus ( TMVcg ) . Exogenous application of 1-aminocyclopropane-1-carboxylic acid ( ACC , a precursor in the ethylene biosynthesis pathway ) enhanced TMVcg accumulation in the infected plants ( Chen et al . , 2013a ) . By contrast , ethylene signaling was shown to be essential for systemic resistance to Chilli veinal mottle virus in tobacco ( Zhu et al . , 2014 ) . Thus , the molecular mechanisms by which ethylene affects host defenses and counter-defenses remain unclear . In plants , S-adenosyl-L-methionine synthetase ( SAMS ) [EC 2 . 5 . 1 . 6] catalyzes the conversion of L-methionine ( L-Met ) and ATP into S-adenosyl-L-methionine ( AdoMet , SAM ) , which serves as a precursor of ethylene and polyamines . The SAMS enzyme is induced by biotic and abiotic stress and is involved in the regulation of development through the histone and DNA methylation pathway ( Roje , 2006; Li et al . , 2011; Chen et al . , 2013a; Gong et al . , 2014; Yang et al . , 2015 ) . A previous study demonstrated that RDV infection perturbed the expression of several ethylene-response genes such as ERFs ( Ethylene Response Factors ) ( Satoh et al . , 2011; Abiri et al . , 2017 ) , indicating that ethylene is involved in the interaction between RDV and rice . However , it is unclear how the ethylene biosynthesis and signaling pathway functions in this interaction . In the current study , we report that the RDV-encoded non-structural protein Pns11 enhances rice susceptibility to RDV by interacting with OsSAMS1 , enhancing its enzymatic activity and leading to increasing production of SAM , ACC , and ethylene . As SAMS and ethylene are key regulators of many biological processes , the capability of RDV-encoded Pns11 to interact specifically with OsSAMS1 and to regulate the ethylene biosynthesis and signaling pathway may represent a novel mechanism by which RDV maximizes its own infection . This study provides a novel mechanism through which ethylene biosynthesis and signaling respond to viral infection . These findings significantly broaden our knowledge of virus–host interactions and provide novel targets for engineered resistance to viruses . RDV-encoded Pns11 protein was previously found to function as a component of viroplasms ( Wei et al . , 2006 ) . To investigate whether Pns11 plays an important role in RDV infection , transgenes encoding Pns11 were introduced into the rice cultivar Zhonghua 11 to generate Pns11-overexpression plants , referred to as S11 OX lines hereafter . Three transgenic lines #3 , #5 , and #11 were chosen for detailed analysis based on the detection of both Pns11 mRNA and the HA-tagged protein ( Figure 1—figure supplement 1A , B ) . No obvious differences in phenotype were observed between the S11 OX lines and wild-type ( WT ) rice except for grain size ( Figure 1—figure supplement 1C–F ) . We then inoculated 30 seedlings ( 14-d-old ) from each S11 OX line with RDV using viruliferous leafhoppers ( Supplementary file 1A ) and observed disease symptoms . At four weeks post inoculation ( 4 wpi ) , the S11 OX lines exhibited more severe RDV infection symptoms with more stunting and chlorotic flecks on the leaves than the WT control plants ( Figure 1A ) . Three RDV RNA genome segments and their encoded proteins were evaluated by northern and western blot assays . The results showed increased accumulation in two of the S11 OX lines ( except S11 OX#3 ) relative to the WT plants ( Figure 1B , C ) . In addition , the infection rates were higher in the S11 OX#5 and S11 OX#11 lines than in the the WT ( Figure 1D , Supplementary file 2A ) . Taken together , these results showed that overexpression of Pns11 compromised rice defense to RDV . The results described above showed that Pns11 overexpression increases rice susceptibility to virus infection . To elucidate the mechanism behind this , we tried to identify rice factors that interact with Pns11 by conducting a yeast two-hybrid screen of a rice cDNA library , with RDV-encoded Pns11 as the bait . This screen identified OsSAMS1 as a strong interaction partner of Pns11 . The rice genome encodes three members of the predicted SAMS family , OsSAMS1 , OsSAMS2 , and OsSAMS3 , which show high levels of sequence identity in their DNA and deduced amino acid sequences ( Li et al . , 2011 ) . However , Pns11 only interacted with OsSAMS1 , and not with OsSAMS2 or OsSAMS3 in yeast . Moreover , OsSAMS1 specifically interacted with Pns11 , but not with other RDV-encoded proteins ( Figure 2A , Figure 2—figure supplement 1A ) . To further test this specific interaction in plant cells , we performed a co-IP experiment by co-expressing hemagglutinin ( HA ) -epitope-tagged Pns11 and FLAG-tagged OsSAMS1 , OsSAMS2 , or OsSAMS3 in a transient expression assay in Nicotiana benthamiana leaves , followed by immunoprecipitation with FLAG-tag antibodies and HA-tag antibodies ( Figure 2B , Figure 2—figure supplement 1B , C ) . This set of experiments confirmed the highly specific interaction between Pns11 and OsSAMS1 . We further verified this interaction using a firefly luciferase ( LUC ) complementation imaging assay . Constructs encoding Pns11 fused with the N-terminus of LUC ( Pns11-nLUC ) and the C-terminus of LUC fused with OsSAMS1 ( cLUC-OsSAMS1 ) were co-infiltrated into N . benthamiana leaves for transient co-expression of these two fusion proteins . A luminescence signal was only detected in Pns11-nLUC/cLUC-OsSAMS1 co-expression regions but not in the negative controls ( Figure 2C ) . Finally , we also performed a in vivo pull-down assay with whole-cell lysates from non-infected controls and RDV-infected rice plants and found that Pns11 and OsSAMS1 interacted in RDV-infected rice cells ( Figure 2D ) . Bimolecular fluorescence complementation ( BiFC ) analysis also demonstrated that Pns11 and OsSAMS1 interacted and were co-localized in both nucleus and cytoplasm ( Figure 2—figure supplement 2 ) . Taken together , our data strongly suggest that Pns11 specifically interacts with OsSAMS1 in vitro and in vivo . To evaluate the biological significance of this specific interaction , we tested the level of OsSAMS1 in S11 OX lines . We used S11 OX rice tissues at three stages ( 5-leaf , 6-leaf , and 10-leaf stage ) for real-time PCR ( qRT-PCR ) measurements and western blot . The OsSAMS1 mRNA and OsSAMS1 protein level did not change in response to Pns11 ( Figure 2—figure supplement 1D , E ) . We then designed an assay to detect the enzymatic activity of OsSAMS1 in vitro . SAMS catalyzes the two-step reaction that produces SAM , pyrophosphate ( PPi ) , and orthophosphate ( Pi ) from ATP and L-Met ( Figure 2E ) . Pns11 fused to maltose-binding protein ( MBP ) ( MBP-Pns11 ) and OsSAMS1 fused to glutathione S-transferase ( GST-OsSAMS1 ) were expressed in Escherichia coli BL21 cells and partially purified . To rule out the effect of the tags , the GST tag was cleaved to obtain pure OsSAMS1 . For unknown reasons , the yield of Pns11 was extremely low if the MBP tag was removed . Therefore , we used MBP-GFP and MBP-P9 ( a structural protein of RDV ) , which did not interact with OsSAMS1 , as negative controls; we also used OsSAMS2 , which did not interact with Pns11 , as another negative control . The addition of L-[35S]-Met mimics the natural substrate and allowed us to quantify the enzymatic activity of OsSAMS1 by measuring the amount of labeled SAM produced . OsSAMS1 was pre-incubated for 20 min with varying amounts of Pns11 ( no Pns11 to a six-fold molar excess of Pns11:OsSAMS1 ) ( Figure 2F ) . Solutions containing ATP , L-[35S]-Met , KCl , and MgCl2 were then added to the reaction mixtures and the reactions were allowed to proceed for another 20 min at 30°C . The reaction was blocked by the addition of EDTA . A reaction with excess OsSAMS1 and no Pns11 was used to label the location of the SAM , another reaction lacking OsSAMS1 and Pns11 was used to label the location of free L-[35S]-Met . Production of SAM and the remaining L-[35S]-Met were monitored by thin-layer chromatography ( Figure 2G ) . The enzymatic activity of OsSAMS1 was enhanced by nearly 60% at a 6:1 molar ratio of Pns11:OsSAMS1 ( Figure 2H ) . In the two negative control reactions with the same molar ratio of Pns11:OsSAMS1 , neither P9 nor GFP affected OsSAMS1 activity ( Figure 2—figure supplement 3 ) . In addition , Pns11 did not affect OsSAMS2 enzymatic activity ( Figure 2—figure supplement 4 ) . The results described above demonstrated that Pns11 only interacts with OsSAMS1 and enhances the activity of OsSAMS1 for SAM synthesis in vitro . The results described above showed that Pns11 specifically interacts with OsSAMS1 and enhances its enzymatic activity to increase SAM production in vitro . SAM serves as the precursor of polyamine and ethylene , and a previous study showed that most ethylene-response genes , such as ERFs and PRs ( Pathogenesis-related genes ) are regulated in RDV-infected rice ( Satoh et al . , 2011; Abiri et al . , 2017; Agrawal et al . , 2001 ) , indicating an important role of ethylene in RDV infection . Thus we wondered whether overexpression of Pns11 in rice would enhance the enzymatic activity of OsSAMS1 and promote the synthesis of SAM , ACC , and ethylene in vivo . S11 OX lines were used for analysis and the results showed that SAM , ACC and ethylene contents increased in two of the S11 OX lines ( but not in S11 OX#3 ) ( Figure 3A–C ) . To further confirm whether ethylene levels were affected by changes in SAM levels , OsSAMS1 was introduced into the rice Zhonghua 11 background to generate OsSAMS1-overexpression ( OX ) lines . We also generated OsSAMS1 RNAi lines ( knockdown ) and OsSAMS1 knockout ( KO ) lines using CRISPR/Cas9 . Positive transgenic rice lines were obtained through antibiotic selection and molecular screening . Among the OsSAMS1-overexpression lines , three ( OsSAMS1 OX#10 , OX#17 , and OX#25 ) were further analyzed . RNAi lines were characterized and classified as strong ( RNAi-S ) or weak ( RNAi-W ) according to the level of downregulation of OsSAMS1 ( Figure 3—figure supplement 1A–C ) . Two independent Ossams1 KO rice lines ( KO#31 and KO#39 ) with a mutation at different codons in the coding sequence were obtained ( Figure 3—figure supplement 1D ) . Seed germination was suppressed in the OsSAMS1 RNAi and KO lines and the suppression could be rescued by supplementation with SAM and ethylene ( Figure 3—figure supplement 1E ) . The RNAi and KO lines also showed developmental defects , including dwarfism and reduced fertility ( Figure 3—figure supplement 1F ) ( Li et al . , 2011 ) . Previous studies have demonstrated that ACC and ethylene contents increased in OsSAMS1 OX lines and decreased in OsSAMS1 RNAi transgenic lines , relative to WT plants ( Chen et al . , 2013b ) . In our study , SAM , ACC , and ethylene contents all increased in the OsSAMS1 OX lines and decreased in the OsSAMS1 RNAi and KO lines ( Figure 3D–F , Figure 3—figure supplement 2 ) . These results further demonstrated that Pns11 enhances the enzymatic activity of OsSAMS1 and alters the expression of SAMS in vivo , resulting in a corresponding change in the production of ACC and ethylene . Overexpression of OsSAMS1 resulted in increased levels of SAM , ACC , and ethylene , and knockout of OsSAMS1 resulted in decreased levels of SAM , ACC , and ethylene in rice . To investigate whether RDV infection , virus accumulation , and the host response is affected by SAM , ACC , and ethylene contents in the OsSAMS1 OX , RNAi , and KO lines , we inoculated 30 seedlings ( 14-d-old ) from each line ( WT , OX#10 , OX#17 , OX#25 , RNAi-S , RNAi-W , KO#31 , and KO#39 ) with RDV using viruliferous leafhoppers and observed the resulting disease symptoms ( Supplementary file 1B , C ) . At 4 wpi , all three OX lines displayed more severe stunting symptoms and chlorotic flecks at the infection site than the WT control plants , suggesting that they were more susceptible to RDV infection , while the RNAi and KO lines showed greater tolerance ( Figure 4A , Figure 4—figure supplement 1A ) . Northern and western blot assays revealed that RDV accumulation was higher in the OX lines than in the WT , but much lower in the RNAi and KO lines ( Figure 4B , C , Figure 4—figure supplement 1B , C ) . The infection rate was also higher in the OX lines , but much lower in the RNAi and KO lines compared to the WT plants ( Figure 4D , Figure 4—figure supplement 1D , Supplementary file 2B , C ) . Taken together , these results suggested that overexpression of OsSAMS1 enhances RDV infection whereas knockout of OsSAMS1 reduces RDV infection , indicating a positive role of ethylene in RDV infection . Our results demonstrated that the endogenous accumulation of ethylene negatively regulates the plant antiviral defense response to RDV infection . However , it is not clear whether ethylene signaling is involved in the response of rice to RDV infection . The rice MHZ7 gene ( named OsEIN2 ) , which encodes a membrane protein homologous to EIN2 , a central component of ethylene signaling in Arabidopsis , also plays a key role in the rice ethylene signaling pathway ( Ma et al . , 2013; Li et al . , 2015 ) . The ethylene signaling mutant mhz7 ( osein2 ) is insensitive to ethylene in both the root and coleoptile . To investigate whether blocking the ethylene signaling pathway affects the rice antiviral defense response , we inoculated the osein2 mutant and two OsEIN2-overexpression lines ( OX#2 and OX#3 ) with RDV using viruliferous leafhoppers and observed the resulting disease symptoms ( Supplementary file 1D ) . At 4 wpi , the OsEIN2 OX#2 and OX#3 lines showed enhanced susceptibility with more severe stunting and chlorotic flecks on the leaves than did the WT control plants . By contrast , the osein2 mutant showed much milder dwarfism and fewer chlorotic flecks on the leaves ( Figure 5A ) . Northern and western blot assays indicated that RDV accumulation was much higher in the OsEIN2 OX lines than in the WT plants , but lower in the mutant lines ( Figure 5B , C ) . RDV infection rates among the OsEIN2 OX lines , the WT , and the mutant lines diminished with time following infection . At 8 wpi , the infection rate of the osein2 mutant was only 43% , which was significantly lower than that of the WT ( 84% ) and the OsEIN2 OX lines ( OX#2 , 96%; OX#3 , 99% ) ( Figure 5D , Supplementary file 2D ) . These results suggested that the ethylene-response mutation enhances the rice antiviral defense response and that overexpression of OsEIN2 increases host susceptibility . This is consistent with above results . To further confirm that antiviral response was conferred through the ethylene signaling pathway , we overexpressed OsSAMS1 in osein2 mutant background rice and obtained three positive transgenic lines J119#1 , J119#2 and J119#3 ( Figure 5—figure supplement 1 ) for further analysis and RDV infection assay . Four weeks post inoculation , we found that the J119 lines , as well as the parental osein2 mutant , were less susceptible to RDV than was WT rice ( Figure 5—figure supplement 2 , Supplementary file 1E , Supplementary file 2E ) . Thus , we conclude that the ethylene-signaling pathway plays an important role in RDV infection and that blocking ethylene signaling would significantly enhance the antiviral defense response in rice . We next investigated whether the interaction and activation between Pns11 and OsSAMS1 affects the levels of SAM , ACC , and ethylene in RDV-infected rice . We first performed a pull-down assay in WT and Ossams1 KO lines , with and without RDV infection , using an anti-OsSAMS1 antibody ( Figure 6A ) . We found the loss of Pns11-OsSAMS1 interaction in Ossams1 KO lines , with or without RDV infection . We then measured the SAM , ACC and ethylene levels in the same set of plants , and found that the RDV-induced increase of SAM , ACC and ethylene levels disappeared in Ossams1 KO plants ( Figure 6B–D ) . RNA-seq experiments on RDV-infected rice , OsSAMS1 OX ( OX#25 ) lines , Ossams1 KO ( KO#39 ) lines and S11 OX ( OX#11 ) lines were carried out and the differentially expressed genes in all comparable pairs were identified ( Supplementary file 3 ) . To determine whether the ethylene pathway was activated , Gene Ontology ( GO ) was used for analysis ( Figure 6—figure supplement 1 ) . Known ethylene-activated pathway genes were highly enriched in both RDV-infected and OsSAMS1 OX transgenic rice , and depleted in Ossams1 KO plants . These data indicate that RDV infection triggers ethylene synthesis and accumulation through the interaction of Pns11 and OsSAMS1 , and the resultant activation of OsSAMS1 . To further elucidate the function of ethylene in RDV infection , 14-d-old seedlings were pretreated with 20 μM ACC , 10 μM AVG ( aminoethoxyvinylglycine , an ethylene biosynthesis inhibitor ) ( Chen et al . , 2013a ) , or H2O as a control for 1 day . We then sampled the treated plants and used qRT-PCR to analyze the expression of OsERF3 , which could be a marker for the response to ethylene ( Qi et al . , 2011 ) . OsERF3 was significantly upregulated by ACC treatment and significantly downregulated by AVG treatment when compared to the H2O-treated control ( Figure 6E ) , demonstrating that the ACC and AVG treatments worked as expected . The treated plants were then inoculated with RDV-carrying or RDV-free leafhoppers ( Supplementary file 1F ) . At 4 wpi , the ACC-treated plants showed greater susceptibility to RDV infection , displaying more severe disease symptoms and virus accumulation than the RDV-infected plants treated with H2O . However , the AVG-treated plants exhibited enhanced disease tolerance , as shown by less virus accumulation and milder disease symptoms , when compared with the RDV-infected plants treated with H2O ( Figure 6F–H ) . In addition , the rate of RDV infection in the ACC-treated plants increased much faster than that in the RDV-infected H2O-treated plants , while the rate of infection in the AVG-treated plants increased more slowly in comparison with that in the H2O-treated plants ( Figure 6I , Supplementary file 2F ) . Taken together , these data suggest that RDV Pns11 interacts with OsSAMS1 and enhances its enzymatic activity , inducing ethylene biosynthesis and accumulation , which in turn enhances viral infection and host susceptibility . Our results demonstrated that ethylene biosynthesis and signaling are critical for RDV infection and rice susceptibility . Overexpression of the RDV non-structural protein Pns11 increases rice susceptibility to viral infection ( Figure 1 ) . Furthermore , we found that Pns11 specifically interacts with OsSAMS1 and enhances its enzymatic activity in vivo and in vitro ( Figures 2 and 3 ) . Overexpression of S11 or OsSAMS1 increases the levels of SAM , ACC , and ethylene , whereas knockdown or knockout of OsSAMS1 by RNAi or CRISPR/Cas9 reduces the level of SAM , ACC , and ethylene ( Figure 3 , Figure 3—figure supplement 2 ) . Our results clearly indicated that an increase in ethylene production by overexpression of OsSAMS1 decreases the host antiviral defense response and enhances RDV infection and accumulation in rice , whereas knockdown or knockout of OsSAMS1 by RNAi or CRISPR/Cas9 reduces ethylene production , diminishes RDV accumulation , and increases the host antiviral defense response ( Figure 4 , Figure 4—figure supplement 1 ) . In addition , plants that have compromised ethylene signaling are more tolerant to RDV infection ( Figure 5 , Figure 5—figure supplement 2 ) . More importantly , RDV infection induces ethylene production , and the accumulation of ethylene increases host susceptibility and enhances RDV infection and replication ( Figure 6 ) . Taken together , these results present a novel mechanism by which the virus highjacks host factors through enhancement of the enzymatic activity of SAMS1 and increasing ethylene production or signaling , thus reducing the host antiviral defense response and enhancing virus infection and accumulation ( Figure 7 ) . These findings provide a novel mechanism , deepen our understanding of the relationship between ethylene and viral infection , and will have a significant impact on our knowledge of the crosstalk between plant hormones and virus-host interactions . We didn’t find a strong difference between WT and S11 OX#3 ( Figure 1 ) , especially in virus accumulation and infection rate . This is probably due to the relatively low expression level of Pns11 in this particular line ( Figure 1—figure supplement 1A , B ) , which may be insufficient to induce a significant increase in ACC and ethylene production ( Figure 3B , C ) . Furthermore , the hyper-susceptibility to RDV in Pns11-overexpressing plants was more prominent prior to 3 wpi ( Supplementary file 2A ) . This is easily explained by the fact that enhanced susceptibility allowed more Pns11 transgenic plants to show more conspicuous symptoms at earlier time points . This does not , however , prevent WT plants from becoming symptomatic at later time points , thus catching up with the transgenic plants in the proportion of plants that are infected . RDV infection affects a number of genes that are interact with the signaling pathways of plant hormones such as JA , ethylene , gibberellin , and auxin . A mutation of a NAC-domain transcription factor , which regulates JA signaling , confers strong tolerance to RDV infection in rice ( Yoshii et al . , 2009 , 2010 ) . Previous studies in our lab have indicated that RDV-encoded P2 interacts with β-ent-kaureen oxidases to reduce gibberellic acid synthesis , resulting in dwarfism ( Zhu et al . , 2005 ) . P2 also reprograms the initiation of auxin signaling through interaction with OsIAA10 , thus enhancing viral infection and pathogenesis ( Jin et al . , 2016 ) . Here , we report a mechanism by which RDV-encoded Pns11 promotes ethylene production to enhance plant susceptibility to viral infection ( Figure 7 ) . RNA-seq of RDV-infected rice also showed a regulation of hormone-responsive genes ( Figure 6—figure supplement 1 ) . Thus , it appears that RDV may interfere with phytohormone pathways to counteract plant immune responses . This complex crosstalk and these hormonal changes may be regulated by RDV infection , especially through interactions with host factors . Ethylene regulates numerous developmental processes and adaptive stress responses in plants ( van Loon et al . , 2006; Broekgaarden et al . , 2015; Kazan , 2015 ) . During biotic stress , ethylene is mainly responsible for defense against necrotrophic pathogens ( Pieterse et al . , 2012; Broekgaarden et al . , 2015 ) and plays a dual role in the plant defense signaling pathway . In some cases , ethylene is used by pathogens as a virulence factor to enhance pathogenesis , whereas in other cases , ethylene aids in the alleviation of stress . Generally , the plant defense responses regulated by ethylene depend on the specific host-pathogen interaction and the crosstalk between multiple signals ( Broekaert et al . , 2006; van Loon et al . , 2006; Denancé et al . , 2013; Alazem and Lin , 2015; Wang , 2015 ) . Although the function of ethylene has been addressed in various host-pathogen interactions ( Iwai et al . , 2006; Shen et al . , 2011; Groen et al . , 2013; Helliwell et al . , 2013; Kim et al . , 2013; Yang et al . , 2017 ) , the role of ethylene and its underlying mechanisms in plant-virus interactions are not well understood , with only a few reports on the involvement of ethylene and ethylene signaling in virus-host interactions ( Marco and Levy , 1979; Knoester et al . , 2001; Huang et al . , 2005; van Loon et al . , 2006; Love et al . , 2007; Endres et al . , 2010; Haikonen et al . , 2013 ) . In order to gain a deeper insight into the mechanism , we analyzed some defense or hormone-responsive genes known to function in SA- , JA- or ethylene-associated pathways and found pathogenesis-related protein 1b ( OsPR1b ) to be highly expressed in osein2 , OsSAMS1 RNAi and KO lines but expressed at reduced levels in OsEIN2 OX , OsSAMS1 OE lines and S11 OX lines ( Figure 5—figure supplement 3 ) ( Shen et al . , 2011 ) . PRs are known to be induced by pathogen infection and involved in responses to many plant phytohormones , disease resistance and general adaptation to stressful environments ( Huang et al . , 2005; Alazem and Lin , 2015 ) . Analysis of RDV microarray data and our RNA-seq data revealed that OsPR1b was also induced after virus infection ( Satoh et al . , 2011 ) ( Supplementary file 3 ) , indicating a role for OsPR1b of rice in defense against RDV . Further studies will improve our understanding of the function of ethylene in plant defense responses and its underlying mechanisms . SAMS is a key enzyme in plants and catalyzes the conversion of ATP and L-Met into SAM ( Roje , 2006 ) . Expression of the SAMS gene is induced by a number of biotic and abiotic stresses and confers increased tolerance to various stresses ( Kawalleck et al . , 1992; Boerjan et al . , 1994; Gómez-Gómez and Carrasco , 1998 ) . Overexpression of SAMS genes in plants alters development ( Boerjan et al . , 1994 ) and confers increased tolerance to abiotic stress . Knockdown of SAMS genes affects plant development ( Boerjan et al . , 1994 ) , leading to late flowering and abnormal methylation in rice ( Li et al . , 2011 ) , and is also related to viral RNA stabilization and accumulation in N . benthamiana ( Ivanov et al . , 2016 ) . We found that knockdown or knockout of OsSAMS1 resulted in abnormal phenotypes ( Figure 3—figure supplement 1 ) . These studies indicate that SAMS is a broad-spectrum signaling molecule that regulates plant responses to various stresses . SAM acts as the precursor in the biosynthesis of polyamines ( PAs ) and ethylene ( Roje , 2006 ) . The involvement of PAs and their metabolism in defense responses against diverse viruses has also been demonstrated ( Yoda et al . , 2003; Mitsuya et al . , 2009 ) . We provide strong evidence that RDV infection activates OsSAMS1 and increases the production of SAM and ethylene ( Figures 2 , 3 and 6 ) , and that disruption of the ethylene signaling pathway enhances rice tolerance to RDV infection ( Figure 5 , Figure 5—figure supplement 2 ) . GO analysis of our RNA-seq data also showed that the class of ethylene-activated pathway genes were highly enriched in both RDV-infected and OsSAMS1 OX transgenic rice , and that knockout of Ossams1 significantly affected the ethylene biosynthetic process . Interestingly , although genes in the hormone-mediated signaling pathway category were enriched in S11 OX transgenic lines , those in the ethylene-activated pathway were not . This is consistent with the observation that Pns11 overexpression enhances OsSAMS1 activity without upregulating its mRNA . In addition , relative to the Pns11 expression level in RDV infection , the transgenically expressed Pns11 level in S11 OX lines was probably low and insufficient to induce significant changes in ethylene pathway genes . Results from early microarray analyses and our RNA-seq data ( Figure 6—figure supplement 1 , Supplementary file 4 ) ( Satoh et al . , 2011; Do et al . , 2013 ) showed no obvious changes in the polyamine pathway in RDV-infected rice compared to the control . These results indicated that the ethylene pathway regulated by Pns11 and OsSAMS1 interaction may be the major determinant of RDV pathogenesis , but an additional mechanism involving other RDV proteins cannot be ruled out at this point ( Figure 7 ) . Here , we report that Pns11 enhances OsSAMS1 activity in vitro and in vivo , but the underlying mechanism remains unknown . A previous study suggested that enhanced substrate affinity may increase enzyme activity ( Toroser et al . , 1999 ) ; but we found that Pns11 did not alter the affinity of OsSAMS1 to the substrate L-Met or ATP ( Figure 2—figure supplement 5 ) . Crystal structures of SAMS have been elucidated from other non-plant organisms , and SAMS isoenzymes appear as homotetramers , dimers , or heterooligomers ( Markham and Pajares , 2009 ) . Although the crystal structure of plant SAMS has not been reported to date , the high sequence similarity of plant SAMS to the known SAMS sequences ( Figure 2—figure supplement 6 ) ( Li et al . , 2013 ) suggests that OsSAMS1 may function as dimers or tetramers . We found that OsSAMS1 exists in a high molecular weight form in RDV-infected rice ( Figure 2—figure supplement 7 ) , and we propose that the interaction of Pns11 with OsSAMS1 may promote oligomerization of OsSAMS1 to the most active form , which may increase its enzymatic activity and lead to the production of more SAM . The mechanism through which Pns11 activates OsSAMS1 remains unknown and requires further exploration . Plant growth and virus inoculation were carried out as previously described ( Wu et al . , 2015; Jin et al . , 2016 ) . Rice seedlings were grown in a greenhouse at 28–30°C for 2 weeks and plants at the third-leaf stage were inoculated with 2–3 viruliferous leafhoppers per plant for 2 days . The insects were then removed and the rice seedlings were maintained under the same growing conditions . At 4 weeks post inoculation , when the viral symptoms appeared on the new leaves , the seedlings were photographed and harvested . A minimum of 30 rice seedlings were used for each sample . The index of non-preference for each line was characterized by the mean number of settled insects on each seedling ( Supplementary file 1 ) as previously described ( Jin et al . , 2016 ) . The number of plants with symptoms for each line was recorded every week and statistical analysis of the infection rates was carried out ( Supplementary file 2 ) . The entire open reading frames ( ORFs ) of OsSAMS1 and S11 were amplified by RT-PCR and then introduced into the pCam2300:Ubi:OCS vector to yield pCam2300:Ubi:Flag OsSAMS1 and pCam2300:Ubi:HA S11 . The pUCC-OsSAMS1 was used to create the OsSAMS1 RNAi knockdown transgenic lines . The OsSAMS1 knockout construct was constructed as previously described ( Miao et al . , 2013 ) . The resulting constructs were used for transformation via Agrobacterium ( BioRun , Wuhan , China ) . All primers used in this assay are listed in Supplementary file 5A and B . Leaves of the same position were cut into 8 cm pieces and six pieces were placed into a 50 mL glass vial with distilled water sealed with a gas-proof septum . After imbibition in a growth cabinet at 28°C for 48 hr , a 0 . 1 mL gas sample was withdrawn from the head space of each bottle using a gas-tight syringe ( Hamilton ) , and the ethylene concentration was determination by gas chromatography ( Agilent 6890N ) equipped with an activated alumina column and flame ionization detectors . A six-point standard ethylene curve with concentration ranging from 0 . 5 to 3 . 0 μL·L−1 was used for the calibration . The quantified data , divided by fresh weight and time , were converted to specific activities . ACC was extracted from the same leaf tissues used for quantifying ethylene contents and ground in liquid nitrogen using a mortar and pestle , then stirred with 80% ( v/v ) ethanol ( 2 mL·g−1 fresh weight ) and the supernatant was evaporated to dryness . The residue was then resuspended in water . The ACC concentration in the supernatant was determined directly by chemical conversion to ethylene as described previously ( Lizada and Yang , 1979; Chen et al . , 2013a ) . SAM was extracted from rice leaves with 5% ( w/v ) trichloroacetic acid ( TCA , Sigma-Aldrich ) . For each extraction , frozen tissue powder ( 0 . 2 g ) was homogenized with extraction solution ( 1 mL ) for 15 min at 4°C and the homogenate was centrifuged at 10 , 000x g for 15 min followed by another centrifugation at 13 , 000x g for 15 min . The supernatant was collected by filtration through a 0 . 45 μm pore-size Millipore filter . All steps were carried out on ice or at 4°C ( during centrifugation ) to prevent SAM degradation ( Van de Poel et al . , 2010 ) . 5 μl of supernatant was used for analysis of SAM using LC-MS/MS ( Agilent UPLC 1290 MS/MS 6495 ) and the conditions are listed in Supplementary file 5C . Five standard concentrations of S-adenosyl-L-methionine solutions ( 0 . 000625 , 0 . 00125 , 0 . 0025 , 0 . 005 , and 0 . 01 mg·mL−1 ) were prepared for the standard curve . The DUALhunter starter kit ( Dualsyetems Biotech ) was used for the yeast two-hybrid assays . All protocols were carried out according to the manufacturer’s manual . The rice cDNA library was constructed in prey plasmid pPR3-N using an EasyClone cDNA library construction kit ( Dualsystems Biotech ) , and the bait plasmid was constructed by inserting full-length RDV-encoded Pns11 into the pDHB1 vector . After library screening , positive clones were selected on SD quadruple dropout ( QDO ) medium ( SD/-Ade/-His/-Leu/-Trp ) and prey plasmids were isolated from these clones for sequencing . To further distinguish positive from false-positive interactions and to confirm the interaction of bait and prey proteins , we co-transformed the two plasmids into yeast strain NMY51 and detected β-galactosidase activity with an HTX Kit ( Dualsystems Biotech ) . The ORF PCR products of S11 and OsSAMS1/2/3 were inserted into the pCam2300:35S:OCS vector ( Wu et al . , 2015 ) to yield pCam2300:35S:HA S11 and pCam2300:35S:Flag OsSAMS1/2/3 . The constructs were then co-infiltrated into N . benthamiana leaves by agroinfiltration . Leaves were harvested 3 days post-infiltration and total proteins were extracted with co-IP buffer ( 50 mM Tris-Cl pH 7 . 5 , 150 mM NaCl , 10% glycerol , 0 . 5 mM EDTA , 0 . 5% NP-40 , and 1 × protease inhibitor cocktail ) . After incubation on ice for 30 min , plant extracts were sonicated and then centrifuged . Cleared extract was combined with anti-Flag or anti-HA antibodies together with recombinant protein G-Sepharose 4B ( Invitrogen ) and incubated for 3 hr at 4°C . After washing five times with co-IP buffer , agarose beads were collected by centrifugation ( 2000x g for 2 min ) and then resuspended in protein extraction buffer . Proteins were separated by SDS-PAGE and detected with the corresponding antibody . The ORFs of S11 and OsSAMS1 were inserted into the pCAMBIA1300-nLUC and pCAMBIA1300-cLUC vectors , respectively ( Jin et al . , 2016 ) . The constructs were then transformed into Agrobacterium tumefaciens strain EHA105 and cultured to OD600 = 0 . 5 , combined with equal volumes of the adjusted culture for specific groups as shown in the figure legends , and incubated at room temperature without shaking for 3 hr followed by infiltrating into N . benthamiana leaves . The LB 985 NightSHADE system ( Berthold Technologies ) was used for luciferase activity detection 3 days after infiltration . The samples were prepared as described in the previous sections using 2 g of rice leaves and 3 mL of co-IP buffer . The lysates were then filtered through a 0 . 22 μm filter . 750 μl of total protein was loaded onto a Superdex 200 10/300 GL column ( GE Healthcare ) and 250 μl fractions were collected at 0 . 3 ml·min−1 . Bimolecular fluorescence complementation ( BiFC ) was carried out using previously described vectors and methods ( Yang et al . , 2011 ) . The ORFs of S11 and OsSAMS1 were inserted into the BiFC expression vectors p2YN and p2YC , respectively . The constructs were mixed 1:1 immediately prior to co-infiltrate into N . benthamiana leaves by agroinfiltration . Leaf tissue was analyzed 3 days post-inoculation by microscopy using a Zeiss LSM710 confocal laser scanning microscope equipped with a C-Apochromat 40X/1 . 20NA water immersion objective . Images were photographed under either white light or UV light and a Chroma filter with a 450- to 490 nm excitation wavelength and 515 nm emission wavelength was used to record YFP . All primers used in this assay are listed in Supplementary file 5A and B . Samples were extracted with IP buffer ( 50 mM Tris-Cl pH 7 . 5 , 150 mM NaCl , 10% glycerol , 0 . 5 mM EDTA , 0 . 5% NP-40 , and 1 × protease inhibitor cocktail ) . After incubation on ice for 30 min , the pull-down assay was performed utilizing a Beaver Beads Protein A/G Matrix Immunoprecipitation kit ( Beaver Nano-Technologies Co . China ) following the manufacturer's instructions with anti-OsSAMS1 antibody ( Abgent , Suzhou , China ) . Proteins were separated by SDS-PAGE and detected with the corresponding antibody . OsSAMS1 and OsSAMS2 were amplified by PCR and then inserted into the pGEX vector ( GE Healthcare Life Sciences ) and expressed as glutathione S-transferase fusion proteins ( GST-OsSAMS1/OsSAMS2 ) in E . coli BL21 cells . After purification by glutathione-agarose chromatography , the GST tag was removed . S11 , S9 , and GFP were also amplified by PCR and introduced into the pMal-p2x vector , which fused a maltose-binding protein at the N-terminal . The constructs were transformed into E . coli BL21 cells and purified by amylose affinity chromatography . Primers used for amplification of OsSAMS1 , OsSAMS2 , S11 , S9 , and GFP prior to insertion in expression vectors are listed in Supplementary file 5A and B . Indirect assays of OsSAMS1 and OsSAMS2 activity were carried out according to the scheme presented in Figure 2E using SAM production by OsSAMS1/2 as a measure of OsSAMS1/2-catalyzed L-Met with ATP yielding SAM . Mixtures containing 30 pmol OsSAMS1/2 and various amounts of Pns11 ( P9 or GFP ) in a total volume of 15 μl were pre-incubated at 30°C for 20 min . Mixtures were then added to reactions containing final concentrations of 100 mM Tris-HCl pH 8 . 0 , 200 mM KCl , 10 mM MgCl2 , 1 mM DTT , 3 . 3 mM ATP , and 5 μCi L-[35S]-methionine ( 1175 Ci/mmol ) . The reactions were incubated at 30°C for another 20 min , then OsSAMS1/2 activity was terminated by addition of 1 . 5 μl of 0 . 5M EDTA . SAM production was analyzed by thin layer chromatography on polyethyleneimine cellulose HPTLC plates developed with n-butyl alcohol:acetic acid:water ( 12:3:5 , v/v ) ( Kim et al . , 2003 ) . After chromatography , radioactive signals on plates were quantitated using a phosphorimager ( Typhoon FLA900 , GE Healthcare ) . For northern blot , 15 μg of total RNA was extracted from rice plants with Trizol ( Invitrogen ) , separated by 1% formaldehyde agarose gel and transferred to Hybond-N +membranes that were then cross-linked and dried as previously described ( Wu et al . , 2015 ) . The 500 bp probes that were partially complementary to RDV-encoded S2 , S8 , and S11 were labeled with α-32P-dCTP . The sequence of the probes and the primers are listed in Supplementary file 5D . For RT-PCR , total RNA ( 2 μg ) was reverse transcribed into cDNA by SuperScript III Reverse Transcriptase ( Invitrogen ) . qRT-PCR amplification was performed in 20 μL reactions containing 4 μL of 20-fold diluted cDNA , 0 . 5 μM of each primer , and 10 μL of SYBR Green PCR Master Mix ( Toyobo ) . The expression was normalized to that of EF-1α . Primer sequences in this assay are listed in Supplementary file 5D . Total RNAs were extracted from RDV-infected rice plants ( 4 wpi , 42-d-old seedlings ) and transgenic rice lines ( 42-d-old seedlings ) using the RNeasy plant mini kit ( Qiagen ) . The RNA-seq analyses were performed at Bionova Company . Libraries were constructed through adaptor ligation and were subjected to pair-ended sequencing with a 150-necleotide reading length . FastQC software was used to access the quality of raw sequencing reads . After removing adaptor and low-quality reads , clean reads were mapped to rice genome MSU7 . 0 using TopHat . Responsive genes were identified by reads per kilobase per million reads ( RPKM ) and edgeR software was used to identify differential expressed genes . The multiple-testing adjusted P-value ( FDR < 0 . 05 ) and fold change ( FC >2 ) was used to determine whether the gene was significantly differentially expressed or not . Three biological replicates were used , and their repeatability and correlation were evaluated by the Pearson’s Correlation Coefficient ( Schulze et al . , 2012 ) . The outputs of RNA-seq analysis used in this study ( series number GSE102366 ) are available at NCBI-GEO . The MST assay was performed as previously described ( Jin et al . , 2016 ) . OsSAMS1 protein was labeled with the red fluorescent dye NHS according to the Monolith NT Protein Labeling Kit RED-NHS instructions ( NanoTemper Technologies GmbH; München , Germany ) . The concentration of NHS-labeled OsSAMS1 was held constant at 1 . 25 μM , whereas the concentrations of L-Met and ATP were gradient-diluted ( L-Met: 200 , 000 nM , 100 , 000 nM , 50 , 000 nM , until it reached 97 . 7 nM; ATP: 50 , 000 nM , 25 , 000 nM , 12 , 500 nM , until it reached 24 . 4 nM ) . After a short incubation , the samples were loaded into MST standard treated glass capillaries . Measurements were performed at 25°C in buffer containing 20 mM Tris pH 8 . 0 and 150 mM NaCl using 40% LED power and 20% MST power . The assays were repeated three times for each affinity measurement . Data analyses were performed using the Nanotemper Analysis and OriginPro 8 . 0 software provided by the manufacturer . In affinity affecting assays , 2 . 5 μM of MBP-Pns11 or MBP was added to 1 . 25 μM NHS-labeled OsSAMS1 and gradient-diluted concentrations of L-Met or ATP . After a short incubation , the samples were loaded into MST standard treated glass capillaries for MST analysis as described above .	Rice provides food for billions of people all over the world , but diseases caused by plant-infecting viruses cause serious risks to the production of rice . As a result , there is an urgent demand for developing new and impactful ways to help defend rice plants from harmful viruses . Toward this goal , it will be important to better understand how viruses actually cause diseases in plants . Plants make chemicals known as hormones to control their own development , and hormone production is often disturbed when viruses infect rice plants . Many viruses cause infected plants to make more of a gaseous hormone called ethylene , which benefits the viruses . Yet , it is still not known how virus infection induces the production of more ethylene . Zhao , Hong et al . have exposed rice plants to infection with a virus called rice dwarf virus . Infected plants made more ethylene than normal , which did indeed help the virus to infect . Further experiments then showed that an enzyme that makes one of the building blocks needed to produce ethylene became more active after infection with this virus . Next , Zhao , Hong et al . engineered rice plants to make more or less of this building block – which is known as S-adenosyl-L-methionine or SAM for short . Plants with too much SAM were less able to defend themselves against the virus , while plants that lacked SAM were better able to fight off viral infection . Zhao , Hong et al . suggest that engineering rice plants to make less of the SAM-producing enzyme could make them more resistant to viruses . Further work will also be needed to find out why ethylene benefits viral infection , and to confirm whether ethylene also performs similar roles when rice is infected with other viruses .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "plant", "biology" ]	2017	A viral protein promotes host SAMS1 activity and ethylene production for the benefit of virus infection
Matrix stiffening with downstream activation of mechanosensitive pathways is strongly implicated in progressive fibrosis; however , pathologic changes in extracellular matrix ( ECM ) that initiate mechano-homeostasis dysregulation are not defined in human disease . By integrated multiscale biomechanical and biological analyses of idiopathic pulmonary fibrosis lung tissue , we identify that increased tissue stiffness is a function of dysregulated post-translational collagen cross-linking rather than any collagen concentration increase whilst at the nanometre-scale collagen fibrils are structurally and functionally abnormal with increased stiffness , reduced swelling ratio , and reduced diameter . In ex vivo and animal models of lung fibrosis , dual inhibition of lysyl oxidase-like ( LOXL ) 2 and LOXL3 was sufficient to normalise collagen fibrillogenesis , reduce tissue stiffness , and improve lung function in vivo . Thus , in human fibrosis , altered collagen architecture is a key determinant of abnormal ECM structure-function , and inhibition of pyridinoline cross-linking can maintain mechano-homeostasis to limit the self-sustaining effects of ECM on progressive fibrosis . Fibrotic diseases are a major cause of morbidity and mortality worldwide and their prevalence is increasing with an ageing population . Within the lung , idiopathic pulmonary fibrosis ( IPF ) is considered the prototypic chronic progressive fibrotic disease ( Raghu et al . , 2011 ) . Treatment options are limited , and with a median survival of less than 3 years from diagnosis , more effective therapies are urgently needed ( Richeldi et al . , 2017 ) . Whilst the exact mechanisms of progressive lung fibrosis are uncertain , IPF is thought to result from repetitive micro-injuries to the alveolar epithelium promoting fibroblast differentiation into extracellular matrix ( ECM ) -producing myofibroblasts . Persistent myofibroblast activation results in ECM deposition which eventually destroys normal alveolar architecture and disrupts gas exchange ( Kirk et al . , 1986 ) . Excessive deposition of fibrillar collagens is considered synonymous with fibrosis . Fibrillar collagens are a major component of lung ECM that form a scaffold to support tissue architecture and are a primary determinant of tissue stiffness ( Senior et al . , 1975; White , 2015 ) . During biosynthesis , collagen molecules acquire several post-translational modifications , including lysine hydroxylation and oxidation that are critical to the structure and biological functions of this protein . Oxidative deamination of lysine or hydroxylysine initiates cross-linking reactions that are essential to stabilise the supramolecular assembly of collagen and produce stable collagen fibrils ( Shoulders and Raines , 2009; Kadler et al . , 1996 ) . The lysyl oxidase ( LOX ) enzymes are a family of five secreted copper-dependent amine oxidases ( LOX and LOX-like ( LOXL ) 1 to 4 ) that are responsible for post-translational modification of collagen in the ECM to initiate covalent cross-linking . LOX family members have been implicated as possible therapeutic targets in cancers and fibrosis ( Trackman , 2016 ) . The type of LOX/LOXL mediated collagen cross-link is determined by the hydroxylation of telopeptidyl and helical lysine residues in collagen prior to cross-link formation ( Yamauchi and Sricholpech , 2012 ) , with increased hydroxylation of telopeptide lysine residues by lysyl hydroxylase 2/procollagen lysine , 2-oxoglutarate 5-dioxygenase 2 ( LH2/PLOD2 ) proposed to be a general fibrotic phenomenon causing increased hydroxyallysine derived pyridinoline cross-links ( Brinckmann et al . , 1999; van der Slot et al . , 2003 ) . Both ECM amount and altered post-translational modifications are postulated to increase matrix stiffness and this stiffening has been proposed to induce self-sustaining mesenchymal cell activation and progressive fibrosis in a positive feedback loop ( Wipff et al . , 2007; Liu et al . , 2010; Zhou et al . , 2013; Booth et al . , 2012; Chen et al . , 2016; Parker et al . , 2014; Shi et al . , 2011; Liu et al . , 2015 ) . Whilst there is a growing understanding of mechanosensitive cellular pathways that are activated by increased ECM stiffness , the specific changes in ECM structure and function that disrupt mechano-homeostasis have not been defined in human fibrosis ( Burgstaller et al . , 2017; Tschumperlin et al . , 2018; Herrera et al . , 2018 ) . To investigate this , we performed the first integrated multiscale structure-function analysis of human fibrosis tissue , and then extended our findings into mechanistic studies in vitro and in vivo . We identify that , at the time of diagnosis , in human lung fibrosis tissue altered collagen architecture rather than collagen concentration is a key determinant of abnormal ECM structure-function , and that targeting pathways which dysregulate collagen architecture may restore ECM homeostasis and so prevent persistent mechanosensitive cellular activation and fibrosis progression . We performed an integrated biochemical and biomechanical characterisation comparing human IPF lung tissue with age-matched control lung tissue . We first used atomic force microscopy ( AFM ) cantilever-based microindentation to assess lung tissue stiffness at the micrometre-scale . IPF tissue was significantly stiffer than control tissue ( Figure 1A ) , with spatially heterogeneous changes in stiffness in IPF tissue including highly localised areas of increased stiffness which were not present in control lung tissue ( Figure 1B and C ) , consistent with the known histopathological heterogeneity of IPF tissue ( Raghu et al . , 2011 ) . We then explored the relative contribution of collagen amount and/or post-translational modifications to these differences in stiffness . Whilst an increase in fibrillar collagen was suggested by second harmonic generation imaging of IPF lung tissue ( Figure 1D ) , quantitation of total collagen concentration by hydroxyproline assay showed no difference in mean total collagen in IPF tissue relative to control tissue following normalisation to either dry weight or to total protein ( Figure 1E and Figure 1—figure supplement 1 ) ; in addition , no dependence of lung tissue stiffness on collagen concentration was found ( Figure 1F ) . In contrast , differences in the expression of collagen cross-linking enzymes were identified in IPF tissue . Whilst mRNA expression levels of LOX and LOXL1 were unchanged ( Figure 2A and B ) , there were significant increases in the relative expression of LOXL2 , LOXL3 , and LOXL4 ( Figure 2C–E ) . This was associated with detection of increased amine oxidase activity in IPF lung tissue sections ( Figure 2F ) . Consistent with previous reports of increased LH2 expression in fibrotic tissue ( Brinckmann et al . , 1999; van der Slot et al . , 2003 ) , we identified increased expression of LH2 in IPF tissue ( Figure 3A and B ) suggesting the potential for altered collagen cross-linking involving hydroxylysine residues . Therefore , we quantified immature ( dihydroxylysinonorleucine ( DHLNL ) and hydroxylysinonorleucine ( HLNL ) ) and mature trivalent ( deoxypyridinoline ( DPD ) and pyridinoline ( PYD ) ) hydroxyallysine-derived collagen cross-links . This revealed a significant increase in the density of immature divalent ( Figure 3C ) and mature trivalent ( Figure 3D ) cross-links in IPF tissue , with an increase in the relative ratio of immature to mature cross-links in IPF tissue ( Figure 3E ) consistent with increased collagen metabolism in IPF tissue and an increase in the DHLNL to HLNL ratio ( Figure 3F ) consistent with increased lysyl hydroxylase activity . No dependence of lung tissue stiffness on immature cross-link density was found ( Figure 3G ) , although a trend towards a correlation with DHLNL cross-link density alone was observed ( correlation 0 . 41 p=0 . 09 ) . In contrast , mature pyridinoline collagen cross-link density was strongly and significantly associated with increasing lung tissue stiffness ( correlation 0 . 72 p=0 . 0007 ) ( Figure 3H ) . Together , these results identify that in IPF lung tissue , differential expression of LH2 and LOXL-family members is associated with increased hydroxyallysine-derived pyridinoline collagen cross-links and this , rather than collagen concentration , is a primary determinant of increased tissue stiffness . Given the finding of increased collagen cross-linking in IPF , the morphological and biomechanical properties of individual collagen fibrils were analysed by AFM cantilever-based nanoindentation , following enzymatic extraction using an established methodology ( Andriotis et al . , 2014 ) ( Figure 4A and B ) . Overall , IPF collagen fibrils were stiffer but exhibited a greater range of stiffness measurements compared with control lung collagen fibrils ( Figure 4C ) . They also showed a skewed size distribution with a significantly smaller median diameter ( Figure 4D and E ) and had a lower fibril swelling ratio ( hydrated to air ) ( Figure 4F ) . These data identify that in human lung fibrosis perturbation of collagen homeostasis occurs at the nanometre-scale , with individual collagen fibrils being structurally and functionally abnormal . To explore the underlying mechanisms of altered fibrillogenesis and collagen structure-function in IPF , we first established a novel long-term 3D in vitro model of lung fibrosis using primary human lung fibroblasts treated with the pro-fibrotic cytokine TGF-β1 . We studied lung fibroblasts from patients with IPF given previous studies identifying aberrations in their fibrogenic responses ( Ramos et al . , 2001 ) . An advantage of this model is that it allows not only direct evaluation of cross-linking and its inhibition but also direct measurement of their influence upon tissue biomechanics . Over 6 weeks of culture , the model shows a progressive increase in collagen content and mature collagen cross-links in association with expression of all LOX/LOXL family members ( Figure 5A–D ) ; histochemical staining of the resultant tissue construct showed fibrillar collagen deposition ( Figure 5E ) similar to that seen in fibroblastic foci , the site of active fibrogenesis in IPF ( Figure 5F ) ( Jones et al . , 2016 ) . Given our finding of differential LOXL enzyme expression , we evaluated the contribution of these enzymes to collagen cross-linking and tissue stiffness using haloallylamine based compounds which have previously been demonstrated to be amine oxidase inhibitors that show selectivity towards individual enzymes ( Schilter et al . , 2015; Chang et al . , 2017 ) . We tested a number of these compounds for inhibition of each LOX/LOXL family member , and identified that one of these inhibitors , PXS-S2A , which was previously reported as a LOXL2-selective inhibitor ( Chang et al . , 2017 ) , is a dual LOXL2 and LOXL3-selective inhibitor . PXS-S2A irreversibly inhibits LOXL2 and LOXL3 ( IC500 . 005 μM and 0 . 016 μM respectively ) but is reversible and more than a hundred times less potent for inhibition of LOX and LOXL1 ( Figure 6A ) . Using the inhibitor in the in vitro model , we observed a dose-dependent inhibition of collagen cross-links ( Figure 6B–D ) , with a greater than 50% reduction in mature pyridinoline cross-links ( Figure 6D ) with 0 . 1 μM PXS-S2A , a concentration which completely inhibits LOXL2 and LOXL3 but has minimal effects on LOX and LOXL1 ( Figure 6A ) . At a pan LOX/LOXL inhibitory concentration of 10 μM , the reduction in cross-links was comparable to that observed with β-aminoproprionitrile ( BAPN ) , a non-selective LOX/LOXL inhibitor ( Trackman , 2016 ) . No reduction in total collagen content was observed following treatment with PXS-S2A or BAPN ( Figure 6E ) . The biomechanical consequence of inhibiting collagen cross-linking was then investigated with parallel plate compression testing . The LOXL2/LOXL3 inhibitor dose-dependently reduced tissue stiffness as measured by Young’s modulus , achieving maximal inhibition of stiffness as compared with BAPN at a pan LOX/LOXL inhibitory concentration of 10 μM . However , almost 70% of maximal reduction in stiffness ( as compared with BAPN ) was already achieved at 0 . 1 μM PXS-S2A which is selective for LOXL2/LOXL3 ( Figure 6F and G , Figure 6—figure supplement 1 , and Video 1 ) , suggesting that inhibition of LOXL2/LOXL3 is sufficient to achieve a substantial reduction in tissue stiffness . Consistent with the IPF tissue findings , stiffness showed no dependency on collagen content ( Figure 6H ) whilst it correlated positively with collagen cross-link density including both immature ( Figure 6I ) and mature pyridinoline cross-links ( Figure 6J ) . Together , these data identify pyridinoline cross-link density to be a significant determinant of tissue stiffness and identify LOXL2/LOXL3 enzyme activities as essential contributors to this process . We next assessed the impact of selective LOXL2/LOXL3 inhibition upon collagen morphology in the in vitro fibrosis model . When visualised by polarised light Picrosirius red microscopy , highly organised collagen fibrils were evident in vehicle-treated fibrotic control cultures as well as in those treated with 0 . 1 μM PXS-S2A; this contrasted with the marked disorganisation observed with BAPN ( Figure 7A ) . Based on these findings , we performed ultrastructural analysis of the collagen fibrils using electron microscopy ( Figure 7B ) . At 0 . 1 μM PXS-S2A , a 12 . 3% mean increase in fibril diameter was observed ( Figure 7C ) , comparable with the 18% difference in fibril size observed in our AFM analyses comparing control and IPF collagen fibrils . Increasing the concentration of PXS-S2A to 0 . 5 μM caused a further small increase in fibril diameter; however , with complete LOX/LOXL inhibition using BAPN there was a broadening of fibril diameter distribution and a marked dysregulation of fibril structure including irregular profiles ( Figure 7B and C ) . In keeping with this , BAPN markedly increased collagen solubility , with only 14% of the total collagen remaining insoluble after proteolytic digestion , whilst with 0 . 1 μM PXS-S2A only a small increase in collagen solubility was observed suggesting that inhibition of LOXL2/LOXL3 did not significantly disrupt fibril stability ( Figure 7D ) . Together , these results identify that LOX/LOXL activity is an essential regulator of collagen fibril assembly and that selective LOXL2/LOXL3 inhibition is sufficient to significantly reduce tissue stiffness and normalise collagen fibril architecture . To extend our in vitro studies , the efficacy of an orally bioavailable small molecule inhibitor ( PXS-S3B , IC500 . 008 μM and 0 . 020 μM for LOXL2 and LOXL3 , compared with 0 . 118 μM , 1 . 130 μM and 1 . 710 μM for LOXL4 , LOX and LOXL1 respectively ) was then assessed at LOXL2/LOXL3-selective doses in an experimental animal model of lung fibrosis . Control experiments using the 3D in vitro model of fibrosis confirmed the effect of PXS-S3B on collagen cross-linking and tissue stiffness was comparable to PXS-S2A ( Figure 8—figure supplement 1 ) . Given the importance of TGF-β as a driver of fibrosis , we used an in vivo model of lung fibrosis driven by transient overexpression of active TGF-β1 by adenoviral vector gene transfer , which results in severe progressive fibrosis and recapitulates several features of pulmonary fibrosis in human disease ( Sime et al . , 1997 ) . Rats received either a replication-defective adenoviral vector producing active TGF-β1 ( AdTGF-β1 ) or vector only control ( AdDL ) on day 0 , and then oral administration of PXS-S3B ( 15 or 30 mg/kg/day ) from day 1 to day 28 at which point lung function was assessed before the animals were euthanised for biochemical and histological analysis of the lungs . There was no effect of PXS-S3B on total collagen content of the lungs ( Figure 8A ) , but there was a significant reduction in fibrosis compared to vehicle-treated AdTGF-β1 rats , as assessed by modified Ashcroft scores ( Figure 8B and C ) , and a significant improvement in lung function ( i . e . a reduction in lung stiffness as measured by pressure-driven pressure volume-loops and elastance ) at day 28 ( Figure 8D and E ) , reflecting our in vitro finding that LOXL2/LOXL3 inhibition reduces tissue stiffness . This was accompanied by a reduction in mature pyridinoline collagen cross-links compared with the AdTGF-β-treated rats ( Figure 8F and G ) , as well as immature ( DHLNL + HLNL ) cross-links ( Figure 8H ) . We also determined that the DHLNL:HLNL ratio was reduced in the treated rats suggesting a reduction in lysine hydroxylation ( Figure 8I ) . Consistent with this , analysis of mRNA expression in the lungs of animals treated with LOXL2/LOXL3 inhibitor , revealed a decrease in LH2 , which is required for telopeptide hydroxyallysine-derived cross-links . In addition , expression of other modulators of fibrillogenesis including collagen V , TGM2 , dermatopontin , fibulin , fibrillin , and periostin ( Figure 8J ) were reduced . Thus , inhibition of LOXL2/LOXL3 not only modifies collagen cross-linking but also modulates fibrillar collagen homeostasis . No adverse effects of treatment with the LOXL2/LOXL3 inhibitor were evident in the model . One of the critical issues in the pathobiology of fibrosis is an insufficient understanding of the structure-function relationship of the ECM in human disease . A major goal of this study was to perform an integrated biomechanical and biochemical characterisation of fibrillar collagen and its pathological role in human fibrosis . We demonstrate that dysregulated collagen fibrillogenesis is a key pathway of human lung fibrosis at the time of diagnosis , with structurally and functionally abnormal collagen – rather than collagen content - increasing lung tissue stiffness as a consequence of pathologic pyridinoline cross-linking . To advance our mechanistic understanding of these findings , we employed small molecule LOXL inhibitors and demonstrate that dual inhibition of LOXL2 and LOXL3 is sufficient to prevent accumulation of these pyridinoline cross-links . This normalises the structure and biomechanical properties of fibrillar collagen , and reduces tissue stiffness without the need to significantly disrupt fibril assembly . Our findings suggest that this targeted inhibition reduces the disruption of mechano-homeostasis to limit the self-sustaining effects of ECM on progressive fibrosis . A thought provoking finding of our studies is that total collagen concentration , as determined by measurement of hydroxyproline that quantifies all collagen irrespective of whether it is assembled into fibrils , was not significantly increased in IPF tissue . Whilst Selman et al identified an increase in collagen of fibrotic lung biopsy samples ( Selman et al . , 1986 ) , other studies have reported no significant increase in collagen deposition ( Nkyimbeng et al . , 2013; Fulmer et al . , 1980; Westergren-Thorsson et al . , 2017 ) , or have reported increased collagen deposition only in late stage disease at autopsy , but not in diagnostic lung biopsy samples ( Kirk et al . , 1986 ) . This suggests that , at the time of diagnosis , any increase in collagen is paralleled by changes in other proteins ( most likely other ECM and myofibroblastic cellular proteins ) , but that by late-stage fibrosis there may be a progressive increase in collagen accumulation , perhaps as the fibrotic tissue becomes more acellular and/or the collagen resists degradation . Furthermore , although we identified an increase in stiffness of IPF lung tissue using AFM micro-indentation , this was not related to collagen concentration . Rather , increased mature fibrillar collagen pyridinoline cross-linking secondary to dysregulated post-translational modification was a primary determinant of tissue stiffness . Fibrillar collagen is a predominant component of the ECM which provides tissues with essential tensile strength by forming a hierarchical assembly of substructures dependent upon their cross-linking: tropocollagen molecules assemble into fibrils and the fibrils further assemble into fibres . Here , we identified that collagen hierarchical assembly is altered at the nanoscale/ultrastructural level in IPF . Collagen fibrils were structurally and functionally abnormal with reduced diameter and increased stiffness , suggesting that at a molecular level they have the potential to influence the microenvironment of individual cells to affect their biomechanical sensing . Covalent cross-links have been proposed to increase fibril stiffness by forming a lateral network of linkages ( Bailey , 2001; Miles et al . , 2005 ) . In models of artificially cross-linked collagen fibrils , it was observed that by drawing the collagen molecules closer together there was a reduction in water between molecules; as a consequence , the density of collagen fibrils increased and this was proposed to affect their mechanical properties ( Bailey , 2001; Miles et al . , 2005 ) . Whether a similar effect occurs after enzymatic cross-linking , especially pathological cross-linking involving pyridinoline cross-links , remains to be determined . However , the reduction in fibril diameter and lesser fibril swelling with hydration of IPF fibrils would be consistent with this concept . Whilst direct experimental evidence is required to determine whether collagen swelling is directly related to cross-linking , collagen hydration alone has been shown to significantly influence individual collagen fibril mechanics and the mechanism of molecular slippage and stretching within fibrils ( Andriotis et al . , 2018; Gautieri et al . , 2011; Yang et al . , 2008 ) . Several different mechanisms have been proposed to contribute to alterations in tissue stiffness including LOX/LOXL and transglutaminase 2 ( TGM2 ) -mediated enzymatic cross-linking and non-enzymatic glycation . Here we have focussed on the LOX and LOXL enzymes which play key roles in the process of fibrillar collagen production , with expression being tightly controlled in normal development ( Trackman , 2016 ) . Whilst LOXL2 has been proposed to have pathologic roles in cancer and fibrosis ( Barker et al . ; Barry-Hamilton et al . , 2010 ) , and increased LOXL2 expression in IPF stroma has been observed previously , less is understood regarding LOXL3 . Previous studies have identified that it is expressed in human IPF lung tissue , has amine oxidase activity for collagens , is induced by fibroblast culture on IPF cell-derived ECM , and is an enhancer and key regulator of integrin signalling in myofibre formation ( Kraft-Sheleg et al . , 2016; Lee and Kim , 2006; Philp et al . , 2018 ) . Recently , both LOXL2 and LOXL3 were identified to be crucial for fibroblast-to-myofibroblast transition in in vitro models of lung fibrosis ( Aumiller et al . , 2017 ) . In keeping with our finding of increased LOXL expression together with increased LH2 expression in IPF tissue , we identified an increase in mature hydroxyallysine-derived cross-links in IPF lung tissue , consistent with a previous report of an increase in mature DPD cross-links in fibrotic lung tissue ( Last et al . , 1990 ) . The extent of lysine hydroxylation can vary from 15 to 90% depending on the collagen types and , even within type I collagen , it varies significantly from tissue to tissue and under physiological or pathological conditions ( Yamauchi and Sricholpech , 2012 ) . Although it is unknown whether LOXL2 and/or LOXL3 have a substrate preference for hydroxylysine , the shift from a skin ( allysine-derived ) to a bone-type ( hydroxyallysine-derived ) of collagen cross-link appears of pathogenetic importance in IPF . Various approaches have investigated the roles of collagen cross-linking by LOX/LOXL enzymes in fibrosis . Multiple studies have tested BAPN as a broad-spectrum LOX/LOXL inhibitor ( Trackman , 2016 ) , although this approach has no therapeutic potential as it does not enable specific enzyme targeting and has undesirable effects on LOX leading to unwanted vascular and skeletal side-effects ( Wawzonek et al . , 1955 ) . A monoclonal antibody which binds and partially inhibits LOXL2 , showed efficacy in the bleomycin mouse model of lung fibrosis ( Barry-Hamilton et al . , 2010 ) . However , recently , a Phase 2 study in IPF of simtuzumab , the humanised version of this antibody , was terminated due to lack of efficacy ( Raghu et al . , 2017 ) . Our finding of differential expression of LOXL2 , LOXL3 and LOXL4 in IPF tissue suggests that therapeutic targeting of additional LOXL family members is required to inhibit pathologic collagen cross-linking . To investigate this finding , we used dual LOXL2 and LOXL3-selective inhibitors which irreversibly inhibit LOXL2 and LOXL3 but are reversible and more than a hundred times less potent for inhibition of LOX and LOXL1 . Characterisation in a novel 3D in vitro model of human fibrosis enabled direct measurement of the influence of LOXL2/LOXL3 inhibition upon pathological changes in collagen structure-function . A key finding of our study is that approximately 70% of maximal reduction in tissue stiffness in vitro occurred following selective inhibition of only LOXL2/LOXL3 when pyridinoline cross-links were significantly reduced and fibril diameter was normalised , whilst fibril architecture was preserved and only a small increase in protease-sensitive collagen solubility was evident . Together our data suggest that in lung fibrosis reducing pathological collagen cross-linking normalises fibril architecture and is sufficient to reduce tissue stiffness without the need to completely disrupt fibril assembly . While LOXL4 was also increased in IPF , our dose-response findings from the parallel plate compression system suggest that it does not make a significant contribution to tissue stiffness . In comparison with selective LOXL2/LOXL3 inhibition in the current study , previous studies treating tendon constructs from days 14 to 21 with the pan LOX/LOXL inhibitor BAPN showed that collagen fibrils were irregular with widely dispersed diameters ( Herchenhan et al . , 2015 ) and we observed a similar effect of BAPN in the current study . Thus , while pan LOX/LOXL inhibition disrupts normal collagen fibrillogenesis , selective LOXL2/LOXL3 inhibition was sufficient to prevent pathological collagen crosslinking and restore fibrillar collagen homeostasis . Although reduction in hydroxyproline measurements of collagen is recommended as a primary measure of anti-fibrotic efficacy during preclinical animal model testing of putative therapeutics for lung fibrosis ( ATS Assembly on Respiratory Cell and Molecular Biology et al . , 2017 ) , our data from human tissue obtained at time of diagnosis identifies that effects on collagen fibrillogenesis are more relevant pathogenic mechanisms with abnormal collagen cross-linking and increased tissue stiffness likely to precede any accumulation of collagen in late-stage human pulmonary fibrosis . Consistent with our proposal , in a rat carbon tetrachloride-induced model of liver injury , increased liver stiffness was associated with increased collagen cross-linking and this was proposed to prime the liver to respond quickly to injury via mechanical feed-forward mechanisms ( Perepelyuk et al . , 2013; Georges et al . , 2007 ) . The relevance of LOXL2/LOXL3-selective inhibition was demonstrated in a TGF-β-driven rat model of lung fibrosis . The model leads to persistent and severe interstitial fibrosis and does not have the confounding robust inflammatory response observed in the commonly studied bleomycin induced mouse model of fibrosis so enabling direct assessment of fibrotic responses ( Sime et al . , 1997; ATS Assembly on Respiratory Cell and Molecular Biology et al . , 2017 ) . Selective targeting of LOXL2 and LOXL3 in this model reduced pyridinoline collagen cross-links , reduced fibrosis , and improved lung function without a significant effect on hydroxyproline content . Whilst this did not completely restore lung function back to normal , there was a significant beneficial effect even in this model of fibrosis which is very strongly driven by TGF-β . An interesting finding was that inhibition of LOXL2 and LOXL3 reduced expression of genes known to influence collagen fibrillogenesis including LH2 , collagen V , TGM2 , dermatopontin , fibulin , fibrillin , and periostin . As these genes , as well as LOXL2 and LH2 are induced by TGF-β , one possible explanation is that TGF-β plays a causal role in fibrosis by disrupting collagen homeostasis to drive abnormal collagen cross-linking and promoting assembly of stiffer collagen fibrils . This change in stiffness initiates self-sustaining mechanosensory signalling pathways in a feed-forward mechanism to drive fibroblast activation and progressive fibrosis ( Herrera et al . , 2018; Tschumperlin et al . , 2018 ) . Given the important contribution of fibrillar collagen in forming a scaffold that supports tissue architecture , our data identify that normalising the biomechanical properties of fibrillar collagen in lung fibrosis may reset mechanosensitive cellular mechanisms and aid restoration of tissue function more effectively than approaches that promote extensive collagen degradation , since significant loss of ECM structure is itself pathological , as evidenced by the emphysematous lung . In summary , this integrated study identifies that altered collagen architecture is a key determinant of abnormal ECM structure-function in human lung fibrosis at time of diagnosis . Dual inhibition of LOXL2 and LOXL3 reduces pathological pyridinoline cross-links , normalises collagen morphology , reduces tissue stiffness , and improves lung function . Thus , targeting pathways which dysregulate collagen architecture may restore ECM homeostasis and so prevent persistent mechanosensitive cellular activation and fibrosis progression . All human lung experiments were approved by the Southampton and South West Hampshire and the Mid and South Buckinghamshire Local Research Ethics Committees ( ref 07/H0607/73 ) , and all subjects gave written informed consent . Clinically indicated IPF lung biopsy tissue samples and age-matched non-fibrotic control tissue samples ( macroscopically normal lung sampled remote from a cancer site in patients undergoing surgery for early stage lung cancer ) deemed surplus to clinical diagnostic requirements were flash frozen and stored in liquid nitrogen . All IPF samples were from patients subsequently receiving a multidisciplinary diagnosis of IPF according to international consensus guidelines ( Raghu et al . , 2011 ) . Unless otherwise indicated , serial 50 μm cryosections were utilised for: enzymatic extraction of collagen for AFM-based nanoindentation experiments ( five sections ) , gene expression analysis ( five sections ) , AFM-based microindentation experiments ( two sections ) , hydrolysis for hydroxyproline and collagen cross-link quantification ( approximately 1 mm3 of adjacent lung tissue following cryosection ) . For second harmonic generation imaging archived formalin-fixed paraffin-embedded samples from control and IPF donors were studied . AFM-based experiments were performed using a NanoWizard ULTRA Speed A AFM system ( JPK Instruments AG , Berlin ) . For AFM-nanoindentation experiments , pyrex nitride cantilevers ( NanoWorld AG , Switzerland ) of 0 . 48 Nm-1 nominal spring constant were used and for microindentation experiments CSC38 cantilevers ( μMasch , Innovative Solutions Bulgaria Ltd . , Bulgaria ) of 0 . 03 Nm-1 spring constant and a 15 μm diameter microsphere were used . The spring constant is defined as the ratio of the force affecting the spring to the displacement caused by it . During mechanical assessment samples were hydrated in phosphate buffered saline ( pH7 . 4 ) . AFM-based microindentation experiments were performed on 50 μm sections . By attaching a microsphere on to the end of the flexible AFM cantilever , microindentation enables assessment of the mechanics of larger structures , such as localised areas on tissue sections composed of hundreds of collagen fibrils ( Kain et al . , 2018 ) . The tip radius of these spheres was approximately 7 . 5 μm , giving a volume interaction between the sphere and the tissue in the order of 10–100 μm3 . Frozen cryosections were warmed to room temperature in PBS containing a protease inhibitor cocktail to minimise tissue degradation and between 80 and 150 force-displacement curve measurements were performed at 6 to 11 sites across each tissue section . For nanoindentation , collagen was enzymatically extracted from tissue sections using 1 mg/ml bovine hyaluronidase and 1 mg/ml trypsin in 0 . 1 M Sorensen's phosphate buffer ( pH 7 . 2 ) at 37°C for 24 hr ( Stolz et al . , 2009; Andriotis et al . , 2014 ) . After thorough washing with deionised water , the samples were smeared on to glass slides to reveal areas with individual fibrils . Samples were then left to dry overnight at 37°C and stored in a desiccating storage box with silica gels until analysis . Extracted collagen fibrils were assessed by employing a conical tip with tip radius of 7–10 nm at the end of a flexible cantilever , giving a volume interaction between the tip and collagen fibril of 0 . 1–1 nm3 . AFM imaging prior to force-displacement acquisition of the dry sample enabled individual collagen fibrils to be identified from the characteristic 67 nm D periodicity . The sample was then hydrated in PBS and force-displacement data were recorded and subsequently analysed with the well-established mathematical Oliver-Pharr model ( Oliver and Pharr , 2004 ) , so enabling determination of the indentation modulus of the measured collagen fibril ( Andriotis et al . , 2014 ) . For each tissue sample , a minimum of four collagen fibrils were studied , with 30 to 50 force-displacement curves measured per fibril . The total acquisition time was kept as low as possible to minimise thermal drift effects . RNA was isolated using a Savant FastPrep FP12 Ribolyser ( Qbiogene , Cedex , France ) and RNeasy Mini kits ( Qiagen , Manchester , UK ) according to the manufacturer’s instructions . For the in vitro lung fibrosis model , RLT buffer also contained Proteinase K ( Sigma , Poole , UK ) to aid lysis . RNA was reverse transcribed to cDNA using a QuantiTect Reverse Transcription Kit or Precision nanoScript2 Reverse Transcription Kit ( Primer design , Southampton , UK ) according to manufacturer’s instructions . Real-time quantitative polymerase chain reaction ( RTqPCR ) was performed using a BioRad CFX96 ( BioRad , Watford , UK ) or a NanoString ( NanoString Technologies , Seattle , USA ) system with a custom CodeSet . Primers and TaqMan probe sets were obtained from Primer Design . Changes in gene expression were compared to two or three constitutively expressed housekeeping genes ( for human samples UBC/A2 and for rat samples Pgk1/Actab/Hprt1 ) . Data were normalised to the mean of the housekeeping genes , and these data normalised to an appropriate baseline control sample . Formalin-fixed paraffin-embedded human lung tissue sections ( 5 μm ) from control and IPF donors ( n = 5 per group ) were dewaxed and imaged using a custom laser scanning microscope ( output of an Optical Parametric Oscillator ( two picosecond pulses , at 80MHz repetition rate Levante Emerald OPO , pumped by Emerald engine , APE , Berlin ) coupled into an inverted microscope ( Ti-U , Nikon , Japan ) via a galvanometric scanner and a short pass excitation dichroic ( 750 nm cut-off , Chroma 750spxrxt ) . Samples were excited at 800 nm and the epi-collected second harmonic signal was further filtered with a 442 nm dichroic mirror ( Semrock , DI-02-R442 ) and a band pass filter centred at 400 nm ( Thorlabs , FB400-40 ) and was finally detected with a photomultiplier tube ( Hamamatsu 10722–210 ) . Samples were imaged with 20 mW of power measured at the sample using the same gain for each sample . For image co-registration , an adjacent 5 μm tissue section was H and E stained and imaged using a Dot-Slide scanning system ( Olympus , Southend-on-Sea , UK ) . Image co-registration was performed in the Fiji distribution ( Schindelin et al . , 2012 ) ( version 1 . 49 p ) of ImageJ using landmark correspondences . An in situ assay for the detection of amine oxidase activity in skin ( Langton et al . , 2013 ) was adapted for use on lung tissue . The assay is based on oxidation of the amine substrate , 1 , 4-diaminobutane and chemiluminescent detection of liberated H2O2 with horseradish peroxidase and luminol . Cryosections ( 5 μm ) of non-fibrotic or IPF lung tissue were placed on glass slides and air dried for 10 min . The substrate/detection mixture was applied to each tissue section and incubated at 37°C for 5 min . After removal of excess substrate solution and covering the tissue section with a coverslip , each section was digitally imaged with chemiluminescence detection ( Amersham Imager 600 , GE Healthcare Life Sciences , Little Chalfont , UK ) . As a control to verify the contribution of LOX/LOXL to the amine oxidase activity in the tissue , a serial section was preincubated with the irreversible LOX/LOXL inhibitor , BAPN ( 300 μM ) for 30 min prior to substrate addition . After imaging , semi-quantitative analysis was performed , with the mean grey scale value of each tissue area following background subtraction calculated using Fiji ( version 1 . 49 p ) , with each sample normalised to the wet weight of 250 μm of adjacent cryosections . Bovine LOX and recombinant LOXL enzymes were tested in an Amplex Red assay using putrescine as substrate over a range of concentrations of inhibitors as previously described ( Schilter et al . , 2015; Chang et al . , 2017 ) . Putrescine was used at 10 mM for LOX and LOXL1 , 5 mM for LOXL2 and 2 mM for LOXL3 and LOXL4 reflecting the different Km values for each enzyme . Bovine LOX was extracted from calf aorta ( Schilter et al . , 2015 ) , recombinant LOXL2 and LOXL3 proteins were purchased from R and D systems , recombinant LOXL1 was stably expressed as a his-tagged protein in NIH-3T3 cells and purified from conditioned media using Nickel-Sepharose affinity purification; recombinant LOXL4 was partially purified from conditioned medium from HEK-293 stably transfected with human LOXL4 using size exclusion enrichment with Amicon 50 kDa centrifugal filter units . The LOXL4 cell line was provided by Dr Fernando Rodríguez Pascual , Universidad Autónoma de Madrid , Spain . All proteins were initially characterised by western blotting and enzymatic activity , in order to confirm identity and to yield comparable Vmax across all assays . Peripheral lung fibroblasts were obtained as outgrowths from surgical lung biopsy tissue ( Davies et al . , 2012 ) of patients ( n = 3 donors ) who were subsequently confirmed with a diagnosis of IPF . All primary cultures were tested and free of mycoplasma contamination . The fibroblasts were seeded in Transwell inserts in DMEM containing 10% FBS . After 24 hr , the media was replaced with DMEM/F12 containing 5% FBS , 10 μg/ml L-ascorbic acid-2-phosphate , 10 ng/ml EGF , and 0 . 5 μg/ml hydrocortisone with or without inhibitors , as indicated; each experiment included a vehicle control ( 0 . 1% DMSO ) . TGF-β1 ( 3 ng/mL ) was added to the cultures , and the medium replenished three times per week . After 6 weeks the spheroids were either harvested into RNAlater ( Sigma ) for RNA extraction , snap frozen for parallel-plate compression testing , analysis of cross-linking , and histochemical staining , or fixed using 4% paraformaldehyde for histochemistry or 3% glutaraldehyde in 0 . 1 M cacodylate buffer at pH 7 . 4 for electron microscopy . 4–7 μm sections were analysed using PicroSirius Red ( Abcam , Cambridge , UK ) or Masson’s Trichrome ( Sigma , Poole , UK ) stain according to the manufacturers’ instructions . Images were acquired using a Dot-Slide scanning system ( Olympus , Southend-on-Sea , UK ) with PicroSirius Red staining visualised under polarised light . PicroSirius Red images were converted to 16-bit and pseudo-coloured through application of the Yellow Lookup Table within Fiji ( version 1·49 c ) . Samples were thawed and reduced with KBH4 before acid hydrolysis in 6M HCl at 102°C for 18 hr . Cooled hydrolysed samples were evaporated to dryness under vacuum and then resuspended in 200 μL HPLC-grade H2O . Total protein was quantified in the hydrolysed samples using a genipin-based amino acid assay ( QuickZyme Biosciences , Leiden , The Netherlands ) , using standard hydrolysed bovine serum albumin as standard . Total collagen content was estimated using either ultra-high performance liquid chromatography-electrospray ionisation tandem mass spectrometry ( UHPLC-ESI-MS/MS ) or by colorimetric assay of hydroxyproline ( Hyp ) based on the reaction of oxidized hydroxyproline with 4- ( Dimethylamino ) benzaldehyde , as per manufacturer’s instruction ( Sigma-Aldrich , Poole , UK ) . The molar content of collagen was estimated from hydroxyproline using a conversion factor of 300 hydroxyprolines per triple helix , and mass of collagen was estimated using a molecular weight of 300 kDa per triple helix ( Miller et al . , 1990 ) . Collagen cross-links were determined using either UHPLC-ESI-MS/MS or ELISA as indicated in the figure legends . Total mature pyridinium cross-links ( PYD +DPD ) were determined using enzyme-linked immunosorbent assay ( ELISA; Quidel Corporation , San Diego , USA ) according to manufacturer’s instructions . For UHPLC-ESI-MS/MS analysis , cross-links were first extracted from the hydrolysate using an automated solid phase extraction system ( Gilson GX-271 ASPECA system ) employing reversed-phase C18-Aq columns ( GracePure , ThermoFisher Scientific , Australia ) followed by SCX strong cation exchange columns . After extraction and drying , the cross-links were converted into heptafluorobutyric acid ( HFBA ) salts for analysis by UHPLC-ESI-MS/MS on a Thermo Dionex UHPLC and TSQ Endura triple quad mass spectrometer . UHPLC separation of cross-links was achieved with an Agilent Rapid Resolution High Definition ( RRHD ) SB-C18 column . UHPLC was performed using a 12-min gradient flow of the mobile phase A ( 10 mM ammonium formate , 0 . 1% formic acid , 0 . 1% HFBA in H2O ) from 96 . 2% to 0% and mobile phase B ( 10 mM ammonium formate , 0 . 1% formic acid , 0 . 1% HFBA in 80% MeOH ) from 3 . 8% to 100% at a flow rate of 0 . 3 mL/min and with a column temperature of 40°C . Positive ESI-MS/MS with Selected Reaction Monitoring ( SRM ) mode was performed using the following parameters: Spray voltage 4000 V; sheath gas 35 ( Arb ) ; aux gas 20 ( Arb ) ; sweep gas 0 ( Arb ) ; ion transfer tube temperature 350°C; vaporiser temperature 300°C . Standards used for quantitation of collagen cross-links or total collagen in hydrolysates by UHPLC-ESI-MS/MS were: DHLNL ( Thermo Fisher Scientific , Scoresby , Australia ) , HLNL ( Toronto Research Chemicals , Toronto , Canada ) , PYD and DPD ( Quidel Corporation , San Diego , USA ) and hydroxyproline ( Sigma-Aldrich , Poole , UK ) . Quantitation of the collagen cross-links and total collagen was achieved by comparing to a standard curve . Sample values were interpolated using GraphPad Prism seven software . To determine the stiffness characteristics ( Young’s modulus , E ) of the 3D in vitro model of fibrosis , cultures were subjected to parallel plate compression testing using a CellScale MicroSquisher fitted with a round tungsten cantilever ( thickness 406 . 4 nm ) and accompanying SquisherJoy V5 . 23 software ( CellScale , Ontario , Canada ) . The fluid bath test chamber was filled with sterile PBS , and the stage and optics calibrated following the manufacturer’s instructions . Example test sequences are shown in Figure 6—figure supplement 1 and Video 1 . Samples were subjected to five testing cycles under hydrated conditions , each achieving 25% engineering strain ( deformation ) over a 15 s compression phase , followed by a 2 s hold , recovery over 15 s , and a 2 s resting period . The first four cycles allowed preconditioning of the tissue , and the resulting force vs time curves could be seen to have stabilised by the fifth compression cycle ( Figure 6—figure supplement 1a ) . Analysis of stress vs strain relationships was carried out using the compression phase of the fifth cycle from where sample stiffness can be inferred . Here , we restrict ourselves to the estimation of the Young’s modulus of the biological material in hydrated state . Force and displacement data were transformed to engineering stress versus engineering strain plots using the horizontal cross-sectional diameter of the sample immediately before the start of each test . Young’s modulus ( E ) , a measure of stiffness , was calculated using a modified Hertzian half-space contact mechanics model for elastic spheres as previously described ( Kim et al . , 2010 ) , averaging between values of E calculated for both the horizontal and vertical radii of the quasi-spherical cultures . This calculation requires the value of the Poisson’s ratio ν . Given the narrow range in which this value lies , the relative insensitivity of the calculations to this material parameter , and significant difficulty in their direct measurement for small biological samples , here we make use of estimated values taken as 0 . 5 , which is realistic for a range of biological matter . All stress-strain curves show an initial 'toe' region , where the stress increases slowly , followed by a linear region , approximately between 10% and 20% strain . The stiffness of the tissue was obtained by averaging the computed values of E within this linear range . Samples were post-fixed sequentially in osmium/ferrocyanide fixative , thiocarbohydrazide solution , osmium tetroxide , uranyl acetate and Walton’s lead aspartate solution before dehydration in graded ethanol and acetonitrile . Samples were embedded in Spurr resin and 100 nm ultra-thin sections visualised using an FEI Tecnai 12 transmission electron microscope ( FEI Company , Hillsboro , OR , USA ) . For measurement of fibril diameters images were acquired at 87 , 000x magnification . Measurements of fibril diameter from two independent experiments were made by two operators in parallel blinded to the treatment conditions using ImageJ ( Schneider et al . , 2012 ) ( Version 1 . 50 g ) . The shortest axis of each fibril was measured with a total of 300 measurements per sample . To assess collagen solubility samples from the 3D in vitro model of lung fibrosis were subjected to sequential collagen extraction under increasingly stringent conditions as described previously ( Liu et al . , 2016 ) . Briefly , samples were subjected to sequential overnight incubations in 1 ) neutral salt ( Tris buffered saline containing protease inhibitors ) ; 2 ) 0 . 5M acetic acid with protease inhibitors; and 3 ) 0 . 5M acetic acid containing 700 units/mL pepsin ( Sigma , Poole , UK ) . Each fraction , and the remaining solid ( insoluble highly cross-linked ) fraction , was hydrolysed in 6M HCl at 102°C for 18 hr and the collagen content of each fraction determined by colorimetric hydroxyproline assay . Studies in rats were performed at McMaster University ( Hamilton , Ontario , Canada ) according to guidelines from the Canadian Council on Animal Care and approved by the Animal Research Ethics Board of McMaster University under protocol # 16-04-14 . Sample sizes were based on previous studies in the laboratory where the studies were performed and were chosen to balance the ability to measure significant differences while reducing the number of animals used . Female Sprague Dawley rats ( Charles River Laboratories , Montreal , Canada ) were randomly assigned to receive 2 . 5 × 108 pfu of either a replication-deficient adenoviral vector carrying an active TGF-β1 ( AdTGF-β1 ) expression construct to induce progressive pulmonary fibrosis ( Sime et al . , 1997 ) or vector only control ( AdDL ) on Day 0 . Treatment groups received LOXL2/LOXL3 inhibitor ( 15 or 30 mg/kg/day dosed daily via the oral route ) from day 1 to day 28; control groups received vehicle ( PBS ) . On day 28 , lung function was assessed using forced oscillation manoeuvres with a FlexiVent system ( SCIREQ Scientific Respiratory Equipment , Montreal , Canada ) to measure pressure-volume loops , resistance ( R ) , compliance ( C ) , and elastance ( E ) of the respiratory system . Animals were then euthanised and the left lung lobe inflated under constant pressure and fixed in 10% formalin for histological examination of tissue sections by staining with picrosirius red and viewing under polarised light . Fibrosis was assessed by six independent , trained individuals who were blinded to the treatment conditions using a modified Ashcroft score ( Bonniaud et al . , 2005 ) . All right lung lobes were combined , flash frozen and ground into a fine powder under liquid nitrogen and stored for quantitation of collagen and its cross links , and for RNA analysis . Statistical analyses were performed in GraphPad Prism v7 . 02 ( GraphPad Software Inc . , San Diego , CA ) unless otherwise indicated . No data were excluded from the studies and for all experiments , all attempts at replication were successful . For each experiment , sample size reflects the number of independent biological replicates and is provided in the figure legend . Normality of distribution was assessed using the D’Agostino-Pearson normality test . Statistical analyses of single comparisons of two groups utilised Student's t-test or Mann-Whitney U-test for parametric and non-parametric data respectively . Where appropriate , individual t-test results were corrected for multiple comparisons using the Holm-Sidak method . For multiple comparisons , one-way analysis of variance ( ANOVA ) with Dunnett’s multiple comparison test or Kruskal-Wallis analysis with Dunn’s multiple comparison test were used for parametric and non-parametric data , respectively . In the in vitro studies , repeated measures ( RM ) ANOVA was used in order to treat data from each donor as a matched set . Sphericity of the repeated measures was assumed since the experiments were designed as randomised blocks . Multivariate correlations were performed using JMP software v13 . 1 . 0 ( SAS , Marlow , UK ) applying the restricted maximum likelihood ( REML ) method to handle missing data . Results were considered significant if p<0 . 05 , where p<0 . 05 , p<0 . 01 , p<0 . 001 , **p<0 . 0001 .	Idiopathic pulmonary fibrosis ( IPF ) is a devastating disease of the lung , which scars the tissue and gradually destroys the organ , ultimately leading to death . It is still unclear what exactly causes this scarring , but it is thought that increasing amounts of proteins in the space surrounding the cells of the lungs , the extracellular matrix , could play a role . These proteins , including collagen , normally form a ‘scaffold’ to stabilize cells , but if they accumulate uncontrollably , they can render tissues rigid . It has been assumed that these changes are a consequence of the disease . However , recent evidence suggests that the increased stiffness itself could stimulate cells to produce even more extracellular matrix , driving the progression of the disease . A better understanding of what exactly causes the tissue to become gradually stiffer may identify new ways to block the progression of IPF . Now , Jones et al . compared measurements of the tissue stiffness and the collagen structure taken from samples of patients with IPF . The results showed that the collagen fibres were faulty and had an abnormal shape . This suggests that these problems , rather than an increased amount of collagen , alter the flexibility of the lung tissue . Jones et al . also found that a specific family of proteins , which helps to connect the collagen fibres , was increased in the tissue of patients with IPF . When these proteins were blocked with a newly developed drug , the collagen structure returned to normal and the stiffness of the tissue decreased . As a consequence , the lung capacity improved . This suggests that treatment approaches that help to maintain a normal collagen structure , may in future prevent the stiffening of the lung tissue and so limit feed-forward mechanisms that drive progressive IPF . Moreover , it indicates that measurements of the structure of collagen rather than the its total concentration could serve as a more suitable indicator for the disease .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2018	Nanoscale dysregulation of collagen structure-function disrupts mechano-homeostasis and mediates pulmonary fibrosis
Across the animal kingdom , environmental light cues are widely involved in regulating gamete release , but the molecular and cellular bases of the photoresponsive mechanisms are poorly understood . In hydrozoan jellyfish , spawning is triggered by dark-light or light-dark transitions acting on the gonad , and is mediated by oocyte maturation-inducing neuropeptide hormones ( MIHs ) released from the ectoderm . We determined in Clytia hemisphaerica that blue-cyan light triggers spawning in isolated gonads . A candidate opsin ( Opsin9 ) was found co-expressed with MIH within specialised ectodermal cells . Opsin9 knockout jellyfish generated by CRISPR/Cas9 failed to undergo oocyte maturation and spawning , a phenotype reversible by synthetic MIH . Gamete maturation and release in Clytia is thus regulated by gonadal photosensory-neurosecretory cells that secrete MIH in response to light via Opsin9 . Similar cells in ancestral eumetazoans may have allowed tissue-level photo-regulation of diverse behaviours , a feature elaborated in cnidarians in parallel with expansion of the opsin gene family . Integration of environmental light information contributes to tight coordination of gamete release in a wide range of animal species . The nature of the photodetection systems involved and their evolutionary origins are poorly undestood . Proposed involvement has mainly focussed on light-entrainment of endogenous clocks , which align many aspects of physiology and behaviour , including reproductive ones , to seasonal , monthly or daily cycles ( Cermakian and Sassone-Corsi , 2002; Tessmar-Raible et al . , 2011 ) . Clock entrainment can involve both of the main families of photo-sensitive proteins ( photopigments ) used for animal non-visual photoreception: the evolutionarily ancient Cryptochrome/Photolyase family , which originated in unicellular eukaryotes ( Oliveri et al . , 2014 ) , and the diverse , animal-specific opsin family of light-sensitive G Protein-Coupled Receptors ( GPCRs ) known best for involvement in visual photodetection ( Cronin and Porter , 2014; Gehring , 2014 ) . Light cues can also provide more immediate triggers for gamete release , which integrate with seasonal and/or circadian regulation ( Kaniewska et al . , 2015; Lambert and Brandt , 1967 ) , but the involvement of specific photopigments in such regulation has not previously been addressed . Members of Hydrozoa , a subgroup of Cnidaria which can have medusae or polyps as the sexual form , commonly display light-regulated sexual reproduction ( Leclère et al . , 2016; Siebert and Juliano , 2017 ) . They have simple gonads in which the germ cells are sandwiched between ectoderm and endoderm , and unlike many other animals they lack additional layers of somatic follicle cells surrounding oocytes in the female ( Deguchi et al . , 2011 ) . Light-dark and/or dark-light transitions trigger the release of mature gametes into the seawater by rupture of the gonad ectoderm ( Freeman and Ridgway , 1988; Miller , 1979; Roosen-Runge , 1962 ) . Gamete release is coordinated with diel migration behaviours in jellyfish to ensure gamete proximity for fertilisation ( Martin , 2002; Mills , 1983 ) . We know that the photodetection systems that mediate hydrozoan spawning operate locally within the gonads , since isolated gonads will spawn upon dark-light or light-dark transitions ( Freeman , 1987; Ikegami et al . , 1978 ) . Opsin gene families have been identified in cnidarians and provide good candidates for a role in this process , with expression of certain opsin genes reported in the gonads both of the hydrozoan jellyfish Cladonema radiatum ( Suga et al . , 2008 ) and of the cubozoan jellyfish Tripedalia cystophora ( Liegertová et al . , 2015 ) . In female hydrozoans , the regulation of spawning is tightly coupled to oocyte maturation , the process by which resting ovarian oocytes resume meiosis to be transformed into fertilisable eggs ( Yamashita et al . , 2000 ) . Oocyte maturation , followed by spawning , is induced by ‘Maturation Inducing Hormones’ ( MIHs ) , released from gonad somatic tissues upon reception of the appropriate light cue ( Freeman , 1987; Ikegami et al . , 1978 ) . The molecular identity of MIH of several hydrozoan species has been uncovered recently as PRPamide family tetrapeptides , which are released from scattered cells with neural-type morphology present in the gonad ectoderm following dark-light transitions ( Takeda et al . , 2018 ) . Here we report the identification of a cnidarian opsin gene expressed in the gonads of the hydrozoan jellyfish Clytia hemisphaerica and show by gene knockout that it is essential for spawning via MIH release in response to light . We further show that this opsin is expressed in the same gonad ectoderm cells that secrete MIH , which thus constitute a specialised cnidarian cell type responsible for mediating light-induced spawning . We discuss how these cells show many features in common with photosensitive deep brain neuroendocrine cells , such as those described in fish and annelids ( Tessmar-Raible et al . , 2007 ) . These cell types may have shared an origin in the common ancestor of the cnidarians and of the bilaterians , an animal clade comprising all the protostomes and deuterostomes . MIH release in Clytia gonads is triggered by a light cue after a minimum dark period of 1–2 hr , with mature eggs being released two hours later ( Amiel et al . , 2010 ) . In order to characterise the light response of Clytia gonads ( Figure 1A ) , we first assessed the spectral sensitivity of spawning . Groups of 3–6 manually dissected Clytia female gonads were cultured overnight in the dark and then stimulated from above with 10 s pulses of monochromatic light across the 340 to 660 nm spectrum . Stimulated ovaries were returned to darkness for one hour before scoring oocyte maturation . Oocyte maturation is accompanied by breakdown of the membrane of the oocyte nucleus ( ‘GV’ for germinal vesicle ) in fully-grown oocytes , followed by polar body emmission and spawning . We found that wavelengths between 430 and 520 nm provoked spawning in at least 50% of gonads , with 470–490 nm wavelengths inducing spawning of ≥75% of gonads ( Figure 1B ) . Oocyte maturation and subsequent spawning of Clytia female gonads is thus preferentially triggered by blue-cyan light ( Figure 1B ) , the wavelength range which penetrates seawater the deepest ( Gehring and Rosbash , 2003; Gühmann et al . , 2015 ) . We identified a total of ten Clytia opsin sequences in transcriptome and genome assemblies from Clytia by reciprocal BLAST searches using known hydrozoan opsin sequences ( Suga et al . , 2008 ) and by the presence of the diagnostic lysine in the seventh transmembrane domain to which the retinal chromophore binds ( Terakita et al . , 2012 ) ( Figure 2—figure supplement 1 ) . We selected candidate opsins for a role in mediating MIH release by evaluating expression in the isolated gonad ectoderm , a tissue which has been shown using Cladonema jellyfish to have an autonomous capacity to respond to light ( Takeda et al . , 2018 ) . Illumina HiSeq reads from Clytia gonad ectoderm , gonad endoderm , growing and fully grown oocyte transcriptome sequencing were mapped against each opsin sequence ( Figure 2A ) . In the gonad ectoderm sample , two opsin mRNAs ( Opsin4 and Opsin7 ) were detected at low levels and one was very highly expressed ( Opsin9 ) , while in the three other gonad samples analysed opsin expression was virtually undetectable ( Figure 2A ) . The high expression of Opsin9 gene in the gonad ectoderm made it a strong candidate for involvement in light-induced spawning . In situ hybridisation of Clytia ovaries revealed Opsin9 expression in a scattered population of gonad-ectoderm cells ( Figure 2B ) . No Opsin9 expression was detected anywhere else in the medusa . We were unable to detect by in situ hybridisation in Clytia gonads either of the two lowly-expressed gonad-ectoderm opsin genes , Opsin4 and Opsin7 ( Figure 2D–E ) . The distribution of Opsin9-expressing cells was highly reminiscent of the expression pattern for PP1 and PP4 ( Figure 2C ) , the two MIH neuropeptide precursor genes co-expressed in a common gonad ectoderm cell population ( Takeda et al . , 2018 ) . Double fluorescent in situ hybridisation using probes for Opsin9 and for PP4 revealed that these genes were expressed in the same cells ( Figure 2F ) ; 99% of Opsin9 mRNA positive cells were also positive for PP4 mRNA , and over 87% of PP4 mRNA-positive cells were also positive for Opsin9 mRNA . This finding raised the possibility that spawning in Clytia might be directly controlled by light detection through an opsin photopigment in the MIH-secreting cells of the gonad ectoderm . To test the function of Opsin9 in light-induced oocyte maturation and spawning , we generated a Clytia Opsin9 knockout ( KO ) polyp colony using a CRISPR/Cas9 approach , which produces very extensive bi-allelic KO of the targeted gene in medusae of the F0 generation ( Momose and Concordet , 2016; see Materials and methods for details ) . This approach is favoured by the Clytia life cycle , in which larvae developed from each CRISPR-injected egg metamorphose into a vegetatively-expanding polyp colony , from which sexual medusae bud clonally ( Houliston et al . , 2010; Leclère et al . , 2016 ) . CRISPR guide RNAs were designed to target a site in the first exon of Opsin9 encoding the third transmembrane domain , and were verified not to match any other sites in available Clytia genome sequences . One of the polyp colonies generated carried a predominant 5 bp deletion , corresponding to a frame-shift and premature STOP codon ( Figure 3A ) . Polyp colony development in this individual , designated Opsin9n1_4 ( see Materials and methods ) , showed no abnormal features . For phenotypic analysis we collected Opsin9n1_4 jellyfish , which were all females , and grew them by twice-daily feeding for two weeks to sexual maturity . Although these Opsin9n1_4 mature medusae initially appeared normal , they did not spawn after the daily dark-light transition , and after a few days displayed grossly inflated ovaries due to an accumulation of unreleased large immature oocytes with intact GVs ( Figure 3B–C ) . In three independent experiments to test the light response of isolated gonads , over 85% of Opsin9n1_4 gonads failed to undergo oocyte maturation and spawning upon light stimulation ( Figure 3D ) . Opsin9n1_4 gonads did , however , release oocytes in response to synthetic MIH peptide ( Figure 3E ) . These oocytes resumed meiosis normally and could be fertilised , although the fertilisation rate was lower than for oocytes spawned in parallel from wild type gonads , possibly a consequence of their prolonged retention in the gonad . Genotyping of individual gonads showed that the rare gonads from Opsin9n1_4 medusae that spawned after light had greater mosaicism of mutations , with a higher ratio of residual non-frameshift mutations and also a significant amount of wild type cells , whereas gonads that failed to spawn carried mainly the predominant 5 bp deletion , a second 21 bp deletion and no detectable wild type cells ( Figure 3—figure supplement 1 ) . The failure of spawning observed in Clytia Opsin9n1_4 mutant jellyfish gonads , together with the absence of detectable Opsin9 expression in non-gonad tissues of the medusa and the autonomous response of isolated wild-type gonads to light , strongly indicates that the gonad photopigment Opsin9 plays an essential role in light-induced oocyte maturation . Since Opsin9 and MIH are co-expressed in the same cells , we reasoned that Opsin9 function was probably required for MIH secretion . This was confirmed by immunofluorescence of Opsin9 mutant gonads ( Figure 4 ) . Quantitative immunofluorescence analyses based on anti-MIH staining were performed in both wild type and Opsin9 mutant gonads , comparing light-adapted and dark-adapted specimens fixed 45 min after white light exposure . MIH-Opsin9 cells were identified by their characteristic organisation of stable microtubules ( Takeda et al . , 2018; see Figure 5E-F ) . Whereas wild type gonads exhibited a significant decrease of MIH fluorescence values in MIH-Opsin9 cells upon light stimulation ( Figure 4A–C ) , Opsin9 mutant gonads maintained similar levels of MIH fluorescence in both conditions ( Figure 4D–F ) . Moreover , average MIH fluorescence levels per cell were significantly higher ( U test at p<0 . 01 ) in Opsin9 mutant than in wild type ovaries suggesting a progressive accumulation of MIH in Opsin9 mutant gonads . These immunofluorescence results indicated that Opsin9 mutant medusae failed to undergo oocyte maturation and spawning because of reduced MIH secretion in response to light . The ability of synthetic MIH peptides to reverse the phenotype of Opsin9 mutant gonads supports this conclusion: Equivalent MIH concentrations to those effective for light-adapted isolated gonads ( Takeda et al . , 2018 ) reliably induced oocyte maturation and spawning in Opsin9 mutant isolated gonads ( Figure 3E ) . The primary defect in these Opsin9 mutants is thus in MIH secretion from the ectoderm following light stimulation , providing strong evidence that the CRISPR-Cas9 approach had specifically targeted the Opsin9 gene . Taken together , these results demonstrate that Opsin9 is required for the MIH-producing cells in the Clytia gonad to release the MIH neuropeptides following dark-light transitions , and thus has an essential role in light-dependent reproductive control . The functional studies described above indicate that Clytia gonad cells that co-express MIH and Opsin9 have photosensory functions , as well as neurosecretory characteristics ( Takeda et al . , 2018 ) . To investigate the morphology of these key cells in more detail we performed immunofluorescence to visualise cortical actin and microtubules in cells producing MIH ( Figure 5 ) . The Clytia gonad ectoderm consists of a monolayer of ciliated epitheliomuscular cells ( Leclère and Röttinger , 2017 ) , tightly joined by apical junctions and bearing basal extensions containing muscle fibres . These basal myofibres form a layer over the oocytes and stain strongly with phalloidin fluorescent probes ( Figure 5A–D ) . The cell bodies of the MIH-secreting cells were found to be positioned within this epithelial layer , scattered between the much more abundant epitheliomuscular cells ( Takeda et al . , 2018; Figure 5A , B , D ) . A variable number of neural-type processes containing prominent microtubule bundles project basally from these cells ( Figure 5E , F ) , and intermingle with the muscle fibres of surrounding epitheliomuscular cells ( Figure 5A–C ) . The cell nuclei were generally located more basally than those of the surrounding epitheliomuscular cells ( Figure 5A , C ) . In regions overlying large oocytes , the gonad ectoderm , including most of the MIH-rich basal processes , was elevated by a characteristic space on the basal side ( Figure 5C ) , probably reflecting the accumulation of extracellular jelly components around oocytes during late stages of growth . This configuration implies that most of the peptide hormone ( MIH ) is secreted at distance from its site of action at the oocyte surface , consistent with a neuroendocrine function . MIH-positive cells did not have extensive apical domains exposed on the external face of the ectodermal epithelium , although MIH staining was in some cases detected very close to this surface at the interstices between surrounding myoepithelial cells ( Figure 5C’ and D” ) . Cilia , which decorated most epitheliomuscular cells , could not be unambiguously associated with the MIH-secreting cells ( Figure 5F , G ) , but basal bodies could be detected in the apical side of these cells with a gamma-tubulin antibody ( Figure 5H–H’ ) . These various immunofluorescence analyses indicate that the MIH-secreting cells have morphological features characteristic of multipolar neurosensory cells in cnidarians ( Saripalli and Westfall , 1996 ) . Other approaches such as electron microscopy will be required to resolve important questions concerning these cells , notably whether closely apposed processes from adjacent cells ( Figure 5E , F ) connect via synapses to form a functional network , and whether they are fully integrated into the epithelial ectoderm through stable junctions and distinct apical surfaces . Never-the-less , based on opsin and neuropeptide expression , general morphology , and biological function , we can confidently propose that this specialised cell type has a dual sensory-neurosecretory nature . With increasing availability of opsin gene sequences covering a widening range of animal taxa , the traditional division into ‘c-opsins’ , expressed in ciliary photoreceptor cells such as vertebrate rods and cones , and ‘r-opsins , ’ as expressed in arthropod rhabdomeric photoreceptors , is clearly no longer adequate as a representation of opsin phylogenetic relationships ( Cronin and Porter , 2014; Feuda et al . , 2012 , 2014 ) . Molecular phylogeny analyses are progressively revealing a complicated evolutionary history for the opsins , involving many gene duplications and losses in separate animal lineages , starting from a set of at least nine genes in the common ancestor of Bilateria , with three or four sequences already present before the cnidarian-bilaterian split ( Ramirez et al . , 2016; Vöcking et al . , 2017 ) . Available cnidarian opsin sequences form three groups , termed ‘anthozoan opsin I’ , ‘anthozoan opsin II’ and ‘cnidops’ . The first two contain only anthozoan ( coral and sea anemone ) sequences . The cnidops group includes some anthozoan opsins and all published opsin genes from the medusozoan clade ( jellyfish and hydra ) , and is probably the sister group to the xenopsin group , which includes opsins from a few disparate protostome taxa ( Ramirez et al . , 2016; Vöcking et al . , 2017 ) . We performed molecular phylogeny for the Clytia opsin genes by adding the ten sequences to two recent datasets of animal opsin amino acid sequences ( Feuda et al . , 2014; Vöcking et al . , 2017 ) and using the LG+Γ amino acid substitution model . All 10 Clytia Opsins fell as expected within the cnidops group , with Opsin9 and the closely-related Opsin10 having more divergent sequences than the other Clytia opsins ( Figure 6A , Figure 6—figure supplement 1 ) . Some trials , for instance using the GTR+Γ model ( conceived for large datasets ) on the Vöcking et al . ( 2017 ) based alignment , produced an alternative tree topology with Clytia Opsin9/10 positioned as sister-group to the anthozoa opsin II group ( Figure 6—figure supplement 2 ) , but we could show that this is a ‘long branch attraction’ artifact . Thus , this topology could be reversed by breaking the long Clytia Opsin9/10 branch using a single Opsin9/10 related sequence from unpublished transcriptome data of the hydrozoan medusa Melicertum octocostatum ( Figure 6—figure supplement 2 ) . More detailed analyses including only medusozoan ( hydrozoan and cubozoan ) opsin sequences available in GenBank ( Figure 6B ) confirmed that Clytia Opsin9 and Opsin10 were indeed amongst the most divergent Cnidops sequences . The draft Clytia genome sequence ( Leclère et al . , 2016 ) further revealed that Clytia Opsin9 and Opsin10 genes contain a distinct intron , unlike all other described medusozoan Cnidops genes , in a distinct position to those found in available Anthozoan Opsin II , Anthozoan Cnidops and Xenopsin genes ( Vöcking et al . , 2017; Liegertová et al . , 2015 ) . Despite this picture of rapid evolutionary divergence of Clytia Opsin9/Opsin10 within the diversifying cnidops family , the Clytia Opsin9 amino acid sequence exhibits all the hallmarks of a functional photopigment ( Figure 2—figure supplement 1 ) . It has conserved amino acids at positions required for critical disulphide bond formation and for Schiff base linkage to the retinal chromophore ( Fischer et al . , 2013; Gehring , 2014; Schnitzler et al . , 2012 ) , including acidic residues at the both potential ‘counterion’ positions , only one of which is largely conserved in cnidarian opsin sequences ( Liegertová et al . , 2015 ) . The Glu/Asp-Arg-Tyr/Phe motif adjacent to the third transmembrane domain , involved in cytoplasmic signal transduction via G proteins ( Fischer et al . , 2013; Kojima et al . , 2000; Schnitzler et al . , 2012 ) is also present in Opsin9 . The exact relationship of Clytia Opsin9/10 to other medusozoan opsins cannot be resolved with the available data . Their phylogenetic position was found to be unstable when analysed using different evolutionary models and datasets ( Figure 6 , Figure 6—figure supplement 1 , Figure 6—figure supplement 2 ) . Extensive sampling from more medusozoan species will be needed to fully resolve their phylogenetic history . We could show unambiguously , however , from our phylogenetic reconstructions ( Figure 6B ) and AU ( ‘approximately unbiased’ ) phylogenetic tests ( Shimodaira , 2002 ) , that Clytia Opsin9 is not orthologous to the opsin genes previously identified as expressed in the gonad of the hydromedusa Cladonema ( AU test: p<1e−7 ) or the cubomedusa Tripedalia ( AU test: p<1e−20 ) . Thus diversification of the cnidops gene family in Medusozoa was accompanied by expression of different opsin paralogs in the gonad in different species . To summarise the results of this study , we have provided the first demonstration , using CRISPR/Cas9–mediated gene knockout , of an essential role for an opsin gene in non-visual photodetection in a cnidarian . Clytia Opsin9 is required for a direct light-response mechanism that acts locally in the gonad to trigger gamete maturation and release ( Figure 7 ) . It occurs in specialised sensory-secretory cells of the Clytia gonad ectoderm to trigger MIH secretion from these cells upon light reception . This peptidic MIH in turn induces oocyte maturation in females , and also release of motile sperm in males ( Takeda et al . , 2018 ) , efficiently synchronising spawning to maximise reproductive success . Comparison of neuropeptide involvement in hydrozoan , starfish , fish and frog reproduction suggested an evolutionary scenario in which gamete maturation and release in ancient metazoans was triggered by gonad neurosecretory cells ( Takeda et al . , 2018 ) . According to this scenario , this cell type would have been largely conserved during cnidarian evolution and function similarly today in hydrozoans , whereas during bilaterian evolution further levels of regulation were inserted between the secreting neurons and the responding gametes . Thus in vertebrates , peptide hormone secretion would have been delocalised to the hypothalamus , and the primary responding cells to the pituitary . Our study on opsins has further revealed a parallel between the MIH-secreting cells of the gonad ectoderm in Clytia and deep brain photoreceptor cells in vertebrates , as well as equivalent cells in various protostome species that regulate physiological responses through neurohormone release in response to ambient light ( Fernandes et al . , 2013; Fischer et al . , 2013; Halford et al . , 2009; Tessmar-Raible et al . , 2007 ) . Like the Clytia MIH-producing cells , TSH-producing cells in birds and fish pituitary ( Nakane et al . , 2013; Vigh et al . , 2002 ) , and vasopressin/oxytocin-expressing cells in fish hypothalamus and annelid forebrain ( Tessmar-Raible et al . , 2007 ) , show opsin-related photodetection , secrete neuropeptide hormones and are implicated in hormonal control of reproduction ( Juntti and Fernald , 2016 ) . We can thus propose that the putative neurosecretory cell type associated with the germ line that regulated gamete release in ancestral metazoans was also photosensitive . The active photopigments in these cells could have been cryptochromes , the most ancient metazoan photopigment family inherited from unicellular ancestors ( Cronin and Porter , 2014; Gehring , 2014 ) or animal opsins , whose gene family is absent in sponges but expanded extensively in both cnidarians and bilaterians from a common ancestral GPCR gene ( Feuda et al . , 2014; 2012; Gehring , 2014; Liegertová et al . , 2015 ) . Under this evolutionary scenario , the essential Opsin9 protein in Clytia , and similarly the gonad-expressed Cladonema ( Suga et al . , 2008 ) and Tripedalia ( Liegertová et al . , 2015 ) opsins , would have replaced the ancestral photopigment in the MIH-secreting cells during cnidarian evolution to provide optimised spawning responses to particular light wavelength and intensity . A parallel can be drawn with the evolution of vision in eumetazoans , in which deployment of animal opsins is thought to have allowed more rapid and precise photoresponses than the ancestral cryptochrome system ( Gehring , 2014 ) . While the scenario above conforms to the evolutionary trend for specialised cell types with distributed functions in bilaterians to evolve from ancestral multifunctional cell types ( Arendt , 2008; Arendt et al . , 2016 ) , an alternative hypothesis is that the Clytia gonad photosensitive–neurosecretory cells and deep brain photoreceptors in bilaterian species arose convergently during evolution . Specifically , a population of MIH-secreting cells in hydrozoan gonads , initially regulated by other environmental and/or neural inputs , may have secondarily acquired opsin expression to become directly photosensitive . More widely , cnidarians may have accumulated a variety of multifunctional cell types as specific populations of neural cells , muscle cells or nematocytes acquired photopigments during evolution ( Porter et al . , 2012 ) . Cnidarians are characterised by the lack of a centralised nervous system , and correspondingly show localised regulation of many physiological processes and behaviours at the organ , tissue or even the cellular levels . In addition to gamete release , other local light-mediated responses include light-sensitive discharge of cnidocyte-associated sensory-motor neurons expressing hydra HmOps2 ( Plachetzki et al . , 2012 ) , diel cycles of swimming activity ( Martin , 2002; Mills , 1983 ) controlled in Polyorchis by photoresponsive cells of the inner nerve ring ( Anderson and Mackie , 1977 ) , and tentacle retraction in corals ( Gorbunov and Falkowski , 2002 ) . Clytia Opsin9 , the first cnidarian opsin to have a demonstrated function , is expressed in the scattered , MIH-secreting cells , which act autonomously in the gonad without any need for input from other parts of the medusa . It remains to be determined if they secrete the MIH peptides from localised cellular sites and are connected by synapses . The nature of these cells may thus be neurosecretory or neuroendocrine , rather than neuronal ( Hartenstein , 2006 ) , fitting with the idea that neuropeptide signalling between epithelial cells may have predated nervous system evolution ( Bosch et al . , 2017 ) . Whether Hydrozoa gonad ectoderm MIH-secreting cells and Bilateria deep brain photosensitive cells derived from an ancestral multifunctional photosensory-secretory cell type , or from non-photoresponsive neural cells remains an open question . Our findings support a scenario in which expansion of the opsin gene family within the hydrozoan clade was accompanied by the local deployment of individual opsins with specific spectral characteristics adjusted to regulate a variety of physiological behaviours in response to light , epitomised by Clytia Opsin9 and its regulation of spawning . Sexually mature medusae from laboratory maintained Clytia hemisphaerica ( ‘Z colonies’ ) were fed regularly with Artemia nauplii and cultured under light-dark cycles to allow daily spawning . Red Sea Salt brand artificial seawater ( ASW ) was used for all culture and experiments . Manually dissected Clytia ovaries in small plastic petri dishes containing Millipore filtered sea water ( MFSW ) were maintained overnight in the dark and then stimulated with monochromatic light , provided by a monochromator ( PolyChrome II , Till Photonics ) installed above the samples , using the set-up described by Gühmann et al . ( 2015 ) , which delivers equivalent levels of irradiance between 400 and 600 nm ( 3 . 2E + 18 to 4 . 3E + 18 photons/s/m2 ) . Monochromatic light excitation was carried out in a dark room . 10 s pulses of different wavelengths , between 340 to 660 nm were applied to separate groups of 3–6 gonads , which were then returned to darkness for one hour before monitoring of oocyte maturation seen as loss of the oocyte nucleus ( Germinal Vesicle ) in fully grown oocytes , followed by spawning . A wavelength was considered to induce maturation if at least one oocyte per gonad underwent maturation and spawning within 30 min of monochromatic light excitation . 10 s exposure times were chosen because initial trials showed that these gave sub-saturating responses at all wavelengths . Gonads that spawned prematurely due to manipulation stress were excluded from analysis . BLAST searches were performed on an assembled Clytia hemisphaerica mixed-stage transcriptome containing 86 , 606 contigs , using published cnidarian opsin sequences or Clytia opsin sequences as bait . The ORFs of selected Clytia opsins were cloned by PCR into pGEM-T easy vector for synthesis of in situ hybridisation probes . Illlumina HiSeq 50nt reads were generated from mRNA isolated using RNAqueous micro kit ( Ambion Life technologies , CA ) from ectoderm , endoderm , growing oocytes and fully grown oocytes manually dissected from about 150 Clytia female gonads . The data have been deposited in NCBI's Gene Expression Omnibus and are accessible through GEO Series accession number GSE101072 ( https://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE101072 ) ( Quiroga Artigas et al . , 2017 ) . Biological replicates for each sample consisted of pooled tissue fragments or oocytes dissected on different days , using the same conditions and jellyfish of the same age . The reads were mapped against the opsin sequences retrieved from a Clytia reference transcriptome using Bowtie2 ( Langmead and Salzberg , 2012 ) . The counts for each contig were normalised per total of reads of each sample and per sequence length . Opsins RNA read counts from each tissue were visualised as a colour coded heat map using ImageJ software . For in situ hybridisation , isolated gonads were processed as previously ( Lapébie et al . , 2014 ) except that 4M Urea was used instead of 50% formamide in the hybridisation buffer as it significantly improves signal detection and sample preservation in Clytia medusa ( Sinigaglia et al . , 2017 ) . Images were taken with an Olympus BX51 light microscope . For double fluorescent in situ hybridisation , female Clytia gonads were fixed overnight at 18°C in HEM buffer ( 0 . 1 M HEPES pH 6 . 9 , 50 mM EGTA , 10 mM MgSO4 ) containing 3 . 7% formaldehyde , washed five times in PBS containing 0 . 1% Tween20 ( PBS-T ) , then dehydrated on ice using 50% methanol/PBS-T then 100% methanol . In situ hybridisation was performed using a DIG-labeled probe for Opsin9 and a fluorescein-labeled probe for PP4 . A three hours incubation with a peroxidase-labeled anti-DIG antibody was followed by washes in MABT ( 100 mM maleic acid pH 7 . 5 , 150 mM NaCl , 0 . 1% Triton X-100 ) . For Opsin9 the fluorescence signal was developed using the TSA ( Tyramide Signal Amplification ) kit ( TSA Plus Fluorescence Amplification kit , PerkinElmer , Waltham , MA ) and Cy5 fluorophore ( diluted 1/400 in TSA buffer: PBS/H2O20 . 0015% ) at room temperature for 30 min . After three washes in PBS-T , fluorescence was quenched with 0 . 01N HCl for 10 min at room temperature and washed again several times in PBS-T . Overnight incubation with a peroxidase-labelled anti-fluorescein antibody was followed by washes in MABT . The anti PP4 fluorescence signal was developed using TSA kit with Cy3 fluorophore . Controls with single probes were done to guarantee a correct fluorescence quenching and ensure that the two channels did not cross-over . Nuclei were stained using Hoechst dye 33258 . Images were acquired using a Leica SP5 confocal microscope and maximum intensity projections of z-stacks prepared using ImageJ software ( Schneider et al . , 2012 ) . Single in situ hybridisations were performed , and gave equivalent results , three times and double in situ hybridisations twice . For co-staining of neuropeptides and tyrosinated tubulin , dissected Clytia gonads were fixed overnight at 18°C in HEM buffer containing 3 . 7% formaldehyde , then washed five times in PBS containing 0 . 1% Tween20 ( PBS-T ) . Treatment on ice with 50% methanol/PBS-T then 100% methanol plus storage in methanol at −20°C improved visualisation of microtubules in the MIH-producing cells . Samples were rehydrated , washed several times in PBS-0 . 02% Triton X-100 , then one time in PBS-0 . 2% Triton X-100 for 20 min , and again several times in PBS-0 . 02% Triton X-100 . They were then blocked in PBS with 3% BSA overnight at 4°C . The day after they were incubated in anti-PRPamide antibody and anti-Tyr tubulin ( YL1/2 , Thermo Fisher Scientific ) in PBS/BSA at room temperature for 2 hr . After washes , the specimens were incubated with secondary antibodies ( Rhodamine goat anti-rabbit and Cy5 donkey anti-rat-IgG; Jackson ImmunoResearch , West Grove , PA ) overnight in PBS at 4°C , and nuclei stained using Hoechst dye 33258 for 20 min . For co-staining of neuropeptides with cortical actin , cilia ( acetylated α-tubulin ) or cilary basal bodies ( ɣ-Tubulin ) , dissected Clytia gonads were fixed for 2–3 hr at room temperature in HEM buffer containing 80 mM maltose , 0 . 2% Triton X-100% and 4% paraformaldehyde , then washed five times in PBS containing 0 . 1% Tween20 ( PBS-T ) . Samples were further washed in PBS-0 . 02% Triton X-100 , then one time in PBS-0 . 2% Triton X-100 for 20 min , and again several times in PBS-0 . 02% Triton X-100 . They were then blocked in PBS with 3% BSA overnight at 4°C . The day after they were incubated in anti-PRPamide antibody and combinations of anti- ɣ-Tubulin ( GTU-88 , Sigma Aldrich ) , anti-acetylated α-tubulin ( 6-11B-1 , Sigma Aldrich ) , in PBS/BSA at room temperature for 2 hr . After washes , the specimens were incubated with appropriate combinations of secondary antibodies ( Rhodamine or Cy5 goat anti-rabbit; fluorescein or Cy5 goat anti-mouse-IgG ) and Rhodamine-Phalloidin overnight in PBS at 4°C , and nuclei stained using Hoechst dye 33258 for 20 min . Images were acquired using a Leica SP8 confocal microscope and maximum intensity projections of z-stacks prepared using ImageJ software . For MIH fluorescence quantification , 5–6 independent gonads for each of the two conditions ( light-adapted and dark-adapted after light stimulation ) and Clytia strains ( WT and Opsin9n1_4 ) were fixed as mentioned above and co-stained for MIH and Tyr-tubulin . All the fixations were done in parallel . Confocal images were acquired using the same scanning parameters ( i . e . magnification , laser intensity and gain ) . In all cases , 10 confocal Z planes were summed over 4 µm depth at the gonad surface using ImageJ software . With ImageJ , we separated the two channels ( MIH and Tyr-tubulin ) and selected the contour of MIH-positive cells using the Tyr-tubulin staining as guidance . Using the ‘Integrated Density’ option , we recovered the ‘RawIntDen’ values of the MIH-stained channel , which refer to the sum of the pixel intensity values in the selected region of interest . These values divided by 1000 correspond to the RFU ( Relative Fluorescence Units ) in Figure 4 . The template for Opsin9 n1 small guide RNA ( Opsin9 n1 sgRNA; sequence in Figure 3—source data 1 ) was assembled by cloning annealed oligonucleotides corresponding 20 bp Opsin9 target sequence into pDR274 ( Hwang et al . , 2013 ) ( 42250 , Addgene ) , which contains tracrRNA sequence next to a BsaI oligonucleotide insertion site . The sgRNA was then synthesised from the linearised plasmid using Megashortscript T7 kit ( Thermo Fisher Scientific ) and purified with ProbeQuant G-50 column ( GE healthcare ) and ethanol precipitation . The sgRNA was dissolved in distilled water at 80 µM and kept at −80°C until use . Purified Cas9 protein dissolved in Cas9 buffer ( 10 mM Hepes , 150 mM KCl ) was kindly provided by J-P Concordet ( MNHN Paris ) and diluted to 5 µM . The sgRNA was added to Cas9 protein in excess ( 2:1 ) prior to injection and incubated for 10 min at room temperature . The final Cas9 concentration was adjusted to 4 . 5 µM and the sgRNA to 9 µM . The mixture was centrifuged at 14 , 000 rpm for 10 min at room temperature . 2–3% of egg volume was injected into unfertilised eggs within 1 hr after spawning , prior to fertilisation . Injected embryos were cultured for 3 days in MFSW at 18°C . Metamorphosis of planula larvae into polyps was induced about 72 hr after fertilisation by placing larvae ( 20–80/slide ) on double-width glass slides ( 75 × 50 mm ) in drops of 3–4 ml MFSW containing 1 µg/ml synthetic metamorphosis peptide ( GNPPGLW-amide; Genscript ) , followed by overnight incubation . Slides with fixed primary polyps were transferred to small aquariums kept at 24°C , to favour the establishment of female colonies ( Carré and Carré , 2000 ) . Primary polyps and young polyp colonies were fed twice a day with smashed Artemia nauplii until they were big enough to be fed with swimming nauplii . Following colony vegetative expansion , a single well-growing colony on each slide was maintained as a founder . After several weeks of growth , polyp colonies were genotyped to assess mutation efficiency and mosaicism , and medusae were collected from the most strongly mutant colony ( colony number four obtained using the n1 guide RNA , designated Opsin9n1_4 ) for further experimentation . Genomic DNA from Clytia polyps and jellyfish gonads was purified using DNeasy blood/tissue extraction kit ( Qiagen ) . The opsin9 target site was amplified by PCR using Phusion DNA polymerase ( New England Biolabs ) . Primers used for genotyping are listed in Figure 3—source data 1 . PCR products were sequenced and mutation efficiency was assessed using TIDE analyses ( Figure 3—figure supplement 1 ) , which estimates the mutation composition from a heterogeneous PCR product in comparison to a wild type sequence ( Brinkman et al . , 2014 ) . We scanned Clytia genome for possible off-targets of Opsin9 sgRNA at http://crispor . tefor . net . From 2 possible off-targets where Cas9 could cut , none was found in coding sequences nor were they right next to a PAM sequence . Sexually mature wild type and Opsin9n1_4 mutant medusae of the same age and adapted to the same day-night cycle were collected for gonad dissection . Individual gonads were transferred to 100 µl MFSW in 96-well plastic plates . Plates were covered with aluminium foil overnight and brought back to white light the following day . For the rescue experiment with synthetic MIH , wild type and opsin9n1_4 mutant gonads adapted to light conditions were dissected and transferred to 96-well plastic plates and acclimatised for two hours . An equal concentration ( 100 µl ) of double concentrated ( 2 × 10−7M ) synthetic WPRPamide ( synthetic MIH; Genscript ) in MFSW was added in each well to give a final concentration of 10−7M . Oocyte maturation was scored after one hour . Spawning followed in all cases where oocyte maturation was triggered . Gonads that spawned prematurely due to manipulation stress were excluded from analysis . Gonad pictures in Figure 3 were taken with an Olympus BX51 microscope . Graphs and statistics for the monochromator assay were prepared using BoxPlotR ( Spitzer et al . , 2014 ) . Fisher’s exact tests and Mann-Whitney U tests were performed at http://www . socscistatistics . com . Fisher’s exact tests were chosen for analysing the spawning results of Figure 3D , E based on 2 × 2 contingency tables . Nonparametric Mann-Whitney U tests were chosen for the MIH fluorescence quantification comparisons ( Figure 4 ) since the results did not follow a normal distribution according to the Shapiro-Wilk test and significance between sample distributions could be appropriately assessed by data transformation into ranks . To assess the relationship of the Clytia opsin amino acid sequences to known opsins , we added them to two recent datasets ( Vöcking et al . , 2017; Feuda et al . , 2014 ) . All incomplete sequences were removed from the untrimmed Vöcking et al . ( 2017 ) dataset before adding the Clytia opsin sequences using profile alignment in MUSCLE ( Edgar , 2004 ) . An Opsin sequence found in our unpublished Melicertum octocostatum transcriptome ( Leptothecata , Hydrozoa ) was added to this dataset using profile alignment . These untrimmed alignments were used in phylogenetic analyses . The same procedure was used for adding the Clytia opsin sequences to the Feuda et al . ( 2014 ) trimmed dataset . Positions containing only Clytia opsin sequences were removed . For more detailed comparison between medusozoan sequences , we generated an alignment including all cubozoan and hydrozoan available opsin protein sequences in GenBank ( March 2017 ) and the 10 Clytia opsin sequences . Cd-hit ( Fu et al . , 2012 ) was run with 99% identity to eliminate sequence duplicates , obtaining a final dataset of 56 Cubozoa and Hydrozoa opsin protein sequences . The alignment was performed using MUSCLE ( Edgar , 2004 ) and further adjusted manually in Seaview v4 . 2 . 12 ( Galtier et al . , 1996 ) to remove the ambiguously aligned N- and C-terminal regions , as well as positions including only one residue . Alignments used for phylogenetic analyses are available as Figure 6—source data 1 . Maximum likelihood ( ML ) analyses were performed using RaxML 8 . 2 . 9 ( Stamatakis , 2014 ) . The GTR+Γ and LG+Γ models of protein evolution were used following Feuda et al . ( 2014 ) and Vöcking et al . ( 2017 ) with parsimony trees as starting trees . ML branch support was estimated using non-parametric bootstrapping ( 500 replicates ) . Bayesian analyses ( MB ) were carried out using MrBayes 3 . 2 . 6 ( Ronquist et al . , 2012 ) , with the LG+Γ model , performing two independent runs of four chains for 1 million generations sampled every 100 generations . MB analyses were considered to have converged when average standard deviation of split frequencies dropped below 0 . 05 . Consensus trees and posterior probabilities were calculated using the 200 . 000 last generations . The resulting trees were visualised with FigTree ( http://tree . bio . ed . ac . uk/software/figtree/ ) . Approximated Unbiased ( AU ) phylogenetic tests ( Shimodaira , 2002 ) were performed as described previously ( Leclère and Rentzsch , 2012 ) . Values presented in the Results section were obtained comparing , among others , the ML tree with Clytia Opsin9 , Opsin10 and either Cladonema or Tripedalia gonad-expressed Opsins constrained as monophyletic using RaxML . Exclusion of the non-gonad-expressed Opsin10 from these monophyletic constraints led to much lower p values ( not shown ) .	Many animals living in the sea reproduce by releasing sperm and egg cells at the same time into the surrounding water . Animals often use changes in ambient light at dawn and dusk as reliable daily cues to coordinate this spawning behavior between individuals . For example , jellyfish of the species Clytia hemisphaerica , which can easily be raised in the laboratory , spawn exactly two hours after the light comes on . Researchers recently discovered that spawning in Clytia and other related jellyfish species is coordinated by a hormone called ‘oocyte maturation-inducing hormone’ , or MIH for short . This hormone is produced by a cell layer that surrounds the immature eggs and sperm within each reproductive organ , and is secreted in response to light cues . It then diffuses both inside and outside of the jellyfish , and triggers the production of mature eggs and sperm , followed by their release into the ocean . However , until now it was not known which cells and molecules are responsible for detecting light to initiate the secretion of MIH . Quiroga Artigas et al . – including some of the researchers involved in the MIH work – now discovered that a single specialised cell type in the reproductive organs of Clytia responds to light and secretes MIH . These cells contain a light-sensitive protein called Opsin9 , which is closely related to the opsin proteins in the human eye well known for their role in vision . When Opsin9 was experimentally mutated , Clytia cells could not secrete MIH in response to light , and the jellyfish failed to spawn . This opsin protein is thus necessary to detect light in order to trigger spawning in jellyfish . A next step will be to examine and compare whether other proteins of the opsin family and hormones related to MIH also regulate spawning in other marine animals . This could have practical benefits for raising marine animals in aquariums and as food resources , and in initiatives to protect the environment . More widely , these findings could help unravel how sexual reproduction has evolved within the animal kingdom .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2018	A gonad-expressed opsin mediates light-induced spawning in the jellyfish Clytia
Sensory systems sequentially extract increasingly complex features . ON and OFF pathways , for example , encode increases or decreases of a stimulus from a common input . This ON/OFF pathway split is thought to occur at individual synaptic connections through a sign-inverting synapse in one of the pathways . Here , we show that ON selectivity is a multisynaptic process in the Drosophila visual system . A pharmacogenetics approach demonstrates that both glutamatergic inhibition through GluClα and GABAergic inhibition through Rdl mediate ON responses . Although neurons postsynaptic to the glutamatergic ON pathway input L1 lose all responses in GluClα mutants , they are resistant to a cell-type-specific loss of GluClα . This shows that ON selectivity is distributed across multiple synapses , and raises the possibility that cell-type-specific manipulations might reveal similar strategies in other sensory systems . Thus , sensory coding is more distributed than predicted by simple circuit motifs , allowing for robust neural processing . Animals rely on their sensory systems to process behaviorally relevant information . One common feature of sensory systems is the sequential processing of information to extract complex features from simple inputs . For example , in the visual system , photoreceptors detect light and then downstream neurons progressively extract distinct features , such as contrast , direction of motion , form , or specific objects ( Gollisch and Meister , 2010; Livingstone and Hubel , 1988 ) . Sensory pathways diverge into pathways that become selective for increasingly specific features . One prominent example is the split into ON and OFF pathways , where individual neurons become selective to either increases ( ON ) or decreases ( OFF ) in a signal . Such an ON/OFF dichotomy enables more efficient coding of stimuli in the visual system ( Gjorgjieva et al . , 2014 ) and occurs across many different species and sensory modalities , such as vision , olfaction , audition , thermosensation , and electrolocation ( Bennett , 1971; Gallio et al . , 2011; Scholl et al . , 2010; Tichy and Hellwig , 2018; Werblin and Dowling , 1969 ) . Examples of how the split into ON and OFF pathways is implemented in sensory information processing have already been described . In the vertebrate retina , ON and OFF pathways split downstream of glutamatergic photoreceptors where ionotropic glutamate receptors on OFF bipolar cells maintain the sign of the response in the OFF pathway , and the metabotropic glutamate receptor mGluR6 , located on ON bipolar cells , inverts the sign in the ON pathway ( Koike et al . , 2010; Masu et al . , 1995; Vardi , 1998 ) . In the olfactory system of C . elegans , an odor response can be split into parallel pathways in which glutamate-gated chloride channels mediate the ON response ( Chalasani et al . , 2007 ) . While these transformations are thought to occur at specific synapses , connectomics data reveals that neural circuits are intricate and that many of the possible neuronal connections are realized ( Eichler et al . , 2017; Takemura et al . , 2013; Zheng et al . , 2018 ) . This argues that important signal transformations might actually be distributed across wider circuit motifs . In the Drosophila visual system , ON and OFF pathways functionally split in the first order lamina interneurons , but the physiological specialization occurs one synaptic layer further downstream . In brief , information travels from the retina , which houses the photoreceptors , through three optic ganglia: the lamina , the medulla , and the lobula complex , comprising lobula and lobula plate ( Figure 1A ) . Contrast is encoded by the transient response of photoreceptors , and downstream lamina neurons amplify the contrast-sensitive signal component ( Laughlin , 1989 ) . Then , distinct ON and OFF pathways are required to detect contrast increments and decrements , respectively ( Joesch et al . , 2010; Strother et al . , 2014 ) . In the lamina , L1 is the major input to the ON pathway , whereas L2 and L3 feed into the OFF pathway ( Clark et al . , 2011; Joesch et al . , 2010; Rister et al . , 2007; Silies et al . , 2013 ) . The assignment of L1 , L2 , and L3 to ON and OFF pathways originates from neuronal silencing studies ( Clark et al . , 2011; Joesch et al . , 2010; Silies et al . , 2013 ) . However , all lamina neurons receiving direct input from photoreceptors depolarize to the offset of light and hyperpolarize to the onset of light ( Clark et al . , 2011; Laughlin , 1989; Silies et al . , 2013; Uusitalo et al . , 1995 ) , thus passing on information about both ON and OFF ( Figure 1B ) . Voltage or calcium signals in most downstream medulla neurons then selectively report only one type of contrast polarity . The major ON pathway medulla neurons Mi1 and Tm3 , for example , selectively respond with depolarization or an increase in calcium signal to ON ( Figure 1B; Behnia et al . , 2014; Strother et al . , 2017; Yang et al . , 2016 ) . In the OFF pathway , most neurons instead selectively respond to OFF stimuli , retaining the response polarity of their lamina inputs ( Figure 1B; Behnia et al . , 2014; Serbe et al . , 2016; Yang et al . , 2016 ) . Therefore , ON selectivity requires a sign inversion between the L1 input and its postsynaptic partners Mi1 and Tm3 . Previous work suggested that the L1 input to the ON pathway is glutamatergic , whereas L2 and L3 , the two major inputs to the OFF pathway , are cholinergic ( Davis et al . , 2018; Takemura et al . , 2011 ) . This suggests that glutamate might also be used as an inhibitory neurotransmitter to implement ON/OFF dichotomy in the fly visual system . However , the molecular and cellular mechanisms implementing this signal transformation are not known in Drosophila visual circuitry . Connectomics data has generated predictions about core circuit motifs ( Shinomiya et al . , 2014; Takemura et al . , 2013 ) . In the ON pathway , L1 makes the largest number of synapses with the medulla intrinsic Mi1 neuron and the transmedullary Tm3 interneurons ( Figure 1A; Takemura et al . , 2013 ) . However , L1 has many other outputs , and Mi1 and Tm3 many additional inputs , such as indirect L1 input via L5 , or the GABAergic neuron C2 , among others ( Takemura et al . , 2013 ) . Thus , coding in the visual system could be distributed across parallel pathways . One synaptic layer downstream , Mi1 and Tm3 medulla neurons project to T4 , the first direction-selective cells of the ON pathway ( Figure 1A; Fisher et al . , 2015a; Maisak et al . , 2013 ) . This core visual circuit motif appears to be surprisingly resilient to perturbations . While genetic silencing of Mi1 or Tm3 leads to some deficits in ON edge motion detection ( Ammer et al . , 2015; Strother et al . , 2017 ) , these flies are not ON motion blind , arguing that other neurons must also play a role in motion detection . Mi4 and Mi9 have now been added to the ON pathway ( Takemura et al . , 2017 ) . These cell types are modulated by octopamine , but silencing Mi4 or Mi9 individually has only subtle phenotypes ( Strother et al . , 2018; Strother et al . , 2017 ) . In the OFF pathway , combinatorial block of more than one cell type aggravates behavioral deficits ( Fisher et al . , 2015a; Serbe et al . , 2016; Silies et al . , 2013 ) . This is also already true for the lamina neuron inputs L1 , L2 and L3 neurons ( Silies et al . , 2013 ) . This could argue that individual neurons might have distinct , but overlapping tuning properties ( Serbe et al . , 2016; Tuthill et al . , 2013 ) . Alternatively , encoding of a single aspect of a feature might already be distributed across parallel pathways . Here , we show that ON selectivity in the Drosophila visual system is mediated by a glutamate-gated chloride channel , GluClα , and that all ON responses are lost upon pharmacological block or genetic loss of GluClα in the entire brain . At the same time , ON responses are robust to cell-type-specific perturbations of GluClα in individual neurons postsynaptic to L1 , arguing for the existence of parallel functional pathways . Furthermore , we found that GABAergic inhibition also plays a role in mediating ON responses downstream of the glutamatergic L1 input . Together , our results indicate that ON selectivity is a multisynaptic computation that depends on both glutamatergic and GABAergic inhibition . This suggests that a seemingly simple computation can be implemented in a multisynaptic manner , allowing for greater functional robustness . To test if medulla neurons in the ON pathway receive glutamatergic input resembling the L1 response , we used the genetically encoded glutamate sensor iGluSnFR ( Marvin et al . , 2013; Richter et al . , 2018 ) . We selectively expressed iGluSnFR in the two major postsynaptic targets of L1: the medulla neurons Mi1 and Tm3 . Using in vivo two-photon imaging , we measured visually evoked responses on Mi1 and Tm3 dendrites reflecting their glutamatergic inputs in medulla layers M1 and M5 . We also recorded iGluSnFR signals in the neurons’ output layer , M9/10 . In M1 and M5 , both Mi1 and Tm3 neurons showed an increase in iGluSnFR signal in response to light OFF and a decrease in iGluSnFR signal in response to light ON , showing that rectification happens downstream of the glutamatergic input ( Figure 1C , D ) . These signals were of the same polarity as intracellular calcium signals recorded within the presynaptic L1 axon terminals , and of the opposite polarity to calcium signals in the same layers of Mi1 and Tm3 ( Figure 1B–D ) . In the proximal medulla ( layer M9/10 ) , Mi1 and Tm3 neurons showed weak iGluSnFR signals that increased in response to both light ON and OFF ( Figure 1C , D ) . This data shows that the major postsynaptic targets of L1 receive glutamatergic input that provides information about both ON and OFF signals , which is in line with graded inputs coming from the L1 input to the ON pathway . Other glutamatergic inputs might further shape medulla neuron properties in the proximal medulla . We hypothesized that glutamatergic inhibition mediates the sign inversion between the dendritic extracellular glutamate signals and intracellular calcium or voltage signals measured in these neurons ( Figure 1; Behnia et al . , 2014 ) . Glutamatergic inhibition can be mediated either by metabotropic glutamate receptors or by ionotropic glutamate-gated chloride channels ( Collins et al . , 2012; Cully et al . , 1996; Liu and Wilson , 2013; Parmentier et al . , 1996 ) . To determine which of these receptor types mediates ON responses , we recorded in vivo calcium signals in response to visual stimuli in the two major postsynaptic partners of L1 while pharmacologically inhibiting each of the two receptor classes . When flies expressing GCaMP6f in Mi1 neurons were shown 5 s full-field flashes , Mi1 showed a transient increase in calcium signals in response to the ON step that decayed to reach a plateau response within 2 s ( Figure 2—figure supplement 1A ) . Bath application of 2-Methyl-6- ( phenylethenyl ) pyridine hydrochloride ( MPEP ) , a selective blocker of metabotropic glutamate receptors , to the same flies did not reduce the responses to visual light flashes in Mi1 neurons ( Figure 2—figure supplement 1A , B ) . Before drug application , Tm3 responses to ON flashes showed a transient light response . Similar to Mi1 , Tm3 responses were not affected by MPEP application ( Figure 2—figure supplement 1C , D ) . We next applied picrotoxin ( PTX ) , a drug that is known to inhibit glutamate-gated chloride channels at high concentrations ( Cully et al . , 1996; Etter et al . , 1999 ) , but which also affects GABAARs at much lower concentration ( Takeuchi and Takeuchi , 1969 ) . In vivo studies in Drosophila had previously used concentrations of 1–5 μM PTX to effectively block GABA-gated hyperpolarization in the olfactory system and GABAergic inhibition in Drosophila visual system neurons ( Fisher et al . , 2015b; Wilson and Laurent , 2005 ) . In contrast , 100 μM PTX was used to block GluCls in the olfactory system ( Liu and Wilson , 2013 ) . Upon bath application of 100 μM PTX , visual responses were completely abolished in Mi1 neurons ( Figure 2A ) . Surprisingly , when we tested the effect of low concentrations ( 2 . 5 μM ) of PTX , ON responses were also lost in Mi1 ( Figure 2A ) . To test this effect more precisely , we used a range of PTX concentrations and observed a loss of visual responses at concentrations ranging from 2 . 5 μM to 100 μM PTX in Mi1 ( Figure 2B ) . When we performed the same experiments in Tm3 , we again found all ON responses to be eliminated in 100 μM PTX and strongly reduced at low concentrations of PTX ( Figure 2C , D ) . This effect was consistent across all medulla layers ( Figure 2A–D , Figure 2—figure supplement 2A–D ) . Dendritic Mi1 and Tm3 regions even showed a small decrease in calcium in response to light at the highest PTX concentrations ( Figure 2—figure supplement 2A–D ) . To evaluate if the PTX effect is ON-pathway selective and to measure the compound effect on ON and OFF pathway responses , we next imaged calcium signals in the direction-selective T4 and T5 axon terminals , the stage at which many medulla neuron inputs converge . When we measured flash responses in T4/T5 cells expressing GCaMP6f , these neurons hardly showed any response to full-field flashes before toxin application , due to surround inhibition ( Figure 2E , F; Fisher et al . , 2015a ) . Upon bath application of 2 . 5 μM PTX , these flash responses were disinhibited , and T4/T5 neurons responded with an increase in calcium signal to both light ON and OFF ( Figure 2E , F; Fisher et al . , 2015b ) . Thus , T4/T5 neurons still show flash responses under conditions in which all responses of their predominant Mi1 and Tm3 inputs are abolished ( Figure 2A–D ) . This suggests , that at least under low PTX concentrations , other neurons ( Takemura et al . , 2017 ) and a lack of local inhibition ( Mauss et al . , 2015 ) can contribute to T4 responses . After increasing the concentration of PTX to 100 μM within the same fly , all ON responses were abolished ( Figure 2E , F ) , showing that T4 no longer receives any functional inputs at high PTX concentrations . In contrast , OFF responses were unaffected relative to the 2 . 5 μM phenotype ( Figure 2E , F ) , arguing that the effect on the ON pathway is specific . This is in line with the idea that glutamate-gated chloride channels mediate ON responses in the visual system . Importantly , the L1 input still responded to visual stimuli even at the highest PTX concentrations used ( Figure 2G , H ) and the iGluSnFR signal on the dendrites of Mi1 or Tm3 were largely unaltered , or even slightly increased at concentrations at which all Mi1 and Tm3 calcium responses were abolished ( Figure 2—figure supplement 3 ) , demonstrating that the glutamatergic input to the ON pathway was still intact . Taken together , our findings show that a systemic disruption of glutamate-gated chloride channels abolishes ON responses in Mi1 and Tm3 , and suggest that GABAA receptors might play a role in mediating ON responses at the L1 to Mi1/Tm3 synapses in the fly visual system . To explore the possibility that both glutamate- and GABA-gated chloride channels mediate ON selectivity , we first looked at the expression of candidate genes . The only glutamate-gated chloride channel in the fly genome is encoded by the GluClα gene . A GluClα protein tagged with GFP ( GluClαMI02890 . GFSTF . 2 ) was found to be widely expressed in the visual system , including the lamina , medulla , lobula and lobula plate ( Figure 3A ) . Expression was stronger in some proximal medulla layers , but the broad expression of this GFP trap did not allow expression to be assigned to specific cell types ( Figure 3A ) . Two recently published cell-type-specific RNA sequencing datasets allowed us to assess candidate gene expression at cellular resolution ( Davis et al . , 2018; Konstantinides et al . , 2018 ) . GluClα mRNA was strongly expressed in all major ON pathway medulla neurons: Mi1 , Tm3 , Mi4 , and Mi9 ( Figure 3B , Figure 3—figure supplement 1 ) . Furthermore , GluClα was also expressed in OFF pathway neurons ( Tm1 , Tm2 , Tm4 , and Tm9 ) , albeit weaker in Tm9 ( Figure 3B , Figure 3—figure supplement 1 ) . The metabotropic glutamate receptor mGluR did not show expression in all ON pathway medulla neurons ( Figure 3B , Figure 3—figure supplement 1 ) , consistent with mGluR not playing a broad role in mediating ON selectivity ( Figure 2—figure supplement 1 ) . Of the three genes known to encode GABAARs , Grd mRNA was not detectable in medulla neurons and Lcch3 was only weakly expressed . Interestingly , the Rdl gene was strongly expressed in all major ON and OFF pathway medulla interneurons ( Figure 3B ) . Thus , the glutamate- and GABA-gated chloride channel GluClα and Rdl are widely expressed in the visual system , including all ON pathway medulla neurons . We next wanted to determine whether the PTX-induced loss of responses was due to inhibition of the glutamate receptor GluClα , the GABA receptor Rdl , or both . We therefore added molecular specificity to the pharmacological approach using alleles that are insensitive to PTX . For Rdl , a single point mutation has been described that leaves channel function intact but renders GABAAR insensitive to PTX ( Ffrench-Constant et al . , 1993 ) . We hypothesized that if ON responses are mediated by the Rdl receptor , the PTX-insensitive RdlMDRR allele should rescue the effect of PTX on visual responses in ON pathway medulla neurons . To ensure that Rdl channels were exclusively composed of the PTX-insensitive RdlMDRR subunit , experiments were performed in trans to an Rdl null mutant ( Rdl1/RdlMDRR ) , or in homozygosity ( RdlMDRR/RdlMDRR ) . We tested rescue of visual responses by the RdlMDRR mutant at 2 . 5 μM PTX , as this was the lowest toxin concentration that resulted in a loss of ON responses in both Mi1 and Tm3 . We individually quantified the amplitude of the maximum response to the ON step , the amplitude of the plateau response , and the integrated response during the ON step ( Figure 4—figure supplement 1A ) . Control Mi1 neurons showed significantly reduced Mi1 peak responses and an eliminated sustained component upon application of PTX ( 2 . 5 μM ) , similar to PTX application in wild type ( Figure 4A , Figure 4—figure supplement 1B ) . Importantly , when channels were only composed of the RdlMDRR insensitive subunit , Mi1 responses were partially rescued and the sustained component of the response was present ( Figure 4B , C Figure 4—figure supplement 1C ) . Whereas the rescue of the peak ON response was only significant in layer M1 ( Figure 4C ) , the integrated response or the plateau response were also prominently rescued in other layers ( Figure 4—figure supplement 1D ) . In Tm3 neurons , PTX application also significantly reduced ON responses in controls ( Figure 4D ) . This response was partially rescued by the presence of the PTX insensitive RdlMDRR allele ( Figure 4E ) . This effect was again strongest in layer M1 ( Figure 4E , F , Figure 4—figure supplement 1E–G ) . The fact that the RdlMDRR allele does not fully rescue all ON responses in Mi1 or Tm3 suggests that PTX might also be acting on GluClα in this context . At the same time , our findings argue that responses in medulla neurons Mi1 and Tm3 are indeed mediated at least in part by the GABAA receptor Rdl . All lamina neurons downstream of photoreceptors were shown to be GABA negative by immunostaining , whereas the lamina feedback neurons C2 and C3 are GABA positive ( Kolodziejczyk et al . , 2008 ) . RNAseq data support this notion , since L1 expresses high levels of genes involved in glutamate synthesis and does not express any GABA synthesis enzymes ( Figure 4—figure supplement 2A; Davis et al . , 2018 ) . Expression of the vesicular GABA transporter dVGAT appears high , but this gene is highly expressed in all neurons , and could be non-specific . Furthermore , although it has been shown that neurons can maintain inhibitory signaling via uptake of GABA ( Tritsch et al . , 2014 ) , this requires expression of the plasma membrane GABA transporter Gat , which is again not expressed in L1 ( Figure 4—figure supplement 2A ) . Finally , GABA immunostaining is not visible in the terminals or cell bodies of L1 neurons , but can be seen in C2/C3 neurons ( Figure 4—figure supplement 2B , C; Kolodziejczyk et al . , 2008 ) . Thus , there is no evidence for L1 co-releasing GABA in addition to glutamate . This suggests that visual responses to ON stimuli in Mi1 and Tm3 do not arise solely through a monosynaptic connection with the L1 inputs as previously thought ( Figure 4Gi ) , but that a GABAergic synapse involving Rdl is likely involved in circuitry upstream of Mi1 and Tm3 . In summary , Rdl-dependent circuits parallel to the glutamatergic L1 to medulla neuron synapse can also mediate ON responses ( Figure 4Gii ) . We next wanted to test if and to what degree GluClα contributes to visual ON responses . No PTX-insensitive GluClα allele has been isolated in Drosophila , but the crystal structure of GluClα has been solved for the C . elegans homolog and the amino acid side chains interacting with the toxin have been identified ( Hibbs and Gouaux , 2011 ) . PTX interacts with specific residues of the M2 transmembrane alpha helix ( Figure 5A; Hibbs and Gouaux , 2011 ) . We aligned the M2 amino acid sequences of histamine , glutamate or GABA-gated chloride channels from different species ( D . melanogaster , C . elegans , M . domestica ) with known PTX sensitivities ( Cully et al . , 1994; Ffrench-Constant et al . , 1993; Hirata et al . , 2008; Horoszok et al . , 2001; Zheng et al . , 2002 ) . This region is highly conserved among different channels and among species , with the exception of a single variable amino acid , corresponding to amino acid S278 in D . melanogaster GluClα ( red , Figure 5B ) . The identity of this single amino acid correlates strongly with the PTX sensitivity of the channel ( Figure 5B ) . Mutations in this amino acid have been shown to change the PTX sensitivity of the channel . For example , the A > S substitution in the D . melanogaster GABAAR allele RdlMDRR exhibits reduced sensitivity to PTX ( Figure 4 , Ffrench-Constant et al . , 1993; Fisher et al . , 2015a ) . This prompted us to generate a potentially PTX-insensitive version of GluClα by introducing a point mutation leading to an S278T exchange . We first characterized mutant GluClαS278T heterologously in Xenopus oocytes . Two-electrode voltage-clamp recordings of wild type GluClα-expressing oocytes revealed fast activating and rapidly inactivating glutamate-induced currents , similar to inhibitory glutamate currents recorded in vivo in honeybees ( Barbara et al . , 2005 ) . This current was sensitive to PTX ( Figure 5C , E ) . Expression of the GluClαS278T mutant led to glutamate-induced currents that were less inactivating compared to wild-type GluClα controls . Importantly , the glutamate-induced currents in the GluClαS278T mutant were insensitive to PTX ( Figure 5D , E ) . We next generated GluClαS278T mutant flies , targeting the endogenous GluClα gene locus using CRISPR/Cas9-based genome editing ( see Materials and methods for details ) . In the absence of toxin , GluClαS278T flies responded to visual stimuli with the typical peak and plateau response , arguing that the altered kinetics of the GluClαS278T mutant observed in oocytes was not a problem under these stimulus conditions ( Figure 6A , B ) . We then tested if the GluClαS278T allele could rescue PTX-induced phenotypes in vivo in the visual system . Because the PTX-insensitive RdlMDRR allele rescued visual responses only partially at 2 . 5 μM , we first tested if GluClα could also account for a loss of responses at such low PTX concentrations previously thought to only block GABAARs . Upon application of low concentrations of PTX ( 2 . 5 μM ) , calcium responses were lost in Mi1 neurons of heterozygous GluClα controls carrying a deficiency ( Df ) uncovering the GluClα locus ( Figure 6A ) . In flies only expressing the GluClαS278T and no wild type protein , the ON responses were partially rescued . The rescue was specifically prominent for the step response , which was significantly rescued by GluClαS278T in all medulla layers ( Figure 6B , C , Figure 6—figure supplement 1A–C ) . This shows that GluClα in vivo is sensitive to lower concentrations of PTX than previously thought , arguing that there is no specific concentrations to only block Rdl , and highlighting the usefulness of these PTX-insensitive alleles for molecular specificity . Furthermore , the rescue by GluClαS278T demonstrates that GluClα mediates ON responses in Mi1 in the fly visual system . When testing if GluClαS278T could also rescue Tm3 responses in the presence of the toxin , this batch of 2 . 5 μM PTX gave a comparably mild phenotype in heterozygous controls ( Figure 6D ) . The integrated ON response was still significantly rescued in a GluClαS278T background in layer M1 ( Figure 6E , F ) . In other medulla layers , 2 . 5 μM PTX more prominently blocked the peak Tm3 ON response in controls , but not in a GluClαS278T-insensitive background ( Figure 6—figure supplement 1D–F ) . This shows that GluClαS278T also partially mediates ON responses in Tm3 . At high concentrations of PTX ( 100 μM ) , the PTX-insensitive GluClαS278T allele did not rescue ON responses in either Mi1 or Tm3 ( Figure 6—figure supplement 1G–L ) . This could suggest that GluClαS278T does not confer PTX sensitivity at such high PTX concentrations in vivo . Alternatively , if GluClαS278T was fully insensitive , this would further argue that PTX blocks other channels that are required for ON responses , and would thus underline the importance of Rdl . While we cannot fully distinguish between these two possibilities , we argued that if the loss of Mi1 and Tm3 responses were due to the role of Rdl , GluClαS278T might still rescue the 100 μM PTX phenotype in T4/T5 . As we showed above , unlike Mi1 and Tm3 responses , ON responses in T4/T5 were specifically blocked by high but not low concentrations of PTX ( Figure 2E , F ) . To test if GluClαS278T can rescue T4/T5 responses at high concentrations , we recorded calcium signals in T4/T5 cells in a GluClαS278T background . Indeed , ON responses to full-field light flashes in 100 μM PTX were rescued ( Figure 6G–I ) . Although this experiment does not tell us which cell types this rescue is coming from , this data shows that GluClαS278T can be effective to rescue responses in some cell types at high PTX concentrations in vivo . Thus , the use of the toxin-insensitive GluClαS278T mutant demonstrates that GluClα also mediates ON responses in vivo in the fly visual system . Together , our data provides support for a combinatorial role of glutamatergic and GABAergic inhibition in mediating ON responses . Because of the glutamatergic L1 input , GluClα is likely to be the receptor on all neurons postsynaptic to L1 ( Figure 6J ) . Rdl could function downstream of GluClα . This suggests that a pathway parallel to a monosynaptic glutamatergic circuit can also mediate ON responses in Mi1 and Tm3 ( Figure 6J ) . Our findings lead to a model in which GluClα mediates responses to glutamatergic inputs in neurons downstream of L1 and in which a GABAergic pathway additionally drives responses in the ON pathway medulla neurons Mi1 and Tm3 ( Figure 6J ) . Given that there is no evidence for L1 being GABAergic , Mi1 and Tm3 responses might not depend solely on monosynaptic L1 input . If this hypothesis is correct , GluClα should not exclusively function in a cell-autonomous manner in neurons downstream of L1 , suggesting that Mi1 and Tm3 might still be able to respond to ON signals when GluClα function is only disrupted within the respective cell type . However , since pharmacological perturbations always targeted the entire visual system , more specific targeting would be required to address this possibility . To test the above hypothesis , we generated a GluClα loss-of-function specifically in either Mi1 or Tm3 . We inserted a FlpStop exon ( Fisher et al . , 2017 ) in the non-disrupting orientation ( GluClαFlpStop . ND ) , in which splicing occurs normally unless the FlpStop exon is inverted by Flp recombinase expression ( Fisher et al . , 2017 ) . Upon pan-neuronal inversion of the FlpStop cassette into the non-disrupting orientation ( GluClαFlpStop . D ) quantification of GluClα expression levels using qRT-PCR showed that transcription was disrupted by half when the FlpStop exon was inverted in a heterozygous background , arguing for a full loss of function in GluClαFlpStop . D ( Figure 7A ) . To selectively disrupt GluClα function in Mi1 neurons , we expressed Flp recombinase in Mi1 ( Figure 7B ) . Expression of tdTomato , a marker for the FlpStop inversion event , further confirmed efficient inversion of the cassette ( Figure 7B ) . When we recorded calcium signals in Mi1 in this cell-type-specific FlpStop background , visual responses to ON flashes were still present in Mi1 neurons and did not differ from controls ( Figure 7C ) . To corroborate these findings , we next used cell-type-specific RNAi . Pan-neuronal expression of GluClαdsRNA reduced GluClα mRNA to 16 . 4 ± 7 . 4% of controls ( Figure 7D ) . When knocking down GluClα in either Mi1 or Tm3 , ON responses were again not abolished and did not differ significantly from controls ( Figure 7E , F ) . These results demonstrate that ON selectivity is not mediated monosynaptically , but that pathways parallel to the L1-Mi1 or L1-Tm3 connection might be sufficient to mediate ON responses . Our results argue that ON responses are encoded in a multi-synaptic fashion . Because L1 is the major input to the ON pathway and is glutamatergic ( Figure 4—figure supplement 2A; Takemura et al . , 2011 ) , all ON responses might still depend on GluClα at the first synapse postsynaptic to L1 . This leads to the hypothesis that Mi1 or Tm3 responses would be lost in a full GluClα mutant background . To test this , we used a FlpStop allele inserted in the disrupting orientation . In this background , expression should be fully disrupted in all cells normally expressing GluClα . Quantification of expression levels of GluClα using qRT-PCR showed that transcription was fully disrupted in GluClαFlpStop . D ( 3 . 7 ± 0 . 6% mRNA compared to wild type , Figure 7G ) . Furthermore , in heterozygous animals , GluClαFlpStop . D transcripts were reduced roughly by half ( 40 . 1 ± 2 . 2% ) and to the same amount as in a GluClα deficiency ( 40 . 2 ± 8 . 2% ) lacking the entire gene locus ( Figure 7G ) . These findings confirm that GluClαFlpStop . D is a null allele . GluClαFlpStop . D mutant flies eclosed but showed locomotor deficits . The viability of the GluClα mutant allowed us to conduct calcium imaging experiments in this null mutant background . Whereas Mi1 and Tm3 neurons in heterozygous GluClα mutant or deficient flies responded normally to light flashes , Mi1 and Tm3 responses were both dramatically affected in full GluClα mutants . No increase in calcium signal was detectable in GluClα null mutants , and light responses were largely absent in all layers ( Figure 7H , I ) . Instead , Mi1 even showed a small and transient decrease in calcium signal in response to light ON , which could potentially be attributed to the additional presence of excitatory glutamate receptors or reveal inputs from the OFF pathway ( Figure 7H ) . Thus , normal ON responses are lost whenever GluClα function is disrupted in the entire visual system , but not when it is disrupted in a cell-type-specific manner . These data are further consistent with the results of pharmacological experiments in wild type and PTX-insensitive alleles . Taken together , these results demonstrate that ON selectivity is a multisynaptic computation that is robust to perturbations at individual synapses . To generalize the role of GluClα function for ON responses , we next asked how the output of the system is affected in a full GluClα loss-of-function by recording T4/T5 responses in the GluClαFlpStop . D mutant . When imaging flies expressing GCaMP6f in both T4 and T5 , individual cell type responses can be separated by showing individual moving ON and OFF edges that activate T4 and T5 , respectively ( Figure 8A; Fisher et al . , 2015b; Maisak et al . , 2013 ) . T4/T5 neurons project to one of the four layers of the lobula plate , and the four layers show distinct directional tuning . In heterozygous controls , T4/T5 neurons responded to both moving ON and OFF edges and the four layers responded preferentially to front to back ( layer A ) , back to front ( layer B ) , upward ( layer C ) , and downward ( layer D ) motion ( Figure 8A ) . Responses to ON edges were completely abolished for motion in all directions in the GluClα null mutant , showing that all T4 inputs depend on GluClα ( Figure 8B , C ) . In contrast , T4/T5 neurons still responded to OFF edges moving in different directions ( Figure 8A–D ) . Both response amplitude and direction selectivity of the OFF response were unaffected in GluClα mutants ( Figure 8D ) . When recording responses to light flashes , T4/T5 axon terminals also no longer showed an increase in calcium in response to the ON step . Interestingly , the calcium signal even decreased , indicating inhibition ( Figure 8E , G ) . Inhibition was previously shown to be an important part of motion computation ( Fisher et al . , 2015a; Gruntman et al . , 2018; Haag et al . , 2016; Leong et al . , 2016; Salazar-Gatzimas et al . , 2016 ) . This could argue that , in the absence of all ON inputs , feed-forward inhibition onto T4 is revealed . Alternatively , T5 neurons might have lost rectification and respond to ON with a decrease in calcium . To distinguish between these two possibilities , we recorded flash responses in layer M10 of the medulla . Here , T4 and T5 projections do not overlap and calcium signals will stem exclusively from T4 dendrites . We found that T4 dendrites also show a negative calcium signal in response to ON ( Figure 8F , H ) , revealing that this inhibition is present in T4 neurons but masked in the presence of GluClα . Furthermore , there was an increase in calcium signal during OFF in T4 dendrites ( Figure 8F ) , suggesting a loss of GluClα-dependent inhibition that is normally active during OFF . Taken together , our results show that GluClα function is critical for ON selectivity in the fly visual system . Our work shows that visual responses in the first ON-selective neuron of the Drosophila visual system uses a combination of GluClα and Rdl receptors . This reveals a new biophysical mechanism through which ON and OFF pathway dichotomy can be established . While pharmacology can be used to deduce the function of specific molecular mechanisms , these approaches are often not specific to one protein . GluCls and GABARs belong to the same receptor family of ligand-gated chloride channels and have closely related structure and phylogeny ( Betz , 1990; Lynagh et al . , 2015 ) . All known noncompetitive antagonists like Picrotoxin , γ-HCH , dieldrin , EBOB and fibronil target both receptor types although the actions are weaker in GluCls compared to GABARs ( Eguchi et al . , 2006 ) . Along these lines , PTX was thought to affect GABAA receptor at low concentrations , and additionally affect GluCls at high concentrations in vitro and in vivo ( Liu and Wilson , 2013; McCavera et al . , 2009; Takeuchi and Takeuchi , 1969; Wilson and Laurent , 2005 ) . Here , the use of PTX-insensitive alleles for glutamate and GABA-gated chloride channels allowed us to deduce that , in vivo , GluClα is already blocked by PTX at lower concentrations than previously thought , and that both GluClα and Rdl play critical roles for ON responses in the Drosophila visual system . These pharmacogenetic experiments using toxin-insensitive alleles prove to be a powerful tool to unambiguously assign specific effects to individual channels . One benefit of the use of two inhibitory transmitter systems might be the distribution of sensory coding across parallel synapses . GluClα and Rdl also appear to have very different channel dynamics ( Cully et al . , 1996; Ffrench-Constant et al . , 1993 ) . Interestingly , PTX-insensitive GluClα and Rdl alleles predominantly rescue different aspects of the visual responses . Whereas GluClαS278T predominantly rescued the peak response in all medulla layers , RdlMDRR mainly rescued the plateau response . This is consistent with our results and with previous oocyte recordings revealing that GluClα is fast desensitizing ( Figure 5; Cully et al . , 1996 ) . It is also consistent with in vivo recordings of inhibitory glutamate currents in the honeybee ( Barbara et al . , 2005 ) . In contrast , GABA receptors stay open throughout the period in which the transmitter is present ( Ffrench-Constant et al . , 1993 ) . Thus , the use of different inhibitory receptors might allow different aspects of a temporally structured stimulus to be encoded . This is consistent with the finding that two different types of inhibition are also in place in the vertebrate retina . There , GABAergic and glycinergic inhibition diversify the response properties of bipolar cells through a direct influence on temporal and spatial features ( Franke et al . , 2017 ) . While both receptors appear to be broadly expressed in many cell types of the visual system , they could be co-expressed with different transporters and channels , and interact with different molecular partners , further diversifying their role . Another common strategy to generate functional diversity is the bringing together of different receptor subunits with certain homology . Both mammalian GlyR and GABAA receptors can function as hetero-oligomers made up of different subunits and thus generating functional diversity ( Betz , 1990 ) . There are at least three different GluCl subtypes in C . elegans that can be combined ( Cully et al . , 1994; Horoszok et al . , 2001 ) . In Drosophila , only one gene coding for a glutamate-gated chloride channel has been identified . Although alternative splicing and post-transcriptional modifications could alter channel function , all known isoforms are identical in their functional domains . However , heteropentameric channels composed of mixed Rdl and GluClα subunits have been suggested biochemically ( Ludmerer et al . , 2002 ) . Such a potential presence of hybrid channels might also explain the higher in vivo sensitivity of GluClα to PTX in some cell types ( Figure 6 , S7 ) . Finally , two distinct inhibitory transmitter systems might be suitable for individual changes during evolution , allowing for adaptation to specific contextual constraints . Our experiments revealed that GluClα is not exclusively required in a cell-autonomous manner for ON responses , since loss of GluClα function in Mi1 or Tm3 individually does not lead to a loss of ON responses . It is unlikely that this is due to an incomplete loss of function , since independent genetic tools ( FlpStop and RNAi ) that both disrupted GluClα expression substantially at the mRNA level ( Figure 7 ) gave the same result . Furthermore , the same FlpStop allele effectively abolished all ON responses when GluClα function was disrupted within its entire expression pattern . Additionally , a PTX-resistant Rdl channel can mediate ON responses in a PTX background , although L1 is not GABAergic . Together , these results suggest that ON selectivity is not a monosynaptic computation , but that parallel functional pathways can even compensate for the loss of the major synaptic connection that links L1 directly to Mi1 or Tm3 . Thus , the emergence of ON selectivity is more distributed than suggested by minimal core circuit motifs . One synaptic layer further downstream , optogenetic activation of Mi1 and Tm3 most strongly contributes to T4/T5 responses ( Strother et al . , 2017 ) . However , our data further show that T4/T5 neurons still respond to ON stimuli when both Mi1 and Tm3 responses are completely blocked by PTX , arguing that other neurons also significantly contribute to T4/T5 responses under visual stimulation and suggesting that coding is again more distributed at this stage . Based on connectomics , one can speculate about candidates for the implementation of these parallel circuit motifs between L1 and Mi1 and Tm3 . The lamina neuron L5 and the GABAergic feedback neurons C2 and C3 receive L1 inputs and could be part of an interconnected local microcircuit ( Takemura et al . , 2013 ) . Intercolumnar neurons , not present in the current connectome datasets , like Pm or Dm neurons , might also be involved and are likely glutamatergic ( Davis et al . , 2018; Raghu and Borst , 2011 ) . In fact , there are close to 100 cell types in the visual system and ~60 medulla neurons , but their role is so far unknown . Sensory pathway splits in the periphery are one of the most fundamental steps in sensory processing . Turning this into a process that parallel pathways can achieve might make this important feature extraction step robust to perturbations . T4 flash responses in a GluClα-deficient background show an increase in calcium signal during the OFF epoch and a decrease during the ON epoch ( Figure 8 ) . For a long time , the mechanisms that generate direction-selective responses in T4/T5 neurons were thought to rely on feedforward excitatory mechanisms ( Fisher et al . , 2015b; Silies et al . , 2014; Yang and Clandinin , 2018 ) . Recently , it was suggested that these direction-selective cells in the fly visual system also implement mechanisms that rely on null-direction suppression ( Haag et al . , 2016; Leong et al . , 2016; Salazar-Gatzimas et al . , 2016 ) . Whereas electrophysiological recordings showed inhibition in T4 when the trailing edge of the receptive field was specifically stimulated ( Gruntman et al . , 2018 ) , whole-cell recording experiments of T4/T5 neurons are daunting and this is the first time that calcium imaging data directly reveals inhibition in response to single ON flashes . Since glutamatergic inhibition via GluClα was disrupted in this experimental context , our data suggests that this is due to GABAergic inhibition . Several neuronal candidates could make inhibitory synapses onto T4 dendrites . Based on connectomics and neurotransmitter identity , neurons like Mi4 , C3 , CT1 or TmY15 give direct input and are GABAergic ( Meier and Borst , 2019; Takemura et al . , 2017 ) . Alternatively , this decrease in calcium signal in T4 might come from a lack of excitatory inputs in a GluClα mutant background . Interestingly , Mi1 and Tm3 themselves show inhibition in response to light when GluClα is blocked . However , this effect is more pronounced at their dendrites than in their output layer and shows different kinetics . Our work might thus help uncover a GABAergic inhibitory input to T4 that is more strongly apparent in the absence of Mi1 and Tm3 excitation , and could ultimately reveal the circuit implementation for the inhibitory component of T4/T5 receptive fields ( Leong et al . , 2016; Salazar-Gatzimas et al . , 2016 ) . Furthermore , our data also reveals an increase in calcium during OFF stimulation . The major inputs to T4 are themselves rectified ( Behnia et al . , 2014 ) . However , rectification in T4 might not be purely inherited by its inputs but also further strengthened at the T4 dendrites . Our findings thus suggest that glutamatergic inhibition contributes to establishing or maintaining contrast selectivity in T4 . Both GluClα and Rdl are ionotropic ligand-gated receptors . While ionotropic receptors also implement the ON and OFF pathway split in C . elegans chemosensation ( Chalasani et al . , 2007 ) , examples in vertebrate vision , olfaction and gustation require metabotropic receptors ( Chandrashekar et al . , 2006; Masu et al . , 1995; Nei et al . , 2008 ) . Ionotropic receptors appear to be more common in insects than in vertebrates ( Silbering and Benton , 2010 ) . Furthermore , glutamate-gated chloride channels have independently arisen three times within invertebrate clades and are present in arthropods , molluscs and flatworms ( Lynagh et al . , 2015 ) , arguing for a strong evolutionary benefit . Ionotropic receptors mediate rapid transduction events at scales smaller than a millisecond , whereas metabotropic ones are in the millisecond to second range and last longer , from seconds to several minutes , due to an enzymatic secondary cascade previous to channel opening ( Betz , 1990; Shiells , 1994 ) . The evolutionary choice of the specific glutamatergic inhibitory system needs to match the sensory processing speed required for accurate behavioral responses in these species . For example , at the photoreceptor level , invertebrate phototransduction is faster than vertebrate phototransduction thanks to sophisticated molecular strategies ( Hardie and Raghu , 2001; Katz and Minke , 2009 ) . Also , the latency of olfactory sensory neurons responses in mammals is longer than that observed in insects ( Sato et al . , 2008; Silbering and Benton , 2010 ) . One advantage that metabotropic receptors have over ionotropic receptors is further amplification of the signal ( Shiells , 1994 ) . The distributed circuit architecture proposed here might therefore strengthen signaling in a system that uses ionotropic signaling . Here we showed that ON selectivity is not a monosynaptic process as described in other systems ( Chalasani et al . , 2007; Masu et al . , 1995 ) . Although acute pharmacological block or a systemic loss of function of GluClα abolished all ON responses in different neurons , cell-type-specific mutants retained intact ON responses , revealing that sensory coding is distributed in the fly visual system . This not only highlights the power of fly genetics but sheds new light onto the mechanisms of ON selectivity in other systems , since conclusions about ON and OFF pathway splits being mediated by specific monosynaptic processes in systems such as the vertebrate retina or the C . elegans chemosensory system relied on systemic loss-of-function approaches ( Chalasani et al . , 2007; Masu et al . , 1995 ) . Several of these systems allow for cell-type-specific manipulations using genetic approaches . It will be interesting to revisit these systems and ask if coding is similarly distributed across multiple synapses in different sensory systems and organisms . Flies were raised at 25°C and 55% humidity on molasses-based food on a 12:12 hr light:dark cycle . Imaging experiments were done at room temperature ( 20°C ) . Genotypes of all Drosophila strains used for experiments are listed in the Key Resources Table . Female flies were dissected 3–5 days after eclosion . Brains were removed in dissection solution and fixed in 2% paraformaldehyde in phosphate buffered lysine ( PBL ) for 50 min at room temperature . Subsequently , the brains were washed 3x for 5 min in phosphate buffered saline containing 0 . 3% Triton X-100 ( PBT ) adjusted to pH 7 . 2 . For antibody staining , the samples were blocked in 10% normal goat serum ( NGS , Fisher Scientific GmbH , Schwerte , Germany ) in PBT for 30 min at room temperature followed by incubation for 24 hr at 4°C in the primary antibody solution ( mouse mAb nc82 , 1:25 , DSHB; chicken anti-GFP , 1:2000 , Abcam ab13970; rabbit anti-GABA , 1:200 , Sigma-Aldrich , A2052 ) . Primary antibodies were removed by washing in PBT 3 times for 5 min and the brains were incubated in the secondary antibody ( anti-chicken-Alexa488 , anti-mouse-Alexa594 , anti-rabbit-Alexa594 , all 1:200 , Dianova ) in the dark at 4°C overnight . The samples were further washed with PBT ( 3 × 5 min ) and mounted in Vectashield ( Vector Laboratories , Burlingame ) . Serial optical sections were taken on a Zeiss LSM710 microscope ( Carl Zeiss Microscopy GmbH , Germany ) equipped with an oil immersion Plan-Apochromat 40x ( NA = 1 . 3 ) objective and using the Zen 2 Blue Edition software ( Carl Zeiss Microscopy , LLC , United States ) . Z-stack images were taken at 1 µm intervals and 512 × 512 pixel resolution . Confocal stacks were rendered into two-dimensional images using Fiji ( Schindelin et al . , 2012 ) . The images were then further processed using Illustrator CS5 . 1 ( Adobe ) or Inkscape version 0 . 92 . 1 ( The Inkscape Team ) . Transgenic lines carrying the FlpStop cassette ( Fisher et al . , 2017 ) for conditional gene control were generated according to standard procedures . In brief , embryos carrying the Mi02890 insertion ( y[1] w[*]; Mi{y[+mDint2]=MIC}GluClαMI02890/TM3 , Sb1 ) were injected with the FlpStop cassette and PhiC31 integrase . Embryos were dechorionated in 50% bleach ( DanKlorix ) for 3 min , followed by washing in a buffer ( 100 mM NaCl , 0 . 02% Triton X-100 ) for 3 min . Injections were done on a Nikon AZ100 microscope using a FemtoJet 4i ( Eppendorf AG , Hamburg , Germany ) . The injection mix ( 20 µl ) consisted of 10 µg of the FlpStop construct , 6 µg of helper DNA ( pBS130 containing the PhiC31 integrase ) and 4 µl of 5x injection buffer ( 25 mM KCl , 0 . 5 mM NaH2PO4 , pH 6 . 8 , 1% phenol red [Sigma Aldrich] ) . Injection needles were pulled from quartz glass microcapillaries ( 10 cm length , 1 . 0 mm outside diameter , 0 . 5 mm inside diameter , Sutter Instruments , USA ) using a P-2000 micropipette puller ( Sutter Instruments , USA ) . Needles were sharpened using a capillary grinder ( Bachofer , Germany ) . After injection , embryos were covered with 10S Voltalef oil and incubated at 18°C until larval hatching . Successful recombinase-mediated cassette exchange was scored by the loss of the yellow marker ( y[+] ) of the MiMIC cassette and verified by single fly PCR , testing for the loss of the MiMIC cassette and the orientation of the inserted FlpSTOP cassette , as in Fisher et al . ( 2017 ) . Raw sequencing reads and TPM tables from published datasets were taken from ( Konstantinides et al . , 2018; GSE 103772 ) and ( Davis et al . , 2018; GSE 116969 ) . To estimate transcript abundance , we used Kallisto ( v0 . 43 . 1; Bray et al . , 2016 ) to pseudo-align reads to dm6 annotation ( ENSEMBLE release 91 derived from FlyBase release version 2017_04 ) . The TPM matrix was processed further in R studio ( R version 3 . 4 . 4 ) . The TPMs were summarized at the level of genes averaged across cell type replicates . Heat maps of the gene expression in selected cell types was generated using MultiExperiment Viewer ( MeV ) 4 . 9 . 0 ( Howe et al . , 2011 ) . All data processing was performed offline using MATLAB R2017a ( The MathWorks Inc , Natick , MA ) . To correct for motion artifact , individual images were aligned to a reference image composed of a maximum intensity projection of the first 30 frames . The average intensity for manually selected ROIs was computed for each imaging frame and background subtracted to generate a time-trace of the response . ROI identities were kept for matching identical ROIs before and after toxin application for paired analysis . All responses and visual stimuli were interpolated at 10 Hz and trial averaged . Neural responses are shown as relative fluorescence intensity changes over time ( ΔF/F0 ) . The mean and the standard error of the mean ( SEM ) were calculated across flies after averaging over ROIs for each fly . A two-tailed Student t test for paired or unpaired ( independent ) samples was used for statistical analysis between two groups . For comparisons between more than two groups in which one independent variable was manipulated ( here: PTX concentrations ) , one-way ANOVA followed by two-tailed t-test with Bonferroni-Holm correction for multiple comparisons was used . For multiple comparisons between groups in which two independent variables were manipulated ( here: PTX concentration and genotype ) , an unbalanced two-way ANOVA followed by Tukey’s Honestly Significant Difference Procedure for multiple comparisons was used . For all the data , normality was tested with a Lilliefors test . For electrophysiological recordings , GluClα constructs ( isoform O , NP_001287409 ) were heterologously expressed in Xenopus laevis oocytes . Oocytes were harvested from our own colony . Frogs were housed according to the German law of animal protection and the district veterinary office . Oocytes were harvested following standard procedures and in agreement with the animal testing approval 84–02 . 04 . 2016 . A077 . GluClα constructs were cloned in the pGEMHE vector . The vector was linearized with NheI and transcribed using the T7 mMessage mMachine kit ( Ambion , Austin , TX ) . Xenopus oocytes were injected with 50 nl RNA ( 0 . 01–0 . 2 μg/µl ) and incubated at 14–16°C for 1–2 days in ND96 medium containing ( in mM ) : 96 NaCl , 2 KCl , 1 . 8 CaCl2 , 1 MgCl2 , 10 4- ( 2-hydroxyethyl ) piperazine-1-ethanesulfonic acid ( HEPES ) , 5 Na-pyruvate , and 100 mg/l gentamicin , adjusted to pH 7 . 5 with NaOH . Electrophysiological experiments were performed at room temperature ( 22–25°C ) . Oocytes were placed in an RC-3Z recording chamber ( Warner Instruments , Hamden , CT ) under a Discovery V8 stereoscope ( Zeiss , Oberkochen , Germany ) and continuously perfused with ND96 by a PC-controlled gravity-driven system with a flux rate of 7 ml/min . Electrodes were pulled from 1 . 5 mm thick borosilicate glass capillaries ( Hilgenberg , Malsfeld , Germany ) on a DMZ puller ( Zeitz Instruments GmbH , Martinsried , Germany ) and filled with 3 M KCl . The resulting initial electrode resistance was 0 . 5–5 MΩ in ND96 . Currents were recorded in the two-electrode voltage-clamp mode at a holding potential of −70 mV with a Gene Clamp 500 amplifier ( Molecular Devices , San Jose , CA ) , connected via a USB-6341 acquisition board ( National Instruments , Austin , TX ) to a PC running WinWCP ( Strathclyde , University of Glasgow , UK ) . L-glutamate was dissolved in ND96 and was repeatedly applied for 20 s , with an interstimulus interval of 1–2 min to ensure full recovery from desensitization . Picrotoxin stock solution was first prepared in dimethyl sulfoxide ( DMSO ) and then diluted in ND96 . The peak current amplitude of the glutamate-evoked response , in the presence of the antagonist , was normalized to the mean peak current amplitude evoked by glutamate after picrotoxin wash-out . Data are shown as the mean ± SD . N indicates the number of cells . Data was analyzed with Igor Pro ( Wavemetrics , Portland , OR ) . An unpaired t test was used for statistical analysis of the residual current between WT and the S278T mutant allele .	We rely on our senses to capture information about the world around us . Sense organs convert sensory information – such as light or sound waves – into patterns of neuronal activity . In the mammalian retina , for example , specialized neurons called photoreceptors detect individual photons of light as they hit the back of the eye . The photoreceptors then pass on this information to neurons called bipolar cells for further processing . During darkness , all photoreceptors release the same chemical signal onto bipolar cells , namely a molecule called glutamate . But bipolar cells respond to glutamate in different ways depending on which proteins are present in their outer membrane . So-called ON cells respond to glutamate by decreasing their activity , and thus effectively become more active when light levels increase . By contrast , OFF cells respond to glutamate by increasing their activity . This ON/OFF binary code enables later stages of the visual system to detect more complex visual features , such as shape and movement . A new study in fruit flies , however , suggests that the ON/OFF code may be more complex than previously thought . While fruit fly eyes look very different to our own , the two have much in common . By studying fruit flies , researchers can also take advantage of a variety of genetic and pharmacological tools to manipulate cells and neuronal circuits . Using such tools , Molina-Obando et al . show that the ON/OFF signal separation in fruit flies uses two different molecular mechanisms . The first involves a gene called GluCl-alpha , which encodes a receptor for glutamate . The second involves a gene called Rdl , which encodes a receptor for another brain chemical , GABA . Deleting the gene for GluCl-alpha from the entire fly brain prevented ON cells from responding to an increase in light levels . However , deleting this gene from specific ON cells alone did not . This suggests that flies can use more than one type of neuronal connection to detect an increase in light . Moreover , if one pathway fails , the other can take over . This makes the system more robust . The results of Molina-Obando et al . are consistent with findings from anatomical studies that have mapped connections between neurons . Future studies should explore whether the same mechanisms exist in other sensory systems , and other animals . These experiments could take advantage of the molecular tools developed as part of the current work , which allow precise manipulation of neural networks .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2019	ON selectivity in the Drosophila visual system is a multisynaptic process involving both glutamatergic and GABAergic inhibition
Reverse replay of hippocampal place cells occurs frequently at rewarded locations , suggesting its contribution to goal-directed path learning . Symmetric spike-timing dependent plasticity ( STDP ) in CA3 likely potentiates recurrent synapses for both forward ( start to goal ) and reverse ( goal to start ) replays during sequential activation of place cells . However , how reverse replay selectively strengthens forward synaptic pathway is unclear . Here , we show computationally that firing sequences bias synaptic transmissions to the opposite direction of propagation under symmetric STDP in the co-presence of short-term synaptic depression or afterdepolarization . We demonstrate that significant biases are created in biologically realistic simulation settings , and this bias enables reverse replay to enhance goal-directed spatial memory on a W-maze . Further , we show that essentially the same mechanism works in a two-dimensional open field . Our model for the first time provides the mechanistic account for the way reverse replay contributes to hippocampal sequence learning for reward-seeking spatial navigation . The hippocampus plays an important role in episodic memory and spatial processing in the brain ( O'Keefe and Dostrovsky , 1971; Scoville and Milner , 1957 ) . Because an episode is a sequence of events , sequential neural activity has been extensively studied as the basis of hippocampal memory processing . In the rodent hippocampus , firing sequences of place cells are replayed during awake immobility and sleep ( Carr et al . , 2011; Pfeiffer , 2018 ) and these reactivations are crucial for performance in spatial memory tasks ( Girardeau et al . , 2009; Jadhav et al . , 2012; Singer et al . , 2013 ) . Replay can be either in the same firing order as experienced ( forward replay ) or in the reversed order ( reverse replay ) . Forward replay is observed during sleep after exploration ( Lee and Wilson , 2002; Wikenheiser and Redish , 2013 ) or in immobile states before rats start to travel towards reward ( Diba and Buzsáki , 2007; Pfeiffer and Foster , 2013 ) , hence forward replay is thought to engage in the consolidation and retrieval of spatial memory . In contrast , reverse replay presumably contributes to the optimization of goal-directed paths because rewarded spatial paths are replayed around the timing of reward delivery ( Diba and Buzsáki , 2007; Foster and Wilson , 2006 ) , and the occurrence frequency is modulated by the presence and the amount of reward ( Ambrose et al . , 2016; Singer and Frank , 2009 ) . Because sequences are essentially time asymmetric , sequence learning often hypothesizes asymmetric spike-timing-dependent plasticity ( STDP ) found in CA1 ( Bi and Poo , 2001 ) which induce long-term potentiation ( LTP ) for pre-to-post firing order and long-term depression ( LTD ) for post-to-pre firing order . This type of STDP enables a recurrent network to reactivate sequential firing in the same order with the experience . However , in the hippocampal area CA3 , it was recently reported that the default form of STDP is time symmetric at recurrent synapses ( Mishra et al . , 2016 ) . Because CA3 is the most likely source of hippocampal firing sequences ( Middleton and McHugh , 2016; Nakashiba et al . , 2009 ) , this finding raises the question whether and how STDP underlies sequence learning in hippocampus . A symmetric time window implies that a firing sequence equally strengthens both forward synaptic pathways leading to the rewarded location and reverse pathways leaving away from the rewarded location in CA3 recurrent network . However , reward-based optimization requires selective reinforcement of forward pathways as it will strengthen prospective place-cell sequences in subsequent trials and forward replay events in the consolidation phase . Such bias toward forward sequences in post-experience sleep ( Wikenheiser and Redish , 2013 ) and goal-directed behavior ( Johnson and Redish , 2007; Pfeiffer and Foster , 2013; Wikenheiser and Redish , 2015 ) is actually observed in hippocampus . How this directionality arises in replay events and how reverse replay enables the learning of goal-directed navigation remain unclear . In this paper , we first show how goal-directed path learning is naturally realized through reverse replay in a one-dimensional chain model . To this end , we hypothesize that the contribution of presynaptic spiking for STDP is attenuated in CA3 by short-term depression , as was revealed in the rat visual cortex ( Froemke et al . , 2006 ) . Under this condition , symmetric STDP and a rate-based Hebbian plasticity rule bias recurrent synaptic weights toward the opposite direction to the propagation of a firing sequence , implying that the combined rule virtually acts like anti-Hebbian STDP . We also show that accumulation of afterdepolarization ( Mishra et al . , 2016 ) in postsynaptic neurons results in the same effect . By simulating the model with various spiking patterns , we confirm this effect for a broad range of spiking patterns and parameters of plasticity rules , including those observed in experiments . Based on this mechanism , we built a two-dimensional recurrent network model of place cells with the combination of reverse replay , Hebbian plasticity with short-term plasticity , and reward-induced enhancement of replay frequency ( Ambrose et al . , 2016; Singer and Frank , 2009 ) . We first demonstrate that the network model can learn forward pathways leading to reward on a W-shaped track . Further , we extend the role of reverse replay to unbiased sequence propagations from reward sites on a two-dimensional open field , which enable learning of goal-directed behavior in the open field . Unlike the previous models for hippocampal sequence learning ( Blum and Abbott , 1996; Gerstner and Abbott , 1997; Jahnke et al . , 2015; Jensen and Lisman , 1996; Sato and Yamaguchi , 2003; Tsodyks and Sejnowski , 1995 ) in which recurrent networks learn and strengthen forward sequences through forward movements , our model proposes goal-directed path learning through reverse sequences . We first simulated a sequential firing pattern that propagates through a one-dimensional recurrent neural network , and evaluated weight changes by rate-based Hebbian plasticity rules . The network consists of 500 rate neurons , which were connected with distance-dependent excitatory synaptic weights modulated by short-term synaptic plasticity ( STP ) ( Romani and Tsodyks , 2015; Wang et al . , 2015 ) . In addition , the network had global inhibitory feedback to all neurons . A first external input to a neuron at one end ( #0 ) elicited traveling waves of neural activity propagating to the opposite end ( Figure 1A ) . Here , we regard these activity patterns as a model of hippocampal firing sequences ( Romani and Tsodyks , 2015; Wang et al . , 2015 ) . A second external input to a neuron at the center of the network triggered firing sequences propagating to both directions ( Figure 1A ) because synaptic weights were symmetric ( Figure 1B ) . Here , we implemented a standard Hebbian plasticity rule , which potentiated excitatory synaptic weights by the product of postsynaptic and presynaptic neural activities . During the propagation of the first unidirectional sequence , this rule potentiated synaptic weights symmetrically without creating any bias in the synaptic weights ( Figure 1B ) . However , when synaptic weights were changed by a modified Hebbian plasticity rule ( see Materials and methods ) , in which the long-term plasticity is also regulated by STP at presynaptic terminals ( Froemke et al . , 2006 ) , the second firing sequence only propagated to the reverse direction of the first one ( Figure 1C ) . This selective propagation occurred because the first firing sequence potentiated synaptic weights asymmetrically in the forward and reverse directions , thus creating a bias in the spatial distribution of synaptic weights ( Figure 1D ) . This means that firing sequences strengthen the reverse synaptic transmissions more strongly than the forward ones in this model , and reverse sequences are more likely to be generated after forward sequences . Why does this model generate such a bias to the reverse direction ? To explain this , we show the packet of neural activity during the first firing sequence ( at t=300 ms ) in Figure 1E , top . We also plotted neurotransmitter release from the presynaptic terminal of each neuron ( presynaptic outputs ) , which is determined by the product of presynaptic neural activity and the amount of available neurotransmitters in the combined Hebbian rule ( Figure 1E , bottom ) . Because neurotransmitters are exhausted in the tail of the activity packet , presynaptic outputs are effective only at the head of the activity packet . Due to this spatial asymmetricity of presynaptic outputs , connections from the head to the tail are strongly potentiated but those from the tail to the head are not ( Figure 1F ) . This results in the biased potentiation of reverse synaptic transmissions in this model ( Figure 1G ) . In other words , the packet of presynaptic outputs becomes somewhat ‘prospective’ ( namely , slightly skewed toward the direction of activity propagation ) , so the weight changes based on the coincidences between presynaptic outputs and postsynaptic activities result in the selective potentiation of connections from the ‘future’ to the ‘past’ in the firing sequence . This mechanism was not known previously and enables forward sequences to potentiate the synaptic transmissions responsible for reverse replay , and vice versa . In the above simulations , STP contributed crucially to the asymmetric potentiation of synaptic connections between neurons . Meanwhile , similar asymmetricity , and hence the potentiation of reverse synaptic transmission , can also emerge when the effect of postsynaptic activity on Hebbian plasticity accumulates through time ( Figure 1—figure supplement 1 ) . Because STDP in CA3 depends on the afterdepolarization ( ADP ) in dendrites ( Mishra et al . , 2016 ) , this phenomenon will occur if the effect of ADPs accumulates over multiples postsynaptic spikes as afterhyperpolarization modulates STDP in the visual cortex ( Froemke et al . , 2006 ) . Because we can obtain essentially the same results for STP-dependent and ADP-dependent plasticity rules , we only consider the effect of STP in the rest of this paper . The potentiation of reverse synaptic transmissions also occurs robustly in spiking neurons with STDP . To show this , we constructed a one-dimensional recurrent network of Izhikevich neurons ( Izhikevich , 2003b , 2004 ) connected via conductance-based AMPA and NMDA synaptic currents ( see Materials and methods ) . We chose Izhikevich model because its frequency adaptation induces instability and enhances the generation of moving activity bumps . However , we note that the learning mechanism itself does not depend on a specific model of spiking neurons . Initial synaptic weights , STP , and global inhibitory feedback were similar to those used in the rate neuron model . We tested two types of STDP: asymmetric STDP in which pre-to-post firing leads to potentiation and post-to-pre firing leads to depression , and symmetric STDP in which both firing orders result in potentiation if two spikes are temporally close or depression if two spikes are temporally distant . Experimental evidence suggests that recurrent synapses in hippocampal CA3 undergo symmetric STDP ( Mishra et al . , 2016 ) . Among these STDP types , only symmetric STDP showed a similar effect to the rate-based Hebbian rule . Under symmetric STDP without modulation by STP , synaptic transmissions were potentiated in both directions ( Figure 2A and B ) . However , symmetric STDP modulated by STP biased the weight changes to the reverse direction ( Figure 2C ) , and accordingly the second firing sequence selectively traveled towards the opposite direction to the first sequence ( Figure 2D ) . This effect did not depend significantly on synaptic time constants , and we could obtain the same effect when we turned off NMDA current and shortened time constants for AMPA and inhibition ( Figure 2—figure supplement 1A , B ) . In contrast , under asymmetric STDP without modulation by STP , firing sequences strengthened the forward sequence propagation ( Figure 2—figure supplement 1C ) and the related synaptic connections ( Figure 2—figure supplement 1D ) , as expected . Introducing modulation by STP did not change the qualitative results ( Figure 2—figure supplement 1E , F ) . These results indicate that a greater potentiation of reverse synaptic transmissions in CA3 occurs under the modified symmetric STDP . We further confirmed the bias effect of the modified symmetric STDP in broader conditions than in the above network simulation . To this end , we generated sequential firing patterns along the one-dimensional network by sampling from a Poisson process , while manually controlling the number of propagating spikes per neuron , the mean inter-spike interval ( ISI ) of Poisson input spike trains , and time lags in spike propagation ( i . e . time difference between the first spikes of neighboring neurons ) ( Figure 3A ) . The amount of neurotransmitter release by each presynaptic spike was calculated by the STP rule ( see Materials and methods ) , and the magnitude of long-term synaptic changes was calculated by a Gaussian-shaped symmetric STDP ( Figure 3B ) ( Mishra et al . , 2016 ) . The net effect of synaptic plasticity was given as the product of the two quantities , as in the previous simulations . The parameter values of short-term and long-term plasticity were adopted from experimental results ( Figure 3B , C ) ( Guzman et al . , 2016; Mishra et al . , 2016 ) . We calculated the long-term weight changes in synapses sent from the central neuron in the network , and defined a weight bias as the difference in synaptic weights between forward and reverse directions ( in which positive values mean bias to the reverse direction ) . We then obtained the mean bias and the fraction of positive biases ( P ( bias >0 ) ) over 100 different realizations of spike trains generated with the same parameter values . For each number of spikes per neuron ( 2 , 3 , 4 and 5 spikes ) , we ran 1000 simulations using different mean ISIs and time lags randomly sampled from the interval [5 , 50 ms] . Significant biases toward the reverse direction were observed in broad simulation conditions ( Figure 3D ) . The biases were statistically significant ( p<0 . 01 in Wilcoxon signed rank test for mean bias or binomial test for P ( bias >0 ) ) already in a part of conditions with two spikes per neuron , and the effect became prominent as the number of spikes increased . In general , the magnitude of synaptic changes is larger for a faster spike propagation ( Figure 3D , Table 1 ) , which is reasonable because the potentiation of symmetric STDP becomes stronger as presynaptic firing and postsynaptic firing get closer in time . On the other hand , P ( bias >0 ) was greater for a smaller mean ISI , and it did not depend on the propagation speed ( Figure 3D , Table 1 ) . Especially , all simulations showed statistically significant biases to the reverse direction regardless of the propagation speed when the number of spikes per neuron is 4 or five and the mean ISI <20 ms . This implies that the bias toward the reverse direction is the most prominent when the neural network propagates a sequence of bursts with intraburst ISIs less than 20 ms . Such bursting is actually observed in CA3 in vivo ( Mizuseki et al . , 2012 ) and simulations of a CA3 recurrent network model suggest that bursting plays a crucial role in propagation of firing sequences ( Omura et al . , 2015 ) . Parameters that regulate STP also influence the bias toward the reverse direction . Actually , these parameters largely change in the hippocampus depending on experimental settings ( Guzman et al . , 2016 ) . Therefore , we also performed simulations with randomly sampled values of the initial release probability of neurotransmitters ( U ) and the time constants of short-term depression and facilitation ( τSTD and τSTF ) . Here , the number of spikes per neuron was fixed to five , and the ISI and time lag were independently sampled from the interval [5 , 20 ms] in every trial . As shown in Figure 4 and Table 1 , both mean bias and P ( bias>0 ) clearly depend on the initial release probability . Prominent bias was observed for U>0 . 3 , and it gradually disappeared as U was decreased . The bias was weakly correlated with τSTD , but there was almost no correlation between the bias and τSTF ( Table 1 ) . In sum , the bias toward the reverse direction occurred robustly for a sufficiently high intraburst firing frequency and a sufficiently high release probability of neurotransmitters . In contrast to symmetric STDP , asymmetric STDP was not effective in potentiating reversed synaptic transmissions even if it was modulated by STP . In our simulations with parameters taken from experiments in CA1 ( Bi and Poo , 2001 ) , such a potentiation effect was never observed for asymmetric STDP with all-to-all spike coupling for any parameter value of STP and the number of spikes per neuron ( 5 or 15 spikes ) ( Figure 4—figure supplement 1 ) . However , asymmetric STDP with nearest-neighbor spike coupling ( Izhikevich and Desai , 2003a ) , in which only the nearest postsynaptic spikes before and after a presynaptic spike were taken into account , generated statistically significant biases toward the reverse direction in some parameter region ( Figure 4—figure supplement 1 ) . In this case , large biases required large values of U and τSTD , and a large number of spikes per neuron ( 15 spikes ) . Thus , the condition that asymmetric STDP generates the biases to the reverse direction is severely limited , although we cannot exclude this possibility . We also tested whether the realistic activity pattern of place cells during run can induce the directional bias . When a rat passes through place fields of CA3 place cells , they typically shows bell-shape activity patterns duration of which is about 1 s and mean peak firing rate is about 13 Hz ( Mizuseki and Buzsáki , 2013; Mizuseki et al . , 2012 ) . Firing of place cells is phase-locked to theta oscillation and the firing phase gradually advances as a rat moves through place fields ( theta-phase precession ) . Furthermore , firing sequences of place cells scan the path from behind to ahead of the rat in every theta cycle , which is a phenomenon called theta sequence ( Dragoi and Buzsáki , 2006; Foster and Wilson , 2007; Huxter et al . , 2008; O'Keefe and Recce , 1993; Wang et al . , 2015; Wikenheiser and Redish , 2015 ) . Some models of hippocampal sequence learning hypothesized that these compressed sequential activity patterns enhance memory formation through STDP ( Jensen and Lisman , 1996 , 2005; Sato and Yamaguchi , 2003 ) . Here , we simulated weight biases induced by Poisson spike trains that mimic place-cell activities when a rat is running through 81 equidistantly-spaced place fields . We modified both mean peak firing rates and magnitude of theta phase-locking ( phase selectivity ) , as shown in Figure 5A ( see Materials and methods ) . We found that weight biases primarily depended on coarse-grained firing rates , but phase selectivity had no significant effect in our model ( Figure 5B , Table 2 ) . Noticeable effects of phase selectivity on weight biases emerged for a narrow STDP time window of 10 ms ( Figure 5—figure supplement 1 , Table 2 ) , suggesting that experimentally observed broad time window of STDP ( 70 ms ) masks small differences in firing phases . As for the mean peak firing rate , 13 Hz was not high enough to induce statistically significant bias in these simulations . However , in two scenarios , weight biases can become strong during movement in our model . First , the summation of synaptic inputs from ten place cells with identical ( or overlapped ) place fields amplified weight biases , and the bias became significant at 13 Hz in this case ( Figure 5C , Table 2 ) . Second , firing rates of place cells obey a log-normal distribution and a small fraction of place cells exhibits extremely high firing rates ( >30 Hz ) ( Mizuseki and Buzsáki , 2013 ) . These cells created large weight changes and significant directional biases ( Figure 5B and C ) . We note that the bias effects of replay sequences can be also enhanced in the above scenarios . Taken together , the bias effect in our model is weaker during run than in replay events because mean firing rate is lower and theta phase-locking is not effective for learning with broad symmetric STDP . However , non-negligible bias effects may arise in some biologically plausible situations . By the mechanism described above , our network model can potentiate forward synaptic pathway through reverse replay . However , whether reverse replay creates strong bias to the forward direction depends crucially on two parameters , that is , the slow time constant of long-term plasticity and the strength of short-term depression . Here , we demonstrate this by simulating the one-dimensional recurrent network model similar to Figure 1 in three different conditions . In all the cases ( Figure 6A , B , C ) , we repeatedly triggered firing sequences at the beginning of each simulation trial ( 0 s < time < 5 s ) , which are regarded as ‘forward’ sequences corresponding to repeated sequential experiences . After this ( time >10 s ) , we repeatedly stimulated central neurons in the network to induce firing sequences , which selectively traveled along the reverse synaptic pathway strengthened by the forward sequences , as was demonstrated in Figure 1 of the manuscript . These sequences overwrote the weight bias induced by forward sequences and eventually reversed it into the forward direction on neuron #100 ( Figure 6D ) . In the end , the bias converges to some value for which reverse replay could not propagate a long distance . On the other hand , neuron #400 was not recruited in reverse replay and hence the reversal of weight bias did not occur ( Figure 6E ) . In the condition 1 ( Figure 6A ) , modifications of synaptic weights were relatively slow ( time constant was 5000 ms ) and the depression effect of STP was the same as in Figure 1 . Due to the slow modifications of synaptic weights , a large number of reverse replay was generated before the bias was overwritten . Therefore , accumulated weight bias to the forward direction became the largest in the final state ( Figure 6D , red ) . In the condition 2 ( Figure 6B ) , the time constant for weight changes was shorter ( 500 ms ) and accordingly each reverse replay rapidly changed the weight bias . Consequently , reverse replay stopped earlier and the weight bias became smaller than the condition 1 ( Figure 6D , blue ) . However , the bias still converges to the forward direction because firing sequences can propagate even when synaptic weights were weakly biased to the opposite direction . In the condition 3 ( Figure 6C ) , we weakened STP in addition to the short time constant for weight changes ( see Materials and methods ) . Because short-term depression enhances the generation of firing sequences ( Romani and Tsodyks , 2015 ) , this manipulation further reduced the number of reverse replay and consequently the final value of weight bias ( Figure 6D , green ) . These results demonstrate that enhanced firing propagation by STP and relatively slow long-term plasticity are necessary to create strongly biased forward synaptic pathways through reverse replay . Strong short-term depression may be replaced by other mechanisms such as dendritic spikes , which also enhance the propagation of firing sequences ( Jahnke et al . , 2015 ) . However , this possibility was not pursued in the present study . We now demonstrate how reverse replay events starting from a rewarded position enables the learning of goal-directed paths . We consider the case where an animal is exploring on a W-maze ( Figure 7A ) . During navigation , the animal gets a reward at the one end of the arm ( position D2 ) , but not at the opposite end ( position D1 ) and other locations . In each trial , the animal starts at the center arm ( position A ) and runs into one of the two side arms at position B . In the present simulations , the animal visits both ends alternately: it reaches to D1 in ( 2n+1 ) -th trials and D2 in ( 2n ) -th trials , where n is an integer . After reaching either of the ends ( i . e . D1 or D2 ) , the animal stops there for 7 s . We constructed the 50 × 50 two-dimensional ( 2-D ) place-cell network associated to the 2-D space that the animal explored ( Figure 7B ) , using the rate neuron model . Each place cell had a place field in the corresponding position on the 2-D space and received global inhibitory feedback proportional to the overall network activity . Neighboring place cells were reciprocally connected with excitatory synapses , which were modulated by short-term and long-term plasticity rules as in Figure 1 and Figure 6 . During the delivery of reward , we mimicked dopaminergic modulations by enhancing the inputs to CA3 and increasing the frequency of triggering firing sequences ( Ambrose et al . , 2016; Singer and Frank , 2009 ) . Under this condition , a larger number of reverse replay was generated in the rewarded position . Thus , larger potentiation of synaptic pathways toward reward is expected in our model . After a few traversals on the W-maze , the network generated reverse replay at D1 and D2 ( Figure 7D and E , red arrows ) and forward replay at A ( Figure 7D and E , black arrows ) during immobility , and theta oscillation induced theta sequences along the animal’s path ( see Figure 7—video 1 ) . Notably , in trial five and later trials , forward replay selectively traveled towards the rewarded position D2 . Furthermore , firing sequences that started from the non-rewarded position D1 propagated to the rewarded position D2 but not to the start A ( Figure 7D and E , blue arrows ) . We note that the animal never traveled directly from D2 to D1 in our simulation . Thus , the network model could combine multiple spatial paths to form a synaptic pathway that has not been traversed by the animal . All these properties of firing sequences look convenient for the goal-directed learning of spatial map . We statistically confirmed the above-mentioned biases in firing sequences . We performed independent simulations of 10 model rats , in which five rats visited the two arms in the above-mentioned order , and the other five rats visited the arms in the reversed sequential order . In each simulation , we counted the number of firing sequences propagating along different synaptic pathways . Propagation of firing sequences triggered at the start A was significantly biased to the rewarded position D2 ( Figure 7F , Wilcoxon’s signed rank test , p=5 . 86×10-3 ) . While firing sequences from D2 tended to be reverse replay which propagates to A ( Figure 7G , Wilcoxon’s signed rank test , p=5 . 86×10-3 ) , the majority of firing sequences from D1 propagated prospectively to D2 and hence were goal-directed sequences ( Figure 7H , Wilcoxon’s signed rank test , p=5 . 57×10-3 ) . To visualize how the recurrent network was optimized for the goal-directed exploratory behavior , we defined ‘connection vectors’ from recurrent synaptic weights . For each place cell , we calculated the weighted sum of eight unit vectors each directed towards one of the eight neighboring neurons , using the corresponding synaptic weights ( Figure 8A ) . These connection vectors represent the average direction of neural activity transmitted from each neuron , and the 2-D vector field shows the flow of neural activities in the 2-D recurrent network and hence in the 2-D maze . We note that these vectors bias the flow , but actual firing sequences can sometimes travel in different directions from the vector flow . Initially , synaptic connections were random and the connection vector field was not spatially organized ( Figure 8—figure supplement 1 ) . However , after the exploration , the vector field was organized so as to route neural activities to those neurons encoding the rewarded position on the track ( Figure 8B ) . A similar route map was also obtained when we reversed the sequential order of visits to the two arms ( Figure 8—figure supplement 2 ) but was abolished when we removed reward ( Figure 8—figure supplement 3 ) or the effect of STP on Hebbian plasticity ( Figure 8—figure supplement 4 ) . As demonstrated previously , direct synaptic pathways from D1 to D2 were also created . The emergence of direct paths relies on two mechanisms in this model . First , as seen in Figure 8—figure supplement 3 , connections are biased from goal to start when there is no reward at the goal because theta sequences enhance synaptic pathways opposite to the direction of animal's movement . In non-rewarded travels , these directional biases are not overwritten by reverse replay . Thus , the relative preference of a synaptic pathway in hippocampal sequential firing decreases for exploration that does not result in reward . Second , some of reverse replay sequences from D2 propagates into D1 instead of the stem arm ( Figure 7E and G and Figure 7—video 1 ) , and such joint replay enhances biases towards goal through unexperienced spatial paths . Thus , our model creates a map not only for the spatial paths experienced by the animal , but also for their possible combinations if they guide the animal directly to the rewarded position from a point in the space . In this sense , our model optimizes the cognitive map of the spatial environment . We also examined the role of theta oscillation in our network model . When we turned off theta oscillation , the network model generated replay-like long-lasting firing sequences not only during immobility , but also during run ( Figure 7—figure supplement 1 ) . These sequences propagated randomly in both forward and reverse directions . In our model , theta oscillation offers periodic hyperpolarization of the membrane potentials that terminates firing sequences in a short period and localizes place-cell activity . Although the absence of theta oscillation does not impair place-cell sequences during run if we weaken recurrent connection weights ( e . g . the model in Wang et al . , 2015 ) , such model does not show replay events . At least in our model , theta oscillation during run is useful to realize the robust generation of local place-cell activity during run and replay events simultaneously . We further explore the role of theta sequences in the next section . So far , we have shown the role of reverse replay for goal-directed learning in a 1-D environment . However , whether a similar mechanism works in a 2-D environment remains unclear . Previous models produce reverse replay of recent paths by transient upregulation of the excitability of recently activated neurons ( Foster and Wilson , 2006; Molter et al . , 2006 ) , which may also work in a 2-D space . Such an effect has been experimentally observed , but the bias to recent paths disappears rapidly ( Csicsvari et al . , 2007 ) . In an open arena , prospective place-cell sequences tend to propagate from the current position to reward sites , but sequence propagation from the goal position was not biased to recent paths ( Pfeiffer , 2018; Pfeiffer and Foster , 2013 ) . These observations suggest that firing sequences during immobility in a 2-D space propagate isotropically from trigger points , rather than reverse replay of recent paths . Our model predicts that such firing sequences triggered at reward sites are beneficial for goal-directed path learning in a 2-D space . To show this , we simulated the same 2-D neural network model as in the previous section , increasing initial connection weights . Because we connected only neighboring neurons , these strong neuronal wiring reflected the topological structure of 2-D square space . We intermittently stimulated the central place cells to trigger firing sequences , which homogenously propagated through the 2-D neural network ( Figure 9A ) . Because sequence propagation potentiates synaptic pathways in the opposite direction , these divergent firing sequences created the connection vector field converging to the center ( Figure 9B ) . This result generalizes the role of 1-D reverse replay to higher dimensional spaces: isotropic sequence propagation from reward sites achieves goal-directed sequence learning in a 2-D ( or even 3-D ) open field . We further demonstrate how this learning mechanism works for goal-directed navigation in an open arena in a task similar to Morris water maze task ( Foster et al . , 2000; Morris et al . , 1986; Vorhees and Williams , 2006 ) and a 2-D foraging task ( Pfeiffer and Foster , 2013 ) . In the simulations , an animal started from random positions in a 2-D square space to search for a reward placed at one of the four candidate reward sites ( Figure 10A ) . In each trial , the animal stayed at the starting position for 3 s , ran around the 2-D space at a constant speed until it found the reward , and stayed at the reward site for 15 s . We triggered replay sequences every 1 s during immobility and theta oscillation induced theta sequences during run . At each time , we calculated a vector from the animal’s current position to the gravity center of the neural activity ( corresponding to the current position expressed by the neural network ) , which we call the activity vector ( Figure 10B ) . We used this vector to rotate the angle of the velocity vector of the animal’s movement . The velocity vector was also updated during immobility to determine the direction of the animal’s movement at the next start . Consequently , goal-directed replay sequences or theta sequences bias the animal’s movement toward the goal . Such a relationship between the animal’s movement and hippocampal firing sequences has been suggested in several experiments ( Huxter et al . , 2008; Pfeiffer and Foster , 2013; Wikenheiser and Redish , 2015 ) . One simulation set consisted of 20 trials , and the position of reward was changed every five trials . Therefore , memory of the previous trials could guide the animal to the reward in trials 2–5 , 7–10 , 12–15 , and 17–20 ( REPEAT trials ) , but not in trials 6 , 11 , 16 ( SWITCH trials ) . We performed 10 independent simulation sets , and 10 control simulation sets in which learning was disabled and the animal’s behavior was similar to a random search . The model quickly learned efficient goal-directed navigation after every SWITCH trial ( Figure 10C ) . Animals took relatively short paths from start to goal in REPEAT trials ( Figure 10E ) , and accordingly exploration time was significantly shorter in REPEAT trials than in other trials ( Figure 10D , Wilcoxon’s rank sum test , p<10-10 for both REPEAT-SWITCH and REPEAT-CONTROL ) . In contrast , exploration time in SWITCH trials was longer than that in control simulations ( Figure 10D , Wilcoxon’s rank sum test , p=3 . 43×10-6 ) because animals typically explored around the previous reward site in SWITCH trials ( Figure 10E ) . The exploration time around the previous reward site in SWITCH trials was significantly longer in simulations with learning than in control simulations ( Wilcoxon’s rank sum test , p=3 . 21×10-8 ) , which is consistent with rodents’ behavior in Morris water maze task ( Vorhees and Williams , 2006 ) . To analyze biases in sequence propagation , we calculated the angle between the instantaneous activity vector and a reference vector . The reference vector was a vector from the animal’s current position to a goal ( reward ) before and during exploration , or a vector from the animal’s current position to the recent path ( the animal’s average position within 3 s before reaching the goal ) at a goal . Bias to the goal or recent paths is strong if the angle is small . However , because of the small size of the network and the 2-D space , the angles were not exactly uniform even in control simulations . To remove this effect , we calculated a mean angular displacement in each behavioral state ( start , run and goal ) in control simulations and subtracted these baseline values . In REPEAT trials , the bias to reward before and during exploration was stronger than the bias to recent path at the goal ( Figure 10F , paired sample t-test , p=2 . 98×10-7 for Start-Goal , p=5 . 02×10-3 for Run-Goal ) . The bias at the goal was almost zero , that is , the same level as a random search . These results suggest that the bias from start to goal was stronger than that from goal to recent paths ( reverse replay ) in REPEAT trials , which is consistent with experimental observation ( Pfeiffer , 2018; Pfeiffer and Foster , 2013 ) . Furthermore , the bias during exploration suggests that weight biases also affected propagation of theta sequences . In contrast , in SWITCH trials , the bias to recent paths at the goal was significantly stronger than the bias to the goal in other periods ( Figure 10G , paired sample t-test , p=5 . 9×10-4 for Start-Goal in SWITCH , p=4 . 83×10-5 for Run-Goal in SWITCH ) . This bias to recent paths in SWITCH trials can be explained by the fact that the animal typically reached reward after visiting the previous reward site which strongly attracted firing sequences ( Figure 10E and F ) . Notably , this bias gives an efficient way to update connection weights for the new reward site because the connection vector fields converging to the two goals differ only in the space between the goals and updating weight biases is unnecessary outside this space . Thus , our model predicts that the bias of sequence propagation from a novel goal position to previously learned goals appears when the animal should update previous memories to a new memory . Taken together , these results suggest that divergent sequences create weight biases for convergent sequence propagation to goals in our model and this learning mechanism can be a basis for efficient goal-directed navigation in the 2-D space . Our results have several implications for spatial memory processing by the hippocampus . Suppose that the animal is rewarded at a spatial position after exploring a particular path . Reverse replay propagating backward from the rewarded location will strengthen the neuronal wiring in CA3 that preferentially propagates forward firing sequences to this location along the path . Because the frequency of reverse replay increases at rewarded positions ( Ambrose et al . , 2016; Singer and Frank , 2009 ) , reward delivery induces the preferential potentiation of forward synaptic pathways , which in turn results in an enhanced occurrence of forward replay in the consolidation phase . Thus , our model predicts that reverse replay is crucial for the reinforcement of reward-seeking behavior in the animal and gives , for the first time , the mechanistic account for the way reverse replay enables the hippocampal prospective coding of reward-seeking navigation . Furthermore , if the occurrence of reverse replay is modulated not only by reward but also by other salient events for the animal , this model is immediately generalized to memorization of important paths to be replayed afterwards . An interesting example would be learning of spatial paths associated with fear memory ( Wu et al . , 2017 ) . Our computational results are qualitatively consistent with the experimentally observed properties of forward and reverse replay events ( Ambrose et al . , 2016; Carr et al . , 2011; Diba and Buzsáki , 2007; Foster and Wilson , 2006; Pfeiffer , 2018; Pfeiffer and Foster , 2013; Singer and Frank , 2009 ) and matches the recent finding of symmetric STDP time windows in CA3 ( Mishra et al . , 2016 ) . Importantly , our model reinforces unexperienced spatial paths by connecting the multiple paths that were previously encoded by separate experiences . In the simulations on the W-maze , the network model not only learned actually traversed paths from the start ( A ) to the goal ( D2 ) , but also remembered paths from other locations ( C1 and D1 ) to the goal despite that the animal had not experienced these paths . This reinforcement occurs because reverse replay sequences starting from the visited arm occasionally propagate or bifurcate into an unvisited arm at the branching point . The hippocampus can generate replay along joint paths ( Wu and Foster , 2014 ) even when the animal has no direct experience ( Gupta et al . , 2010 ) . Therefore , the above mechanism is biologically possible . In 1-D tracks , reverse replay is observed immediately after the first lap ( Foster and Wilson , 2006; Wu and Foster , 2014 ) . Our model shows that such a replay creates a bias to the forward direction ( i . e . toward the reward ) even in the very early stage of learning . Consistently , a weak bias to forward replay was observed in the first exposure to a long 1-D track ( Davidson et al . , 2009 ) . Our model predicts that the bias to the goal-directed sequences will be suppressed ( or enhanced ) if we selectively block ( or enhance ) reverse replay at the goal . Such experiments are possible by using the techniques of real-time decoding feedback ( Ciliberti and Kloosterman , 2017; Sodkomkham et al . , 2016 ) . The most critical assumption in our model is the rapid modulation of STDP coherent to the presynaptic neurotransmitter release of STP . Such a modulation was actually reported in the visual cortex ( Froemke et al . , 2006 ) . Although short-term depression also exists in CA3 ( Guzman et al . , 2016 ) , STDP is modulated in a slightly different fashion in the hippocampus: the strong modulation arises from the second presynaptic spike rather than the first one ( Wang et al . , 2005 ) . However , the experiment was performed in a dissociated culture in which hippocampal sub-regions were not distinguished . Moreover , the modulation of STDP was not tested for more than two presynaptic spikes , and whether a third presynaptic spike further facilitates or rather depresses STDP remains unknown . Therefore , the contributions of STP to STDP should be further validated in CA3 . The proposed role of short-term depression in biasing replay events may also be examined by pharmacological blockade or enhancement of STP in the hippocampus ( Froemke et al . , 2006 ) . In addition , we showed that the dendritic ADP accumulated over multiple spikes causes a similar phenomenon ( Figure 1—figure supplement 1 ) . This can be directly tested in CA3 by modifying the protocol described previously ( Mishra et al . , 2016 ) . Moreover , a bias to the reverse direction can be further strengthened if some neuromodulator expands the time window of symmetric STDP selectively toward the anti-causal temporal domain ( tpre>tpost ) . Our model also predicts this type of metaplasticity in CA3 Neuromodulations , especially reward-triggered facilitation of replay events ( Ambrose et al . , 2016; Singer and Frank , 2009 ) , play important roles for the proposed mechanism of goal-directed learning with reverse replay . CA3 primarily receives dopaminergic input from the locus coeruleus ( LC ) which signals novelty and facilitate learning in the hippocampus ( Takeuchi et al . , 2016; Wagatsuma et al . , 2018; Walling et al . , 2012 ) . Therefore , sequence learning in CA3 is also affected by novelty and salience of events ( Lisman and Grace , 2005; Lisman et al . , 2011 ) . Consequently , any place in which the animal experiences behaviorally important events can be a potential goal for the hippocampal path learning , and valence of the goal is encoded in downstream areas ( de Lavilléon et al . , 2015; Redondo et al . , 2014 ) . This may explain sequence learning for fear memory ( Wu et al . , 2017 ) and vicarious trial and error in hippocampus , that is , evaluation of potential paths before decision making ( Johnson and Redish , 2007; Singer et al . , 2013 ) . In this case , reverse replay in CA3 contributes to selective forward replay of paths informative for decision making . In our model , modulation of triggering replay events ( or sharp wave ripples ) crucially affects the learning with reverse replay . Therefore , other hippocampal sub-regions may also participate in goal-directed path learning . The area CA2 encodes the current position during immobility ( Kay et al . , 2016 ) and can trigger sharp wave ripples ( Oliva et al . , 2016 ) . Thus , dopaminergic enhancement in CA2 may increase replay events from the current position as in our simulation settings . Dentate gyrus intensively encodes reward sites and triggers sharp wave ripples in CA3 in working memory task ( Sasaki et al . , 2018 ) . It is interesting to examine whether these ripples accompany replay sequences , which remains unclear at present . If this is indeed the case , our results suggest that CA2 and dentate gyrus also play active roles in the goal selection of hippocampal path learning . Our model also predicts that the release probability of neurotransmitter strongly affects the magnitude and probability of bias toward the reverse direction ( Figure 4 ) . Therefore , modulation of neurotransmitter release in CA3 can regulate the behavioral impact of hippocampal firing sequences . For example , acetylcholine suppresses neurotransmitter release at recurrent synapses ( Hasselmo , 2006 ) , which may abolish the directional biases created during movement ( see Figure 4 and Figure 8—figure supplement 3 ) . Presynaptic long-term plasticity ( Costa et al . , 2015 ) may also affect the directional biases at longer timescales . In our simulation , the strength of theta phase-locking of place cell activity did not have significant effects on learning ( Figure 5 and Figure 5—figure supplement 1 ) . This result poses a question against the long-standing hypothesis that theta sequences are essential for sequence learning ( Jensen and Lisman , 1996; 2005; Sato and Yamaguchi , 2003 ) . However , it is possible that theta phase-locking is effective in learning CA3-to-CA1 connections at which STDP is asymmetric and more sensitive to time differences between presynaptic and postsynaptic activities than at CA3 recurrent synapses ( Bi and Poo , 2001 ) . Furthermore , our simulations in the 2-D space showed that theta sequences can act as readout of weight biases to reward , and hence are useful for planning future trajectories during exploration . This role of theta sequences is consistent with experimental findings ( Huxter et al . , 2008; Wikenheiser and Redish , 2015 ) . Short-term facilitation also had only limited effects on the proposed learning mechanism ( Figure 4 ) ; however , it contributed to the generation of theta sequences ( Wang et al . , 2015 ) . Therefore , in our model , short-term facilitation also enhances the readout of spatial information during run . By extending the role of reverse replay in 1-D space , we showed that divergent sequences that isotropically propagate from reward sites helps goal-directed sequence learning and hence efficient navigation in 2-D space . Similar foraging tasks can be solved by temporal difference ( TD ) learning model ( Dayan , 1993; Sutton , 1988 ) and the relevance of TD learning to hippocampal information processing has been proposed ( Foster et al . , 2000; Stachenfeld et al . , 2017 ) . Thus , how our learning mechanism is related to TD learning should be mathematically investigated in the future . A conceptual model has been proposed for goal-directed sequence learning with symmetric STDP based on the gradient field of connection strength ( Pfeiffer , 2018 ) . The conceptual model concluded that goal-directed sequence learning would not occur if sequences homogenously propagate through the entire space . However , our model shows that this is not the case for STP-modulated symmetric STDP and proposes a network mechanism to implement the conceptual model based on reverse replay . While the present model could demonstrate goal-directed path learning , the model has yet to be improved to learn context-dependent switching of behavior such as navigation on an alternating T-maze . In our simulation , animals learn only reward delivered at the same location . However , in alternating T-maze tasks , the animal has to remember recent experiences to change its choices based on the stored memory . Furthermore , the experiment in Pfeiffer and Foster ( 2013 ) also contains working-memory-based switching between predictable and unpredictable reward searches . Thus , the consistency between our simulations and experiments is still limited . A straightforward extension of our model is to maintain multiple charts representing the alternating paths ( Samsonovich and McNaughton , 1997 ) and switch them according to the stored short-term memory or certain context information . Each chart will be selectively reinforced by reverse replay along one of the alternating paths . This switching of CA3 activity may be supported by the dentate gyrus ( Sasaki et al . , 2018 ) . How to protect the previous memories from overwriting with novel reward experiences is another important issue that is left unsolved by the present model . If the animal can memorize all four candidate reward sites simultaneously , the animal can efficiently search reward even when the reward position is changed in every trial . One solution for this issue is triggering remote replay ( Gupta et al . , 2010; Karlsson and Frank , 2009 ) at the previous reward sites . Thus , we predict that bias of starting points of remote replay affects persistence of multiple memories . While 2-D hippocampal place cells are omnidirectional , the majority of 1-D place cells are unidirectional ( Buzsáki , 2005 ) . We did not take into account this property of 1-D place cells in this study . Because we only simulated unidirectional movements in the W-maze , the network may describe an ensemble of place cells for one direction and another neuron ensemble is necessary for the opposite direction . In this case , place cells for the path B→D1 and those for the opposite path D1→B are different . Thus , the directional bias learned in our network may not implicate direct paths from D1 to D2 for the animal . To learn unexperienced paths , continuous replay events of multiple unidirectional place-cell ensembles is necessary , which has been experimentally observed ( Davidson et al . , 2009; Gupta et al . , 2010; Wu and Foster , 2014 ) . Neural network models also demonstrated that Hebbian plasticity can connect multiple unidirectional place cells at the junction points ( Brunel and Trullier , 1998; Buzsáki , 2005; Káli and Dayan , 2000 ) . Relating to this , replay of long paths and complex spatial structures is often accompanied by concatenated sharp wave ripples ( Davidson et al . , 2009 ) . Interestingly , each sharp wave ripple corresponds to a segment in the spatial structure ( Wu and Foster , 2014 ) . The effect of this segmentation to our learning mechanism is not obvious . Elucidating the underlying mechanism will reveal how the hippocampal circuit segment and concatenate sequential experiences . Reverse replay has not been found in the neocortical circuits . For instance , firing sequences in the rodent prefrontal cortex are reactivated only in the forward directions ( Euston et al . , 2007 ) . To the best of our knowledge , neocortical synapses obey asymmetric STDP ( Froemke et al . , 2006 ) . This seems to be consistent with the selective occurrence of forward sequences because , as suggested by our model , the sensory-evoked forward firing sequences should only strengthen forward synaptic pathways under asymmetric STDP . However , if dopamine turns asymmetric STDP into symmetric STDP , which is actually the case in hippocampal area CA1 ( Brzosko et al . , 2015; Zhang et al . , 2009 ) , forward firing sequences will reinforce the propagation of reverse sequences . Whether reverse sequences exist in the neocortex and , if not , what functional roles replay events play in the neocortical circuits are still open questions . Whether the present neural mechanism to combine experienced paths into a novel path accounts for cognitive functions other than memory is an intriguing question . For instance , does this mechanism explain the transitivity rule of inference by neural networks ? The transitive rule is one of the fundamental rules in logical thinking and says , ‘if A implies B and B implies C , then A implies C . ’ This flow of logic has some similarity to that of joint forward-replay sequences , which says , ‘if visiting A leads to visiting B and visiting B leads to visiting C , then visiting A leads to visiting C . ’ Logic thinking is more complex and should be more rigorous than spatial navigation , and little is known about its neural mechanisms . The proposed neural mechanism of path learning may give a cue for exploring the neural substrate for logic operations by the brain . In sum , our model proposes a biologically plausible mechanism for goal-directed path learning through reverse sequences . In the dynamic programming-based path finding methods such as Dijkstra’s algorithm for finding the shortest path ( Dijkstra , 1959 ) and Viterbi algorithm for finding the most likely state sequences in a hidden Markov model ( Bishop , 2010 ) , a globally optimal path is determined by backtrack from the goal to the start after an exhaustive local search of forward paths . Our model enables similar path finding mechanism through reverse replay . Such a mechanism has been suggested in machine learning literature ( Foster and Knierim , 2012 ) , but whether and how neural dynamics achieves it remained unknown . In addition , our model predicts the roles of neuromodulators that modify plasticity rules and activity levels in sequence learning . These predictions are testable by physiological experiments . We simulated the network of 500 neurons . Firing rate of neuron i was determined asri=frate ( Iiexc−Iinh+Iiext ) The function frateI was threshold linear functionfrate ( I ) =max{0 , ρ ( I−ϵ ) }where ρ=0 . 0025 and ϵ=0 . 5 . Excitatory synaptic current Iiexc and Inhibitory feedback Iinh followedI˙iexc=−Iiexcτexc+∑jwijrjDjFjI˙inh=−Iiinhτinh+winh∑jrjDjFjwhere τexc=τinh=10 ms , and winh=1 . Variables for short-term synaptic plasticity Dj and Fj obeyed the following equations ( Wang et al . , 2015 ) :D˙j=1−Dj τSTD−rjDjFjF˙j=U−FjτSTF+U ( 1−Fj ) rj Parameters were τSTD=500 ms , τSTF=200 ms , and U=0 . 6 . External input Iiext was usually zero and changed to 5 for 10 ms when the cell was stimulated . We stimulated neurons 0≤i≤10 at the beginning of simulations , and neurons 245≤i≤255 at 3 s after the beginning . We determined initial values of excitatory weights wij as ( 7 ) wij=wmaxexp⁡ ( −∣i−j∣d ) where wmax=27 and d=5 . We set self-connections wii to zero throughout the simulations . The weights were modified according to the rate-based Hebbian synaptic plasticity asw˙ij=ΔijτwΔ˙ij=−Δij+ηrirjDjFjwhere η=20 and τw=1000 ms . When we simulated Hebbian synaptic plasticity without the modulation by short-term plasticity , we removed DjFj from this equation and changed the value of η to 4 . In the simulation with accumulation of ADP ( Figure 1—figure supplement 1 ) , we calculated smoothed postsynaptic activity pi by solvingτADPp˙i=−pi+riwhere τADP=80ms , and changed Hebbian plasticity ( 9 ) toτwΔ˙ij=−Δij+ηpirj The value of η was also changed to 4 . We used Izhikevich model ( Izhikevich , 2003b ) for the simulation of spiking neurons:v˙i=0 . 04vi2+5vi+140−ui+Iisyn−Iinh+Iiextu˙i=a ( bvi−ui ) If the membrane voltage vi≥30 mV , the neuron emits a spike and the two variables were reset as vi←c and ui←ui+d . Parameter values were a=0 . 02 , b=0 . 2 , c=-65 and d=8 . Excitatory synaptic current was ( 14 ) Iisyn=giAMPA0-vi+fNMDAvigiNMDA0-vi , and the synaptic conductance followed ( 15 ) g˙iAMPA=−giAMPAτAMPA+∑j , kwijAMPADjFjδ ( t−tjkf−tdelay ) , ( 16 ) g˙iNMDA=−giNMDAτNMDA+∑j , kwijNMDADjFjδ ( t−tjkf−tdelay ) , where tjkf is the timing of the k-th spike of neuron j and parameter values were τAMPA=5 ms , τNMDA=150 ms and tdelay=2 ms . The voltage dependence of NMDA current ( Izhikevich et al . , 2004 ) wasfNMDA ( V ) = ( V+8060 ) 21+ ( V+8060 ) 2 Inhibitory feedback Iiinh was calculated asI˙inh=−Iiinhτinh+winh∑j , kDjFjδ ( t−tjkf−tdelay ) where winh=1 and τinh=10 ms . Short-term synaptic plasticity obeyed the following dynamics:D˙j=1−Dj τSTD−DjFjδ ( t−tjf ) F˙j=U−FjτSTF+U ( 1−Fj ) δ ( t−tjf ) Parameter values were τSTD=500 ms , τSTF=200 ms and U=0 . 6 . External input Iiext was the same as that in the rate neuron model . We determined initial values of synaptic weights wijAMPA as ( 21 ) wijAMPA=wmaxexp⁡ ( −∣i−j∣d ) where wmax was 0 . 3 and d=5 . We set self-connections wiiAMPA to zero throughout the simulations . The weights of NMDA current wijNMDA were determined as wijNMDA=0 . 2wijAMPA and fixed at these values throughout the simulations . The weights of AMPA current were modified by STDP asw˙ijAMPA=ΔijAMPAτwΔ˙ijAMPA=−ΔijAMPA+η∑k , lfSTDP ( tikf , tjlf ) DjFjδ ( t−tjkf ) We simulated two different STDP types by changing the function fSTDPtpost , tpre as follows . Asymmetric STDP:fSTDP ( tpost , tpre ) ={A+exp⁡ ( −tpost−tpreτ+ ) if tpost≥tpre−A−exp ( −tpre−tpostτ− ) iftpost<tpre Symmetric STDP:fSTDP ( tpost , tpre ) =A+exp⁡ ( −∣tpost−tpre∣τ+ ) −A−exp ( −∣tpost−tpre∣τ− ) Parameter values were η=0 . 05 , τw=1000 ms , A+=1 , A-=0 . 5 , τ+=20 ms and τ-=40 ms for all STDP types . We took into account contributions of all spike pairs were in the simulations . When we simulated STDP without the modulation by short-term plasticity , we removed DjFj from the above equation and changed the value of η to 0 . 01 . In the simulation with short time constants ( Figure 2—figure supplement 1A ) , we changed the values of parameters as τAMPA=2 . 5 ms , τinh=5 ms , wmax=0 . 35 and wijNMDA=0 . In each simulation , we sampled spike trains of 21 neurons ( neuron #1 - #21 ) 100 times for given values of the number of spikes per neuron Nspike , mean inter-spike interval ( ISI ) tISI , and firing propagation speed tspeed . We set first spikes of neuron #n to tn , 1f=n-1tspeed . Following a first spike , Poisson spike train for each neuron was simulated by sampling ISI ( Δtn , kf=tn , kf-tn , k-1f ) from the following exponential distribution:P ( Δtn , kf ) =1tISIexp⁡ ( −Δtn . kftISI ) , ( k=2 , 3 , . . . , Nspike ) We induced an absolute refractory period by resampling ISI if it was shorter than 1 ms . After we generated spike trains , we simulated neurotransmitter release by solving Equations ( 19 ) and ( 20 ) for each neuron . In Figure 3 , parameter values of short-term plasticity were τSTD=150 ms , τSTF=40 ms and U=0 . 37 . In Figure 4 , we sampled the values of U , τSTD and τSTF from [0 . 1 , 0 . 6] , [50 , 500 ms] and [10ms , 300 ms] , respectively . We calculated changes in the weight from the neuron in the center ( j=11 ) to the neuron i asΔij=∑k , lfSTDP ( tikf , tjlf ) DjFjδ ( t−tjkf ) In all-to-all STDP , we calculated the above summation over all spike pairs . In nearest-neighbor STDP ( Izhikevich and Desai , 2003a ) , we considered only pairs of a presynaptic spike and the nearest postsynaptic spikes before and after the presynaptic spike . For symmetric STDP , fSTDPtpost , tpre was Gaussian ( Mishra et al . , 2016 ) fSTDP ( tpost , tpre ) =Aexp⁡ ( −12 ( tpost−tpreτ ) 2 ) where A=1 and τ=70 ms . For asymmtric STDP , fSTDPtpost , tpre was the same as the Equation ( 24 ) except that parameter values were changed as A+=0 . 777 , A-=0 . 273 , τ+=16 . 8 ms and τ-=33 . 7 ms ( Bi and Poo , 2001 ) . We calculated weight biases for each spike train as ∑i=110Δij-∑i=1221Δij . In each simulation , we sampled one-second-long spike trains of 81 neurons ( neuron #1 - #81 ) 100 times . Calculation of firing probabilities and sampling of spikes were performed every 1 ms time bin . We assumed a constant speed of the rat , and expressed the animal’s current position with current time t [s] . We defined theta phase at time t as θt=2π×8t+c , where c is a random offset . Place field of each neuron was given by normalized Gaussian functionPFi ( t ) =12πσPFexp⁡ ( −12 ( t−μiσPF ) 2 ) The place-field center of neuron i was μi=2×i-180-0 . 5 which spanned from -0 . 5 to 1 . 5 , and σPF=0 . 2 . We simulated theta phase-locking with von-Mises distribution function ( Bishop , 2010 ) , which is periodic version of Gaussian functionPLi ( t ) =12πI0 ( β ) exp⁡ ( β cos ( θ ( t ) −pi ( t ) ) ) where I0β is the zeroth-order Bessel modified function of the first kind . The mean firing phase of neuron i at time t changed through time as pit=πμi-t . Using these functions , we determined the firing rate of neuron i at time t as αPFitPLit [kHz] . In this setting , α and β were the tuning parameters that control the peak firing rates and phase selectivity , respectively . The range of sampling was 0 . 01≤α≤0 . 15 and 0 . 1≤β≤10 , and the peak firing rates were calculated by smoothing spike trains of the central neuron ( #41 ) with Gaussian kernel of the standard deviation 50 ms . Changes and biases of weights from the central neuron ( j=41 ) were calculated in the same way with the previous section . In the evaluation of the biases summed over 10 overlapping place cells , we sampled 10 spike trains for the central neuron ( #41 ) , and summed weight biases calculated from these samples . We used the same network model as in Figure 1 and additionally implemented normalization of synaptic weights for homeostasis . We calculated the sum of incoming weights on each neuron ( ∑jwij ) at each time step and normalized the weights as wij←∑jwijinit∑jwijwij , where {wijinit} denotes initial synaptic weights . We changed the time constant of long-term plasticity ( τw ) to 5000 ms in the condition 1 , and 500 ms in the conditions 2 and 3 . In the condition 3 , we also changed the values of parameters for short-term plasticity and initial synaptic weights as τSTD=200 ms , U=0 . 3 and wmax=30 . Neurons were stimulated every 1 s . Weight biases were calculated as ∑i<jwij−∑i>jwij for two neurons ( j=100 , 400 ) . We simulated an animal moving on a two-dimensional space spanned by x and y ( 0≤x≤50 , 0≤y≤50 ) . Coordinates of the six corners ( A , B , C1 , C2 , D1 , D2 ) of W-maze were zA= ( 25 , 15 ) , zB= ( 25 , 35 ) , zC1= ( 45 , 35 ) , zD1= ( 45 , 15 ) , zC2= ( 5 , 35 ) and zD2= ( 5 , 15 ) . In each set of trials with time length T=15 s , we determined the position of the animal zpos at time t'=t mod T aszpos={zA ( 0 s≤t′<2 s ) ( t′−2 ) ( zB−zA ) +zA ( 2 s≤t′<4 s ) ( t′−3 ) ( zX−zB ) +zB ( 4 s≤t′<6 s ) ( t′−4 ) ( zY−zX ) +zX ( 6 s≤t′<8 s ) zY ( 8 s≤t′<15 s ) We set zX=zC1 and zY=zD1 in the ( 2n+1 ) -th trials and zX=zC2 and zY=zD2 in the ( 2n ) -th trials . The neural network consisted of 2500 place cells that were arranged on a 50 x 50 two-dimensional square lattice . The place field centers of neurons in the i-th column ( x-axis ) and the j-th row ( y-axis ) were denoted as zi , j=i , j . Each place cell received excitatory connections from eight surrounding neurons . The connection weight from a neuron at i-k , j-l to a neuron at i , j were denoted as wi , jk , l , where the possible combinations of ( k , l ) were given as S=1 , 1 , 1 , 0 , 1 , -1 , 0 , 1 , 0 , -1 , -1 , 1 , -1 , 0 , -1 , -1 . Initial connection weights were uniformly random and normalized such that the sum of eight connections obeys ∑k , l∈Swi , jk , l=0 . 5 . We simulated activities of place cells ri , j asri , j=frate ( Ii , jE ) I˙i , jE=−Ii , jEτ+∑ ( k , l ) ∈Swi , jk , lri−k , j−lDi−k , j−lFi−k , j−l−Iinh−Itheta+Ii , jplace+Ii , jnoise Time constant τ was 10 ms . The function frateI was a threshold linear functionfrate ( I ) =max{0 , ρ ( I−ϵ ) }where ρ=1 and ϵ=0 . 002 . Inhibitory feedback Iinh followedI˙inh=−Iinhτinh+winh∑i , jri , jDi , jFi , jwhere τinh=10 ms and winh=0 . 0005 . Variables for short-term synaptic plasticity Di , j and Fi , j obeyedD˙i , j=1−Di , j τSTD−ri , jDi , jFi , jF˙i , j=U−Fi , jτSTF+U ( 1−Fi , j ) ri , jwith parameter values τSTD=300 ms , τSTF=200 ms , and U=0 . 4 . We induced theta oscillation byItheta=B2 ( sin ( 2πtttheta ) +1 ) where B=0 . 005 kHz and ttheta=10007 ms . Ii , jnoise was independent Gaussian noise with the standard deviation 0 . 0005 kHz . We determined place-dependent inputs for each neuron Ii , jplace from the place field center of each neuron zi , j and the current position of the animal zpos:Ii , jplace=Cexp⁡ ( −12d2 ( zpos−zi , j ) T ( zpos−zi , j ) ) where d = 2 . The parameter C was set as 0 . 005 kHz when the animal was moving and 0 . 001 kHz when the animal was stopping at D2 ( the position of reward ) . When the animal was stopping at other positions , C was set at zero but occasionally changed to 0 . 001 kHz for a short interval of 200 ms . The occurrence of this brief activation followed Poisson process at 0 . 1 Hz , but it always occurred one second after the onset of each trial to trigger prospective firing sequences . We implemented the Hebbian synaptic plasticity asw˙i , jk , l=Δi , jk , lτwΔ˙i , jk , l=−Δi , jk , l+ηri , jri−k , j−lDi−k , j−lFi−k , j−lwhere η=1 and τw=30 s . If the sum of synaptic weights on each neuron ( ∑k , l∈Swi , jk , l ) was greater than unity , we renormalized synaptic weights by dividing them by the sum . When we simulated Hebbian synaptic plasticity without modulations by short-term plasticity , Di-k , j-lFi-k , j-l was removed from the above equation and the η value was changed to 0 . 1 . We calculated 'connection vector' of each neuron ui , j by the following weighted sum of unit vectors vk , l=kk2+l2 , lk2+l2:ui , j=∑ ( k , l ) ∈SwI+k , j+lk , lvk , l We used a similar two-dimensional space and a similar 50 x 50 neural network model to those used in the simulation of the W-maze . We made some minor changes: ( 1 ) We normalized initial connection weights as ∑k , l∈Swi , jk , l=1 . ( 2 ) In order to create finite-length firing sequences , we added an external inhibitory input Iextinh to the Equation ( 33 ) . When the animal was immobile , we kept excitatory input Ii , jplace nonzero ( C=0 . 001 kHz ) , and triggered sequences by disinhibiting the network ( Iextinh=0 ) every 1 s and terminated sequences ( Iextinh=0 . 1 ) at 800 ms after each trigger . In the simulation of divergent sequences ( Figure 9 ) , parameters of Hebbian plasticity were changed as η=0 . 5 and τw=10 s . We triggered firing sequences at the point x , y=25 , 25 for 30 s . In the simulation of foraging task ( Figure 10 ) , parameters of Hebbian plasticity were changed as η=0 . 1 and τw=10 s . Four candidate reward sites were positioned at x , y=15 , 15 , 15 , 35 , 35 , 15 , 35 , 35 . The reward position was randomly determined in each simulation . The animal could explore the space defined by 5≤x≤45 , 5≤y≤45 . We randomly determined a starting position in this area such that the initial distance to the reward position was longer than 10 , and the animal began to explore after 3-s-long immobility . When the animal reached within distance 3 from the reward , the animal was set immobile for 15 s , and then reset to the starting position of the next trial . We terminate a trial when the animal did not reach reward within 300 s , and regarded the exploration time of these trials as 300 s . We excluded these trials from the analysis of angular displacements . We calculated the activity vector in the 2-D coordinate system asa=∑i , jri , jzi , j∑i , jri , j−zposwhen ∑i , jri , j>0 . We changed the animal’s position zpos through the velocity vector v as ( 44 ) z˙pos=γvvand rotated the velocity vector towards the direction of the activity vector:v˙=γaa\|\|a\|\| +γnoiseanoisewhere v was normalized to unity at each time and anoise is a two-dimensional independent normal Gaussian noise . The speed γv was 0 . 01 or 0 during exploration and immobility , respectively . Other values of parameters were γa=0 . 01 and γnoise=0 . 05 throughout the simulation . We calculated angular displacements from cosine similarity between a and the reference vector ρ every 20 ms . Before and during exploration , ρ was a vector from zpos to the reward position . At the reward position , ρ was a vector from the current zpos to the mean of zpos within 3 seconds before reaching the reward position . We excluded the periods in which the maximum firing rate in the network was below 0 . 01 kHz or the length of a was below 1 . Due to the small size of the network and the 2-D space , angular displacements were not uniform even in the control simulations . To compensate this background biases , we calculated the mean angular displacements of the control simulations and subtracted those baseline values from the mean angular displacements obtained in each condition . We applied paired sample t-test after checking normality of the data by Shapiro Wilk test ( p>0 . 05 ) . Simulations and visualization were written in C++ and Python 3 . The codes are available at https://github . com/TatsuyaHaga/reversereplaymodel_codes ( Haga , 2018; copy archived at https://github . com/elifesciences-publications/reversereplaymodel_codes ) .	To find their way around , animals – including humans – rely on an area of the brain called the hippocampus . Studies in rodents have shown that certain neurons in the hippocampus called place cells become active when an animal passes through specific locations . At each position , a different set of place cells fires . A journey from A to B will thus be accompanied by a sequential activation of place cells corresponding to a particular point . Rats can learn new routes to a given place . Every time they take a specific way , the connections between the activated place cells become stronger . After learning , the hippocampus replays the sequence of place cell activation both in the order the rat has experienced and backwards . This is known as reverse replay , which occurs more often when the animals find rewards at their destination . This suggests that reverse replay may help animals learn the routes to locations where food is available . To test this idea , Haga and Fukai built a computer model that simulates the hippocampal activity seen in a rat running through a maze . In contrast to previous models , which featured only forward replay , the new simulation also included reverse replay . The results confirmed that reverse replay helps the rodents to learn routes to rewarded locations . It also enables the hippocampus to combine multiple past experiences , which may teach animals that a combination of previous paths will lead to a reward , even if they have never tried the combined route before . The hippocampus has a central role in many different types of memory . The findings by Haga and Fukai may therefore provide a framework for studying the mechanisms of memory and decision-making . The results could even offer insight into the mechanisms of logical thinking . After all , the ability to combine multiple known paths into a new route bears some similarity to joining up thought processes such as ‘If Sophie oversleeps , she will miss the bus’ with ‘If Sophie misses the bus , she will be late for school’ to reason that ‘If Sophie oversleeps , she will be late for school’ . Future studies should test whether reverse replay helps with this process of deduction .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2018	Recurrent network model for learning goal-directed sequences through reverse replay
Social insects frequently engage in oral fluid exchange – trophallaxis – between adults , and between adults and larvae . Although trophallaxis is widely considered a food-sharing mechanism , we hypothesized that endogenous components of this fluid might underlie a novel means of chemical communication between colony members . Through protein and small-molecule mass spectrometry and RNA sequencing , we found that trophallactic fluid in the ant Camponotus floridanus contains a set of specific digestion- and non-digestion related proteins , as well as hydrocarbons , microRNAs , and a key developmental regulator , juvenile hormone . When C . floridanus workers’ food was supplemented with this hormone , the larvae they reared via trophallaxis were twice as likely to complete metamorphosis and became larger workers . Comparison of trophallactic fluid proteins across social insect species revealed that many are regulators of growth , development and behavioral maturation . These results suggest that trophallaxis plays previously unsuspected roles in communication and enables communal control of colony phenotypes . Many fluids shared between individuals of the same species , such as milk or semen , can exert significant physiological effects on recipients ( Poiani , 2006; Liu and Kubli , 2003; Liu et al . , 2014; Bernt and Walker , 1999 ) . While the functions of these fluids are well known in some cases ( Liu and Kubli , 2003; Liu et al . , 2014; Bernt and Walker , 1999; Perry et al . , 2013 ) , the role ( s ) of other socially exchanged fluids ( e . g . , saliva ) are less clear . The context-specific transmission and interindividual confidentiality of socially exchanged fluids raise the possibility that this type of chemical exchange mediates a private means of chemical communication . Social insects are an interesting group of animals to investigate the potential role of socially exchanged fluids . Colonies of ants , bees and termites are self-organized systems that rely on a set of simple signals to coordinate the development and behavior of individual members ( Bonabeau et al . , 1997 ) . While colony-level phenotypes may arise simply from the independent behavior of individuals with similar response thresholds , many group decisions require communication between members of a colony ( LeBoeuf and Grozinger , 2014 ) . In ants , three principal means of communication have been described: pheromonal , acoustic , and tactile ( Hölldobler and Wilson , 1990 ) . Pheromones , produced by a variety of glands , impart diverse information , including nestmate identity and environmental dangers . Acoustic communication , through substrate vibration or the rubbing of specific body parts against one another , often conveys alarm signals . Tactile communication encompasses many behaviors , from allo-grooming and antennation to the grabbing and pulling of another ant’s mandibles for recruitment to a new nest site or resource . Ants , like many social insects and some vertebrates ( Boulay et al . , 2000a; Greenwald et al . , 2015; Malcolm and Marten , 1982; Wilkinson , 1984 ) , also exhibit an important behavior called trophallaxis , during which liquid is passed mouth-to-mouth between adults or between adults and juveniles . The primary function of trophallaxis is considered to be the exchange of food , as exemplified by the transfer of nutrients from foragers to nurses and from nurses to larvae ( Wilson and Eisner , 1957; Buffin et al . , 2009; Cassill and Tschinkel , 1995; Cassill and Tschinkel , 1996; Wheeler , 1918; Wheeler , 1986 ) . The eusocial Hymenopteran forgeut has evolved a specialized distensible crop and a restrictive proventriculus ( the separation between foregut and midgut ) enabling frequent fluid exchange and regulation of resource consumption ( Terra , 1990; Eisner and Brown , 1958; Lanan et al . , 2016; Hunt , 1991 ) . In addition to simple nourishment , trophallaxis can provide information for outgoing foragers about available food sources ( Grüter et al . , 2006; Farina et al . , 2007 ) . Trophallaxis also occurs in a number of non-food related contexts , such as reunion with a nestmate after solitary isolation ( Boulay et al . , 2000b ) , upon microbial infection ( Hamilton et al . , 2011 ) , and in aggression/appeasement interactions ( Liebig et al . , 1997 ) . Furthermore , adult ants have been suggested to use a combination of trophallaxis and allo-grooming to share cuticular hydrocarbons ( CHCs ) which are important in providing a specific ‘colony odor’ ( Boulay et al . , 2000a; Soroker et al . , 1995 ) , although the presence of CHCs in trophallactic fluid has not been directly demonstrated . Given these food-independent trophallaxis events , and the potential of this behavior to permit both ‘private’ inter-individual chemical exchange as well as rapid distribution of fluids over the social network of a colony , we tested the hypothesis that trophallaxis serves as an additional means of chemical communication and/or manipulation . To identify the endogenous molecules exchanged during this behavior , we used mass-spectrometry and RNA sequencing to characterize the contents of trophallactic fluid , and identified many growth-related proteins , CHCs , small RNAs , and the insect developmental regulator , juvenile hormone . We also obtained evidence that some components of trophallactic fluid can be modulated by social environment , and may influence larval development . Analysis of the molecules exchanged during trophallaxis necessitated development of a robust method for acquiring trophallactic fluid ( TF ) . We focused on the Florida carpenter ant , Camponotus floridanus , which is a large species whose genome has been sequenced ( Bonasio et al . , 2010 ) . We first attempted to collect fluid from unmanipulated pairs of workers engaged in trophallaxis , but it was impossible to predict when trophallaxis would occur , and events were usually too brief to collect the fluid being exchanged . We found that after workers were starved and isolated from their colony , then fed a 25% sucrose solution and promptly reunited with a similarly conditioned nestmate , approximately half of such pairs displayed trophallaxis within the first minute of reunion ( as observed previously [Boulay et al . , 2000a; Dahbi et al . , 1999] ) , and were more likely to remain engaged in this behavior for many seconds or even minutes . Under these conditions , it was sometimes possible to collect small quantities of fluid from the visible droplet between their mouthparts ( referred to as ‘voluntary’ samples ) . However , even under these conditions , trophallaxis was easily interrupted , making this mode of collection extremely low-yield . To obtain larger amounts of TF and avoid the confounding factors of social isolation and feeding status , we developed a non-lethal method to collect the contents of the crop by lightly-squeezing the abdomen of CO2-anesthetised ants ( referred to as ‘forced’ samples , similar to [Hamilton et al . , 2011] ) . This approach yielded a volume of 0 . 34 ± 0 . 27 µL ( mean ± SD ) of fluid per ant . To determine whether the ‘forced’ fluid collected under anesthesia was similar to the fluid collected from ants voluntarily engaged in trophallaxis , and ensure that it was not contaminated with hemolymph or midgut contents , we also collected samples of these fluid sources . To compare the identities and quantities of the different proteins found in each fluid , all samples were analyzed by nanoscale liquid chromatography coupled to tandem mass spectrometry ( nano-LC-MS/MS ) ( Figure 1A ) . Hierarchical clustering of normalized spectral counts of the proteins identified across our samples revealed high similarity between voluntary and forced TF , but a clear distinction of these fluids from midgut contents or hemolymph ( Figure 1A , for protein names and IDs see Figure 1—figure supplement 1 ) . While the voluntary TF samples contained fewer identified proteins than the forced TF samples – likely due to lower total collected TF volume per analyzed sample ( < 1 µL voluntary vs . > 10 µL forced ) – the most abundant proteins were present across all samples in both methods of collection ( Figure 1A ) . To investigate whether the few differences observed between the voluntary and forced TF in the less abundant proteins might be due to the starvation and/or social isolation conditions used to collect voluntary TF , we isolated groups of 25–30 ants from their respective queens and home colonies for 14 days , with constant access to food and water , and collected TF both directly before and after the period of isolation . Social isolation affected the ratios of proteins in TF , with five of the top 40 proteins in TF becoming significantly less abundant and one more abundant when ants were socially isolated ( Figure 1B; see Figure 1—figure supplement 2 for names and IDs ) . Three of the proteins down-regulated in social isolation were also significantly less abundant in voluntary TF samples ( from socially isolated ants ) relative to forced TF samples ( from within-colony ants , Figure 1—figure supplement 2 ) . Together these results provide initial evidence that the composition of this fluid is influenced by social and/or environmental experience of an ant , and support the validity of our methodology to force collect TF . Of the 50 most abundant proteins found in TF ( Figure 1C ) , many are likely to be digestion-related ( e . g . three maltases , one amylase , various proteases , two glucose dehydrogenases , one DNase ) and three Cathepsin D homologs might have immune functions ( Hamilton et al . , 2011 ) . However , at least 10 of the other proteins have putative roles in the regulation of growth and development , including two hexamerins ( nutrient storage proteins [Zhou et al . , 2007] ) , a yellow/major royal jelly protein ( MRJP ) homolog ( likely nutrient storage [Drapeau et al . , 2006] ) and an apolipophorin ( vitellogenin-domain containing lipid-transport protein [Kutty et al . , 1996] ) . Most notably , TF contained several proteins that are homologous to insect juvenile hormone esterases ( JHEs ) and Est-6 in Drosophila melanogaster ( Figure 1D ) . JHEs are a class of carboxylesterases that degrade the developmental regulator , juvenile hormone ( JH ) ( Kamita and Hammock , 2010; Nijhout et al . , 2014 ) , and Est-6 is an abundant esterase in D . melanogaster seminal fluid ( Mane et al . , 1983; Richmond et al . , 1980; Younus et al . , 2014 ) . While most insects have only one or two JHE-like proteins ( Nijhout et al . , 2014; Qu et al . , 2015 ) , C . floridanus has an expansion of more than 10 related proteins , eight of which were detected in TF ( Figure 1D , Figure 1—figure supplements 1 and 2 ) and two in hemolymph ( Figure 1—figure supplement 1 ) . Three of the abundant JHE-like proteins are significantly down-regulated in the TF of ants that have undergone social isolation ( Figure 1—figure supplement 2 ) . Thirty-three of the 50 most abundant TF proteins had predicted N-terminal signal peptides , suggesting that they can be secreted directly into this fluid by cells lining the lumen or glands connected to the alimentary canal ( Figure 5—figure supplement 1 ) . Moreover , half of the proteins without such a secretion signal had gene ontology terms indicating extracellular or lipid-particle localization , which suggests that they may gain access to the lumen of the foregut through other transport pathways . Many small RNAs have been found in externally secreted fluids across taxa , such as seminal fluid , saliva , milk and royal jelly ( Sarkies and Miska , 2013; Weber et al . , 2010; Guo et al . , 2013 ) . Although the functions of extracellular RNAs remain unclear ( Turchinovich et al . , 2012 ) , we investigated if TF also contains such molecules by isolating and sequencing small RNAs from C . floridanus TF . After filtering out RNA corresponding to potential commensal microorganisms ( including the known symbiont , Blochmannia floridanus; Supplementary file 1 ) and other organic food components , we detected 64 miRNAs . Forty-six of these were identified based upon their homology to miRNAs of the honey bee Apis mellifera ( Guo et al . , 2013; Søvik et al . , 2015; Greenberg et al . , 2012 ) , while 18 sequences ( bearing the structural stem-loop hallmarks of miRNA transcripts ) were specific to C . floridanus ( Figure 2 , Figure 2—source data 1 ) . The most abundant of the 64 miRNAs was miR-750 , followed by three C . floridanus-specific microRNAs . The role of miR-750 is unknown , but the expression of an orthologous miRNA in the Asian tiger shrimp , Penaeus monodon , is regulated by immune stress ( Kaewkascholkul et al . , 2016 ) . Notably , 16 of the miRNAs detected in the C . floridanus TF are also present in A . mellifera worker jelly and/or royal jelly ( Guo et al . , 2013 ) , which are oral secretions fed to larvae to bias them toward a worker or queen fate . Trophallaxis has long been suggested to contribute to the exchange and homogenization of colony odor ( Crozier and Dix , 1979 ) . The observation that different ant species have distinct blends of non-volatile , cuticular hydrocarbons ( CHCs ) – which also display quantitative variation within species – have made these chemicals prime candidates for conveying nestmate recognition cues ( Sharma et al . , 2015; Bos et al . , 2011; Hefetz , 2007; Lahav et al . , 1999; Ozaki et al . , 2005; van Zweden and d’Ettorre , 2010 ) . CHC profiles of ants within a colony are similar , while individuals isolated from their home colony often vary in their CHC make-up ( Boulay and Lenoir , 2001 ) , suggesting that the profiles are constantly unified between nestmates ( Boulay et al . , 2000a ) . However , it remains unclear if CHCs are exchanged principally by trophallaxis , or through other mechanisms such as allo-grooming and contact with the nest substrate ( Boulay et al . , 2000a; Soroker et al . , 1995; Bos et al . , 2011 ) . We therefore investigated whether CHCs are present in TF , and how they relate to those present on the cuticle . To do this , we analyzed by gas chromatography mass spectrometry ( GC-MS ) the TF and cuticular extracts collected from the same five groups of 20–38 workers . We identified 61 molecules in TF ( Table 1 ) , including fatty acids and fatty acid esters , linear alkanes , double bonded hydrocarbons , branched hydrocarbons , and a cholesterol-like molecule . The most abundant TF compounds comprised 27 or more carbons . Cuticular extracts also contained predominantly multiply branched alkanes with 27 or more carbons , corresponding to previous observations of CHCs in this species ( Sharma et al . , 2015; Endler et al . , 2004 ) ( Figure 3 ) . All the highly abundant hydrocarbons in TF ( marked in Figure 3 ) were also present on the cuticle of the workers analyzed , supporting the potential for trophallaxis to mediate inter-individual CHC exchange . Moreover , there was a significantly greater ( t-test , p<0 . 0003 ) similarity across colonies in the hydrocarbon profiles of TF than hydrocarbon profiles of the cuticle ( similarity was measured by cross-correlation between GC-MS profiles , Figure 3D–E ) . Altogether , these observations indicate that while CHCs are likely to be exchanged by trophallaxis , additional mechanisms are probably involved in generating the colony-specific bouquet of these compounds . Given the abundance of the expanded family of JHE-like proteins in TF , we asked whether JH itself is also present in this fluid . The primary JH found in Hymenoptera , JH III , is thought to circulate in the hemolymph after being produced by the corpora allata ( Wigglesworth , 1936 ) . To detect and quantify JH in both TF and hemolymph , we employed a derivatization and purification process prior to GC-MS analysis ( Brent and Vargo , 2003; Shu et al . , 1997; Bergot et al . , 1981 ) . While both sets of measurements were highly variable across samples , we found high levels of JH in TF with concentrations of the same order of magnitude as those found in hemolymph ( Figure 4A ) . Because JH is an important regulator of development , reproduction and behavior ( Nijhout and Wheeler , 1982 ) , we investigated whether the dose of JH that a larva receives by trophallaxis with a nursing worker might be physiologically relevant . If an average worker has approximately 0 . 34 μL of TF in her crop at a given time ( as measured in our initial experiments; see above ) , this corresponds to a dose of approximately 31 pg of JH . The analysis of 35 larvae collected midway through development ( i . e . , third instar out of four worker instars , mean ± SD: 4 . 0 ± 0 . 19 mm long and 1 . 4 ± 0 . 12 mm wide ) revealed that they contained 100–700 pg of JH ( Figure 4B ) . Thus , the amount of JH received during an average trophallaxis event amounts to 5–31% of the JH content of a third instar larva . While it is difficult to determine how much of the JH fed to larvae remains in the larval digestive tract , these results indicate that there is potentially sufficient JH in a single trophallaxis-mediated feeding to shift the titer of a recipient larva . We next determined whether adding exogenous JH to the food of nursing workers could change the growth of the reared larvae . We created groups of 25–30 workers and allowed them to each rear 5–10 larvae to pupation , in the presence of food and sucrose solution that was supplemented with either JH or only a solvent . Larvae reared by JH-supplemented workers grew into larger adults than those reared by solvent-supplemented controls ( Figure 4C , GLMM , p<9 . 01e−06 ) . Moreover , larvae reared by JH-supplemented caretakers were twice as likely to successfully undergo metamorphosis relative to controls ( Figure 4D , binomial GLMM , p<7 . 39e−06 ) . These results are consistent with previous studies in other species of ants and bees using methoprene ( a non-hydrolyzable JH analog ) , whose external provision to the colony can lead to larvae developing into larger workers and even queens ( Nijhout and Wheeler , 1982; Libbrecht et al . , 2013; Wheeler and Nijhout , 1981 ) . To expand our survey of TF , we collected this fluid from other species of social insects: a closely related ant ( C . fellah ) , an ant from another sub-family ( the fire ant Solenopsis invicta ) and the honey bee A . mellifera . Nano-LC-MS/MS analyses identified 79 , 350 and 136 proteins in these three species , respectively ( 84 were identified in C . floridanus ) . We assigned TF proteins from all four analyzed species to 138 distinct groups of predicted orthologous proteins ( see Materials and methods ) ; of these , 72 proteins were found in the TF of only one species ( Figure 5 , Figure 5—source data 1 ) . Only eight ortholog groups contained representatives present in the TF of all four species ( Figure 5B ) . Most of these appear to have functions related to digestion , except for apolipophorin , which is involved in lipid/nutrient transport ( Kutty et al . , 1996 ) . For ortholog groups found in the TF of the three ant species , most were also digestion-related with the notable exception of CREG1 , a secreted glycoprotein that has been implicated in cell growth control ( Di Bacco and Gill , 2003 ) and insect JH response ( Li et al . , 2007; Barchuk et al . , 2007; Zhang et al . , 2014 ) . Genus- and species-specific proteins were frequently associated with growth or developmental roles . For example , A . mellifera TF contained 12 distinct major royal jelly proteins ( MRJP ) , which are thought to be involved in nutrient storage and developmental fate determination ( Drapeau et al . , 2006; Kamakura , 2011 ) , while S . invicta TF contained a highly abundant JH-binding protein and a vitellogenin . The two Camponotus species shared many orthologous groups , consistent with their close phylogenetic relationship . Five of the seven JHE/Est-6 proteins found by Scaffold in C . floridanus TF also had orthologs present in C . fellah TF . Additionally , these two species shared a MRJP/Yellow homolog , and an NPC2-related protein , which , in D . melanogaster , is involved in sterol binding and ecdysteroid biosynthesis ( Huang et al . , 2007 ) . Finally , 26 ortholog groups contained representatives from multiple species , but not the most closely related ones ( e . g . , A . mellifera and S . invicta , or S . invicta and only one of the two Camponotus species ) . One-third of these were associated with growth and developmental processes ( e . g . , three hexamerins , two MRJP/yellow proteins , imaginal disc growth factor 4 , vitellogenin-like Vhdl and an additional NPC2 ) . Together these analyses indicate that approximately half of all TF protein ortholog groups appear to be digestion-related , consistent with TF being composed of the contents of the foregut . However , many TF proteins have putative roles in growth , nutrient storage , or the metabolism and transport of JH , vitellogenin or ecdysone . We have characterized the fluid that is orally exchanged during trophallaxis , a distinctive behavior of eusocial insects generally considered as a means of food sharing . Our results reveal that the transmitted liquid contains much more than food and digestive enzymes , and includes non-proteinaceous and proteinaceous molecules implicated in chemical discrimination of nestmates , growth and development , and behavioral maturation . These findings suggest that trophallaxis underlies a private communication mechanism that can have multiple phenotypic consequences . More generally , our observations open the possibility that exchange of oral fluids ( e . g . , saliva ) in other animals might also serve functions not previously suspected ( Humphrey and Williamson , 2001; Ribeiro , 1995 ) . In ants , trophallaxis has long been thought to be a mode of transfer for the long-chain hydrocarbons that underlie nestmate recognition ( Boulay et al . , 2000a Dahbi et al . , 1999; Lahav et al . , 1999; van Zweden and d’Ettorre , 2010; Soroker et al . , 1995; Lalzar et al . , 2010 ) . However , previous work has analyzed only the passage of radio-labeled hydrocarbons between individuals ( Boulay et al . , 2000a; Dahbi et al . , 1999; Soroker et al . , 1995 ) and between individuals and substrates ( Bos et al . , 2011 ) . Without explicitly sampling the contents of the crop , it is not possible to differentiate between components passed by trophallaxis or by physical contact . Our study is , to our knowledge , the first to demonstrate that TF contains endogenous long-chain hydrocarbons . Interestingly , we found differences in the hydrocarbon profiles of the TF and the cuticle of the same ants , suggesting the existence of other processes regulating the relative proportion of CHCs and/or additional sources of these compounds . Further work manipulating and comparing hydrocarbon profiles of TF and cuticle of multiple individual ants within the same colony is necessary to understand the impact of the observed variation in TF and cuticular hydrocarbons . Consistent with the view that the gut is one of the first lines of defense in the body’s interface with the outside world ( Lemaitre and Hoffmann , 2007; Söderhäll and Cerenius , 1998 ) , the four TF proteomes analyzed in this study include many potential defense-related proteins . Homologs of the Cathepsin D family of proteins , which have been implicated in growth , defense and digestion ( Saikhedkar et al . , 2015; Ahn and Zhu-Salzman , 2009 ) , were present in the TF of both Camponotus species . Prophenoloxidase 2 , an enzyme responsible for the melanization involved in the insect immune response ( Söderhäll and Cerenius , 1998 ) , was found in the TF of A . mellifera and S . invicta . All four species contained several members of the serine protease and serpin families , which have well-documented roles in the prophenoloxidase cascade and more general immune responses in D . melanogaster and other animals ( Lemaitre and Hoffmann , 2007; Lucas et al . , 2009 ) . The TF of two species had orthologs of the recognition lectin GNBP3 which is also involved in the prophenoloxidase cascade . Finally , a handful chromatin-related proteins ( e . g . , histones , CAF1 ) were found in S . invicta , and to a lesser extent in the other two ants . Their presence may simply reflect a few cells being sloughed from the lumen of the foregut , or could be indicative of a defense process termed ETosis , whereby chromatin is released from the nuclei of inflammatory cells to form extracellular traps that kill pathogenic microbes ( Robb et al . , 2014; Brinkmann et al . , 2004 ) . The presence of development- and growth-related components in the TF of diverse social insects suggests that this fluid may play a role in directing larval development . Previous work has shown that the developmental fate of larvae and the process of caste determination in various ant and bee species can be influenced by treating colony members with JH or JH analogs ( Wheeler , 1986; Wheeler and Nijhout , 1981; Cnaani et al . , 1997; Cnaani et al . , 2000; Rajakumar et al . , 2012 ) . Moreover , in some species , workers play an important role in regulating caste determination ( Kamakura , 2011; Linksvayer et al . , 2011; Villalta et al . , 2016; Mutti et al . , 2011 ) but it is unknown how workers might influence the JH titers of larvae . Our finding that TF contains JH raises the possibility of trophallaxis as a direct means by which larval hormone levels and developmental trajectories can be manipulated . This type of mechanism has some precedent: honey bee workers bias larval development toward queens by feeding them royal jelly , an effect that might be mediated by MRJPs ( Kamakura , 2011; Buttstedt et al . , 2016; Kucharski et al . , 2015; Kamakura , 2016 ) . Across the TF of four species of social insect , we have found diverse molecules intimately involved in insect growth regulation: JH , JHE , JH-binding protein , vitellogenin , hexamerin , apolipophorin , and MRJPs . Several of the proteins and microRNAs identified in ant and bee TF are also found in royal and worker jelly ( Guo et al . , 2013; Zhang et al . , 2014 ) . Notably , there are also some similarities between proteins in social insect TF and mammalian milk ( Zhang et al . , 2015 ) , such as the cell growth regulator CREG1 . The between-genera variation in the most abundant growth-related proteins ( e . g . , MRJPs in honey bee , JH-binding protein and hexamerins in fire ant , and JHEs in Camponotus ) indicates that there might be multiple evolutionary origins and/or rapid divergence in trophallaxis-based signals potentially influencing larval growth . The finding that TF contains many proteins , miRNAs , CHCs and JH raises the question of how they come to be present in this fluid . Many non-food-derived components in TF are likely to be either directly deposited into the alimentary canal: over two-thirds of the proteins present in TF had a predicted signal peptide suggesting that they are probably secreted by the cells lining the foregut or glands connected to the alimentary canal ( e . g . , postpharyngeal , labial , mandibular , salivary glands ) . Additionally , TF molecules are likely to be acquired by transfer among nestmates through licking , grooming , and trophallaxis . For example , CHCs are produced by oenocytes , transmitted to the cuticle , ingested , and sequestered in the postpharyngeal gland ( Bos et al . , 2011; van Zweden and d’Ettorre , 2010; Soroker and Hefetz , 2000 ) . Interestingly , JH is synthesized in the corpora allata , just caudal to the brain ( Goodman and Granger , 2009 ) . Although the mechanisms of JH uptake by tissues are still incompletely understood ( Parra-Peralbo and Culi , 2011; Rodríguez-Vázquez et al . , 2015; Engelmann and Mala , 2000; Suzuki et al . , 2011 ) , the TF proteins apolipophorin , hexamerin , JH-binding protein and vitellogenin have all been implicated in JH binding and transport between hemolymph and tissues ( Goodman and Granger , 2009; Rodríguez-Vázquez et al . , 2015; Engelmann and Mala , 2000; Suzuki et al . , 2011; Amsalem et al . , 2014 ) , and these factors may be responsible for transporting JH into the foregut . Unfortunately , even in Drosophila mechanisms of JH uptake by tissues are still unclear ( Rodríguez-Vázquez et al . , 2015; Engelmann and Mala , 2000; Suzuki et al . , 2011; Parra-Peralbo and Culi , 2011 ) . Extracellular miRNAs are secreted and transported through a variety of pathways , but the functional relevance of such molecules is still controversial ( Sarkies and Miska , 2013; Turchinovich et al . , 2016; Masood et al . , 2016; Søvik et al . , 2015; Rayner and Hennessy , 2013 ) . While the simultaneous presence of JH and a set of putative JH degrading enzymes ( the JHE/Est-6 family ) in C . floridanus TF may appear surprising , there are at least two possible explanations . First , this might reflect a regulatory mechanism for JH levels at the individual and colony levels . Many biological systems use such negative feedback mechanisms to buffer signals and enable rapid responses to environmental change ( Alon , 2007 ) . Alternatively , these enzymes may have evolved a different function along with their novel localization in TF . In other insects JHE-related enzymes are typically expressed in the fat body and circulate in the hemolymph ( Hammock et al . , 1975; Ward et al . , 1992; Campbell et al . , 1998 ) . , The expansion of JHE-like proteins in C . floridanus was accompanied by a high specificity in their localization , with two ( E2AM67 , E2AM68 ) being present exclusively in the hemolymph , and four ( E2ANU0 , E2AI90 , E2AJM0 , E2AJL9 ) exclusively in TF . This coexistence of JH and JHEs reveals a striking parallel with the constituents of a different socially exchanged fluid in D . melanogaster . Drosophila Est-6 is highly abundant in seminal fluid , together with its presumed substrate , the male-specific pheromone 11-cis-vaccenyl acetate ( cVA ) . cVA has multiple roles in influencing sexual and aggressive interactions and its transfer to females during copulation potently diminishes her attractiveness to future potential mates ( Poiani , 2006; Chertemps et al . , 2012; Costa , 1989 ) . Given that larvae are fed JH and food , and both are necessary for development , a future goal will be to dissect the relative contribution and potential synergy of these components . Furthermore , JH has many other functions in social insects , including behavioral modulation , longevity , fecundity and immunity ( Flatt et al . , 2005; Jindra et al . , 2013; Dolezal et al . , 2012; Wang et al . , 2012 ) ; the relevance of trophallaxis-mediated JH exchange between adults remains to be explored . Considering that most , if not all , individuals in a colony must share food through trophallaxis , it will also be of interest to understand if and how individual- , or caste-specific TF-based information signals emerge . Recent work on trophallaxis networks ( Greenwald et al . , 2015 ) and interaction networks ( Mersch et al . , 2013 ) , indicate that ants preferentially interact with others of similar behavioral type ( e . g . , nurses with nurses , and foragers with foragers ) . This raises the possibility that different pools of TF with different qualities exist . A key challenge will be to develop specific genetic and/or pharmacological tools to test the biological relevance of molecules transmitted by trophallaxis . Camponotus floridanus workers came from 16 colonies established in the laboratory from approximately 1-year-old founding queens and associated workers collected from the Florida Keys in 2006 , 2011 and 2012 . Ants were provided once a week with fresh sugar water , an artificial diet of honey or maple syrup , eggs , agar , canned tuna and a few D . melanogaster . Maple syrup was substituted for honey and no Drosophila were provided in development and proteomic experiments to avoid contamination with other insect proteins . Colonies were maintained at 25°C with 60% relative humidity and a 12 hr light:12 hr dark cycle . Camponotus fellah colonies were established from queens collected after a mating flight in March 2007 in Tel Aviv , Israel ( Colonies #5 , 28 , 33 ) . The ant colonies were maintained at 32°C with 60% relative humidity and a 12 hr light:12 hr dark cycle . Fire ant workers ( Solenopsis invicta ) were collected from three different colonies , two polygyne , one monogyne , maintained at 32°C with 60% relative humidity and a 12 hr light:12 hr dark cycle . Honeybee workers ( Apis mellifera , Carnica and Buckfast ) were collected from six different hives maintained with standard beekeeping practices . ‘Voluntary’ TF was collected from individuals who had been starved and socially isolated for 1–3 weeks , then fed 25% sucrose solution , and promptly re-introduced to other separately isolated nestmates . Ants were monitored closely for trophallaxis events; when one had begun , a pulled glass pipette was brought between the mouthparts of the 2–4 individuals and fluid was collected . Typically this stopped the trophallaxis event , but on rare occasions it was possible to acquire up to sub-microliter volumes of TF . For ‘forced’ collection , ants were anesthetized by CO2 ( on a CO2 pad; Flypad , FlyStuff , San Diego , CA ) and promptly flipped ventral-side up . The abdomen of each ant was lightly squeezed with wide insect forceps to prompt the regurgitation of fluids from the social stomach . Ants that underwent anesthesia and light squeezing yielded on average 0 . 34 µL of fluid and recovered in approximately 5 min . TF was collected with graduated borosilicate glass pipettes , and transferred immediately to either buffer ( 100 mM NaH2PO4 , 1 mM EDTA , 1 mM DTT , pH 7 . 4 ) for proteomic experiments , to RNase-free water for RNA analyses , or to pure EtOH for GC-MS measurements . TF was collected from C . fellah and S . invicta in the same manner as from C . floridanus; for S . invicta 100–300 ants were used due to their smaller body size and crop volume . Honey bee TF was collected from bees that were first cold-anesthetized and then transferred to a CO2 pad to ensure continual anesthesia during collection of TF as described above . Honey bees yielded higher volumes of TF than did ants ( 0 . 94 µL TF per honey bee , SD = 0 . 54 relative to 0 . 34 µL TF per C . floridanus ant , SD = 0 . 27 ) . Hemolymph was collected from CO2-anesthetized ants by puncturing the junction between the foreleg and the distal edge of the thoracic plate with a pulled glass pipette . This position was chosen over the abdomen in order to ensure that hemolymph was collected and not the contents of the crop . Approximately 0 . 1 µL was taken from each ant . Midgut samples were collected by first anesthetizing an ant , immobilizing it in warm wax ventral-side up , covering the preparation in 1x PBS , and opening the abdomen with dissection forceps and iris scissors . The midgut was punctured by a sharp glass pipette and its contents collected . Because the pipette also contacts the surrounding fluid , some hemolymph contamination was unavoidable . Hydrocarbon analysis was performed on trophallactic fluid from five groups of 20–38 ants , each collected from one of five different colonies ( C11-C15 ) . TF samples were placed directly into 3:1 hexane:methanol . Immediately after TF collection , body surface CHCs were collected by placing anesthetized ants in hexane for 1 min before removal with cleaned forceps . Methanol was added to the cuticular-extraction hexane ( maintaining the 3:1 proportion of the TF samples ) . The TF and body samples were vortexed for 30 s and centrifuged for 7 min . Hexane fractions were collected using a thrice-washed Hamilton syringe . Samples were kept at −20°C until further analysis . A Trace 1300 GC chromatograph interfaced with a TSQ 8000 Evo Triple Quadrupole Mass Spectrometer ( Thermo Scientific , Bremen , Germany ) was used for the study . Hydrocarbons were separated on a 30 m x 0 . 25 mm I . D . ( 0 . 25 mm film thickness ) Zebron ZB-5 ms capillary column ( Phenomenex , Torrance , CA ) using the following program: initial temperature 70°C held for 1 min , ramped to 210°C at 8 °C/min , ramped to 250°C at 2 °C/min , ramped to 300°C at 8 °C/min and held for 5 min . Helium was used as carrier gas at a constant flow of 1 mL/min . Injections of 1 µL of ants’ TF or body extracts were made using splitless mode . The injection port and transfer line temperature were kept at 250°C , and the ion source temperature set at 200°C . Ionization was done by electron-impact ( EI , 70 eV ) and acquisition performed in full scan mode in the mass range 50–550 m/z ( scan time 0 . 2 s ) . Identification of hydrocarbons was done using XCalibur and NIST 14 library . The TIC MS was integrated and the area reported as a function of Retention time ( Rt , min ) for each peak . Characterization of branched alkanes by GC-MS remains a challenge due to the similarity of their electron impact ( EI ) mass spectra and the paucity of corresponding spectra listed in EI mass spectra databases . A typical GC-MS chromatogram ( Figure 3—figure supplement 1A ) reveals the complexity of the TF sample . The workflow described here was systematically used to characterize the linear and branched hydrocarbons present in TF samples summarized in Table 1 . The parent ion was first determined for each peak after background subtraction . Ambiguities remained in some cases due to the low intensity or absence of the molecular ion . Linear alkanes present in TF samples were localized using a standard mixture of C8-C40 alkanes . Then RI values were deduced for all compounds present in the samples based on their retention times ( Figure 3—figure supplement 1G , red ) . To determine the number of methyl branches for alkanes , we examined the distribution of fragment ions in the spectrum by fitting their intensities with an exponential decay function and specifically looking for the fragment ions that do not fit to the calculated exponential decay function . From the experimental mass spectrum ( Figure 3—figure supplement 1B–C ) , the intensity of all fragment ions was extracted and fitted with the function ( Figure 3—figure supplement 1D–E ) . Figure 3—figure supplement 1 clearly shows two different resulting EI mass spectra profiles: on the left a linear alkane corresponding to n-hexacosane ( Rt = 34 . 69 min , MF C26H54 ) ; on the right , a monomethyl-branched structure most likely corresponding to 9-methylnonacosane ( Rt = 43 . 29 min , MF C30H62 ) with two enhanced fragment ions at m/z 141 and 309 emerging from the curve . Extracting and fitting the fragment ion profiles first helped to discriminate between linear and branched hydrocarbons , but also to estimate the number of methyl branches . To confirm for each compound the number of carbons and branches obtained , we used Kovats retention index ( RI ) values in the NIST Chemistry WebBook ( Linstrom and Mallard , 2000 ) . Based on the RI values for similar GC stationary phase from C15 to C38 hydrocarbons , six different curves of RI vs . number of carbons were constructed from linear to pentamethylated alkanes ( Figure 3—figure supplement 1F ) . To construct the curves , average RI values were taken of all hydrocarbons available in the database , with a given number of carbons and a given number of methyl branches . For example , the RI value of 2409 obtained for C25 and two branches is the average of 3 , ( 7/9/11/13 ) -dimethyltricosane , 3 , ( 3/5 ) -dimethyltricosane and 5 , ( 9/11 ) -dimethyltricosane values listed in the NIST Chemistry WebBook for the same stationary phase . Those curves were used to check for every compound that , from the measured RI value and the number of carbons found , the number of ramifications found fit properly with the curves . Once the number of branches and the parent ion mass were known , the position of the different branches could be deduced from the fragment ion values . When ambiguity remained on the position of the branch or double bond , this is indicated with an asterisk in Table 1 . The plot of experimental retention times for each compound as a function of the RI index closely fitted the plot using RI values given by NIST for the identified compounds ( Figure 3—figure supplement 1G ) , bringing additional confidence to the identifications . Table 1 summarizes the proposed structures for hydrocarbons and other compounds detected in TF samples . For each sample , a known quantity of TF or hemolymph was collected into a graduated glass capillary tube and blown into an individual glass vial containing 5 µL of 100% ethanol . Samples were kept at −20°C until further processing . This biological sample was added to a 1:1 mixture of isooctane and methanol , vortexed for 30 s , and centrifuged for 7 min at maximum speed . Avoiding the boundary between phases , the majority of the isooctane layer and the methanol layer were removed separately , combined and stored at –80°C until analysis . Before analysis , 50% acetonitrile ( HPLC grade ) was added . Prior to purification , farnesol ( Sigma-Aldrich , St Louis , MO ) was added to each sample to serve as an internal standard . Samples were extracted three times with hexane ( HPLC grade ) . The hexane fractions were recombined in a clean borosilicate glass vial and dried by vacuum centrifugation . JH III was quantified using the gas chromatography mass spectrometry ( GC–MS ) method of Bergot et al . ( 1981 ) as modiﬁed by Shu et al . ( 1997 ) and Brent and Vargo ( 2003 ) . The residue was washed out of the vials with three rinses of hexane and added to borosilicate glass columns filled with aluminum oxide . In order to filter out contaminants , samples were eluted through the columns successively with hexane , 10% ethyl ether–hexane and 30% ethyl ether–hexane . After drying , samples were derivatized by heating at 60°C for 20 min in a solution of methyl-d alcohol ( Sigma-Aldrich ) and trifluoroacetic acid ( Sigma-Aldrich , St Louis , MO ) . Samples were dried down , resuspended in hexane , and again eluted through aluminum oxide columns . Non-derivatized components were removed with 30% ethyl ether . The JH derivative was collected into new vials by addition of 50% ethyl-acetate–hexane . After drying , the sample was resuspended in hexane . Samples were then analyzed using an HP 7890A Series GC ( Agilent Technologies , Santa Clara , CA ) equipped with a 30 m x 0 . 25 mm Zebron ZB-WAX column ( Phenomenex , Torrence , CA ) coupled to an HP 5975C inert mass selective detector . Helium was used as a carrier gas . JH form was confirmed by first running test samples in SCAN mode for known signatures of JH 0 , JH I , JH II , JH III and JH III ethyl; JH III was confirmed as the primary endogenous form in this species . Subsequent samples were analyzed using the MS SIM mode , monitoring at m/z 76 and 225 to ensure specificity for the d3-methoxyhydrin derivative of JH III . Total abundance was quantified against a standard curve of derivatized JH III , and adjusted for the starting volume of TF . The detection limit of the assay is approximately one pg . To determine the effect of exogenous JH on larval development , ants were taken from laboratory C . floridanus colonies ( Expt 1: C2 , C3 , C5 , C9 , C16 , C17; Expt 2: C4 , C5 , C6 , C18; Expt 3: C1 , C5 , C6 , C7 , C11 , C16 , C19 ) . Approximately 90% of the ants were taken from inside the nest on the brood , while the remaining 10% were taken from outside the nest . Each colony explant had 20–30 workers ( each treatment had the same number of replicates of any given colony ) and was provided with five to ten second or third instar larvae from their own colony of origin ( staged larvae were equally distributed across replicates ) . Each explant was provided with water , and either solvent- or JH III-supplemented 30% sugar water and maple-syrup-based ant diet ( 1500 ng of JH III in 5 µL of ethanol was applied to each 3×3×3 mm food cube and sucrose solution had 83 ng JH III/µL ) . No insect-based food was provided . Food sources were refreshed twice per week . JH was found to transition to JH acid gradually at room temperature , where after 1 week ~ 50% was JH acid ( as measured by radio-assay as in Kamita et al . , 2011 , data not shown ) . Twice weekly before feeding , each explant was checked for pupae , and developing larvae were measured and counts using a micrometer in the reticle of a stereomicroscope . Upon pupation , or cocoon spinning , larvae/pupae were removed from the care of workers and kept in a clean humid chamber until metamorphosis . Cocoons were removed using dissection forceps . The head width of the pupae was measured using a micrometer in the reticle of a stereomicroscope 1–4 days after metamorphosis ( immediately after removal of the larval sheath , head width is not stable ) . Long-term development experiments were stopped when fewer than three larvae remained across all explants and these larvae had not changed in size for 2 weeks . Of larvae that did not successfully undergo metamorphosis , approximately 75% were eaten by nursing workers at varying developmental stages over the course of the experiment . The remaining non-surviving larvae were split between larvae that finished the cocoon spinning phase ( Wallis , 1960 ) but did not complete metamorphosis and larvae that had ceased to grow by the end of the approximately 10-week experiment . In order to identify orthologous proteins across the species TF , we first needed to determine orthology across the four species . Compiling RefSeq , UniProt , and transcriptome protein models from the four species yielded 131 , 122 predicted protein sequences . For C . floridanus , A . mellifera and S . invicta , this was done for both NCBI RefSeq and UniProt databases because there are discrepancies in annotation and thus in protein identification between databases . We determined 21 , 836 groups of one-to-one orthologs using OMA stand-alone ( Altenhoff et al . , 2015 ) v . 1 . 0 . 3 , RRID:SCR_011978 , although only 4538 orthologous groups had members in all four species . Default parameters were used with the exception of minimum sequence length , which was lowered to 30 aa . The 40 proteins with the highest average NSAF value across TF samples of that species were selected . For each of the top proteins , we checked across the other species and databases for orthologous proteins . If an orthologous protein was identified in the proteome , we then checked if that protein was also present in that species’ TF , even at low abundance . The protein sequences of our proteomically identified TF C . floridanus JHE/Est-6 proteins , functionally validated JHEs ( Tribolium castaneum JHE ( UniProt D7US45 ) , A . mellifera JHE ( Q76LA5 ) , Manduca sexta JHE ( Q9GPG0 ) , D . melanogaster JHE ( A1ZA98 ) and Culex quinquefasciatus ( R4HZP1 ) ) and Est-6 proteins from A . mellifera ( B2D0J5 ) and D . melanogaster ( P08171 ) were aligned using PROMALS3D with the crystal structure of the M . sexta JHE ( 2fj0 ) used as a guide . Phylogeny was inferred using Randomized Axelerated Maximum Likelihood ( RAxML , RRID:SCR_006086; Stamatakis , 2006 ) , with 100 bootstrapped trees . The dendrogram was visualized in FigTree ( v1 . 4 . 2 , RRID:SCR_008515 ) . The proteomic sample sizes were determined by the variation observed in protein IDs and abundances in unmanipulated samples . Visualization together with hierarchical clustering was done in R version 3 . 0 . 2 ( www . r-project . org , RRID:SCR_001905 ) using the ‘heatmap . 3’ package in combination with the pvclust package . Heatmap visualization without clustering was done in MATLAB ( 2012b , RRID:SCR_001622 ) . For experiments in Figures 3 and 4 , the number of colonies and number of replicates or ants per colony were determined by both preliminary trials to assess sample variation and the health/abundance of the C . floridanus ant colonies available in the lab . Hydrocarbon GC-MS traces were analyzed in MATLAB normalizing the abundance ( area under the curve ) of each point by maximum and minimum value within the CHC retention time window . Peaks were found using ‘findpeaks’ with a lower threshold abundance of 7% of the total abundance for that sample . To compare across samples , peaks were filtering into 0 . 03 min bins by retention time . Cross-correlation was computed using the ‘xcorr’ function . For long-term development experiments , the number of same-staged larvae per colony was the limiting factor for the number of replicates per experiment . Because of this limitation , the long-term development experiment testing the effect of JH was fully repeated three times . Colonies were hibernated approximately 1 month prior to each of these experiments to maximize number of same-staged brood . General linear mixed models ( GLMM ) were used so that colony , replicate and experiment could be considered as random factors . Models were done in R using ‘lmer’ and ‘glmer’ functions of the lme4 package , and p values were calculated with the ‘lmerTest’ package in R . Overall , no data points were excluded as outliers and all replicates discussed are biological not technical replicates .	Ants , bees and other social insects live in large colonies where all the individuals work together to gather food , rear young and defend the colony . This level of cooperation requires the insects in the colony to be able to communicate with each other . Most social insects share fluid mouth-to-mouth with other individuals in their colony . This behavior , called trophallaxis , allows these species to pass around food , both between adults , and between adults and larvae . Trophallaxis therefore creates a network of interactions linking every member of the colony . With this in mind , LeBoeuf et al . investigated whether trophallaxis may also be used by ants to share information relevant to the colony as a form of chemical communication . The experiments show that in addition to food , carpenter ants also pass small ribonucleic acid ( RNA ) molecules , chemical signals that help them recognize nestmates , and many proteins that appear to be involved in regulating the growth of ants . LeBoeuf et al . also found that trophallactic fluid contains juvenile hormone , an important regulator of insect growth and development . Adding juvenile hormone to the food that adult ants pass to the larvae made it twice as likely that the larvae would survive to reach adulthood . This indicates that proteins and other molecules transferred mouth-to-mouth over this social network could be used by the ants to regulate how the colony develops . The next steps following on from this work will be to investigate the roles of the other components of trophallactic fluid , and to examine how individual ants adapt the contents of the fluid in different social and environmental conditions . Another challenge will be to determine how specific components passed to larvae in this way can control their growth and development .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology", "genetics", "and", "genomics" ]	2016	Oral transfer of chemical cues, growth proteins and hormones in social insects
Hypoxia and ischemia are linked to oxidative stress , which can activate the oxidant-sensitive transient receptor potential ankyrin 1 ( TRPA1 ) channel in cerebral artery endothelial cells , leading to vasodilation . We hypothesized that TRPA1 channels in endothelial cells are activated by hypoxia-derived reactive oxygen species , leading to cerebral artery dilation and reduced ischemic damage . Using isolated cerebral arteries expressing a Ca2+ biosensor in endothelial cells , we show that 4-hydroxynonenal and hypoxia increased TRPA1 activity , detected as TRPA1 sparklets . TRPA1 activity during hypoxia was blocked by antioxidants and by TRPA1 antagonism . Hypoxia caused dilation of cerebral arteries , which was disrupted by antioxidants , TRPA1 blockade and by endothelial cell-specific Trpa1 deletion ( Trpa1 ecKO mice ) . Loss of TRPA1 channels in endothelial cells increased cerebral infarcts , whereas TRPA1 activation with cinnamaldehyde reduced infarct in wildtype , but not Trpa1 ecKO , mice . These data suggest that endothelial TRPA1 channels are sensors of hypoxia leading to vasodilation , thereby reducing ischemic damage . Interruption of regional blood flow within the brain can rapidly cause irreparable neuronal damage . The cerebral circulation exhibits unique capabilities that allows it to adjust to varying environmental conditions and pathophysiological situations so as to maintain optimal perfusion and minimize such injury . G-protein-coupled receptors and ion channels present on the endothelial cells and vascular smooth muscle cells ( SMCs ) that form the walls of cerebral blood vessels initiate many of the signaling cascades that enable these intrinsic adaptive processes . The specific cellular pathways responsible for cerebrovascular homeostasis are of considerable interest as potential therapeutic targets for diseases associated with impaired blood flow regulation within the brain , such as stroke and vascular cognitive impairment; however , these pathways remain incompletely understood . We recently reported that transient receptor potential ankyrin 1 ( TRPA1 ) cation channels are present in the endothelium of arteries within the brain , but not in other arterial beds ( Sullivan et al . , 2015 ) , suggesting a specialized role for these channels in cerebral blood flow regulation . In the current study , we investigated the endogenous adaptive and protective functions of TRPA1 channels within the cerebral vasculature . TRPA1 , the sole member of the mammalian ankyrin TRP subfamily , is a large-conductance , Ca2+-permeable , non-selective cation channel ( Nagata et al . , 2005; Earley and Brayden , 2015; Karashima et al . , 2010 ) . TRPA1 channels are present on perivascular nerves surrounding small mesenteric arteries , and their stimulation causes vasodilation through release of C-protein gene-related peptide ( Bautista et al . , 2005 ) . In addition , TRPA1 channels form a Ca2+-signaling complex with intermediate-conductance , Ca2+-activated potassium channels ( KCa3 . 1 ) at sites of close contact between cerebral artery endothelial cells and underlying SMCs ( Earley et al . , 2009 ) . Consequently , activation of TRPA1 channels , for example with the electrophilic compound allyl isothiocyanate ( AITC ) , derived from mustard oil , induces endothelium-dependent vasodilation of cerebral resistance arteries , an effect that is prevented by blockade of KCa3 . 1 channels ( Earley et al . , 2009 ) . Endogenous regulation of TRPA1 in the cerebral endothelium has been linked to reactive oxygen species ( ROS ) , particularly superoxide anions ( O2- ) generated by NADPH oxidase 2 ( NOX2 ) ( Sullivan et al . , 2015 ) . ROS-induced activation of TRPA1 has been shown to occur through a process that requires generation of lipid peroxidation products that directly activate the channel , such as 4-hydroxynonenal ( 4-HNE ) ( Sullivan et al . , 2015; Trevisani et al . , 2007 ) . Increased generation of ROS and oxidative stress is a hallmark of vascular diseases , but the importance of TRPA1-mediated vasodilation for the regulation of blood flow in the brain under pathological conditions remains unknown . Cerebral blood vessels have the inherent ability to dilate in response to hypoxia ( Kontos et al . , 1978 ) . When hypoxia occurs within specific regions of the brain , this adaptation increases localized blood flow to the affected area , enhancing delivery of oxygen . Although the molecular and cellular mechanisms responsible for this important response are not known , a recent study suggested that TRPA1 channels are sensitive to changes in O2 concentration and are activated by both hyperoxic and hypoxic conditions ( Takahashi et al . , 2011 ) . In addition , hypoxia and ischemia are linked to elevated ROS production and increased generation of 4-HNE ( Schmidt et al . , 1996; Kunstmann et al . , 1996 ) , an endogenous activator of TRPA1 activity . Here , we tested the hypothesis that hypoxia causes activation of TRPA1 channels in the cerebral endothelium , resulting in vasodilation , and further propose that this response constitutes an adaptive response during perfusion deficiency . Using an integrative approach and a newly developed mouse model expressing a Ca2+ biosensor exclusively in endothelial cells , we elucidated the underlying molecular mechanisms responsible for activation of TRPA1 channels in the intact cerebral endothelium during hypoxia and investigated the pathophysiological impact of this novel pathway in vivo using an established model of ischemic strokes . Our findings suggest that TRPA1 channels in the cerebral endothelium are early sensors of hypoxia and initiate an adaptive response to reduce ischemic damage in the brain . A primary goal of this study was to determine the direct effects of hypoxia on TRPA1 channels that are present in the endothelium of intact cerebral arteries ( Figure 1—figure supplement 1 ) . An effective strategy for directly measuring changes in channel activity is to record local , transient elevations in cytosolic Ca2+ levels generated by Ca2+ influx through single TRPA1 channels in the endothelium , detected as TRPA1 sparklets ( Sullivan et al . , 2015 ) , using optical patch-clamp methods ( Sullivan and Earley , 2013 ) ( Figure 1A ) . A previous study reported the use of transgenic mice expressing the Gcamp2 Ca2+ biosensor under the control of the Cx40 promoter to record TRPV4 sparklets in the endothelium of mesenteric arteries ( Sonkusare et al . , 2012 ) . However , we found that the fluorescence intensity of Cx40-based Gcamp2 ( and Gcamp5 ) Ca2+ biosensors was very low in the endothelium of cerebral arteries , preventing reliable recordings from being obtained . We also found that the cerebral endothelium could not be effectively loaded with standard Ca2+ indicator dyes , such as Fluo-4AM . To overcome these limitations , we generated a new transgenic mouse line that expresses Gcamp6f , a Ca2+ biosensor with fast kinetics ( Chen et al . , 2013 ) , exclusively in the endothelium . To accomplish this , we crossed mice heterozygous for the expression of a floxed Stop codon upstream of the Gcamp6f gene ( see Materials and methods ) with mice heterozygous for the expression of cre-recombinase under the control of the endothelial-specific Tek promoter/enhancer ( Tekcre ) . The mT/mG reporter mouse ( Muzumdar et al . , 2007 ) was used to confirm that cre-recombinase is only expressed in the endothelium of cerebral arteries from Tekcre mice ( Figure 1—figure supplement 2 ) . We found that cerebral , mesenteric , skeletal muscle and pulmonary arteries isolated from Tek:Gcamp6f mice expressed the Ca2+ biosensor only in the endothelium and provided excellent signal-to-noise ratio for optical detection of spontaneous and evoked Ca2+ signals in this tissue ( Videos 1–4 ) . To initially characterize TRPA1 sparklets in the intact cerebral endothelium , we mounted pial arteries from Tek:Gcamp6f mice en face as previously described ( Sonkusare et al . , 2012 ) . The endothelium of arteries mounted en face is not subjected to physiological levels of luminal shear stress and intraluminal pressure and cannot be used to study how these stimuli affect endothelial cell function . However , Sonkusare et al . showed that the TRPV4 sparklets recorded from the endothelium of arteries mounted en face are statistically indistinguishable from those recorded from intact arteries pressurized at physiological levels , suggesting that longitudinal incision of the artery does not alter basic ion channel properties ( Sonkusare et al . , 2012 ) . To isolate Ca2+ influx events , we treated initial preparations with the sarcoplasmic/endoplasmic reticulum Ca2+-ATPase ( SERCA ) inhibitor cyclopiazonic acid ( CPA , 30 μM ) to prevent release of Ca2+ from intracellular stores . Vessels used for these Ca2+ signaling experiments were also treated with the cell-permeant Ca2+ chelator EGTA-AM ( 10 μM ) to limit the intracellular diffusion of Ca2+ influx and improve the signal-to-noise ratio . Using these conditions , the endothelium was exposed to a near-maximally effective concentration ( 1 μM ) of the endogenous TRPA1 agonist 4-HNE ( Sullivan et al . , 2015 ) via the superfusing bath . Ca2+ events were recorded at an average of ~40 frames/s in a 512 × 512 pixel field of view ( pixel size = 0 . 27 μm ) containing approximately 40 endothelial cells ( mean area of cerebral artery endothelial cells = 462 ± 24 μm2 , n = 25 cells , N = 5 mice ) . We observed that 4-HNE significantly increased the frequency of highly localized , transient Ca2+ signals by approximately 10-fold compared with vehicle controls ( 0 . 51 ± 0 . 05 Hz vs . 0 . 07 ± 0 . 01 Hz ) ( Figure 1D and Videos 5 and 6 ) . Notably , these signals exhibited distinct amplitude levels and duration reminiscent of single-channel activity recorded using patch-clamp electrophysiology ( Demuro and Parker , 2004 ) ( Figure 1C ) . The selective TRPA1 antagonist A967079 ( 1 μM ) ( Chen et al . , 2011 ) significantly diminished the 4-HNE–induced increase in activity , reducing the frequency of these Ca2+ signals from 0 . 51 ± 0 . 06 Hz to 0 . 06 ± 0 . 02 Hz ( Figure 1E ) . The frequency of Ca2+ signals induced by exposure to 4-HNE was also significantly reduced by removal of extracellular Ca2+ , confirming that these signals are generated by an influx of Ca2+ ( Figure 1—figure supplement 3 ) . Collectively , these data identify the local , transient Ca2+ signals evoked by administration of 4-HNE as bona fide TRPA1 sparklets . Further analyses showed that the number of active sites per field of view significantly increased from 1 . 68 ± 0 . 27 under control conditions to 4 . 41 ± 0 . 52 in the presence of 4-HNE ( Figure 1B and D , data shown as number of sites per cell ) . In addition , sparklet frequency at existing sites significantly increased from 0 . 06 ± 0 . 01 Hz at baseline to 0 . 13 ± 0 . 02 Hz following application of 4-HNE ( Figure 1C and D ) . These findings indicate that the increase in TRPA1 sparklet frequency induced by administration of 4-HNE resulted from both the recruitment of previously inactive sparklet sites as well as an increase in the frequency of previously active sites . TRPA1 inhibition with the selective inhibitor A967079 largely prevented the increases in the number of active sparklets sites as well as the increase in sparklet frequency at previously active sites ( Figure 1B , C and E ) . A plot of the amplitudes ( ΔF/F0 ) of individual TRPA1 sparklets revealed a Gaussian distribution with a mode of 1 . 10 ( Figure 1—figure supplement 4 ) . The mean attack time ( half-time to reach peak signal amplitude ) , decay time ( half-time from peak to loss of signal ) , and duration of mode-amplitude TRPA1 sparklets were 87 ± 7 ms , 84 ± 6 ms , and 373 ± 23 ms , respectively . The mean spatial spread of TRPA1 sparklets was approximately 17 . 2 ± 0 . 6 μm2 , or ~3% of the total surface area of a single endothelial cell in this intact vascular preparation ( Supplementary file 1 ) . We previously reported that stimulation of TRPA1 channels with the electrophilic compound allyl isothiocyanate ( AITC ) increased the frequency of dynamic Ca2+ release from the ER due to Ca2+-induced Ca2+ release ( Qian et al . , 2013 ) . To determine if 4-HNE -stimulated TRPA1 sparklets have a similar effect on ER Ca2+ release , we exposed cerebral arteries from Tek:Gcamp6f mice mounted en face to 4-HNE in the absence of CPA or EGTA-AM . Under these conditions , the frequency of spontaneous transient events was significantly higher compared with arteries treated with CPA and EGTA-AM . 4-HNE significantly increased the frequency of Ca2+ transients ( 0 . 76 ± 0 . 08 vs . 2 . 44 ± 0 . 20 Hz , Videos 7 and 8 , Figure 2C ) , and this response was blocked by the TRPA1 channel inhibitor A967079 ( 1 . 12 ± 0 . 09 Hz , Figure 2C and Video 9 ) . Spontaneous and 4-HNE-evoked Ca2+ transients were not significantly different in amplitude , but were larger in terms of spatial spread and mean duration compared with TRPA1 sparklets ( Supplementary file 1 and 2 ) . The increase in Ca2+ transients appeared to be a consequence of recruitment of previously inactive sites ( number of sites per cell: 0 . 26 ± 0 . 02 vs . 1 . 44 ± 0 . 24 ) , as well as an increase in frequency of Ca2+ release from previously active sites ( 0 . 072 ± 0 . 004 vs . 0 . 121 ± 0 . 005 Hz , Figure 2A–C ) . TRPA1 inhibition with A967079 diminished the recruitment of new sites of Ca2+ transients ( number of sites per cell: 0 . 29 ± 0 . 02 , Figure 2A and C ) and reduced the frequency of Ca2+ transients in previously active sites ( 0 . 098 ± 0 . 003 Hz , Figure 2B and C ) . These data suggest that TRPA1 sparklets induce Ca2+ release from the ER , thereby amplifying the initial Ca2+ influx signal . To test the hypothesis that hypoxia induces production of 4-HNE in the endothelium , en face cerebral arteries were incubated in normoxic or hypoxic conditions . Hypoxia was achieved by superfusing the tissue with physiological saline solution ( PSS ) equilibrated with a hypoxic gas mixture ( 5% O2 , 6% CO2 , 89% N2 ) . The pO2 measured in the recording bath under these conditions was 13 ± 2 mmHg ( n = 5 ) . The pH of the superfusing solution was monitored in real time , and was constantly maintained between 7 . 37 and 7 . 42 during normoxic and hypoxic conditions . We observed an increase in 4-HNE immunolabeling in cerebral arteries superfused with hypoxic PSS when compared to arteries superfused with normoxic PSS ( Figure 3A and Figure 3—figure supplement 1 ) . Arteries from Tek:Gcamp6f mice were used to test the hypothesis that acute hypoxic exposure acts through TRPA1 channels to increase the frequency of Ca2+ influx events in the cerebral endothelium . Acute hypoxic exposure significantly increased the frequency of TRPA1 sparklets , a response that was significantly reduced by A967079 ( Figure 3B–3D and Videos 10 and 11 ) . The increase in TRPA1 sparklet frequency induced by hypoxia reflected an increase in the number of active sparklets sites ( Figure 3B and D ) as well as the frequency of sparklets at active sites ( control , 0 . 049 ± 0 . 007 Hz; hypoxia , 0 . 089 ± 0 . 007 Hz; p<0 . 05 , Figure 3C and D ) ; both increases were prevented by the TRPA1 inhibitor A967079 , which restored TRPA1 sparklet frequency to levels that were not significantly different from normoxic conditions ( 0 . 057 ± 0 . 007 Hz , Figure 3B – D ) . The amplitude , kinetics , and spatial spread of hypoxia-evoked TRPA1 sparklets ( Supplementary file 1 and Figure 3—figure supplement 2 ) did not significantly differ from those of 4-HNE–induced TRPA1 sparklets . These data demonstrate that pathophysiologically relevant levels of hypoxia acutely stimulate TRPA1 activity in the endothelium of intact cerebral arteries . We next investigated the molecular mechanisms responsible for activation of TRPA1 channels in endothelial cells during hypoxia . Our prior study demonstrated that extracellular O2- generated by NOX2 stimulates the formation of 4-HNE , which in turn activates TRPA1 channels in the cerebral endothelium ( Sullivan et al . , 2015 ) . Unexpectedly , we found that extracellular superoxide dismutase ( SOD ) and the NOX inhibitor apocynin did not significantly inhibit hypoxia-induced increases in TRPA1 sparklet frequency ( Figure 4—figure supplement 1 ) , suggesting the involvement of an alternative pathway . Consistent with this latter possibility , we found that hypoxia-induced increases in TRPA1 sparklet frequency and sites per cell were significantly reduced by membrane-permeant PEG-SOD , suggesting that intracellular generation of ROS is essential for this response ( Figure 4B ) . This conclusion is supported by imaging experiments showing that hypoxia caused a time-dependent increase in the fluorescence intensity of the ROS indicator dihydroethidium ( DHE ) in freshly isolated cerebral artery endothelial cells , which was prevented by PEG-SOD ( Figure 4A ) . Having eliminated NOX as a potential source of increased ROS production during hypoxia , we investigated the involvement of other pathways . A previous study suggested that generation of ROS by mitochondrial respiration is increased during hypoxia ( Hernansanz-Agustín et al . , 2014 ) . In agreement with this report , we found that pre-incubation of freshly isolated endothelial cells with the mitochondrial-targeted antioxidant mitoTEMPO ( 500 nM ) significantly inhibited hypoxia-induced increases in DHE fluorescence ( Figure 4A ) , suggesting that mitochondria are the primary source of intracellular ROS generated during hypoxia . To determine if mitochondrial ROS are responsible for the activation of TRPA1 channels during hypoxia , we treated cerebral arteries from Tek:Gcamp6f mice with mitoTEMPO prior to hypoxia exposure . This treatment significantly diminished hypoxia-induced increases in TRPA1 sparklet frequency , indicating a critical role for mitochondrial ROS generation in this response ( Figure 4C ) . Pressure myography studies using intact cerebral pial arteries were carried out to study the influence of TRPA1 channels on hypoxia-induced dilation . Arteries were pressurized to physiological levels ( 60 mmHg ) and allowed to generate spontaneous myogenic tone . Acute exposure to hypoxia ( pO2 = 13 ± 2 mmHg ) induced vasodilation that was significantly diminished by disruption of endothelial cell function by passing an air bubble through the vascular lumen ( Ralevic et al . , 1989 ) ( Figure 5A ) , indicating involvement of the endothelium . Vasodilation in response to hypoxia was also inhibited by Tempol ( 100 µM ) ( Figure 5B ) , but was not significantly reduced by NOX2 inhibition and extracellular SOD ( Figure 5—figure supplement 1 ) , supporting the concept that intracellular ROS generation is required for this response . Further , we found that hypoxia-induced dilation was significantly blunted by mitoTEMPO ( 500 nM ) ( Figure 5C ) , suggesting that mitochondria are the source of ROS required for this response . Control experiments showed that Tempol and mitoTEMPO had no significant effect on cerebral artery dilation in response to administration of 4-HNE , indicating that these drugs do not directly inhibit this TRPA1-dependent response ( Figure 5—figure supplement 2 ) . Vasodilation in response to hypoxic exposure was significantly reduced by blockade of TRPA1 channels with A967079 ( Figure 6A ) and was diminished in cerebral arteries isolated from endothelial cell-specific TRPA1 knockout mice ( Trpa1 ecKO ) ( Sullivan et al . , 2015 ) compared with TRPA1 floxed , cre-recombinase negative , littermate controls ( Trpa1fl/fl , Figure 6B ) . To investigate potential vasodilator mechanisms acting downstream of TRPA1 channels , we blocked KCa3 . 1 channels with TRAM34 and KCa2 . 3 channels with apamin . This treatment significantly reduced hypoxia-induced dilation ( Figure 6C ) . Together , these data demonstrate that hypoxia-induced increases in mitochondrial ROS production stimulate TRPA1-mediated Ca2+ influx in the cerebral artery endothelium , causing vasodilation of pial arteries through activation of KCa3 . 1 and/or KCa2 . 3 channels . We also investigated the effects of TRPA1 activity on vasomotor activity of cerebral penetrating arterioles . Brain slices ( 200 µm in thickness ) from perfusion-fixed Tekgfp mice ( Tg ( TIE2GFP ) 287Sato/J ) reporter mice that express GFP in the endothelium ( Motoike et al . , 2000 ) were immunolabeled for TRPA1 . Penetrating arterioles were identified as branches from the surface pial arteries that entered the underlying brain parenchyma . TRPA1 immunofluorescence was detected in the endothelium of penetrating arterioles ( Figure 7A , left panels , red ) but was not detected when the primary antibody for TRPA1 was omitted , although GFP fluorescence was observed ( Figure 7A , right panels ) . Functional ex vivo pressure myography experiments showed that exposing penetrating arterioles to 4-HNE induced dilation that was significantly diminished by A967079 ( Figure 7B ) . In addition , exposure of pressurized penetrating arterioles to hypoxia caused dilation that was blunted by A967079 ( Figure 7C ) . Together , these data suggest that TRPA1 channels are present and functional in penetrating arterioles and elicit vasodilation in response to 4-HNE and hypoxia . Our data indicate that hypoxia-induced activation of TRPA1 channels in the endothelium of cerebral arteries and penetrating arterioles causes vasodilation . We propose that hypoxia-induced vasodilation of cerebral arteries is an adaptive response that serves to improve collateral perfusion within affected brain regions following ischemic stroke . To test this hypothesis , we induced ischemic strokes in Trpa1 ecKO mice and TRPA1fl/fl control littermates using the middle cerebral artery occlusion ( MCAO ) model ( Longa et al . , 1989 ) without reperfusion . We found that although reduction in cerebral perfusion after MCAO did not significantly differ between Trpa1fl/fl and Trpa1 ecKO mice ( Figure 8—figure supplement 1 ) , infarct size was significantly increased in Trpa1 ecKO mice compared with littermate controls 24 hr after MCAO ( Figure 8A ) . In companion interventional studies , we examined the effects of stimulating TRPA1 activity with cinnamaldehyde ( CinA ) after MCAO . Using pressure myography , we found that CinA induced a concentration-dependent dilation of pressurized cerebral arteries isolated from control mice , a response that was absent in cerebral arteries isolated from Trpa1 ecKO mice ( Figure 8—figure supplement 2A and B ) . CinA ( 30 µM ) also induced dilation of penetrating arterioles , a response that was significantly blunted by A967079 ( Figure 8—figure supplement 2C and D ) . Pharmacological activation of TRPA1 channels in vivo by intraperitoneal ( i . p . ) injection of CinA ( 50 mg/kg ) ( Huang et al . , 2007 ) , administered 15 min after MCAO , significantly reduced infarct size in control C57/bl6 mice ( Figure 8B ) . This protective effect of CinA is partially dependent on endothelial cell TRPA1 channels as infarct size in Trpa1 ecKO mice was not significantly reduced by CinA ( Figure 8C ) . These data suggest that the endogenous activity of TRPA1 channels in the cerebral endothelium reduces cerebral damage associated with ischemic strokes . In this study , we investigated the functional importance of TRPA1 channels in the cerebral endothelium under pathophysiological conditions . We found that acute hypoxic exposure induced an increase in TRPA1 sparklet frequency in the endothelium of intact cerebral pial arteries and penetrating arterioles that caused dilation . Pharmacological activation of TRPA1 in vivo reduced the loss of brain tissue in response to experimentally induced ischemic stroke , whereas conditional knockout of TRPA1 in the endothelium exacerbated this damage ( Figure 8—figure supplement 3 ) . We propose that TRPA1 activity is important in mediating hypoxia-induced dilation of the cerebral vasculature during ischemic stroke , and that this response improves collateral blood flow to the affected region to improve outcomes . TRPA1 channels are activated by electrophilic compounds , such as AITC , allicin and CinA , which target specific cysteine residues located in the cytosolic pre-S1 domain linking the channel’s ankyrin-repeat region to the first membrane-spanning domain ( Hinman et al . , 2006; Macpherson et al . , 2007; Paulsen et al . , 2015 ) . In addition , considerable evidence demonstrates that TRPA1 channels are activated by ROS and/or ROS-derived metabolites ( Sullivan et al . , 2015; Trevisani et al . , 2007; Pires and Earley , 2017 ) . Lipid peroxidation products such as 4-HNE and related substances appear to act on the same cysteine residues that are targeted by electrophilic molecules ( Hinman et al . , 2006; Macpherson et al . , 2007; Pires and Earley , 2017 ) , providing evidence for a common mechanism of activation . It has been suggested that TRPA1 channels are activated by hypoxia , such as reported by Takahashi et al . , who showed that shifting pO2 from 150 mmHg to 80 mmHg stimulated TRPA1 activity , measured using the FLIPPER assay in HEK cells overexpressing TRPA1 . The authors interpreted this finding as evidence of hypoxia-induced activation of TRPA1 . However , under normal physiological conditions in healthy humans , the pO2 of arterial blood is ~80–100 mmHg within the aorta , ~30–70 mmHg in capillaries , and ~20–40 mmHg in the venous system ( Tsai et al . , 2003 ) . Viewed in this context , the data reported by Takahashi et al . appear to suggest that TRPA1 channels are activated by restoration of normoxia following hyperoxic exposure , rather than by hypoxia per se . In our study , the pO2 of the superfusing bath was closely monitored and was maintained at ~80–90 mmHg during control ( normoxic ) conditions and determined to be ~13 mmHg during acute hypoxic challenge . TRPA1 activity , detected as TRPA1 sparklets , was low under normoxic conditions , but was stimulated by change to a hypoxic pO2 . It should be noted that the pO2 in the rat brain following an ischemic stroke varies from near anoxia in the infarct core ( ~1–2 mmHg ) to ~10–15 mmHg in the peri-infarct area ( Liu et al . , 2004 ) — values of pO2 similar to those observed in our tissue bath experiments . Thus , our data suggest that pathophysiologically relevant levels of hypoxia acutely activate TRPA1 channels in the intact cerebral endothelium . In a prior study , we showed that extracellular O2- generated by NOX2 activates TRPA1 channels in the cerebral endothelium through a process that requires lipid peroxidation ( Sullivan et al . , 2015 ) . However , here we found that hypoxia-induced activation of TRPA1 was independent of both NOX activity and extracellular O2- . Instead , we found that intracellular generation of O2- was required for hypoxia-induced activation of TRPA1 . Previous reports have shown that acute hypoxic exposure uncouples the mitochondrial electron transport chain ( Klimova and Chandel , 2008 ) , stimulating a localized burst of O2- ( Hernansanz-Agustín et al . , 2014 ) , which can enter the cytoplasm through anion channels present in the mitochondrial outer membrane ( Han et al . , 2003 ) . In agreement with earlier reports showing that hypoxia increases mitochondrial O2- production in cancer cells ( Chandel et al . , 2000; Sabharwal and Schumacker , 2014; Eales et al . , 2016 ) and cultured endothelial cells ( Hernansanz-Agustín et al . , 2014 ) , we found that exposure to acute hypoxia stimulated ROS generation in native cerebral artery endothelial cells , and that this response was inhibited by a mitochondrial-targeted SOD mimetic and by intracellular PEG-SOD . A recent study reported that mitochondria are located at the base of myoendothelial projections ( Maarouf et al . , 2017 ) , which are regions of near contact between the plasma membrane of endothelial cells and the underlying smooth muscle cells . TRPA1 channels are also concentrated within myoendothelial projection ( Sullivan et al . , 2015; Earley et al . , 2009 ) . Thus , it is possible that hypoxia-induced increases in mitochondria O2- production generates 4-HNE within myoendothelial projections , leading to TRPA1 activation in those sites . Our data provide support for a signaling pathway initiated by mitochondrial-generated O2- , showing that these superoxide ions stimulate the activity of TRPA1 channels in the cerebral endothelium during acute hypoxic exposure . The current studies do not specifically establish whether mitochondrial O2- activates TRPA1 channels directly or if activation requires lipid peroxidation . TRPA1 channels conduct mixed cation currents with a large Ca2+ fraction ( Karashima et al . , 2010; Wang et al . , 2008 ) , and activation of TRPA1 on the plasma membrane of endothelial cells transiently creates small microdomains with high localized Ca2+ concentrations ( Sullivan et al . , 2015 ) . Our prior studies demonstrated that TRPA1-mediated Ca2+ influx stimulated by direct application of AITC ( Earley et al . , 2009 ) or through O2- generated by NOX2 ( 1 ) triggers endothelium-dependent dilation of cerebral arteries , a response that is unaffected by blocking nitric oxide synthase or cyclooxygenase pathways , but is sensitive to inhibition of Ca2+-activated KCa3 . 1 and KCa2 . 3 K+ channels . TRPA1 channels and KCa3 . 1 channels co-localize within myoendothelial projections ( Earley et al . , 2009 ) , such that influx of Ca2+ though TRPA1 channels within these spatially restricted regions is sufficient to increase the local [Ca2+] to levels capable of activating outward K+ currents through KCa3 . 1 and KCa2 . 3 , leading to hyperpolarization of the endothelial cell plasma membrane ( Sullivan et al . , 2015; Earley et al . , 2009 ) . Hyperpolarization of the endothelial cell plasma membrane , in turn , is conducted to SMCs through myoendothelial gap junctions , resulting in relaxation and vasodilation ( Sokoya et al . , 2006; Chadha et al . , 2011 ) . In addition , TRPA1 activation by AITC ( Qian et al . , 2013 ) and 4-HNE stimulates Ca2+ release from intracellular stores , leading to an increase in Ca2+ transients previously linked to activation of KCa3 . 1 ( 39 ) . Our current findings show that activity of KCa3 . 1 and/or KCa2 . 3 channels is required for cerebral artery dilation in response to hypoxia , suggesting that these channels act downstream of TRPA1 in this setting . Our data also show that pharmacological inhibition of TRPA1 channels or Trpa1 knockout in the endothelium did not completely abolish hypoxia-induced dilation of cerebral arteries and arterioles , suggesting the possibility that redundant or overlapping mechanisms contribute to this response . Several possibilities have been reported , including endogenous generation of carbon monoxide ( Leffler et al . , 1999 ) and TRPV3 channel activity . TRPV3 has been shown to be sensitized by acute hypoxia in a ROS-independent manner ( Karttunen et al . , 2015 ) , and cause endothelium-dependent dilation of cerebral pial arteries and penetrating arterioles in response to chemical agonists ( Pires et al . , 2015; Earley et al . , 2010 ) . Ischemic stroke is one of the leading causes of death and disability worldwide . The extent of neuronal loss following an ischemic insult is determined in part by the supply of blood to the region surrounding the infarct core , known as the ischemic penumbra ( Maas et al . , 2009; Hoehn-Berlage et al . , 1995 ) . The penumbra is characterized by structurally intact , but metabolic silent , parenchymal tissue in which perfusion is impaired ( Symon et al . , 1977 ) . Increasing blood flow to the ischemic penumbra has been shown to reduce expansion of the infarct zone and , consequently , improve neurological outcomes ( Cipolla et al . , 2018 ) . Our findings suggest that activation of TRPA1 channels in endothelial cells of cerebral arteries is an early adaptive response to acute reduction in tissue pO2 , leading to a possible increase in perfusion within the penumbra . Loss of TRPA1 signaling in endothelial cells of cerebral arteries increased cerebral infarcts following permanent middle cerebral artery occlusion in mice . Further , pharmacological activation of TRPA1 decreased infarcts , an effect blunted by loss of endothelial TRPA1 channels . Isoflurane , the anesthetic agent used during MCAO surgeries , was previously shown to potentiate TRPA1 activity ( Matta et al . , 2008 ) . Interestingly , isoflurane anesthesia is neuroprotective in various animal models of stroke and subarachnoid hemorrhage ( Altay et al . , 2012a; Altay et al . , 2012b; Li et al . , 2013; Khatibi et al . , 2011 ) , but the mechanistic basis of this effect is not known . The current findings support the concept that isoflurane may provide neuroprotection by potentiating TRPA1-mediated cerebral arterial dilation . In summary , the present study shows that hypoxia stimulates Ca2+ influx through TRPA1 channels in the intact cerebral endothelium via a mechanism that requires generation of intracellular O2- by mitochondria . We also demonstrated that hypoxia-induced TRPA1 activity initiates endothelium-dependent dilation of cerebral pial arteries and penetrating arterioles . Selective deletion of TRPA1 expression in the endothelium exacerbated the loss of brain tissue associated with ischemic stroke , providing evidence that hypoxia-induced activation of TRPA1 in the cerebral endothelium constitutes a novel adaptive response . Adult male and female mice ( 12–16 weeks of age ) were used for all experiments . All animal procedures used in this study were approved by the Institutional Animal Care and Use Committee of the University of Nevada , Reno School of Medicine , and are in accordance with the National Institutes of Health ‘Guide for the Care and Use of Laboratory Animals’ , eigth edition . For isolation of cerebral arteries , mice were deeply anesthetized with 4% isoflurane and euthanized by decapitation and exsanguination . The skull was carefully removed and the brain with the brainstem intact was placed in ice-cold tissue collection solution consisting of ( in mM ) 140 NaCl , 5 KCl , 2 MgCl2 , 10 dextrose , 10 HEPES ( pH 7 . 4 ) , without Ca2+ , supplemented with 1% bovine serum albumin ( BSA ) . Cerebral pial arteries were carefully removed from the brain and cleaned of meningeal membranes . Ca2+ imaging was performed on cerebral arteries isolated from Tek:Gcamp6f mice mounted en face . Arteries were opened longitudinally using fine spring scissors ( Fine Science Tools , Foster , CA ) and mounted on a Sylgard pad using insect pins with the endothelium facing up as previously described ( Sonkusare et al . , 2012; Ledoux et al . , 2008 ) . The tissue was stretched to its in vivo length and maintained in Ca2+-free PSS containing the membrane-permeable Ca2+ chelator EGTA-AM ( 10 μM , ThermoFisher Scientific , Eugene , OR ) at 37°C for 15 min , then washed with warm ( 37°C ) Ca2+ PSS consisting of ( in mM ) : 119 NaCl , 4 . 7 KCl , 21 NaHCO3 , 1 . 18 KH2PO4 , 1 . 17 MgSO4 , four dextrose , and 1 . 8 CaCl2 . The solution was continuously aerated with a normoxic gas mixture ( 21% O2 , 6% CO2 , 73% N2 ) to maintain a constant pH of 7 . 38–7 . 42 and pO2 of ~80 mmHg . The pH was continuously measured with an in-line flow-through pH mini-electrode , and solution oxygenation was assessed using an in-line oxygen microelectrode ( both from Microelectrodes , Inc . , Bedford , NH ) . ER Ca2+ stores were depleted by treating the preparation with the SERCA inhibitor CPA ( 30 μM; Tocris Biosciences , Bristol , UK ) to eliminate Ca2+ signals generated by spontaneous ER Ca2+ release . CPA was maintained in the bath solution throughout the experiment . All pharmacological agents ( 4-HNE , A967079 , PEG-SOD , mitoTEMPO , extracellular SOD and apocynin ) were administered via the superfusing bath solution . Hypoxia was induced by superfusing the tissue with PSS aerated with a hypoxic gas mixture ( 5% O2 , 6% CO2 , 89% N2 ) . The pO2 of this solution was 13 ± 2 mmHg , and the pH in the recording chamber was 7 . 3–7 . 42 . Ca2+ images were obtained using an inverted microscope ( Olympus iX81; Olympus Corp . , Tokyo , Japan ) , modified to allow imaging in the upright configuration ( LSM Tech , Etters , PA ) and equipped with epifluorescence illumination , a 60x water-immersion objective ( numerical aperture 1 . 0 ) , and a highly sensitive iXon Ultra EMCCD camera ( Andor Technology , Belfast , Northern Ireland ) . Each field of view was 512 × 512 pixels ( one pixel = 0 . 27 μm ) , and images were recorded at a rate of 30–55 frames/s for a total of 1000 frames . The duration of each recording was 20–30 s . A separate cohort of Tek:Gcamp6f mice were used to assess the effects of 4-HNE on Ca2+ signaling activity in the cerebral endothelium in the absence of EGTA-AM and CPA . Recordings were analyzed using custom software ( SparkAn , kindly provided by Drs . Adrian Bonev and Mark T . Nelson , University of Vermont , Burlington Vermont ) ( Dabertrand et al . , 2012 ) . An average of the first 10 images of each recording obtained prior to stimulation was used to define baseline fluorescence ( F0 ) . TRPA1 sparklets were identified as localized increases in fluorescence ( ΔF ) within a small region of interest ( 5 . 3 × 5 . 3 μm ) after subtraction of F0 , and manifested as quantal events with characteristic step-wise peaks in plots of fluorescence intensity over time . The frequency of Ca2+ signals with amplitude between 1 . 08 and 1 . 1 ΔF/F0 ( and multiples ) is reported . Each peak was analyzed individually , and amplitude , attack time ( half-time elapsed from the beginning of the event to the peak ) , decay time ( half-time elapsed from the peak to dissipation of the event ) , duration ( total open time of the channel , calculated as [attack + decay] x 2 ) , and spatial spread of the Ca2+ signal were determined . Accumulation of 4-HNE in cerebral arteries exposed to hypoxia was assessed by immunofluorescence labeling . Freshly isolated cerebral arteries were mounted en face as described above , and superfused with normoxic or hypoxic PSS for 15 min . The preparations were then fixed with 4% formaldehyde in PBS for 10 min and incubated in 5% BSA +0 . 1% Triton X-100 for 2 hr at room temperature . Preparations were then incubated with a rabbit polyclonal antibody against 4-HNE ( 1:1000 , Abcam PLC , Cambridge , United Kingdom ) overnight . Preparations were washed with PBS and incubated with a donkey anti-rabbit secondary antibody conjugated to AlexaFluor 568 ( 1:2000 , Thermo Fisher Scientific ) diluted in 2% BSA +0 . 1% Triton X-100 for 90 min at room temperature . Preparations were then washed with PBS and incubated with the endothelial cell-labeling isolectin GS-IB4 conjugated to AlexaFluor 488 ( 1:1000 in PBS , Thermo Fisher Scientific ) for 10 min at room temperature . Preparations were washed and mounted in glass slides with coverslips and fluorescence analyzed by laser scanning confocal microscopy in an Olympus FluoView confocal microscope within the same day . Images were obtained in a field of view of 800 × 800 pixels using a 60x oil immersion objective ( numerical aperture: 1 . 42 ) and a Z-stack spanning the entire thickness of the cerebral artery , from the adventitia to the endothelial cell layer . In order to perform the semi-quantitative assessment of 4-HNE fluorescence , settings ( PMT sensitivity , gain and laser power ) were adjusted for normoxic cerebral arteries , and the same settings were used for cerebral arteries exposed to hypoxia . Quantification of fluorescence intensity was performed by plotting a Z-axis profile on ImageJ to obtain a fluorescence intensity curve , and the area under the curve was calculated to best represent fluorescence intensity of the entire thickness of the preparation . A separate group of mice were used to determine if TRPA1 channels are present in penetrating arterioles in the brain . Endothelial cell reporter mice constitutively expressing GFP under the control of the Tek promoter ( Tekgfp , Jackson labs , stock number 003658 ) were perfusion-fixed with 4% formaldehyde in PBS and the brain was removed from the skull for histological processing . Brain slices ( 200 µm thick ) were acquired using a Leica VT 1200S vibratome ( Leica Biosystems GmbH , Wetzlar , Germany ) and permeabilized with 0 . 1% Triton-X in PBS for 2 hr at room temperature . Slices were then incubated in 10% normal horse serum +0 . 1% Triton-X in PBS for 2 hr at room temperature , followed by incubation with the primary antibodies rabbit anti-TRPA1 ( 1:1000 , Alomone Labs , Jerusalem , Israel ) and goat anti-GFP ( 1:500 , Abcam ) dissolved in blocking solution overnight at room temperature on a shaker . Controls lacking TRPA1 primary antibody were incubated with anti-GFP alone overnight . Slices were washed with PBS 3 times for 10 min each , then incubated with a donkey anti-rabbit conjugated with AlexaFluor 594 ( 1:2000 , Thermo Fisher Scientific ) and a donkey anti-goat conjugated with AlexaFluor 488 ( 1:1000 , Thermo Fisher Scientific ) for 2 hr at room temperature on a shaker . Slices were washed three times with PBS and mounted on glass slides using a Fluoroshield Mounting Medium with DAPI ( Abcam ) . Fluorescence images were taken by an Olympus FluoView laser scanning confocal microscope using a 60x oil-immersion objective ( numerical aperture: 1 . 42 ) in a 1024 × 1024 pixels per field of view . The pixel size was maintained constant at X = 180 nM , Y = 180 nM , Z = 0 . 801 nM . Z-stacks were captured ( ~40 µm in thickness ) to allow for 3-dimensional reconstruction of the arteriolar bed . Superoxide production by native cerebral artery endothelial cells was assessed by confocal imaging of cells stained with the ROS-sensitive dye dyhydroethidium ( DHE ) ( Pires et al . , 2010 ) . Native endothelial cells were isolated by dissecting cerebral arteries and incubating them in tissue collection solution supplemented with 1 mg/mL papain ( Worthington Biochemical Company , Lakewood , NJ ) , 1 mg/mL DL-dithiothreitol and 10 mg/mL BSA for 12 min in a 37°C water bath . Arteries were then washed three times with Ca2+-free PSS and incubated in tissue-collection solution supplemented with 1 mg/mL type II collagenase ( Worthington Biochemical Company ) for 12 min in a 37°C water bath . Thereafter , arteries were washed three times with PSS and triturated by passing the solution through a fire-polished glass Pasteur pipette , as described previously ( Pires et al . , 2015; Pires et al . , 2017 ) . Dissociated cells in the solution were placed in a 35 mm glass-bottom dish ( Greiner Bio-One GmBH , Germany ) and allowed to adhere to the dish for 30 min in a cell culture incubator . After incubating cells with 5 µM DHE ( ThermoFisher ) for 5 min at 37°C , the culture dish was transferred to a spinning-disk confocal microscope for real-time recordings of fluorescence intensity . Attached cells in the dish were exposed to normoxic or hypoxic PSS supplemented with DHE ( 5 µM ) in cells pre-incubated with vehicle , PEG-SOD ( 30 min prior to recording ) or mitoTEMPO ( 15 min prior to recording ) and recorded at a rate of 1 frame/min for 10 min . Recordings were performed in an Olympus IX-71 microscope coupled to a Yokogawa CSU22 Confocal Scanning Unit ( Yokogawa Electric Corporation , Tokyo , Japan ) and an Andor iXon + camera ( Andor Technology ) using a 100x oil-immersion objective ( numerical aperture 1 . 45 ) . Cells were excited using a 488 nm laser and emission was detected using a 500–530 band-pass emission filter . The effect of hypoxia on the diameters of isolated intact cerebral pial resistance arteries was assessed using pressure myography . Cerebral arteries were carefully dissected and cannulated between two glass cannulas in a pressure myograph chamber ( Living Systems Instrumentation , St . Albans , VT ) . Arteries were pressurized to 60 mmHg and maintained in normoxic PSS at 37°C for 30 min to reach equilibration . Luminal diameter was continuously recorded by edge-detection using videomicroscopy . The viability of each preparation was established by briefly incubating arteries in isotonic PSS containing high extracellular K+ ( 60 mM KCl , 59 mM NaCl; all other components held constant ) to induce constriction . Afterwards , arteries were washed with normoxic PSS until spontaneous myogenic tone was generated . Myogenic tone ( % ) was calculated as [1 – ( active luminal diameter/passive luminal diameter ) ] x 100 . Arteries were then exposed to hypoxic PSS for 10 min and subsequently returned to normoxic PSS . All pharmacological experiments were performed in a paired fashion . Passive diameter at 60 mmHg was determined by incubating arteries in Ca2+-free PSS supplemented with EGTA ( 2 mM ) and diltiazem ( 10 μM ) . In a subset of experiments , the endothelium was removed by passing an air bubble through the lumen of the artery ( Ralevic et al . , 1989 ) . This method of endothelium removal was shown to almost completely ablate endothelium-dependent responses , without damaging the underlying smooth muscle cell layer ( Ralevic et al . , 1989 ) . The effects of TRPA1 agonists and hypoxia were also assessed in freshly isolated cerebral penetrating arterioles . Isolation of penetrating arterioles was performed as previously described ( Pires et al . , 2016 ) . Briefly , penetrating arterioles branching out of the middle cerebral artery were carefully isolated from the surrounding parenchyma and mounted onto glass cannulas in a blind-sac experimental configuration . Penetrating arterioles were pressurized to 40 mmHg and superfused with artificial cerebrospinal fluid ( aCSF , in mM: 124 NaCl , 3 KCl , 2 CaCl2 , 2 MgCl2 , 1 . 25 NaH2PO4 , 26 NaHCO3 , four glucose , pH 7 . 4 ) . Arteriolar viability was assessed by exposing the preparation to isotonic aCSF containing high extracellular K+ ( in mM: 67 NaCl , 60 KCl , all other salts remain the same concentration ) . Viable arterioles were washed with regular aCSF and allowed to generate spontaneous myogenic tone , after which they were exposed to 4-HNE , hypoxia or CinA in the presence or absence of the TRPA1 inhibitor A967079 . Permanent cerebral ischemia was induced using the intraluminal suture model of MCAO ( Longa et al . , 1989 ) , modified for mice . Mice were initially anesthetized with 2% isoflurane in oxygen and their body temperature was maintained at 37°C . An incision was made at the top of the head for attachment of a laser Doppler flow probe ( Periflux System 5000; Perimed AB , Järfälla , Sweden ) to measure blood flow to the region supplied by the MCA ( 5 mm lateral and 1 mm posterior to the bregma ) . A midline incision was made at the neck to expose the carotid artery , and the lingual , thyroid and external carotid arteries were tied off . A 6–0 nylon monofilament with a rounded silicone tip ( Doccol Corporation , Sharon , MA ) , ~210 μm in diameter , was inserted into the common carotid artery and advanced through the internal carotid artery to block the MCA where it branches from the circle of Willis . MCAO was verified by a sharp drop in blood perfusion , as measured by laser Doppler flowmetry . Mice that showed less than an 80% drop in perfusion were excluded from the study . Ischemia was maintained for 24 hr , after which mice were anesthetized and euthanized by decapitation . A subset of mice was injected with CinA ( 50 mg/kg body weight [Huang et al . , 2007] ) or an equal volume of vehicle ( 0 . 2–0 . 3 ml of DMSO ) 15 min after MCAO to assess the effects of pharmacological activation of TRPA1 on ischemic stroke outcome . The brain was removed from the skull and sliced into 1 mm-thick sections . Ischemic damage was assessed by subsequently staining sections for 20 min with 2% 2 , 3 , 5-triphenyltetrazolium chloride , which stains viable tissue red , leaving infarcted tissue white . Brain slices were fixed in 4% formaldehyde in PBS , and digital images were taken . The percentage of infarction was determined using the following equation: %HI = [ ( VCVL ) /VC]*100 , where HI is the hemisphere infarcted , VC is the volume of normal tissue in the non-ischemic hemisphere , and VL is the volume of normal tissue in the ischemic hemisphere ( Swanson et al . , 1990 ) . Unless otherwise stated , all chemicals were purchased from Sigma-Aldrich Corporation ( St . Louis , MO ) . All summary data are expressed as means ± SEM . Statistical analyses were performed and graphs were constructed using Prism 6 . 0 ( GraphPad Software , Inc . , La Jolla , CA ) . The significance of differences between two groups was tested by paired or unpaired two-tailed Student’s t-test , depending on the experimental design . Data that did not fit a normal distribution were tested using a non-parametric alternative . Comparisons of three or more groups were tested by one-way analysis of variance ( ANOVA ) with a Sidak post hoc test . Values of p<0 . 05 were considered statistically significant for all experiments . Sample size for experiments were determined by performing a 2-sided power analysis to reach a power of 0 . 8 for a value of α ≤0 . 05 . Using these parameters , we determined that Ca2+ imaging experiments should have a minimum of 16 field of views; experiments assessing 4-HNE fluorescence should be 5 field of views; pressure myography experiments should be performed in at least three arteries; MCAO experiments required at least three mice .	A stroke can cause long-lasting physical and mental disabilities in patients including loss of mobility , speech defects and confusion . Most strokes happen when the blood supply to part of the brain is cut off due to blood clots or clumps of fat blocking blood vessels called arteries . To prevent a blocked blood vessel causing a stroke , the network of blood vessels in the brain contains alternative routes to each area . The arteries in these alternative routes can widen to allow more blood to flow through them and avoid the blockage . When the blood supply to part of the brain is cut off , the level of oxygen in that area decreases . This causes highly reactive molecules known collectively as free radicals to be produced , which can bind to other molecules in cells and stop them from working properly . A protein called TRPA1 is found in the cells that form the inner lining of blood vessels . When it is active , TRPA1 forms a channel that allows signals known as calcium ions to enter the cell , which ultimately leads to arteries in the brain becoming wider . A free radical known as 4-HNE binds to TRPA1 , but it is not clear if this enables the channel to directly sense the levels of oxygen in the brain . Pires and Earley studied TRPA1 channels in brain arteries from mice . The experiments found that decreasing the levels of oxygen in the arteries caused 4-HNE to accumulate and activate TRPA1 , resulting in the blood vessels becoming wider . Chemicals that inhibit the production of free radicals blocked the activity of the TRPA1 channels . Mice that lacked TRPA1 were more likely to sustain damage to the brain during strokes than normal mice . Furthermore , injecting normal mice experiencing a stroke with a drug that activates TRPA1 reduced the amount of damage to the brain . The findings of Pires and Earley suggest that TRPA1 plays an important role in protecting the brain during strokes and other conditions that reduce the brain’s blood supply . Future studies will assess whether drugs that activate TRPA1 have the potential to help reduce long-term disabilities in human patients who have a stroke .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2018	Neuroprotective effects of TRPA1 channels in the cerebral endothelium following ischemic stroke
Primate evolution has been argued to result , in part , from changes in how genes are regulated . However , we still know little about gene regulation in natural primate populations . We conducted an RNA sequencing ( RNA-seq ) -based study of baboons from an intensively studied wild population . We performed complementary expression quantitative trait locus ( eQTL ) mapping and allele-specific expression analyses , discovering substantial evidence for , and surprising power to detect , genetic effects on gene expression levels in the baboons . eQTL were most likely to be identified for lineage-specific , rapidly evolving genes; interestingly , genes with eQTL significantly overlapped between baboons and a comparable human eQTL data set . Our results suggest that genes vary in their tolerance of genetic perturbation , and that this property may be conserved across species . Further , they establish the feasibility of eQTL mapping using RNA-seq data alone , and represent an important step towards understanding the genetic architecture of gene expression in primates . Gene regulatory variation has been shown to make fundamental contributions to phenotypic variation in every species examined to date . This relationship has been demonstrated most clearly at the level of gene expression , which captures the integrated output of a large suite of other regulatory mechanisms . Variation in gene expression levels has been linked to fitness-related morphological , physiological , and behavioral variation in both lab settings and natural populations ( e . g . , Abzhanov et al . , 2004; Hammock and Young , 2005; Tishkoff et al . , 2006; Chan et al . , 2010; reviewed in Wray , 2007 ) , and is a robust biomarker of disease in humans ( e . g . , Golub et al . , 1999; Borovecki et al . , 2005 ) . In addition , patterns of gene expression are often associated with signatures of natural selection ( Rifkin et al . , 2003; Denver et al . , 2005; Gilad et al . , 2006a; Blekhman et al . , 2008 ) , suggesting their functional importance even when their phenotypic significance remains unknown . In primates , the majority of research on the evolution of gene expression has concentrated on cross species comparisons , particularly using humans , chimpanzees , and rhesus macaques ( Enard et al . , 2002; Cáceres et al . , 2003; Khaitovich et al . , 2004; Gilad et al . , 2005 , 2006b; Haygood et al . , 2007; Blekhman et al . , 2008; Babbitt et al . , 2010; Barreiro et al . , 2010; Blekhman et al . , 2010; Brawand et al . , 2011; Perry et al . , 2012 ) . These studies—motivated by a long-standing argument about the importance of gene regulation in primate evolution ( King and Wilson , 1975 ) —have been important for identifying patterns of constraint on gene expression phenotypes over long evolutionary time scales , and for suggesting candidate loci that might contribute to phenotypic uniqueness in humans or other species . For example , gene expression patterns associated with neurological development appear to have experienced an accelerated rate of change in primates relative to other mammals , with axonogenesis-related and cell adhesion-related genes accelerated specifically in the human lineage ( Brawand et al . , 2011 ) . Similarly , differentially expressed genes in human liver are enriched for metabolic function ( Blekhman et al . , 2008 ) , suggesting a potential molecular basis for arguments implicating dietary shifts in the emergence of modern humans ( Kaplan et al . , 2000; Ungar and Teaford , 2002; Wrangham , 2009 ) . Adaptively relevant changes in gene expression levels across species implicate selection on gene expression phenotypes within species , and particularly within populations , the basic unit of evolutionary change . However , in contrast to cross species comparisons , we still know little about the genetic architecture of gene expression levels in natural nonhuman primate populations . No estimates of the heritability of gene expression traits are available , even for populations that have been intensively studied for many decades . We also do not know whether segregating genetic variation that affects gene expression is common or rare , how the effect sizes of such variants are distributed , or whether they carry a signature indicative of natural selection . If gene regulatory variation has indeed been key to primate evolution , as classic arguments suggest ( King and Wilson , 1975 ) , then large gaps therefore remain in our understanding of this process . Three primary reasons combine to account for the absence of such data . First , until relatively recently , the only feasible approach for measuring genome-wide gene expression levels on a population scale was microarray technology . This constraint limited the diversity of systems that could be assessed because cost-effective , commercially available arrays have only been developed for a handful of taxa . Second , genomic resources , especially detailed catalogs of known genetic variants ( e . g . , 1000 Genomes Project Consortium et al . , 2010 , International HapMap Consortium , 2005 ) , are also limited to a small set of species . The lack of such resources creates major barriers to genome-scale studies of the genetics of gene expression in other organisms , which rely on complementary gene expression and genotype data . Finally , for many taxa , samples suitable for gene expression profiling can be challenging to collect . In nonhuman primates , for example , RNA samples are rarely available even for the most intensively studied natural populations . Recently , sequencing-based methods for measuring gene expression levels ( e . g . , RNA-seq ) have eliminated the need for species-specific arrays . Comparative genomic studies using RNA-seq have thus vastly expanded the set of taxa for which genome-wide expression data are available ( including primates: Brawand et al . , 2011; Perry et al . , 2012 ) . Importantly , because fragments of expressed genes are resequenced many times in RNA-seq studies , data on genetic variation are also generated in the process . Although these data can be affected by technical biases , several studies have demonstrated the generally high reliability of genotypes inferred from RNA-seq reads ( Perry et al . , 2012; Piskol et al . , 2013 ) . Such data can provide important insight into genetic diversity in species for which little other information exists ( Perry et al . , 2012 ) . Additionally , they provide the two ingredients necessary for mapping gene expression traits to genotype , at moderate cost and without the requirement for previously ascertained genetic variants . Here , we evaluate the potential for such work in an intensively studied wild primate population , the baboons ( Papio cynocephalus ) of the Amboseli basin in Kenya . 43 years of prior research on this population have established it as an important model for human social behavior , health , and aging ( Alberts and Altmann , 2012 ) , and have facilitated the development of protocols for collecting samples appropriate for gene expression analysis ( Tung et al . , 2009; Babbitt et al . , 2012; Runcie et al . , 2013 ) . We generated RNA-seq data for 63 individually recognized members of the Amboseli study population . We used these data to explore the frequency , impact , and potential selective relevance of variants associated with variation in gene expression levels , using complementary expression quantitative trait locus ( eQTL ) mapping and allele-specific expression ( ASE ) approaches . We found evidence for abundant functional regulatory variation in the Amboseli baboons , and a surprising amount of power to detect these variants even with a modest sample size . We also found that functional variants are depleted in highly conserved genes , consistent with constraint on gene expression patterns . However , among genes with eQTL , we did not find strong support for a relationship between effect size and minor allele frequency . Such a relationship would be consistent with pervasive negative selection on gene expression phenotypes ( i . e . , selection against variants that produce large perturbations in gene expression levels ) and has been suggested by work in humans ( Battle et al . , 2014 ) . Finally , we used our data set to provide the first estimates of the heritability of gene expression levels in wild primates , including the relative contributions of cis-acting and trans-acting genetic variation . We obtained blood samples from 63 individually recognized adult baboons in the Amboseli population ( Figure 1—figure supplement 1 ) . From these samples , we produced a total of 1 . 89 billion RNA-seq reads ( mean of 30 . 0 ± 4 . 5 s . d . million reads per individual , with 8 . 6 ± 1 . 8 s . d . million reads uniquely mapped to exons: Supplementary file 1A ) . On average , 67 . 2% of reads mapped to the most recent release of the baboon genome ( Panu2 . 0 ) , 69 . 2% of which could be assigned to a unique location . We used the set of uniquely mapped reads to estimate gene-wise gene expression levels for NCBI-annotated baboon RefSeq genes . After subsequent read processing and normalization steps ( ‘Materials and methods’ , Figure 1—figure supplement 1 and Figure 1—figure supplement 2 ) , we considered variation in gene expression levels for 10 , 409 genes expressed in whole blood ( i . e . , all genes for which we could test for cis-acting genetic effects on gene expression ) . We also used the RNA-seq reads to identify segregating genetic variants in the Amboseli population . We considered only high confidence sites that were variable within the Amboseli population ( ‘Materials and methods’; Figure 1—figure supplement 3 ) . As expected ( Piskol et al . , 2013 ) , these sites were highly enriched in annotated gene bodies ( Figure 1; Figure 1—figure supplement 4 ) . Based on parallel analyses applied to human RNA-seq data , we estimated approximately 97% of these sites to be true positives , and a median correlation between true genotypes and inferred genotypes of 98 . 7% ( ‘Materials and methods’; Figure 1—figure supplements 5–6 ) . To identify putative expression quantitative trait loci ( eQTL ) , we focused on variants that passed quality control filters , within 200 kb of the gene of interest . Such variants represent likely cis-acting eQTL , which are more readily identifiable in small sample sizes than trans-eQTL . To identify cases of allele-specific expression , which provides independent but complementary evidence for functional cis-regulatory variation , we focused on genes for which multiple heterozygotes were identified for variants in the exonic regions of expressed genes . We also required a minimum total read depth at exonic heterozygous sites of 300 reads ( which should provide high power to detect modest ASE: Fontanillas et al . , 2010 ) , resulting in a total set of 2280 genes tested for ASE . 10 . 7554/eLife . 04729 . 003Figure 1 . Baboon eQTLs are enriched in and near genes . The locations of all SNPs tested in the eQTL analysis are shown in gold relative to the 5′ most gene transcription start site ( TSS ) and the 3′ most gene transcription end site ( TES ) for all 10 , 409 genes . SNPs detected as eQTL are overplotted in blue , and are enriched , relative to all SNPs tested , near transcription start sites , transcription end sites , and within gene bodies . Gray shaded rectangle denotes the region bounded by the TSS and TES , with gene lengths divided into 20 bins for visibility ( because the gene body is thus artificially enlarged , SNP density within genes cannot be directly compared with SNP density outside of genes ) . Note that SNPs that fall outside of one focal gene may fall within the boundaries of other genes . Inset: distribution of all SNPs tested relative to the location of genes , highlighting the concentration of SNPs in genes ( the peak at the center of the plot ) . See Figure 1—figure supplements 1–14 for additional details on workflow , variant calling validation , location of all analyzed SNPs relative to genes , agreement between eQTL and ASE detection , and effects of local structure . DOI: http://dx . doi . org/10 . 7554/eLife . 04729 . 00310 . 7554/eLife . 04729 . 004Figure 1—figure supplement 1 . Detailed workflow for gene expression level estimation . DOI: http://dx . doi . org/10 . 7554/eLife . 04729 . 00410 . 7554/eLife . 04729 . 005Figure 1—figure supplement 2 . Elimination of GC bias via quantile normalization . Each plot shows gene GC content ( x-axis ) vs the log of the ratio of the individual's RPKM for that gene to mean RPKM across all individuals . Data for three individuals are shown in pairs ( A and B , C and D , E and F ) for prior to ( left ) and after ( right ) quantile normalization . DOI: http://dx . doi . org/10 . 7554/eLife . 04729 . 00510 . 7554/eLife . 04729 . 006Figure 1—figure supplement 3 . Detailed workflow for SNP genotyping . DOI: http://dx . doi . org/10 . 7554/eLife . 04729 . 00610 . 7554/eLife . 04729 . 007Figure 1—figure supplement 4 . Location of analyzed SNPs relative to genes . The locations of all SNPs tested in the eQTL analysis are shown in gold relative to the 5′ most gene transcription start site ( TSS ) and the 3′ most gene transcription end site ( TES ) for all 10 , 409 genes . The location of all SNPs tested in association with eQTL genes is overplotted in blue . Gray shaded rectangle denotes the region bounded by the TSS and TES , with gene lengths divided into 20 bins for visibility . DOI: http://dx . doi . org/10 . 7554/eLife . 04729 . 00710 . 7554/eLife . 04729 . 008Figure 1—figure supplement 5 . Accuracy of genotype calls for SNPs independently typed in HapMap3 . ( A ) Distribution of correlations between SNPs called using RNA-seq data and SNPs called independently by HapMap3 ( n = 9919 variants ) . ( B ) Estimated homozygosity levels for n = 69 YRI individuals at the same set of sites; outliers ( denoted with red stars ) reflect those individuals with the lowest correlation between RNA-seq-based genotypes and HapMap3 genotypes . The four starred outliers in ( B ) include the three lowest accuracy individuals in the boxplots in ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04729 . 00810 . 7554/eLife . 04729 . 009Figure 1—figure supplement 6 . PCA projection of YRI samples using the RNA-seq-based pipeline vs independently typed SNPs . PCA projection of genotype data from the RNA-seq-based pipeline and the HapMap3 data place individual samples very close together . ( A ) and ( B ) show the same data , but ( B ) zooms in on the central cluster for better visibility . DOI: http://dx . doi . org/10 . 7554/eLife . 04729 . 00910 . 7554/eLife . 04729 . 010Figure 1—figure supplement 7 . Agreement between eQTL and ASE approaches for identifying functional variants . ( A ) Venn diagram depicting the overlap between genes with significant eQTL and ASE , among genes tested in both cases ( note that the number of genes with eQTL is smaller in this figure than in the overall data set because we consider only the set of genes that were testable for both eQTL and ASE , n = 2280 instead of n = 10 , 409 ) . Genes with significant eQTL are more likely to have significantly detectable ASE and vice-versa ( n = 2280; p < 10−25 ) . ( B ) eQTL SNPs in exonic regions that could also be tested for ASE reveal correlated effect sizes ( n = 123; p < 10−20 ) . ( C ) Similarly , ASE SNPs exhibit effect sizes that are correlated with evidence for eQTL at the same sites ( n = 510; p < 10−45 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04729 . 01010 . 7554/eLife . 04729 . 011Figure 1—figure supplement 8 . Power to detect ASE vs eQTL . ( A ) Detection of ASE is favored for genes with higher expression levels ( p = 3 . 99 × 10−209 ) , ( B ) whereas detection of eQTL is favored for genes with greater cis-regulatory SNP density ( p = 1 . 05 × 10−73 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04729 . 01110 . 7554/eLife . 04729 . 012Figure 1—figure supplement 9 . Characteristics of YRI eQTL identified in the RNA-seq vs conventional pipelines . Boxplot differences between eQTL identified in the YRI data set using chip-based genotype data vs RNA-seq-based genotype data for ( A ) gene expression levels in RPKM ( Wilcoxon test p = 6 . 53 × 10−9 ) ; ( B ) conservation levels measured by average phyloP per gene ( p = 0 . 707 ) ; ( C ) conservation levels measured using Homologene conservation scores ( p = 0 . 600 ) ; and ( D ) magnitude of the eQTL effect size ( p = 0 . 137 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04729 . 01210 . 7554/eLife . 04729 . 013Figure 1—figure supplement 10 . Differences in the magnitude of ASE vs distance between sites . ( A ) Difference in the magnitude of ASE estimated for pairs of tested sites ( i . e . , absolute difference of the absolute values of z-scores ) , by distance between sites . ( B ) Difference in the magnitude of ASE estimated for pairs of tested sites for genes with significant ASE only , where one site in the pair is the site with the best ASE support for the gene . In both plots , distance categories reflect the range from the previous category to the labeled max value . DOI: http://dx . doi . org/10 . 7554/eLife . 04729 . 01310 . 7554/eLife . 04729 . 014Figure 1—figure supplement 11 . Location of eQTL SNPs relative to genes with and without controlling for local structure . The locations of all eQTL SNPs ( n = 1787 ) identified in the main eQTL analysis are shown in gold relative to the 5′ most gene transcription start site ( TSS ) and the 3′ most gene transcription end site ( TES ) . eQTL SNPs detected in a parallel analysis controlling for local structure ( n = 1583 ) are overplotted in blue . Gray shaded rectangle denotes the region bounded by the TSS and TES , with gene lengths divided into 20 bins for visibility . Note that SNPs that fall outside of one focal gene may fall within the boundaries of other genes . Inset: Quantile–quantile plot of eQTL locations in models that do and do not control for local structure ( Kolmogorov-Smirnov test , p = 0 . 577 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04729 . 01410 . 7554/eLife . 04729 . 015Figure 1—figure supplement 12 . Number of eQTL identified by PCs removed from the gene expression data set . DOI: http://dx . doi . org/10 . 7554/eLife . 04729 . 01510 . 7554/eLife . 04729 . 016Figure 1—figure supplement 13 . Coverage by genotype call . Mean coverage by genotype class for ( A ) all SNPs tested in the baboon eQTL analysis ( n = 64 , 432 ) , and ( B ) SNPs identified as eQTL ( n = 1693 ) . QQ plot of mean coverage in homozygotes for the reference allele vs homozygotes for the alternate allele for ( C ) all SNPs and ( D ) SNPs identified as eQTL . The magnitude of increased coverage in reference allele homozygotes indicates the degree of systematic reference allele mapping bias ( dashed line shows the expectation for no mapping bias ) . Reference allele homozygotes tend to have higher coverage , on average , than alternate allele homozygotes ( K-S test: p < 2 . 2 × 10−16 for all SNPs; p = 3 . 9 × 10−5 for eQTL SNPs ) , suggesting some degree of mapping bias; however the effect is actually smaller for eQTL SNPs than for all SNPs ( K-S D = 0 . 167 for all SNPs; K-S D = 0 . 084 for eQTL SNPs ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04729 . 01610 . 7554/eLife . 04729 . 017Figure 1—figure supplement 14 . Detection of ASE is not dependent on number of heterozygotes , conditional on total read depth . SNPs within the set tested for ASE ( n = 8145 ) were divided into deciles based on total read depth . The evidence for a relationship ( −log10 of the p-value from a Wilcoxon test ) between number of heterozygous individuals at each site and detection of significant ASE is shown on the y-axis for each decile . Dashed line shows a nominal significance threshold of p = 0 . 01 . Blue numbers above each point show the number of sites that fall within the decile; purple numbers below each point show the maximum total read depth for that decile ( minimum total read depth is the maximum depth for the previous decile , or 300 for the lowest decile ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04729 . 017 Both analyses converged to reveal extensive segregating genetic variation affecting gene expression levels in the Amboseli population . At a 10% false discovery rate , we identified eQTL for 1787 ( 17 . 2% ) of the genes we analyzed , and evidence for ASE for 510 ( 23 . 4% ) of tested genes . Consistent with reports in humans ( e . g . , Veyrieras et al . , 2008; Pickrell et al . , 2010a ) , eQTL were strongly enriched near gene transcription start sites and in gene bodies ( Figure 1; controlling for the background distribution of sites tested , which were also enriched in and around genes ) . Within gene bodies , eQTL were particularly likely to be detected near transcription end sites; this potentially reflects enrichment in 3′ untranslated regions , which are poorly annotated in baboon . Also as expected , genes with eQTL were more likely to exhibit significant ASE and vice-versa ( hypergeometric test: p < 10−25; Figure 1—figure supplement 7 ) . The magnitude and direction of ASE and eQTL were significantly correlated when an eQTL SNP could also be assessed for ASE ( n = 123 genes; r = 0 . 719 , p < 10−20 , Figure 1—figure supplement 7 ) , and when ASE SNPs were assessed as eQTL ( n = 510 genes; r = 0 . 575 , p < 10−45 , Figure 1—figure supplement 7 ) . Detection of ASE was most strongly favored for highly expressed genes ( i . e . , higher RPKM: Wilcoxon test: p < 10−208; Figure 1—figure supplement 8 ) , whereas detection of eQTL was most strongly favored for genes with high local SNP density ( p < 10−72; Figure 1—figure supplement 8 ) . The number and effect sizes of the eQTL we detected indicate that our power to detect eQTL in the Amboseli population was surprisingly high , especially given that our genotyping data set was limited only to those sites represented in RNA-seq data ( i . e . , primarily within transcribed regions of moderately to highly expressed genes ) . Further , while thousands of cis-eQTL have been mapped in single human populations , doing so has generally required sample sizes several fold larger than ours ( Lappalainen et al . , 2013; Battle et al . , 2014 ) . To provide a more informative estimate of the difference in power to detect eQTL in baboons relative to humans , we applied the same mapping , data processing , variant calling , and eQTL modeling pipeline to a similarly sized RNA-seq data set on 69 Yoruba ( YRI ) HapMap samples , in which samples were sequenced to a similar depth ( Pickrell et al . , 2010a ) . Using our approach for estimating and modeling the gene expression data , but obtaining the genotype data from an independent array platform , we could identify 700 genes with significant eQTL in the YRI data set at a 10% FDR . Approximately half ( 51% ) could be recovered if we only focused on SNPs in transcribed regions . This number ( n = 357 ) therefore reflects the likely theoretical limit of detection for performing eQTL mapping in which SNPs are called based on RNA-seq data . Indeed , when eQTL mapping for the YRI was conducted using genotype data obtained from RNA-seq reads ( i . e . , the same pipeline used for the baboons ) , we identified 290 genes with eQTL ( 41 . 4% of those identified using independently collected genotype data ) . eQTL identified in the RNA-seq pipeline do not differ from those identified only in the conventional pipeline in either effect size or in surrounding sequence conservation , but do tend to fall in more highly expressed genes ( Wilcoxon test on RPKM values: p = 6 . 53 × 10−9; Figure 1—figure supplement 9 ) , suggesting that sequencing coverage considerations reduce the number of identifiable eQTL below the theoretical maximum . The RNA-seq-based pipeline therefore reduces the number of genes with detectable eQTL by 50–60% , suggesting that if genotyping array data had been available for the baboons , we might have identified eQTL for ∼3500–4000 genes , comparable to results from human data sets with more than 350 samples ( Lappalainen et al . , 2013 ) . To better understand the reasons behind this difference , we investigated three possible explanations . Interestingly , we found that genes harboring eQTL in baboons were also more likely to have detectable eQTL in the YRI ( hypergeometric test , p = 2 . 39 × 10−7 ) . Given the sample size limitations of the data sets we considered , this overlap suggests that large effect eQTL tend to be nonrandomly concentrated in specific gene orthologues . This pattern could arise if the regulation of some genes has been selectively constrained over long periods of evolutionary time , whereas others have been more permissible to genetic perturbation . Indeed , we found that the mean per-gene phyloP score calculated based on a 46-way primate comparison was significantly reduced ( reflecting less conservation ) for genes with detectable eQTL in both species , and greatest for genes in which eQTL were not detected in either case ( p < 10−53; Figure 3A ) . We obtained similar results using phyloP scores based on a 100-way vertebrate comparison ( p < 10−21; Figure 3—figure supplement 1 ) . 10 . 7554/eLife . 04729 . 020Figure 3 . Mixed evidence for negative selection on variants affecting gene expression level . ( A ) Genes that harbor detectable eQTL in baboons , the YRI , or both are more likely to be conserved across long stretches of evolutionary time , based on mean phyloP scores in a 46-way primate genome comparison ( n = 7268; p < 10−53 ) . ( B ) These genes are also more likely to be lineage-specific , based on Homologene annotations ( n = 7065; p = 1 . 78 × 10−8 ) . ( C ) Although we detect a strong negative correlation between eQTL effect size and eQTL minor allele frequency , in support of pervasive selection against alleles with large effects on gene expression levels , this correlation also appears when simulating constant eQTL effect sizes , suggesting winner's curse effects . See Figure 3—figure supplement 1 for phyloP results based on a 100-way vertebrate genome comparison . DOI: http://dx . doi . org/10 . 7554/eLife . 04729 . 02010 . 7554/eLife . 04729 . 021Figure 3—figure supplement 1 . Correlation between eQTL detection and mean phyloP scores based on 100-way vertebrate comparison . Genes with eQTL in both data set or one data set are less conserved across vertebrates than genes for which no eQTL were detected ( n = 7 , 268 , p < 10−19 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04729 . 021 eQTL were more likely to be identified for genes with higher genetic diversity ( Figure 1—figure supplement 8 ) , which may account for the relationship between phyloP score and eQTL across species: highly conserved genes are less likely to contain many variable sites . More conserved genes also tend to have slightly lower average minor allele frequencies ( p = 0 . 002 ) , which might reduce the power to detect eQTL ( although the effect size is small: r2 = 0 . 001 ) . However , genes with eQTL in both species were also less likely to have orthologues in deeply diverged species , based on conservation in Homologene ( β = −0 . 036 , p = 1 . 78 × 10−8; Figure 3B ) . Genetic diversity within the baboons is very weakly correlated with Homologene conservation ( r2 = 0 . 004 ) and uncorrelated with average minor allele frequency ( p = 0 . 38 ) . Thus , sequence-level conservation scores and depth of homology across species combine to suggest that eQTL—or at least those with relatively large effect sizes—are least likely to be detected for strongly conserved loci , and most likely to be detected for lineage-specific , rapidly evolving genes . Consistent with this idea , genes involved in basic cellular metabolic processes were under-enriched among the set of genes with eQTL in both species , and enriched among the set of genes for which no eQTL were detected in either species ( Supplementary file 1B–C ) . The set of genes with eQTL in either or both species , on the other hand , were enriched for loci involved in antigen processing , catalytic activity , and interaction with the extracellular environment ( e . g . , receptors , membrane-associated proteins ) . Widespread selective constraint on gene expression levels has been suggested in previous eQTL analyses in humans , with evidence supplied by a strong negative correlation between minor allele frequency and eQTL effect size ( Battle et al . , 2014 ) . This pattern could arise if selection acts against large genetic perturbations , such that variants of large effect would be present only at low frequencies . Consistent with this idea , plotting eQTL effect size vs MAF in the baboons results in a very strong , highly significant negative correlation ( r = −0 . 723 , p < 10−280; Figure 3C ) , with no large effect eQTL detected at higher MAFs . However , such a relationship could also be a consequence of the so-called winner's curse ( in which sampling variance leads to upwardly biased effect size estimates: Zöllner and Pritchard , 2007 ) because the degree of bias in effect size estimation is itself negatively correlated with MAF . Indeed , when we simulated sets of eQTL with constant small effect sizes ( β = 0 . 75 , close to the mean effect size detected for SNPs with MAF ≥0 . 4 ) , we found that the relationship between estimated effect size and MAF among detected eQTL almost perfectly recapitulated the observed negative correlation . Hence , the correlation between estimated eQTL effect size and MAF in the baboons does not provide strong support for widespread negative selection on gene expression phenotypes within species . We note , however , that our sample size of individuals is much smaller than that used for a similar analysis in humans ( Battle et al . , 2014: n = 922 individuals ) , and larger sample sizes should attenuate winner's curse effects . Finally , we took advantage of our data set to generate the first estimates of genetic , demographic ( age and sex ) , and environmental contributions to gene expression variation in wild nonhuman primates ( Supplementary file 1D ) . While our limited sample size leads to high variance around estimates for any individual gene , the median estimates across genes should be unbiased ( Zhou et al . , 2013 ) , so we concentrated on these overarching patterns . We focused specifically on three social environmental variables of known importance in this population , all of which have been extensively investigated as models for human social environments . These were: ( i ) early life social status , which predicts growth and maturation rates ( Altmann and Alberts , 2005; Charpentier et al . , 2008 ) ; ( ii ) maternal social connectedness to other females , which predicts both adult lifespan and the survival of a female's infants ( Silk et al . , 2003 , 2009 , 2010; Archie et al . , 2014 ) ; and ( iii ) maternal social connectedness to males , based on recent evidence that heterosexual relationships have strong effects on survival as well ( Archie et al . , 2014 ) . Overall , we found that genetic effects on gene expression levels tended to be far more pervasive than demographic and environmental effects . Specifically , the median additive genetic PVE was 28 . 4% , similar to , or slightly greater than , estimates from human populations ( Monks et al . , 2004; McRae et al . , 2007; Emilsson et al . , 2008; Price et al . , 2011; Wright et al . , 2014 ) . We applied a Bayesian sparse linear mixed model ( BSLMM: Zhou et al . , 2013 ) to further partition this additive genetic PVE into two components: a component attributable to cis-SNPs ( here , all SNPs within 200 kb of a gene ) and a component attributable to trans-SNPs ( all other sites in the genome ) . Again similar to humans ( Price et al . , 2011; Wright et al . , 2014 ) , we found that more of the additive genetic PVE is explained by the trans component ( median PVE = 23 . 8% ) than the cis component ( median PVE = 2 . 9% ) ( Figure 4 ) . Unsurprisingly , we estimated a larger cis-acting component for genes in which functional cis-regulatory variation was detected in our previous analysis ( median PVE = 10 . 2% among eQTL genes and median PVE = 5 . 0% among ASE genes ) . 10 . 7554/eLife . 04729 . 022Figure 4 . Genetic contributions to variance in gene expression levels in wild baboons . Proportion of variance in gene expression levels estimated for all genes , genes without detectable eQTL , and genes with detectable eQTL . Additive genetic effects on gene expression variation , especially cis-acting effects , are larger for eQTL genes than for other genes . See Figure 4—figure supplements 1–3 for related results on percent variance explained by genetic , environmental , and demographic variables and results using an alternative set of SNPs for estimating ptrans . DOI: http://dx . doi . org/10 . 7554/eLife . 04729 . 02210 . 7554/eLife . 04729 . 023Figure 4—figure supplement 1 . PVE explained by demographic and early environmental variables . QQ plots of PVE explained by a variable of interest vs PVE explained by that variable with permuted data , for ( A ) age; and ( B ) maternal social connectedness to males ( SCI-M ) . Bottom panels show the difference between evidence for significant PVE by sex for ( C ) genes on autosomes vs ( D ) genes on the X chromosome ( bottom right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04729 . 02310 . 7554/eLife . 04729 . 024Figure 4—figure supplement 2 . Distribution of PVE explained by additive genetic variance , age , sex , and maternal social connectedness to males across all genes . DOI: http://dx . doi . org/10 . 7554/eLife . 04729 . 02410 . 7554/eLife . 04729 . 025Figure 4—figure supplement 3 . Genetic contributions to variance in gene expression levels , with ptrans based on SNPs on other chromosomes only . DOI: http://dx . doi . org/10 . 7554/eLife . 04729 . 025 In contrast to the substantial genetic effects we detected , the median PVE explained by age and sex were 1 . 89% and 0 . 82% , respectively ( Figure 4—figure supplements 1–2 ) . The distribution of PVE explained by age was significantly greater than expected by chance ( Kolmogorov–Smirnov test on binned PVEs , in comparison to permuted data: p < 10−11 ) , whereas that explained by sex was not ( p = 0 . 100 ) ; large sex effects tended to be constrained to a small set of genes on the X chromosome ( Figure 4—figure supplement 1 ) . Of the early environmental variables we investigated , only maternal social connectedness to males explained more variance in gene expression levels than expected by chance ( p = 4 . 19 × 10−3 ) , with a median PVE of 1 . 9% . Notably , while social connectedness to males ( i . e . , heterosexual bonds ) and social connectedness to females ( i . e . , same-sex bonds ) are both known predictors of longevity in the Amboseli baboons , previous analyses suggest that their effects are largely independent ( Archie et al . , 2014 ) . Our result extends this observation to the early life effects of maternal social connectedness on variance in gene expression levels . Taken together , our data suggest that while almost all genes are influenced by genetic variation , the effects of demographic and environmental parameters are generally modest for any single aspect of the environment . However , in at least some cases , we find evidence that early environmental effects on gene expression levels appear to persist across the life course , as has previously been demonstrated in laboratory settings and in response to severe early adversity in humans ( e . g . , Weaver et al . , 2004; Miller et al . , 2009; Cole et al . , 2012 ) . Much of what we know about genetic contributions to variation in gene expression levels in primates ( and vertebrates more generally ) come from the extensive body of research on humans . However , increasing evidence indicates that humans are demographically unusual: compared to other primates , humans exhibit low levels of neutral genetic diversity and a low long-term effective population size ( Chen and Li , 2001; Hernandez et al . , 2007; Perry et al . , 2012 ) . Further , humans are distinguished from other primates by recent explosive population growth ( Keinan and Clark , 2012; Tennessen et al . , 2012 ) . While late Pleistocene population expansion has been suggested for some nonhuman primates , including chimpanzees and Chinese-origin rhesus macaques ( Hernandez et al . , 2007; Wegmann and Excoffier , 2010 ) , none have undergone the extreme levels of population increase that characterized humans . Indeed , evidence from microsatellite data suggests that the long-term effective population size of baboons actually may have contracted during this period ( Storz et al . , 2002 ) . These differences are not simply of historical interest , but also important for understanding the genetic architecture of traits measured in the present day . Differences in demographic history not only affect overall levels of genetic variation and the minor allele frequency spectrum , but also the mean effect size of sites that contribute to phenotypic variation ( Lohmueller , 2014 ) . Interestingly , demographic history does not impact overall trait heritability ( Lohmueller , 2014; Simons et al . , 2014 ) , perhaps explaining why we estimated mean additive genetic PVEs for gene expression levels in baboons that are similar to those estimated for humans . However , demographic history can influence the power to detect individual genetic contributions to phenotypic variation . Large-scale population expansion of the type that occurred in human history appears to reduce power to identify genotype-phenotype correlations for fitness-related traits ( Lohmueller , 2014 ) . This observation may account , in part , for our ability to identify many more functional regulatory variants in the baboons than we expected based on previous studies in humans . However , while our analysis extends previous observations that large effect eQTL are nonrandomly distributed , we found mixed evidence for widespread negative selection on gene expression levels . Specifically , within the baboons alone , we found that the negative relationship between eQTL effect size and minor allele frequency was explicable based on winner's curse effects alone . Thus , increased power to identify functional regulatory variants in the baboons is probably not due to pervasive associations between gene expression levels and fitness . In contrast , stronger evidence for selection on gene expression patterns stems from our cross species comparisons . In particular , we observed that genes with eQTL in baboons significantly overlapped with genes with eQTL in humans , and that these genes as a class also tended to be less constrained at the sequence-level ( consistent with observations for analyses of cis-eQTL in humans alone: Popadin et al . , 2014 ) . This result suggests that genes vary in their tolerance of functional regulatory genetic variation , and , intriguingly , that gene-specific robustness to genetic perturbation may be a conserved property across species . Because no comparable data are yet available for other large mammal populations , including for other baboons , it is unclear whether our results are typical or instead a consequence of the Amboseli population's own unique history . In particular , the population has experienced recent admixture between yellow baboons , the dominant taxon , and closely related anubis baboons ( P . anubis ) ( Alberts and Altmann , 2001; Tung et al . , 2008 ) . Admixture , which appears to be relatively common in natural populations ( Mallet , 2005 ) , can have important consequences for genetic diversity and LD patterns . While it appears to have had a modest impact on the relative ability to map gene expression phenotypes in baboons vs the YRI data set , comparison to a non-admixed baboon population could help resolve this question further . More generally , our results encourage further investigation of the relationship between demography and trait genetic architecture in other populations , as has been suggested for humans ( Lohmueller , 2014 ) but could also be profitably extended to nonhuman model systems . Such comparisons would provide an empirical basis for testing predicted relationships between demographic history and the power to identify genotype-phenotype associations . From an applied perspective , they could also help identify animal models that favor more highly powered association mapping studies , a strategy that has already been heavily exploited in domestic dogs ( Karlsson et al . , 2007; Karlsson and Lindblad-Toh , 2008 ) and suggested for rhesus macaques ( Hernandez et al . , 2007 ) . While the same sites will probably rarely be associated with the same traits across species , this strategy could help identify molecular mechanisms that are conserved across humans and animal models ( e . g . , Lamason et al . , 2005 ) . Comparisons that use matched sample types will be particularly informative: our study compared eQTL detection in whole blood ( from the baboons ) with eQTL from lymphoblastoid cell lines ( YRI ) , which exhibit highly correlated , but not identical , patterns of overall gene expression ( Spearman's rho = 0 . 645 , p < 1 × 10−16 ) —these differences could affect rates of eQTL detection as well . Finally , our data—the first profile of genome-wide gene expression levels in a wild primate population—serve as a useful proof of principle of the ability to concurrently generate genome-wide gene expression phenotype and genotype data , and to relate them to each other using eQTL and ASE approaches . Intensively studied natural primate populations—some of which have been studied continuously for 30 or more years—have emerged as important phenotypic models for human behavior , health , and aging . The approach we used here provides a way to leverage these models for complementary genetic studies as well , especially if eQTL prove to be strongly enriched for sites associated with other traits , as in humans ( Nicolae et al . , 2010 ) . Although preliminary , our results highlight the increasing feasibility of integrating functional genomic data with phenotypic data on known individuals in the wild . For example , our data set revealed a number of genes in which variation in gene expression levels could be mapped to an identifiable eQTL , validated using an ASE approach , and also linked to early life environmental variation . Such cases suggest the potential for future investigations of the molecular basis of persistent environmental effects , including whether genetic and environmental effects act additively or interact . Study subjects were 63 individually recognized adult members ( 26 females and 37 males ) of the Amboseli baboon population . All study subjects were recognized on sight by observers based on unique physical characteristics . To obtain blood samples for RNA-seq analysis , each baboon was anesthetized with a Telazol-loaded dart using a handheld blowpipe . Study subjects were darted opportunistically between 2009 and 2011 , avoiding females with dependent infants and pregnant females beyond the first trimester of pregnancy ( female reproductive status is closely monitored in this population , and conception dates can be estimated with a high degree of accuracy ) . Following anesthetization , animals were quickly transferred to a processing site distant from the rest of the group . Blood samples for RNA-seq analysis were collected by drawing 2 . 5 ml of whole blood into PaxGene Vacutainer tubes ( Qiagen , Valencia , CA ) , which contain a lysis buffer that stabilizes RNA for downstream use . Following sample collection , study subjects were allowed to regain consciousness in a covered holding cage until fully recovered from the effects of the anesthetic . They were then released within view of their social group; all subjects promptly rejoined their respective groups upon release , without incident . Blood samples were stored at approximately 20°C overnight at the field site . Samples were then shipped to Nairobi the next day for storage at −20°C until transport to the United States and subsequent RNA extraction . For each RNA sample ( one per individual ) , we constructed an RNA-seq library suitable for measuring whole genome gene expression using Dynal bead poly-A mRNA purification and a standard Illumina RNA-seq prep protocol . Each library was randomly assigned to one lane of an Illumina Genome Analyzer II instrument and sequenced to a mean depth of 30 million 76-base pair reads ( ±4 . 5 million reads s . d . , Supplementary file 1A ) . The resulting reads were mapped to the baboon genome ( Panu2 . 0 ) using the efficient short-read aligner bwa 0 . 5 . 9 ( Li and Durbin , 2009 ) , with a seed length of 25 bases , a maximum edit distance of two mismatches in the seed , a read trimming quality score threshold of 20 , and the default maximum edit distance ( 4% after trimming ) . To recover reads that spanned putative exon–exon junctions , and therefore could not be mapped directly to the genome , we used the program jfinder on reads that did not initially map ( Pickrell et al . , 2010b ) . Finally , we filtered the resulting mapped reads data for low quality reads ( quality score <10 ) and for reads that did not map to a unique position in the genome . To assign reads to genes , we used the RefSeq exon annotations for Panu 2 . 0 ( ref_Panu_2 . 0_top_level . gff3 , downloaded September 6 , 2012 ) . We considered the total read counts for each gene and individual as the sum of the number of reads for that individual that overlapped the union of all exon base pairs assigned to a given gene . In downstream analyses , we considered only highly expressed genes that had non-zero counts in more than 10% individuals , and that had mean read counts greater than or equal to 10 ( excluding the gene for beta-globin ) . We then performed quantile normalization across samples followed by quantile normalization for each gene individually , resulting in estimates of gene expression levels for each gene that were distributed following a standard normal distribution . This procedure effectively removed GC bias in gene expression level estimates ( Figure 1—figure supplement 2 ) . For eQTL mapping , ASE analysis , and PVE estimation for sex and age we used all 63 individuals . For PVE estimation for maternal rank and social connectedness , missing data meant that we conducted our analysis on n = 52 and n = 47 individuals , respectively . To identify genetic variants in the baboon data set , we used the Genome Analysis Toolkit ( v . 1 . 2 . 6; McKenna et al . , 2010; DePristo et al . , 2011 ) . Because no validated reference set of known genetic variants are available for baboon , we performed an iterative bootstrapping procedure for base quality score recalibration . Specifically , we performed an initial round of base quality score recalibration and identified a set of variants using GATK's UnifiedGenotyper and VariantFiltration walker . From this call set , we constructed a set of high confidence variants with quality score ≥100 that passed all filters for variant confidence ( variants failed if QD < 2 . 0 ) , mapping quality ( variants failed if MQ < 35 . 0 ) , strand bias ( variants failed if FS > 60 . 0 ) , haplotype score ( variants failed if HaplotypeScore >13 . 0 ) , mapping quality ( variants failed if MQRankSum < −12 . 5 ) and read position bias ( variants failed if ReadPosRankSum < −8 . 0 ) . We used this high confidence set as the set of ‘known sites’ in a second round of base quality score recalibration , repeating this procedure until the number of variants identified in consecutive rounds of recalibration stabilized . In the final call set , we removed all sites ( i ) that were monomorphic in the Amboseli samples; ( ii ) for which genotype data were missing for more than 12 individuals ( 19% ) in the data set; ( iii ) that deviated from Hardy–Weinberg equilibrium; and ( iv ) that failed the above quality control filters . We further filtered the data set to contain only sites with a minimum quality score of 100 that were located within 200 kb of a gene of interest , and that were sequenced at a mean coverage ≥5× across all samples . We validated our quality control and filtering steps by performing the same procedure on an RNA-seq data set from the HapMap Yoruba population ( see below ) . These steps resulted in a set of 64 , 432 single nucleotide polymorphisms carried forward into downstream analysis ( 30 , 938 for the YRI ) . For eQTL mapping analysis , missing genotypes in this final set were imputed using BEAGLE ( Browning and Browning , 2009 ) . To estimate genome-wide LD , we followed the approach of Eberle et al . ( 2006 ) , which uses allele frequency-matched SNPs to calculate pair-wise LD . Specifically , we selected SNPs with MAFs greater than 10% and divided them into four subgroups ( MAF between 10%–20%; MAF between 20%–30%; MAF between 30%–40%; and MAF between 40%–50% ) . We then calculated pair-wise r2 for all SNP pairs within 100 kb in each subgroup using VCFtools ( Danecek et al . , 2011 ) and combined values from all four subgroups . To assess the accuracy of the RNA-seq-based genotyping calls we performed in the baboons , we investigated a similarly sized data set of RNA-seq reads from a human population ( Pickrell et al . , 2010a ) . Because this data set focused on samples from the HapMap consortium ( n = 69 members of the Yoruba population from Ibadan , Nigeria ) , we were able to compare genotypes called using the RNA-seq pipeline to independently collected genotype data from HapMap Phase 3 ( r27 ) ( International HapMap Consortium , 2010 ) . To do so , we focused on 9919 variants that were genotyped in both data sets . We then calculated the correlation between genotypes called in the RNA-seq-based pipeline and genotypes from HapMap , for each individual ( Figure 1—figure supplement 5A ) . We also found that low accuracy was correlated with the level of apparent homozygosity in the genotype data ( Figure 1—figure supplement 5B ) . In the baboon data , we had no individuals with unusually low homozygosity , but six individuals with unusually high homozygosity ( >80% of genotype calls ) . These outliers were missing a median of 10 . 6% of data in the unimputed genotype data set , whereas all other individuals were missing a median of 0 . 6% data . However , removing these six individuals from our analysis resulted in very similar results as using the full data set: 87 . 6% of eQTL genes ( n = 1566 ) identified when using all individuals were also identified with this subset . Importantly , the available data from humans also support accurate variant discovery . Of the 30 , 938 sites that we identified from the RNA-seq data and that passed all of our filters , only 3 . 1% ( 967 ) did not have an assigned rsID in dbSNP release 138 . These sites were likely enriched for false positives , as the transition/transversion ratio for this set was 1 . 42 , vs 2 . 80 for the set of 30 , 938 sites as a whole . To identify cis-acting eQTLs in the baboon data set , we used the linear mixed model approach implemented in the program GEMMA ( Zhou and Stephens , 2012 ) . This model provides a computationally efficient method for eQTL mapping while explicitly accounting for genetic non-independence within the sample; in our case , some individuals in the data set are related ( although overall relatedness was low: the median kinship coefficient across all pairs was 0 . 015; mean = 0 . 024 ± 0 . 033 s . d . ) . For each gene , we considered all variants within 200 kb of the gene as candidate eQTLs . For each variant , we fitted the following linear mixed model:y=μ+xβ+u+ε , u∼MVN ( 0 , σu2K ) , ε∼MVN ( 0 , σe2I ) , and tested the null hypothesis H0: β = 0 vs the alternative H1: β ≠ 0 . Here , y is the n by 1 vector of gene expression levels for the n individuals in the sample . Gene expression values were first corrected for hidden factors that could act as sources of global structure ( e . g . , batch effects or ancestry- or environment-related trans effects ) by regressing out the first 10 principal components of the gene expression data . Consistent with previous results ( e . g . , Pickrell et al . , 2010a ) , this procedure greatly improves our ability to detect eQTL ( Figure 1—figure supplement 12 ) . In the model , μ is the intercept; x is the n by 1 vector of genotypes for the variant of interest; and β is the variant's effect size . The n by 1 vector of u is a random effects term to control for individual relatedness and other sources of population structure , where the n by n matrix K = XXT/p provides estimates of pairwise relatedness derived from the complete 63 × 64 , 432 genotype data set X . Residual errors are represented by ε , an n by 1 vector , and MVN denotes the multivariate normal distribution . We took the variant with the best evidence ( i . e . , lowest p-value ) for association with gene expression levels for each gene , and then calculated corrected gene-wise q-values ( with a 10% false discovery rate threshold ) via comparison to the same values obtained from permuted data ( similar to Barreiro et al . , 2012; Pickrell et al . , 2010a ) . We evaluated the potential for eQTL mapping based on RNA-seq data to introduce four possible confounds . First , for genes with large effect cis-eQTLs , reads from heterozygotes at eQTL-linked sites might be biased towards the allele associated with higher gene expression levels . If so , heterozygotes might be mistakenly genotyped as homozygotes for the high expressing allele , resulting in an underrepresentation of heterozygous genotypes relative to neutral expectations . To control for this possibility , we eliminated sites that violated Hardy-Weinberg expectations ( n = 2386 ) from our analyses . We note , however , that this scenario would not introduce false positives . Instead , it would lead to more conservative detection of additive eQTL effects , with the direction of an estimated eQTL effect still consistent with the true effect . Second , SNP calling might be biased towards the reference allele . If so , more reads would be required to support a genotype call of homozygote alternate than a genotype call of homozygote reference . This bias would result in higher apparent expression levels for alternate allele homozygotes and lower expression levels for reference allele homozygotes , which could create false positive eQTLs . However , we observe no evidence for this scenario in our data set . For all tested SNPs ( n = 64 , 432 ) and for eQTL SNPs only ( n = 1693 ) , alternate allele homozygotes tend to have slightly lower coverage than reference allele homozygotes , and heterozygotes tend to have the highest coverage ( because more reads are required to support inference of heterozygosity ) ( Figure 1—figure supplement 13 ) . Thus , coverage and genotype do not covary additively , and this potential confound is unlikely to produce false positive eQTLs . Third , read mapping might be biased towards the reference allele , such that reads carrying the alternate allele are less likely to map because they contain more mismatches to the reference genome . This possibility is consistent with our observation that alternate allele homozygotes tend to have slightly less coverage than reference allele homozygotes ( Figure 1—figure supplement 13 ) . While this difference in coverage is significant ( Kolmogorov-Smirnov test: p < 2 . 2 × 10−16 for all SNPs; p = 3 . 9 × 10−5 for eQTL SNPs ) , the magnitude of the effect itself is modest ( Figure 1—figure supplement 13 ) , probably because we allowed reads to map with up to three mismatches: Wittkopp and colleagues have shown that reference allele mapping bias is largely obviated by allowing reads to map with more mismatches ( Stevenson et al . , 2013 ) . Further , systematic calling of false positive eQTLs due to biased read mapping would predict a bias towards negative effect sizes ( i . e . , eQTL effects suggesting that the alternate allele is associated with lower expression levels ) . Our data are not consistent with such a pattern: 47% of eQTL betas are negative , whereas 53% are positive . Reference allele mapping biases are , however , more likely to affect ASE analysis , producing a pattern of greater expression in the reference allele . Indeed , we do observe a bias towards negative betas in the ASE analysis ( 67 . 2% of n = 510 genes ) , although the overall magnitude and direction of ASE data agree well with eQTL evidence . Fourth , lower mean coverage in homozygotes of either type relative to heterozygotes could induce false positive eQTLs in which the major allele was associated with lower gene expression levels . To test this possibility , we recoded eQTL effects to reflect the effect of the major allele instead of the effect of the alternate allele ( i . e . , a genotype of 0 = homozygous minor and a genotype of 2 = homozygous major ) . We observed a modest excess of eQTL for which the major allele was associated with lower gene expression levels ( 56% , binomial test p = 1 . 15 × 10−7 ) . This bias did not differ depending on whether the major allele was the reference allele or the alternate allele ( Fisher's Exact Test , p = 0 . 28 ) , supporting minimal read mapping biases in our data . Instead , it appears to be primarily driven by SNPs with low minor allele frequencies ( proportion of negative betas for the lowest quartile of MAFs = 62 . 8% , p = 7 . 49 × 10−8; highest quartile of MAFs = 48 . 6% , p = 0 . 602 ) . At these sites , eQTL inference relies primarily on two genotype classes ( the major allele homozygotes and heterozygotes ) rather than three genotype classes . Because heterozygotes tend to have slightly higher coverage than homozygotes of both classes , spurious relationships between genotype and gene expression levels are much less likely to be observed when both types of homozygotes are well represented ( i . e . , MAFs are larger ) . Along with the high genotype accuracy rates estimated from the Yoruba data , our analyses thus indicate that the set of eQTL we identified are largely robust to RNA-seq-specific confounds . The eQTL identified in YRI in the conventional pipeline vs the RNA-seq pipeline offer a further source of comparison . We find that eQTL identified through the RNA-seq pipeline tend to be associated with more highly expressed genes ( providing greater power to call genotypes: Wilcoxon test p = 6 . 53−9 ) , but otherwise do not differ in sequence conservation ( phyloP scores: p = 0 . 707; Homologene scores: p = 0 . 603 ) or in estimated effect size ( p = 0 . 137 ) ( Figure 1—figure supplement 9 ) . Further , effect size magnitude is highly correlated across pipelines when eQTL are discovered in both pipelines ( r = 0 . 874 , p < 10−57 ) . When eQTL were discovered only in the RNA-seq pipeline ( n = 104 ) , they tended to be high on the ranked list of eQTL evidence in the conventional pipeline as well ( median rank of 1395 , where the top 700 were significant and 10 , 615 genes were tested ) , suggesting that many of them did not pass the threshold for eQTL detection in that analysis . Thus , the most salient source of error stems from low MAF sites , which are also the cases most vulnerable to sampling error and winner's curse effects more generally ( Figure 3 ) —a problem that is not confined to RNA-seq-based eQTL mapping . Taken together , these analyses argue that , as a general rule , eQTL associated with lower MAF SNPs should be treated with increased caution . To identify ASE , we focused on SNPs within gene exons with Phred-scaled quality scores greater than 10 . We further required that these sites have more than five reads in more than two individuals and more than 300 total reads across all heterozygous individuals . This threshold is based on the observation that the power to detect ASE is dependent on sequencing read coverage at heterozygous sites ( Fontanillas et al . , 2010 ) . Indeed , in our data set , power to detect ASE appeared to scale primarily with total read coverage rather than number of heterozygous individuals . Sites with more reads tended to have more heterozygotes ( r = 0 . 266 , p < 10−100 ) ; however , when sites were partitioned by total read depth ( in deciles ) , sites with significant ASE were not more likely to harbor more heterozygotes in any decile ( Wilcoxon test comparing number of heterozygotes in significant sites vs background; Figure 1—figure supplement 14 ) . After these filtering steps , we retained 8154 SNPs associated with 2280 genes for ASE analysis . For ASE analysis , we did not take into account possible recombination between exonic SNPs and the ( unknown ) cis-regulatory variants whose effects they capture , as we did not have detailed data on recombination rates across the baboon genome . However , recombination between exonic SNPs and the true causal regulatory SNPs would decrease our power to detect ASE . For each variant , we considered a beta-binomial distribution ( following Pickrell et al . , 2010a ) to model the number of reads from the ( + ) haplotype ( denoted as xi+ ) or the number of reads from the ( − ) haplotype ( denoted as xi− ) , conditional on the number of total reads ( denoted as yi = xi+ + xi− ) , for each individual i , orxi+\|yi∼binomial ( yi , θ ) , θ∼beta ( α , β ) . We tested the null hypothesis H0: α = β vs the alternative H1: α ≠ β using a likelihood ratio test . For both the null model and the alternative model , beta distribution parameters ( α and β ) were estimated via a maximum likelihood approach , using the R function optim . Again , we took the variant with the lowest p-value for each gene , and then calculated corrected gene-wise q-values ( using a 10% false discovery rate threshold ) via comparison to the same values obtained from an empirical null distribution . To construct the empirical null distribution , we performed the same analysis after substituting the xi+ value for each variant of interest , for each heterozygous individual , with a randomly selected xi+ value from a heterozygous site elsewhere in the genome ( contingent on that site having the same number of total reads , yi ) . To assess the relative power of eQTL mapping in baboons vs the YRI data set , we randomly selected 10% of the genes in each data set to harbor eQTL . For each of these simulated eQTL genes , we then randomly chose a SNP among all the cis-SNPs tested ( i . e . , all variable sites that passed quality control filters and fell within 200 kb of a gene of interest ) and assigned it as a causal eQTL . The impact of the eQTL was simulated using either effect size , in which we simulated a constant effect size between 0 . 25 and 2 . 5 ( in intervals of 0 . 25 ) or PVE , in which we chose an effect size that explained a specific proportion of variance in gene expression levels ( from 5% to 50% , in intervals of 5% ) . We then simulated gene expression levels by adding the effect of the simulated cis-eQTL SNP to residual errors drawn from a standard normal distribution . To calculate the FDR , we also simulated a set of genes with no eQTL . For each combination of effect sizes and population ( baboon or YRI ) , and for each simulation scenario ( e . g . , with the causal SNP masked or unmasked , with SNP density thinned in the baboons , or using PVE vs a constant effect size ) , we performed 10 replicates . For each replicate , we calculated the power to detect eQTL as the proportion of simulated eQTL genes recovered at a 10% empirical FDR . To investigate whether admixture might drive our power to detect eQTL in the baboon data set , we performed three analyses . First , we asked whether evidence for ASE remained similar across longer distances ( i . e . , between sites separated by more base pairs ) in the baboons vs in the YRI . Such a pattern might be expected if long-distance , admixture-driven LD explained our other observations . However , the pattern of ASE similarity ( the magnitude of the difference between ASE estimates ) by distance between sites was highly congruent between the YRI and baboon data sets ( Figure 1—figure supplement 10 ) . Second , we investigated whether adding a control for local structure ( i . e . , population structure in cis to a gene of interest , and based only on variants located on the same chromosome ) asymmetrically reduced evidence for eQTL in the baboon data set relative to the YRI data set . To do so , we regressed out the top two PCs for variants on the same chromosome as the gene of interest from the gene expression data prior to fitting mixed effects models . We found that this approach modestly reduced the number of eQTL discoveries in the baboon data set ( n = 1583 from n = 1787 , an 11 . 4% difference ) . However , this number was still 5 . 4× larger than the number of eQTL detectable in the YRI , and when we applied the same local structure control to the YRI data , a comparable drop in the number of discoverable eQTL also occurred ( n = 216 from n = 290 , resulting in a ∼7× fold increase in eQTL in baboons vs YRI ) . Third , we compared the spatial distribution of eQTL in baboon between the models with and without local structure controls . We reasoned that if admixture drove most of the signal in the data set , controlling for local structure should shift the location of discovered eQTL closer to the gene of interest , where the strongest cis effects are generally identified . However , the locations of eQTL were very similar under both models ( Kolmogorov-Smirnov test , p = 0 . 577; Figure 1—figure supplement 11 ) . We investigated the relationship between conservation level and the presence of detectable eQTL in the Amboseli baboons or the YRI using phyloP conservation scores ( Pollard et al . , 2010 ) and Homologene conservation of orthology across species . For the former , we extracted the per-site phyloP score from the 46-way primate comparison or 100-way vertebrate comparison on the UCSC Genome Browser for each base contained within the annotated exons ( including untranslated regions ) used for mapping RNA-seq reads in the YRI . We then calculated the average phyloP score across all exons associated with a given gene . We obtained Homologene scores from the CANDID database ( Hutz et al . , 2008 ) . In both cases , we used linear models to test for a relationship between conservation level and three categories of genes: those with no detectable eQTL in either the baboons or YRI; those with a detectable eQTL in one of the two species; and those with a detectable eQTL in both species . To investigate whether the correlation between minor allele frequency and eQTL effect size could be a result of winner's curse effects , we extracted the results from our simulations in which the causal variant was masked and the true effect size was fixed at a small value ( beta = 0 . 75 ) . We then calculated the correlation between the estimated effect size ( β ) from these simulations against minor allele frequency , for detected eQTL only . We used the Bayesian sparse linear mixed model ( BSLMM ) approach implemented in the GEMMA software package ( Zhou and Stephens , 2012 ) to estimate the genetic contribution to gene expression variation . Specifically , for each gene , we fit the following model:y=μ+xcisβcis+xtransβtrans+ε , βcis , i∼πN ( 0 , σa2 ) + ( 1−π ) δ0 , βtrans , i∼N ( 0 , σb2 ) , where y is the n by 1 vector of gene expression levels for n individuals; μ is the intercept; xcis is an n by pcis matrix of genotypes for pcis cis-SNPs and βcis are the corresponding effect sizes; xtrans is an n by ptrans matrix of genotypes for ptrans trans-SNPs and βtrans are the corresponding effect sizes; and ε is an n by 1 vector of i . i . d . residual errors . We used different priors for cis-acting effects and trans-acting effects to capture different properties for the two components . Specifically , the spike-slab prior on the cis effects βcis captures our prior belief that only a small proportion of local SNPs has cis effects and these effects are relatively large . The normal prior on the trans effects captures our prior knowledge that trans-acting SNPs tend to be relatively difficult to find and have relatively small effects . In addition , because pcis is small and ptrans approximately equals p , the number of total SNPs , we used p instead of ptrans to facilitate computation ( i . e . , ptrans was based on all genotyped sites used in our analyses , n = 64 , 432 ) . Results are qualitatively similar if ptrans is calculated based on sites that must act in trans ( i . e . , sites located on a different chromosome than the chromosome containing the gene of interest: Figure 4—figure supplement 3 ) . We used Markov chain Monte Carlo ( MCMC ) to fit the model with 1000 burn-in and 10 , 000 sampling steps . We obtained posterior samples of βcis and βtrans to calculate the PVE attributed by each of the two components , as well as the total additive genetic PVE contributed by both components . To calculate PVE values for demographic and environmental predictors , we again used the linear mixed model approach implemented in GEMMA to control for additive genetic effects . Sex was known from direct observation of the study subjects . Ages were known to within a few days' error for 52 of the 63 individuals in the data set; six animals had birth dates estimated to be accurate within 1 year , four animals had birth dates estimated to be accurate within 2 years , and one had a birth date estimated to be less accurate than 2 years . Early social status was measured using the proportional dominance rank of the individual's mother , at the time of that individual's conception . Dominance ranks are assigned monthly using ad libitum observations of dyadic agonistic ( aggressive or competitive ) encounters within social groups ( Hausfater , 1974; Alberts et al . , 2003 ) . Maternal social connectedness values were defined as the social connectedness of the individual's mother , in the year of that female's life during which the focal individual was born . Social connectedness is calculated on a yearly basis as the frequency with which a female was involved in affiliative interactions , relative to the median for all females in the population at the same time and controlling for observer effort ( see Runcie et al . , 2013; Archie et al . , 2014 ) . Social connectedness is measured for females , but can focus on either female–female relationships ( SCI-F ) or a female's relationship with adult males ( SCI-M ) , which have independent effects on longevity in this population ( Archie et al . , 2014 ) . For SCI-F , affiliative interactions included both grooming interactions and close spatial proximity to other females . For SCI-M , only grooming interactions were used . For each gene , we fit the following model:y=μ+xβ+u+ε , u∼MVN ( 0 , σu2K ) , ε∼MVN ( 0 , σe2I ) , where x is the n by 1 vector of values for the demographic or environmental predictor of interest and β is its coefficient . The n by 1 vector of u is a random effects term with K = XXT/p controlling for additive genetic effects . We calculated the PVE estimate as var ( xβ ) /var ( y ) , where var denotes the sample variance .	Our genes contain the instructions needed to make all aspects of the body . These instructions can be changed by altering the sequence of the DNA that makes up the genes , which can account for many of the different characteristics found in humans and other animals . However , our characteristics can also be altered by changing how often the genes issue their instructions , which is known as gene expression . For example , it is thought that changes in the expression of some genes in primates may account for the expansion of brain sizes over evolutionary time , particularly in the ancestors of modern humans . Most studies into gene expression in primates have compared different species or focused on humans . It is less clear how many , and what type of , genes vary in expression between individuals of the same species in other natural populations . Here , Tung , Zhou et al . used a technique called RNA-sequencing to study gene expression in a population of wild baboons that have been studied for over four decades by the Amboseli Baboon Research Project in Kenya . This involved collecting blood samples from 63 individually recognized adult baboons . After RNA-sequencing , Tung , Zhou et al . were able to identify specific sections of the baboon genome where the DNA sequence an individual baboon carried could predict how highly individual genes were expressed . These sections are known as ‘expression quantitative trait loci’ ( or eQTLs for short ) . Tung , Zhou et al . found that there was a lot of genetically controlled variation in gene expression across the 63 baboons . Most of the eQTLs were found to be in genes that are rapidly evolving or are relatively new . There were fewer eQTLs in genes that are shared across a wide variety of species , possibly because keeping the expression of these genes stable is important for processes that are essential for life . Many of the eQTLs found in the baboons were in genes where eQTLs are also found in humans . This suggests that the set of genes where genetic variation affects gene expression in the baboons may also be a similar set in humans . Tung , Zhou et al . also examined how the age , sex , and social integration of the baboons affected the variation in gene expression observed in the population . They found that for most genes , these factors had only small effects on gene expression levels . However , for some genes , these factors could affect the level of expression throughout the life of the individual . These findings demonstrate that it is feasible to study gene expression patterns in wild primates . The next challenge is to investigate how environmental and genetic factors combine to influence gene expression , and the evolutionary impact of these effects for animals as a whole .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology" ]	2015	The genetic architecture of gene expression levels in wild baboons
Tumor suppressor p53 prevents early death due to cancer development . However , the role of p53 in aging process and longevity has not been well-established . In humans , single nucleotide polymorphism ( SNP ) with either arginine ( R72 ) or proline ( P72 ) at codon 72 influences p53 activity; the P72 allele has a weaker p53 activity and function in tumor suppression . Here , employing a mouse model with knock-in of human TP53 gene carrying codon 72 SNP , we found that despite increased cancer risk , P72 mice that escape tumor development display a longer lifespan than R72 mice . Further , P72 mice have a delayed development of aging-associated phenotypes compared with R72 mice . Mechanistically , P72 mice can better retain the self-renewal function of stem/progenitor cells compared with R72 mice during aging . This study provides direct genetic evidence demonstrating that p53 codon 72 SNP directly impacts aging and longevity , which supports a role of p53 in regulation of longevity . Aging is a complex process of time-dependent series of progressive loss of functions and structures of all systems , which leads to an increased vulnerability to death ( López-Otín et al . , 2013 ) . Cancer is an age-associated disease , which can lead to both premature death and age-associated increase in morbidity and mortality ( Campisi and Yaswen , 2009 ) . Tumor suppressor p53 plays a pivotal role in tumor prevention ( Feng et al . , 2008; Vousden and Prives , 2009 ) . Loss or disruption of p53 function is often a prerequisite for tumor initiation and development . In humans , more than 50% of all human tumors contain mutations in the p53 gene ( Olivier et al . , 2002 ) . In mice , loss of both Trp53 alleles ( p53-/- ) leads to the development of tumors early in life and a reduced lifespan compared with wild type mice ( Donehower et al . , 1992 ) . Therefore , p53 ensures longevity by preventing cancer development early in life . Longevity depends on the balance between tumor suppression and tissue renewal mechanisms ( Campisi and Yaswen , 2009 ) . While genomic instability is a hallmark of aging , stem cell exhaustion is another important hallmark of aging ( López-Otín et al . , 2013 ) . It has been indicated that the anti-proliferative function of p53 which is crucial for tumor suppression could affect self-renewal function of stem/progenitor cells and contribute to aging ( van Heemst et al . , 2005; Donehower , 2002 ) . However , the precise role of p53 in aging process and longevity has not been clearly established . Inconsistent results on aging and longevity have been reported in different mouse models in which the p53 activity has been manipulated through different strategies ( Tyner et al . , 2002; Dumble et al . , 2007; Maier et al . , 2004; Liu et al . , 2010; García-Cao et al . , 2002; Mendrysa et al . , 2006; Matheu et al . , 2007 ) . Specifically , transgenic mice with constitutively elevated p53 activity by expression of certain p53 mutants or a short p53 isoform showed increased cancer resistance but premature aging phenotypes ( Tyner et al . , 2002; Dumble et al . , 2007; Maier et al . , 2004; Liu et al . , 2010 ) . The ‘super p53’ mice with a regulated hyperactive p53 activity by having an extra copy of the wild type Trp53 gene were resistant to cancer but did not exhibit signs of accelerated aging ( García-Cao et al . , 2002; Mendrysa et al . , 2006 ) . Interestingly , the ‘super p53’ mice with an extra copy of Ink4/Arf showed extended longevity ( Matheu et al . , 2007; Matheu et al . , 2009 ) . It is worth noting that mouse models used in these studies did not reflect p53 activation under physiological conditions . It is therefore critical to address the role of p53 in the aging process and longevity using a proper mouse model reflecting the p53 activity under physiological conditions . TP53 is a haplo-insufficient gene , a little decrease in p53 levels or activity ( e . g . 2-fold difference ) significantly impacts tumorigenesis ( Venkatachalam et al . , 2001; Berger and Pandolfi , 2011; Bond et al . , 2004 ) . p53 protein levels and activity are under tight regulation in cells ( Feng et al . , 2008; Vousden and Prives , 2009 ) . In humans , naturally occurring single nucleotide polymorphisms ( SNPs ) in the p53 pathway , which modulate the activity or levels of p53 , have been found to significantly impact cancer risk ( Bond et al . , 2004; Whibley et al . , 2009; Lin et al . , 2008; Basu and Murphy , 2016 ) . p53 codon 72 SNP is a common coding SNP in the TP53 gene , which results in either an arginine ( R72 ) or a proline ( P72 ) residue at codon 72 . We and others have reported that compared with the R72 allele , the P72 allele displays a weaker p53 transcriptional activity towards a group of its target genes , many of which are involved in apoptosis and suppressing cell transformation ( Dumont et al . , 2003; Jeong et al . , 2010 ) . Studies in human populations indicate that p53 codon 72 SNP may modify cancer risk , but currently the consensus has not been reached on this in the literature ( van Heemst et al . , 2005; Whibley et al . , 2009 ) . Several studies of aged or general human populations indicate that the P72 carriers have an increased lifespan despite an increased mortality from cancer ( van Heemst et al . , 2005; Bojesen and Nordestgaard , 2008; Smetannikova et al . , 2004 ) . These epidemiological results support the dual functions of p53 in longevity , and suggest that codon 72 SNP may have an impact upon aging and longevity . Considering the genetic background variations of human populations and environmental factors in epidemiological studies , the precise role of p53 codon 72 SNP in aging and longevity remains elusive . In this study , we employed a mouse model with knock-in of human TP53 gene ( Hupki ) carrying codon 72 SNP to directly investigate the impact of p53 codon 72 SNP upon longevity and its underlying mechanism . The Hupki mice carrying codon 72 SNP recapture the impacts of codon 72 SNP upon p53 transcriptional activity and function in tumor suppression , which is widely used for studies on p53 and codon 72 SNP ( Feng et al . , 2011; Kung et al . , 2016; Azzam et al . , 2011; Reinbold et al . , 2008; Frank et al . , 2011; Leu et al . , 2013 ) . We found that despite the increased cancer risk , P72 mice that have escaped tumor development have a longer lifespan than R72 mice and display a delay of age-associated phenotypes compared with R72 mice . Mechanistically , P72 mice have a better ability to retain the self-renewal function of stem/progenitor cells compared with R72 mice during the aging process . Long-term stem cells from aging P72 mice have better engraftment and repopulation abilities than aging R72 mice . In turn , P72 mice have less expansion of long-term stem/progenitor cells than R72 mice during the aging process . Taken together , our study provides direct genetic evidence demonstrating that human p53 codon 72 SNP has a direct impact upon aging and longevity in vivo , which supports the role of p53 in longevity . To investigate the impact of human p53 codon 72 SNP upon aging and the lifespan , Hupki mice with knock-in of human TP53 gene carrying codon 72 SNP in place of the corresponding mouse Trp53 gene were employed ( Kung et al . , 2016; Frank et al . , 2011; Leu et al . , 2013 ) . It has been reported that p53 protein levels in different tissues are comparable between R72 and P72 mice , which was confirmed in this study ( Figure 1—figure supplement 1A ) ( Kung et al . , 2016; Frank et al . , 2011; Leu et al . , 2013 ) . Previous studies including ours showed that the P72 allele in these mice has a weaker transcriptional activity towards a subset of p53 target genes than the R72 allele , suggesting that these mice retain the function of p53 codon 72 SNP in human ( Feng et al . , 2011; Kung et al . , 2016; Azzam et al . , 2011 ) . Because the lifespan of mice varies among different inbred strains , Hupki mice with p53 codon 72 SNP were backcrossed to mice with different genetic backgrounds , including 129SVsl and C57BL/6J , for ten generations to establish p53 codon 72 SNP Hupki mice in 129SVsl and C57BL/6J backgrounds , respectively . The lifespan of mice with p53 codon 72 SNP in 129SVsl and C57BL/6J backgrounds was measured in a cohort of ~150 mice for each genotype . The median survival age was 740 days in 129SVsl mice and 490 days in C57BL/6J mice , respectively ( Figure 1—figure supplement 1B and C ) , which is consistent with previously reported lifespans of these two mouse strains ( Storer , 1966 ) . In 129SVsl mice , P72 mice showed an overall longer lifespan compared with R72 mice; the median survival age was 759 days for P72 mice and 697 days for R72 mice , respectively ( Log-rank test: p<0 . 0001 ) ( Figure 1A ) . The causes of death included tumor , inflammation ( including dermatitis ) , ocular lesion , urinary syndrome , nephropathy , etc . , which are common causes of death in 129SVsl mice as reported by previous studies ( Marx et al . , 2013; Brayton et al . , 2012; Radaelli et al . , 2016 ) ( Table 1 ) . For those mice died from non-neoplastic events , P72 mice showed a significantly longer lifespan than R72 mice; the median survival was 768 days for P72 mice and 673 days for R72 mice , respectively ( Log-rank test: p<0 . 0001 ) ( Figure 1B ) . For those mice died from neoplastic diseases , R72 mice ( with a median survival of 774 days ) showed a longer lifespan than P72 mice ( with a median survival of 756 days ) ( Log-rank test: p=0 . 015 ) ( Figure 1C ) . Further analysis of mice older than 18 months , which are equivalent to humans older than 60 years ( Dutta and Sengupta , 2016 ) , showed that P72 mice had a longer lifespan ( with a median survival of 780 days ) than R72 mice ( with a median survival of 715 days ) ( Log-rank test: p<0 . 0001 ) ( Figure 1D ) . Similar results were observed in C57BL/6J mice . P72 mice had an overall longer lifespan ( with a median survival of 495 . 5 days ) than R72 mice ( with a median survival of 481 days ) ( Log-rank test: p=0 . 015 ) ( Figure 2A ) . For those mice died from non-neoplastic events , P72 mice had a significantly longer lifespan ( with a median survival of 564 . 5 days ) than R72 mice ( with a median survival of 438 days ) ( Log-rank test: p<0 . 0001 ) ( Figure 2B ) . For those mice died from neoplastic diseases , R72 mice ( with a median survival of 566 days ) had a longer lifespan than P72 mice ( with a median survival of 411 days ) ( Log-rank test: p=0 . 0084 ) ( Figure 2C ) . Further analysis of mice older than 18 months showed that P72 mice had a longer lifespan ( with a median survival of 693 days ) than R72 mice ( with a median survival of 657 days ) ( Log-rank test: p<0 . 0001 ) ( Figure 2D ) . Our results that mice carrying different p53 codon 72 SNP have different lifespans suggest that p53 codon 72 SNP impacts the aging process . Therefore , several aging-associated phenotypes were examined in R72 and P72 mice at different ages . During the aging process , mice develop lordokyphosis which is characterized by an increased curvature of the spine ( López-Otín et al . , 2013 ) . In this study , skeleton structures of 129SVsl mice at different ages were imaged and reconstructed by a micro-CT scan . A narrowing of the spine angle indicates an increase in lordokyphosis . In 6-month-old mice , lordokyphosis was not observed , and there was no significant difference in spinal curves between R72 and P72 mice ( Figure 3A&B ) . In 18-month-old mice , lordokyphosis was observed . Notably , 18-month-old R72 mice developed more pronounced lordokyphosis compared with age-matched P72 mice ( Figure 3B ) . A similar phenotype was observed in C57BL/6J mice ( Figure 3G&H ) . Another aging-related phenotype in both humans and mice is osteoporosis ( López-Otín et al . , 2013 ) . The mouse tibias bone structure and density were examined by a micro-CT scan followed by 3D reconstruction . The structure and density of the tibias bone between 6-month-old 129SVsl R72 and P72 mice were morphologically identical , and showed no sign of osteoporosis ( Figure 3C ) . Osteoporosis was observed in both R72 and P72 mice at the age of 18 months . Notably , R72 mice displayed a more obvious sign of osteoporosis than P72 mice ( Figure 3C ) . Analysis of tibias bone structure and density of mice at different ages showed aging-related changes , including decreased bone volume/total volume ( BV/TV ) , decreased trabecular number and increased trabecular spacing during aging ( Figure 3D–F ) . P72 mice showed a delayed development of all these aging-related changes compared with R72 mice , with the most obvious differences observed at the age of 18 months ( Figure 3D–F ) . Similar results were obtained in C57BL/6J mice ( Figure 3I–K ) . Decreases in the skin dermal thickness and subcutaneous adipose tissues occur during the aging process ( López-Otín et al . , 2013 ) . Older 129SVsl mice ( 12–18 month-old ) had a thinner dermal layer and less subcutaneous adipose tissues than young mice ( 6-month-old ) ( Figure 4A–C ) . There was no obvious difference in the dermal thickness and the amount of subcutaneous adipose tissues between young R72 and P72 mice . Notably , in older mice , R72 mice showed more significant decreases in both skin dermal thickness and subcutaneous adipose thickness compared with P72 mice ( Figure 4A–C ) . One of the hallmarks of aging is the stem cell exhaustion , which leads to the reduced ability of tissue repair ( López-Otín et al . , 2013 ) . Under stress , such as skin wounds , epidermal stem cells exhibit a highly organized and complex self-renewal process to restore the integrity and function of the skin . This ability dampens down as both humans and mice age ( López-Otín et al . , 2013 ) . Therefore , the cutaneous repair ability of mice was examined by measuring the wound healing process which reflects the function of the skin stem cell ( Shaw and Martin , 2009 ) . Three-mm wounds were introduced in the mouse skin by punch and the wound diameters were measured daily . Both R72 and P72 129SVsl young mice ( 6-month-old ) showed an efficient wound healing ability ( Figure 4D ) . However , 12- and 18-month-old R72 mice showed a more pronounced decrease in the wound healing ability than age-matched P72 mice ( Figure 4D ) . Similar results were obtained in C57BL/6J mice ( Figure 4E ) . These results demonstrate that P72 mice exhibited a delayed aging process compared with R72 mice . Stem cell exhaustion is considered as a hallmark of the aging process ( López-Otín et al . , 2013 ) . During aging in both humans and mice , the regeneration ability of stem cells gradually diminishes ( López-Otín et al . , 2013 ) . Ample studies on stem cell aging process have focused on hematopoietic stem cell ( HSC ) ( Chambers and Goodell , 2007; Seita and Weissman , 2010 ) . Studies using mouse models demonstrated a HSC aging phenotype with the characteristic of the increase of the pool of stem/progenitor cells and the reduction of their self-renewal abilities during the aging process ( Dumble et al . , 2007; Chambers et al . , 2007 ) . p53 has been indicated to play a critical role in regulating the function of stem/progenitor cells ( Dumble et al . , 2007; Kaiser and Attardi , 2018 ) . Here , we investigated the impact of p53 codon 72 SNP upon HSC pool size and self-renewal function during aging . To this end , the numbers of long term-HSCs ( LT-HSCs ) as well as proliferating HSCs , which represent HSC pool size and self-renewal function , respectively , were measured in R72 and P72 mice at different ages . Bone marrow cells were isolated from mouse hind limb bones and stained with mature hematopoietic lineage markers . The numbers of LT-HSCs ( Lin-/low , Sca1+ , c-kit+ and CD34- , Flk2- ) were determined by FCM analysis ( Figure 5—figure supplement 1 ) . Consistent with previous reports ( Akunuru et al . , 2016; Morrison et al . , 1996 ) , the percentage of LT-HSCs in bone marrow cells clearly increased during the aging process in both 129SVsl and C57BL/6J mice ( Figure 5A–C ) . R72 mice showed a more rapid increase in the numbers of LT-HSCs than P72 mice during aging . While there was no significant difference in LT-HSC numbers between young 129SVsl R72 and P72 mice , much higher LT-HSC numbers were observed in R72 mice than P72 mice at the age of both 12 and 18 months ( Figure 5A&B ) . Similar results were obtained in C57BL/6J mice ( Figure 5C ) . These results demonstrated that P72 mice showed a delayed HSC expansion during aging . To determine the population of functional/proliferating HSCs in R72 and P72 mice at different ages , BrdU-labeled proliferating HSCs were quantified by FCM analysis . As shown in Figure 5D&E , the number of proliferating HSCs decreased during aging in both 129SVsl and C57BL/6J mice , which is consistent with previous reports ( Dumble et al . , 2007; Chambers and Goodell , 2007 ) . Notably , the decrease of proliferating HSC numbers was more rapid in R72 mice than P72 mice during aging . In 129SVsl mice , the percentage of proliferating HSCs in all HSCs in R72 mice decreased from 45% at the age of 6 months to 28% at the age of 22 months , whereas the decrease of proliferating HSCs in P72 mice was less pronounced: from 44% at the age of 6 months to 34% at the age of 22 months ( Figure 5D ) . Similar results were obtained in C57BL/6J mice; the decrease of proliferation HSCs was more rapid in R72 mice than P72 mice during aging ( Figure 5E ) . To further evaluate the self-renewal and repopulation function of HSCs in mice with p53 codon 72 SNP during aging , the engraftment and repopulation abilities of HSCs of mice were determined by bone marrow transplantation assays . Bone marrow cells isolated from donor CD45 . 2 mice with different p53 codon 72 SNP at different ages were transplanted into lethally irradiated recipient CD45 . 1 mice along with CD45 . 1 bone marrow cells ( Figure 6A ) . The long-term HSC engraftment and repopulation abilities were evaluated by analyzing CD45 . 1 or CD45 . 2 cell surface markers of peripheral blood cells at 16 weeks after transplantation . As shown in Figure 6B , while bone marrow cells from 6-month-old R72 and P72 mice showed similar abilities in engraftment and contribution to mature peripheral lymphocytes , bone marrow cells from 18-month-old P72 mice showed a significantly higher engraftment ability than R72 mice . For donors from 129SVsl mice , ~76% vs . ~68% of lymphocytes were derived from 18-month-old P72 and R72 donors , respectively . Similar results were obtained in C57BL/6J mice;~74% vs ~56% of lymphocytes were derived from 18-month-old P72 and R72 donors , respectively ( Figure 6B ) . Taken together , these results demonstrated that P72 mice displayed a delayed aging process in HSC number and function compared with R72 mice , which contributes to the delayed aging phenotypes in P72 mice . In response to stress , p53 transcriptionally regulates a group of target genes that can lead to different cell fates through inducing growth arrest or apoptosis , etc . Here , 129SVsl mice were employed to examine whether p53 codon 72 SNP differentially regulates the basal expression of its target genes involved in cell cycle arrest ( p21 ) and apoptosis ( Puma and Noxa ) , which in turn impacts the number and function of stem cells . As shown in Figure 6C , p21 mRNA expression levels in the bone marrow were slightly higher in P72 mice compared with R72 mice as determined by real-time PCR assays with this difference being more obvious in older mice than young mice . This difference in p21 expression levels was confirmed at the protein level as determined by Western-blot assays ( Figure 6C ) . In contrast , the bone marrow from P72 mice displayed slightly lower expression levels of Puma and Noxa than that from R72 mice ( Figure 6C ) . These results demonstrate the differential regulation of the basal expression levels of p21 , Puma and Noxa by p53 codon 72 SNP in the bone marrow , which may contribute to the delayed aging process in HSC number and function observed in P72 mice . While the role of p53 in assuring longevity through prevention of early cancer development has been well established , its role in regulating aging and longevity aside from cancer prevention has not been well established . Divergent results have been obtained from different mouse models in which the p53 activity was manipulated through different strategies . The increased p53 activity was reported to lead to accelerated aging in some mouse models , but do not affect the lifespan or even prolong the life span in other mouse models ( Tyner et al . , 2002; Dumble et al . , 2007; Maier et al . , 2004; Liu et al . , 2010; García-Cao et al . , 2002; Mendrysa et al . , 2006; Matheu et al . , 2007 ) . These results indicate that p53 can be pro-aging or pro-longevity depending on the context of its regulation and activity . The precise role of p53 in intrinsic aging process , especially under the physiological condition , remains unclear . Longevity depends on a balance between tumor suppression and tissue renewal mechanisms ( López-Otín et al . , 2013; Campisi , 2003a ) . Declines in stem cells self-renewal and differentiation are critical components of aging ( Campisi and Yaswen , 2009 ) . The anti-proliferative function of p53 , which is crucial for suppression of cancer cells , plays a crucial role in eliminating damaged cells including stem cells ( TeKippe et al . , 2003; Shounan et al . , 1996 ) . The pleiotropic antagonism theory suggests that certain cellular processes that provide beneficial effects in youth , may compromise organismal fitness later in life ( Campisi , 2003b ) . Currently , it is unclear whether p53 has the antagonistic pleiotropy and how the balance of p53 for tumor surveillance and stem cell regulation is regulated . In humans , functional SNPs have been identified in both p53 and its signaling pathway , such as p53 codon 72 SNP and SNP309 in p53 negative regulator MDM2 . These SNPs alter the levels and/or function of p53 . Some of these SNPs , including p53 codon 72 SNP , appear to have undergone the natural selection , which suggests that p53 has evolutionarily-conserved functions other than tumor suppression ( Atwal et al . , 2007; Atwal et al . , 2009 ) . It is possible that these SNPs modulate the function of p53 in maintaining the balance between tumor surveillance and stem cell regulation , which are important genetic modifiers for human longevity . Up till now , majority studies on p53 codon 72 SNP have focused on its impact upon cancer risk . However , there is no consensus in the literature as to the impact of p53 codon 72 SNP upon cancer risk ( van Heemst et al . , 2005; Whibley et al . , 2009 ) . Further , the role of p53 codon 72 SNP in stem cell regulation and aging process remains unclear . Several epidemiological studies of human populations indicate that p53 codon 72 SNP may influence human longevity . A perspective study with an aging human population ( ≥85 years , n = 1226 ) reported that individuals homozygous for the P72 allele have a 41% increased survival ( p=0 . 032 ) despite a 2 . 54-fold increased mortality from cancer ( van Heemst et al . , 2005 ) . Another perspective study of the Danish general population ( ages 20–95 , n = 9219 ) reported that the p53 P72 allele is associated with an increased overall survival rate ( Bojesen and Nordestgaard , 2008 ) . Similarly , a study of long-lived individuals in Novosibirsk and Tyumen Regions ( n = 131 ) reported the enrichment of the p53 P72 allele in the long-lived group ( Smetannikova et al . , 2004 ) . These findings suggest that p53 activity is reversely associated with aging , and p53 codon 72 SNP may impact the lifespan in humans . In this study , we used a genetic approach to investigate whether p53 codon 72 SNP modulates longevity through regulating the balance of p53 functions in tumor surveillance and stem cell regulation by using Hupki mice carrying p53 codon 72 SNP . To exclude the effect of mixed mouse genetic backgrounds on the lifespan and the aging process , mice carrying p53 codon 72 SNP were backcrossed with 129SVsl mice and C57BL/6J mice , respectively , for 10 generations . As shown in Figure 1—figure supplement 1B and C and Table 1 , the lifespan and causes of death of these two mouse strains are very different . p53 codon 72 SNP showed a clear impact upon the lifespan in both strains of mice; P72 mice have a longer lifespan compared with their R72 littermates , although P72 mice have a higher risk for tumor development . This result is consistent with the observation in human populations showing that P72 carriers have a longer lifespan ( van Heemst et al . , 2005; Bojesen and Nordestgaard , 2008; Smetannikova et al . , 2004 ) . Further , P72 mice display delayed aging-associated phenotypes compared with R72 mice . Results from this study further showed that P72 mice have a better self-renewal ability of stem cells and a delay of compensatory expansion of the stem cell pool compared with R72 mice . Stem cells from older P72 mice have better engraftment and repopulation ability compared with older R72 mice as determined by the bone marrow transplantation assays ( Figure 6D ) . It has been suggested that p53 codon 72 SNP can influence the basal expression levels of some p53 target genes in humans , including p21 and PAI-1 ( Salvioli et al . , 2005 , Testa et al . , 2009 ) . For instance , it was reported that dermal fibroblasts from P72 carriers display a higher expression of p21 ( Salvioli et al . , 2005 ) . In plasma samples from healthy populations , the P72 allele plays an important role in determining PAI-1 levels in aging populations ( Testa et al . , 2009 ) . Results from this study showed that the bone marrow from P72 mice displays higher expression levels of p21 but lower expression levels of Puma and Noxa compared with R72 mice . The differential expression of these target genes may modulate the p53 decision on cell fates towards survival or death , which may contribute to the delayed aging process in HSC function observed in P72 mice . In addition to p53 codon 72 , a group of functional SNPs have been identified in the p53 gene and important genes in the p53 pathway , such as MDM2 . A very recent study reported that a patient affected by a segmental progeroid syndrome has a germline mutation in the MDM2 gene ( Lessel et al . , 2017 ) . This mutation abrogates MDM2 function and leads to increased p53 levels and function , which might be the driving cause for the premature aging phenotype of this patient ( Lessel et al . , 2017 ) . It will be of interest to study how these SNPs in the p53 pathway ( individual or in combination ) impact aging and longevity in future studies . Taken together , results from this study provided the genetic evidence showing that functional p53 codon 72 SNP , which regulates the activity of p53 , influences aging and longevity through the regulation of self-renewal function of stem cells . Results from this study strongly support a role of p53 in regulation of stem/progenitor cell function and longevity . Hupki mice carrying either the P72 or R72 allele were generous gifts from Dr . Maureen Murphy ( The Wistar Institute ) ( Kung et al . , 2016 ) . Hupki mice in 129SVsl and C57BL/6J backgrounds were produced by backcrossing Hupki mice ten times to 129SVsl and C57BL/6J , respectively . C57BL/6J CD45 . 1 mice ( RRID:IMSR_JAX:002014 ) were purchased from The Jackson Laboratory ( Bar Harbor , ME ) . All animal experiments were approved by the IACUC committee of Rutgers University . Mice were anesthetized for CT scanning of whole body skeletons using the INVEON PET/CT system ( Siemens Healthcare ) . The images were reconstructed using INVEON Research Workplace software ( Siemens Healthcare , Tarrytown , NY ) . The bone microstructure measurement was carried out as previously described ( Ell et al . , 2013 ) . In brief , mouse tibias were scanned by micro-CT . The images were reconstructed with Beam Hardening Correction and Hounsfield calibrated before being analyzed using INVEON Research Workplace software . The 3D images were generated corresponding to the trabecular bone regions . CT scans were carried out at the Preclinical Imaging Shared Resource of Rutgers Cancer Institute of New Jersey . Paraffin-embedded skin specimens were sectioned with 5 µm thickness and stained with hematoxylin and eosin ( H and E ) . The thickness of the dermal and adipose layers from the skin samples were determined by taking three random measurements along the length of each skin sample using ImageJ software . Cutaneous wound healing assays were carried out as previously described ( Tyner et al . , 2002 ) . In brief , mice were anesthetized and a full-thickness wound was generated in the mouse dorsal skin using a 3 mm biopsy punch ( Integra , York , PA ) . Wound diameters were measured daily . Wound areas = 0 . 25 × π ×width × length . LT-HSC numbers were determined as previously described ( Dumble et al . , 2007 ) . In brief , bone marrow cells were flushed out from mouse hind limb bones with PBS and stained with a cocktail of antibodies ( BD bioscience Pharmingen ) , including an anti-lineage-APC antibody ( RRID:AB_1645213 ) , an anti-Sca-1-PE-Cy7 antibody ( RRID:AB_647253 ) , an anti-c-kit-PE-CF594 antibody ( RRID:AB_11154233 ) , an anti-CD34-FITC antibody ( RRID:AB_395017 ) and an anti-Flk-2-PE antibody ( RRID:AB_395079 ) . LT-HSCs which were selected as Lin-/low , Sca1+ , c-kit+ and CD34- , Flk2- cells were quantified by FCM analysis using a Beckman-Coulter Cytomics FC500 Flow Cytometer ( Indianapolis , IN ) . To determine the HSC proliferation ability , mice were injected intraperitoneally with 1 mg BrdU ( BD bioscience Pharmingen ) at 16 hr before the collection of bone marrow . Bone marrow cells were stained with a cocktail of antibodies , including an anti-lineage-APC antibody , an anti-Sca-1-PE-Cy7 antibody , an anti-c-kit-PE-CF594 antibody and an anti-Flk-2-PE antibody . After cell surface staining , a BrdU-FITC Flow Kit ( BD bioscience Pharmingen; RRID:AB_2617060 ) was used to identify cycling cells according to the manufacturer’s instructions . Proliferating HSCs were identified as Lin-/low , Sca1+ , c-kit+ , Flk2- and BrdU+ by FCM analysis . Bone marrow transplantation assays were carried out as previously described ( Dumble et al . , 2007 ) . In brief , bone marrow cells from 6-month-old and 18-month-old ‘donor’ CD45 . 2 mice were mixed with bone marrow cells from 6-month-old ‘competitor’ CD45 . 1 mice at a ratio of 2:1 . Recipient CD45 . 1 mice at the age of 6 to 12 week-old were irradiated with a lethal dose of 10 Gy the day before bone marrow transplantation . n = 6/group . The mixture of bone marrow cells was injected into recipient mice via the tail vein . Sixteen weeks after transplantation , peripheral white blood cells of recipient mice were analyzed for CD45 . 1 and CD45 . 2 cell surface markers using an anti-CD45 . 1-PE antibody ( RRID:AB_395044 ) and an anti-CD45 . 2-FITC antibody ( RRID:AB_395041 ) , respectively ( BD Biosciences Pharmingen ) . Standard Western-blot assays were used to analyze protein expression in tissues . The following antibodies were used for assays: anti-p53 ( FL393 , Santa Cruz Biotechnology; RRID; AB_653753 ) , anti-p21 ( Santa Cruz Biotechnology; RRID:AB_628073 ) , and β-actin ( Sigma; RRID:AB_476744 ) . Total RNA was prepared by using an RNeasy kit ( Qiagen ) . All probes were purchased from Applied Biosystems . Real-time PCR was done in triplicate with TaqMan PCR mixture ( Applied Biosystems ) . The expression of genes was normalized to the β-actin gene . The data were present as mean ± SD . The lifespan of mice were summarized by Kaplan-Meier plots and compared using the log-rank test using GraphPad Prism software . All other p values were obtained using the Student’s t-test . Based on survival data of p53 codon 72 SNP mice , we hypothesized that the P72 mice have a delayed development of aging associated phenotypes . Therefore , one-tailed Student’s t-test was used for majority of data analysis related to the development of aging associated phenotypes . Values of p<0 . 05 were considered to be significant .	How long most animals live depends on the balance between the biological processes that allow them to regenerate their tissues when damaged and those that prevent them from developing cancer . Regeneration relies mostly on cells , in particular stem cells , dividing to make new cells , while cancer occurs when cell division becomes uncontrolled . Tumor suppressor genes protect against cancer . One such gene encodes a protein called p53 that eliminates damaged cells before they can become cancerous . The p53 protein is also believed to be involved in regulating how quickly an animal ages and how long it lives , but this second role has not yet been clearly established . Previous studies using different strategies to change the activity of p53 in several mouse models have led to inconsistent results . However , the mouse models used in these earlier studies did not reflect how p53 works under normal conditions . Zhao et al . have now used mice in which the mouse gene for p53 was replaced with one of two versions of the equivalent human gene to study its impact on lifespan and the aging process . The two versions of p53 only differ slightly; a single building block of the protein , the amino acid at position 72 , is a proline in one version but an arginine in the other . This difference makes one version of p53 weaker than the other; in other words , it is less able to eliminate damaged cells . Zhao et al . revealed that the mice with the weaker p53 lived for longer and appeared to age more slowly too . Further experiments showed that the stem cells in the mice with a weaker p53 were able to keep dividing and create new cells for longer . This is important because a decline in this activity – which is known as self-renewal – is a hallmark of aging . Together these findings show that a small yet common change in p53 impacts both aging and lifespan , possibly by altering how stem cells are regulated . Further work is now needed to better understand why the different versions of p53 have different effects on stem cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cancer", "biology" ]	2018	A polymorphism in the tumor suppressor p53 affects aging and longevity in mouse models
PAK1 inhibitors are known to markedly improve social and cognitive function in several animal models of brain disorders , including autism , but the underlying mechanisms remain elusive . We show here that disruption of PAK1 in mice suppresses inhibitory neurotransmission through an increase in tonic , but not phasic , secretion of endocannabinoids ( eCB ) . Consistently , we found elevated levels of anandamide ( AEA ) , but not 2-arachidonoylglycerol ( 2-AG ) following PAK1 disruption . This increased tonic AEA signaling is mediated by reduced cyclooxygenase-2 ( COX-2 ) , and COX-2 inhibitors recapitulate the effect of PAK1 deletion on GABAergic transmission in a CB1 receptor-dependent manner . These results establish a novel signaling process whereby PAK1 upregulates COX-2 , reduces AEA and restricts tonic eCB-mediated processes . Because PAK1 and eCB are both critically involved in many other organ systems in addition to the brain , our findings may provide a unified mechanism by which PAK1 regulates these systems and their dysfunctions including cancers , inflammations and allergies . It is generally accepted that normal brain function is dependent upon a balance of excitation and inhibition ( i . e . balanced E/I ratio ) and that altered E/I ratios are associated with , and thought to cause , a wide range of neurological and mental disorders , including autism and schizophrenia ( Eichler and Meier , 2008; Kehrer et al . , 2008; Marín , 2012; Yizhar et al . , 2011 ) . Recent studies indicate that GABA-mediated synaptic inhibition is particularly important to maintain an appropriate E/I ratio and that inhibitory postsynaptic currents ( IPSCs ) mediated by GABA receptors are frequently altered in various brain diseases and their animal models ( Braat and Kooy , 2015; Lewis et al . , 2005 ) . Although subjected to multiple regulations , GABA transmission is strongly inhibited by endocannabinoids ( eCBs ) , a group of neuromodulatory lipids known to affect a wide range of physiological processes and medical conditions ( Katona and Freund , 2012; Morena et al . , 2016; Piomelli , 2003 ) . In the brain , eCBs are produced and secreted from postsynaptic neurons and activate presynaptic cannabinoid 1 ( CB1 ) receptors to reduce the release of a multitude of neurotransmitters , including GABA ( Katona and Freund , 2012; Morena et al . , 2016; Piomelli , 2003 ) . Two types of eCB-mediated suppression of GABA release have been studied: tonic eCB release that regulates basal synaptic transmission , and phasic eCB release , induced by postsynaptic depolarization or receptor-mediated eCB production , which mediates transient decreases in synaptic transmission during short-term plasticity ( Katona and Freund , 2012; Morena et al . , 2016 ) . Although the metabolic process of the eCBs and the enzymes involved in their regulation have been a focus of extensive research ( Katona and Freund , 2012; Morena et al . , 2016; Piomelli , 2003 ) , cellular signaling mechanisms that regulate eCB signaling , particularly tonic eCB signaling , remain poorly understood . p21-activated kinases ( PAKs ) are a family of serine/threonine protein kinases that are activated by multiple synaptic proteins including Ras and Rho GTPases . Extensive studies have indicated that PAKs are involved in a number of cellular processes , particularly in the regulation of gene expression and cellular cytoskeleton ( Bokoch , 2003; Zhao and Manser , 2012 ) . Accordingly , changes in PAKs are found to be associated with a wide range of physiological and pathological conditions including various forms of cancers and PAK inhibitors are being actively exploited as therapeutic agents to treat these diseases ( Kelly and Chernoff , 2012; Kumar et al . , 2006 ) . Recent human studies have also revealed that PAKs are linked to a number of devastating neurological and mental disorders including autism , intellectual disability , Huntingtin’s diseases and Alzheimer’s diseases ( Gilman et al . , 2011; Ma et al . , 2012 ) . Animal studies have indeed shown that PAKs , particularly PAK1 , the predominant member of the PAK family expressed in the brain , are involved in the regulation of excitatory synaptic function , including spine structure , synaptic plasticity and memory formation ( Asrar et al . , 2009; Hayashi et al . , 2004; Huang et al . , 2011; Meng et al . , 2005 ) . Most remarkably , more recent studies demonstrate that inhibition of PAK1 , either genetically or pharmacologically , can ameliorate the cognitive and social deficits in several animal models of neurodevelopmental disorders , particularly autism , including genetic models targeting fragile X syndrome and neurofibromatosis ( Dolan et al . , 2013; Hayashi et al . , 2007; Molosh et al . , 2014 ) . However , the mechanism by which PAK1 exerts such diverse therapeutic effects remains elusive . Quite interestingly , many of these same animal models of neurodevelopmental disorders also exhibit pronounced alterations in eCB signaling ( Földy et al . , 2013; Jung et al . , 2012 ) , and several reports have now suggested that , like PAK1 , targeting eCB signaling may provide benefit in these conditions ( Busquets-Garcia et al . , 2013; Qin et al . , 2015 ) . As these animal models share a common deficit in E/I balance , which appear to involve critical roles of both PAK1 and eCB signaling , we have hypothesized that PAK1 might be a critical player in the regulation of E/I homeostasis through an interaction with eCB signaling . Consistent with this hypothesis , our data indicate that PAK1 restricts tonic eCB signaling in the hippocampus through the regulation of synaptosomal cyclooxygenase-2 ( COX-2 ) expression , a non-canonical but relevant pathway in the metabolism of eCB signaling . In turn , this ability of PAK1 to restrict tonic eCB signaling confers an alteration in the E/I homeostasis of the hippocampus through the regulation of tonic GABA transmission . Given the overlapping importance of PAK1 , COX-2 and eCB signaling in an array of physiological and pathophysiological processes , the identification of this functional signaling interaction likely has significant implications for a multitude of disease processes , such as autism , inflammatory conditions and cancer . Since all the animal models of brain disorders that are functionally rescued by manipulations of PAK1 share a common deficit in E/I balance ( Braat and Kooy , 2015; Dolan et al . , 2013; Eichler and Meier , 2008; Gao and Penzes , 2015; Hayashi et al . , 2007; Kehrer et al . , 2008; Lewis et al . , 2005; Marín , 2012; Molosh et al . , 2014; Yizhar et al . , 2011 ) , we therefore examined whether disrupting PAK1 would affect the E/I ratio by performing whole cell patch-clamp recordings in CA1 pyramidal neurons of hippocampal slices acutely prepared from PAK1 KO mice and their WT littermates ( Figure 1a ) . Excitatory and inhibitory postsynaptic currents ( EPSCs and IPSCs ) were pharmacologically isolated by using respective inhibitors specific to glutamate or GABA receptors ( i . e . EPSC recorded by including 100 μM GABAα receptor antagonist picrotoxin and IPSC by including 10 μM AMPAR antagonist NBQX plus 50 μM NMDAR antagonist D-APV ) . First , we measured the E/I ratio by sequentially recording evoked synaptic responses ( Figure 1b ) , first in the absence of any inhibitors to obtain total synaptic currents ( i . e . eEPSC+eIPSC ) , then in the presence of NBQX/APV to obtain eIPSC , and finally in the presence of NBQX/APV/picrotoxin to verify the eIPSC component . As shown in Figure 1c , the E/I ratio was significantly increased in PAK1 KO compared to the WT littermates . Because PAK1 KO mice show no deficits in basal excitatory synaptic strength ( Asrar et al . , 2009 ) , the increased E/I ratio in the KO mice is likely due to impaired inhibitory transmission . To test this possibility directly , we performed input/output experiments of eIPSC and as shown in Figure 1d , the amplitude of eIPSC was significantly smaller in the KO compared to the WT control over a wide range of stimulus intensities . To exclude the possibility that the KO mice may have suffered developmental compensations that could contribute to the reduced eIPSC , we tested the effect of the group1 PAK inhibitor IPA3 ( 10 μM ) ( Rudolph et al . , 2013 ) . As shown in Figure 1e , bath application of IPA3 caused a rapid and significant decrease in the amplitude of eIPSC in WT , but not in PAK1 KO neurons . To further corroborate this result , we employed an independent short peptide known to specifically inhibit PAK1 ( Shin et al . , 2013 ) . As shown in Figure 1f , inclusion of this peptide in the postsynaptic neurons ( 20 μg/ml ) also significantly decreased the amplitude of eIPSC in WT , but not in PAK1 KO neurons . The fast acting nature of these inhibitors ( within minutes ) indicate that the effect of PAK1 on eIPSC is not likely due to a developmental effect but rather direct involvement of PAK1 at the synapse . These results also indicate that PAK1 disruption specifically in the postsynaptic CA1 neurons is sufficient to cause impaired inhibitory synaptic transmission . 10 . 7554/eLife . 14653 . 003Figure 1 . Genetic ablation of PAK1 enhances E/I ratio by selectively suppressing inhibitory synaptic responses . ( a ) Diagram of a hippocampal slice showing the placement of stimulating and recoding electrodes . ( b ) A representative whole-cell recording experiment and samples traces at indicated time points showing the time course of evoked synaptic currents in the absence or presence of various inhibitors to determine the E/I ratio . Scale bar: 100 pA/25 ms . ( c ) Left: sample traces of various components of synaptic currents . Right: summary data showing an increased E/I ratio in PAK1 KO compared to WT control ( WT = 3 . 92 ± 0 . 39 , n = 8 ( 5 ) ; KO = 7 . 39 ± 1 . 13 , n = 6 ( 4 ) ; *p=0 . 007; t-test ) . Scale bar: 100 pA/25 ms . ( d ) Whole-cell recordings of input-output curves showing significantly reduced amplitude of evoked IPSC ( eIPSC ) in PAK1 KO compared to WT neurons ( genotype: F ( 1 , 27 ) = 5 . 946 , p=0 . 022; stimuli: F ( 7 , 189 ) = 70 . 983 , **p<0 . 001; repeated measures two-way ANOVA [also see Figure 1—source data 1]; at 1 . 5 mA stimulus: WT = 453 . 39 ± 49 . 41 pA , n = 15 ( 5 ) ; KO = 304 . 08 ± 46 . 54 pA , n = 14 ( 5 ) ; p=0 . 037; t-test ) . Scale bar: 125 pA/25 ms . ( e ) Whole-cell recordings of eIPSC showing that bath application of IPA3 caused a rapid decrease in eIPSC amplitude in WT , but not in PAK1 KO neuron ( genotype: F ( 1 , 10 ) = 5 . 615 , p=0 . 039; time: F ( 3 , 30 ) =16 . 332 , p<0 . 001; repeated measures two-way ANOVA [also see Figure 1—source data 2]; at 21–30 min post IPA3 perfusion: WT = 61 . 84 ± 4 . 01% , n = 7 ( 4 ) ; KO = 88 . 65 ± 6 . 08% , n = 5 ( 3 ) ; p=0 . 003; t-test ) . Scale bar: 20 pA/25 ms . ( f ) Whole-cell recordings of eIPSC showing that intracellular infusion of the PAK1 inhibitory peptide specifically in the postsynaptic neurons caused a rapid decrease in eIPSC amplitude in WT , but not in PAK1 KO neurons ( genotype: F ( 1 , 9 ) = 62 . 66 , *p<0 . 001; time: F ( 2 , 18 ) = 30 . 720 , p<0 . 001; repeated measures two-way ANOVA [also see Figure 1—source data 3]; at 21–30 min after whole-cell break-in: WT = 47 . 07 ± 6 . 63% , n = 5 ( 4 ) ; KO = 96 . 17 ± 1 . 67% , n = 6 ( 5 ) ;*p<0 . 001; t-test ) . Scale bar: 40 pA/25 ms . DOI: http://dx . doi . org/10 . 7554/eLife . 14653 . 00310 . 7554/eLife . 14653 . 004Figure 1—source data 1 . Statistical data summary for Figure 1d: input/output curves of eIPSC using repeated measures two-way ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 14653 . 00410 . 7554/eLife . 14653 . 005Figure 1—source data 2 . Statistical data summary for Figure 1e: IPA3 effect on eIPSC using repeated measures two-way ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 14653 . 00510 . 7554/eLife . 14653 . 006Figure 1—source data 3 . Statistical data summary for Figure 1f: PAK1 inhibitory peptide effect on eIPSC using repeated measures two-way ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 14653 . 006 To investigate whether the reduced inhibitory transmission is caused by pre- and/or postsynaptic changes , we recorded spontaneous IPSC ( sIPSC ) and miniature IPSC ( mIPSC ) . As shown in Figure 2a–f , the frequency of sIPSC ( Figure 2c ) and mIPSC ( Figure 2f ) was significantly reduced in PAK1 KO compared to WT neurons . The amplitude of sIPSC ( Figure 2b ) and mIPSC ( Figure 2e ) were not altered in the KO mice . Consistent with previous results ( Asrar et al . , 2009 ) , neither the frequency nor the amplitude of spontaneous or miniature excitatory synaptic currents ( sEPSC or mEPSC ) was altered in PAK1 KO mice ( Figure 2g–i ) . To test whether these changes were specific to PAK1 KO mice , we analyzed KO mice lacking ROCK2 ( Zhou et al . , 2009 ) , a closely related kinase also activated by the Rho family small GTPases , but found no significant changes in any of these parameters in these mice ( Figure 2—figure supplement 1 ) . These results suggest that the release property at the inhibitory synapse is selectively impaired in PAK1 KO neurons . To investigate this further , we examined transmitter release induced by sustained high frequency stimulations . As shown in Figure 2—figure supplement 2 , synaptic depression induced by 3 min of 5 Hz stimulations was significantly slower in PAK1 KO mice compared to the WT littermates . These results indicate that PAK1 regulates inhibitory synaptic transmission likely through a presynaptic mechanism . 10 . 7554/eLife . 14653 . 007Figure 2 . PAK1 deletion specifically reduces the frequency , but not the amplitude of inhibitory synaptic responses . ( a ) Sample traces of sIPSC recordings . ( b , c ) Summary graphs of ( a ) showing normal distribution and mean value of the amplitude ( b: WT = 19 . 07 ± 2 . 44 pA , n = 13 ( 5 ) ; KO = 16 . 44 ± 1 . 62 pA , n = 20 ( 6 ) ; p=0 . 35 ) , but decreased frequency ( c: WT = 1 . 50 ± 0 . 17 Hz , n = 13 ( 7 ) ; KO 1 . 03 ± 0 . 11 Hz , n = 20 ( 6 ) ; p=0 . 021 ) of sIPSCs in PAK1 KO compared to WT control . ( d ) Sample traces of mIPSC recordings . ( e , f ) Summary graphs of ( d ) showing normal distribution and mean value of the amplitudes ( e: WT = 10 . 69 ± 0 . 75 pA , n = 10 ( 4 ) ; KO = 10 . 40 ± 0 . 64 pA , n = 16 ( 4 ) ; p=0 . 773 ) , but decreased frequency ( f: WT = 0 . 91 ± 0 . 12 Hz , n = 10 ( 4 ) ; KO = 0 . 60 ± 0 . 08 Hz , n = 16 ( 4 ) ; p=0 . 034 ) of mIPSCs in PAK1 KO compared to WT control . ( g ) Sample traces of sEPSC recordings . ( h , i ) Summary graphs of ( g ) showing normal amplitude ( h: WT = 12 . 37 ± 0 . 30 pA , n = 20 ( 5 ) ; KO = 12 . 43 ± 0 . 46 pA , n = 14 ( 5 ) ; p=0 . 897 ) and frequency ( i: WT = 0 . 43 ± 0 . 08 Hz , n = 20 ( 5 ) ; KO = 0 . 30 ± 0 . 06 Hz , n = 14 ( 5 ) ; p=0 . 272 ) of sEPSCs in PAK1 KO compared to WT control . ( j ) Sample traces of mEPSC recordings . ( k , l ) Summary graphs of ( j ) showing normal amplitude ( k: WT = 9 . 79 ± 0 . 06 pA , n = 7 ( 3 ) ; KO = 10 . 30 ± 0 . 06 pA , n = 9 ( 3 ) ; p=0 . 545 ) and frequency ( l: WT = 0 . 36 ± 0 . 05 Hz , n = 7 ( 3 ) ; KO = 0 . 33 ± 0 . 07 Hz , n = 9 ( 3 ) ; p=0 . 806 ) of mEPSCs in PAK1 KO compared to WT control . All scale bars: 50 pA/1 s . DOI: http://dx . doi . org/10 . 7554/eLife . 14653 . 00710 . 7554/eLife . 14653 . 008Figure 2—figure supplement 1 . Normal inhibitory transmission in ROCK2 KO mice . DOI: http://dx . doi . org/10 . 7554/eLife . 14653 . 00810 . 7554/eLife . 14653 . 009Figure 2—figure supplement 1—source data 1 . Statistical data summary for Figure 2—figure supplement 1: Normal inhibitory transmission in ROCK2 KO mice using one-way ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 14653 . 00910 . 7554/eLife . 14653 . 010Figure 2—figure supplement 2 . Impaired transmitter depletion in response to sustained synaptic activation . DOI: http://dx . doi . org/10 . 7554/eLife . 14653 . 010 Similarly , to exclude the possibility of developmental compensations in the KO animals , we also tested the effect of IPA3 ( 10 μM ) on sIPSC and mIPSC . Again the drug was included in the recording electrode to specifically inhibit PAK1 in the postsynaptic neurons . As shown in Figure 3a–c , inclusion of IPA3 caused a significant decrease in the frequency ( Figure 3c ) , but not the amplitude ( Figure 3b ) of sIPSC in WT neurons . IPA3 also significantly reduced the frequency ( Figure 3d , f ) , but not the amplitude ( Figure 3d , e ) of mIPSC in WT neurons . The frequency of sIPSC and mIPSC in IPA3 treated WT neurons was similar to that of PAK1 KO neurons ( Figure 2c , f; Figure 3c ) . Importantly , IPA3 had no effect on the frequency of sIPSC and mIPSC in PAK1 KO mice ( Figure 3a , c , d , f ) , again confirming that the effect of IPA3 is mediated by PAK1 . Neither the frequency nor the amplitude of sEPSC or mEPSC was affected by IPA3 ( Figure 3g–i ) . The rapid and selective effect of IPA3 on the frequency of sIPSC and mIPSC again suggests that disruption of postsynaptic PAK1 is sufficient to reduce presynaptic release of inhibitory neurotransmitters . 10 . 7554/eLife . 14653 . 011Figure 3 . Acute disruption of postsynaptic PAK1 also selectively impairs the frequency , but not the amplitude of inhibitory synaptic responses . ( a ) Sample traces of sIPSC recordings . Scale bar: 60 pA/1 s . ( b , c ) Summary graphs of ( a ) showing normal amplitude ( b: WT+DMSO = 16 . 14 ± 0 . 65 pA , n = 13 ( 4 ) ; WT+IPA3 = 16 . 83 ± 0 . 95 pA , n = 15 ( 4 ) ; KO+DMSO = 16 . 46 ± 1 . 41 pA , n = 9 ( 3 ) ; KO+IPA3 = 17 . 74 ± 1 . 98 pA , n = 9 ( 3 ) ; genotype: F ( 1 , 42 ) = 0 . 250 , p=0 . 620; drug: F ( 1 , 42 ) = 0 . 646 , p=0 . 426; two-way ANOVA ) , but reduced frequency ( c: WT+DMSO = 1 . 84 ± 0 . 26 Hz , n = 13 ( 4 ) ; WT+IPA3 = 0 . 95 ± 0 . 15 Hz , n = 15 ( 4 ) ; KO+DMSO = 0 . 93 ± 0 . 20 Hz , n = 9 ( 3 ) ; KO+IPA3 = 0 . 88 ± 0 . 24 Hz , n = 9 ( 3 ) ; genotype: F ( 1 , 42 ) = 4 . 908 , p=0 . 032; drug: F ( 1 , 42 ) = 4 . 524 , p=0 . 039; also see Figure 3—source data 1 for t-tests between groups ) of sIPSCs in the PAK1 inhibitor IPA3 treated compared to vehicle ( DMSO ) treated WT neurons . ( d ) Sample traces of mIPSC recordings . ( e , f ) Summary graphs of ( d ) showing normal amplitude ( e: WT+DMSO =9 . 10 ± 0 . 47 pA , n = 15 ( 6 ) ; WT+IPA3 = 9 . 02 ± 0 . 46 pA , n = 11 ( 6 ) ; KO+DMSO = 9 . 02 ± 0 . 36 pA , n = 10 ( 4 ) ; KO+IPA3 = 9 . 13 ± 0 . 36 pA , n =8 ( 3 ) ; genotype: F ( 1 , 40 ) = 0 . 001 , p=0 . 969; drug: F ( 1 , 40 ) = 0 . 001 , p=0 . 978; two-way ANOVA ) , but decreased frequency of mIPSCs ( f: WT+DMSO = 0 . 89 ± 0 . 07 Hz , n = 15 ( 6 ) ; WT+IPA3 = 0 . 46 ± 0 . 08 Hz , n = 11 ( 6 ) ; KO+DMSO = 0 . 40 ± 0 . 09 Hz , n = 10 ( 4 ) ; KO+IPA3 = 0 . 48 ± 0 . 10 Hz , n = 8 ( 3 ) ; genotype: F ( 1 , 40 ) = 7 . 703 , p=0 . 008; drug: F ( 1 , 40 ) = 4 . 259 , p=0 . 046 , two-way ANOVA; also see Figure 3—source data 2 for t-tests between groups ) in IPA3 treated compared DMSO treated WT neurons . ( g ) Sample traces of sEPSC recordings . ( h , i ) Summary graphs of ( g ) showing that IPA3 had no effect on either amplitude ( h: WT+DMSO = 11 . 21 ± 0 . 49 pA , n = 8 ( 4 ) ; WT+IPA3 = 11 . 52 ± 0 . 55 pA , n = 12 ( 5 ) ; p=0 . 693; t-test ) or frequency ( i: WT+DMSO = 0 . 52 ± 0 . 08 Hz , n = 8 ( 4 ) ; WT+IPA3 = 0 . 43 ± 0 . 05 pA , n = 12 ( 5 ) ; p=0 . 322; t-test ) compared to DMSO treated WT neurons . ( j ) Sample traces of mEPSC recordings . ( k , l ) Summary graphs of ( j ) showing that IPA3 had no effect on either amplitude ( k: WT+DMSO = 11 . 40 ± 0 . 71 pA , n = 9 ( 5 ) ; WT+IPA3 = 11 . 52 ± 0 . 39 pA , n = 10 ( 5 ) ; p=0 . 882; t-test ) or frequency ( l: WT+DMSO =0 . 46 ± 0 . 07 Hz , n = 9 ( 5 ) ; WT+IPA3 = 0 . 50 ± 0 . 08 pA , n = 10 ( 5 ) ; p=0 . 754; t-test ) of mEPSCs compared to DMSO treated WT neurons . Scale bars for d , j and l: 20 pA/1 s . DOI: http://dx . doi . org/10 . 7554/eLife . 14653 . 01110 . 7554/eLife . 14653 . 012Figure 3—source data 1 . Statistical data summary for Figure 3b , c: Effect of IPA3 on frequency and amplitude of sIPSC of WT and PAK1 KO neurons using two-way ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 14653 . 01210 . 7554/eLife . 14653 . 013Figure 3—source data 2 . Statistical data summary for Figure 3e , f: Effect of IPA3 on frequency and amplitude of mIPSC of WT and PAK1 KO neurons using two-way ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 14653 . 013 Although the above results that postsynaptic inhibition of PAK1 causes a reduction in the frequency , but not the amplitude of IPSCs , suggest a presynaptic mechanism , it is possible that a reduction in postsynaptic GABA receptors , which could result in silent synapses , may also lead to a reduced frequency in sIPSC and mIPSC . To address this possibility , we first examined the number of GABA-positive neurons and inhibitory synapses , and the level of GABA receptor associated proteins , but found no significant differences between WT and PAK1 KO mice ( Figure 4a–e; also see Figure 4—figure supplement 1 ) . To determine if GABA receptors were functionally equivalent , we recorded IPSCs evoked by a brief puff of GABA ( 1 mM , 100 ms ) in a co-culture system where WT and PAK1 KO neurons were grown on the same coverlips in order to minimize the differences in culturing conditions between genotypes ( see Materials and methods ) , but again found no differences between WT and PAK1 KO neurons ( Figure 4f , g ) . To further examine if other postsynaptic processes could contribute to the effect of PAK1 , we analyzed the effect of cytochalasin D ( 5 μM ) and NSC23766 ( 250 μM ) , pharmacological inhibitors for actin polymerization and Rac1 activation respectively . Both actin and Rac1 are key targets of PAK1 signaling ( Bokoch , 2003; Zhao and Manser , 2012 ) . As shown in Figure 5a–f , no effect on sIPSCs was observed for either inhibitor . It is important to note that consistent with previous results ( Meng et al . , 2002; Zhou et al . , 2011 ) , these two inhibitors ( at the same concentrations used here ) had profound effects on excitatory synaptic function , including basal synaptic transmission ( Figure 5g–i ) and metabotropic glutamate receptor ( mGluR ) -dependent long-term depression induced by 50 μM DHPG ( Figure 5j , k ) . Taken together , we concluded that postsynaptic PAK1 regulates inhibitory synaptic transmission likely through a retrograde mechanism to modulate GABA release . 10 . 7554/eLife . 14653 . 014Figure 4 . Normal GABAergic neurons , synapses , GABA receptor function and postsynaptic actin network in PAK1 KO mice . ( a ) Confocal images of hippocampal sections stained with the nucleus marker DAPI and GABA and summary graph ( b ) showing similar number of GABAergic neurons in PAK1 KO and WT control mice ( WT = 30 ± 4 . 3 neurons/section , n = 8 ( 3 ) ; KO = 31 . 73 ± 3 . 1 6 neurons/section , n = 11 ( 4 ) ; p = 0 . 744; t-test ) . Scale bar: 100 μm . ( c ) Confocal images of hippocampal sections stained with the GABAergic presynaptic marker VGAT and postsynaptic marker gephyrin and summary graph ( d ) showing normal synapse number in PAK1 KO mice ( WT = 10 . 50 ± 1 . 18 puncta/image , n = 8 ( 4 ) ; KO = 13 . 13 ± 1 . 79 , n = 8 ( 4 ) ; p=0 . 241; t-test ) . Scale bar: 10 μm . ( e ) Western blots of hippocampal lysate and summary graph showing no differences in the level of total ( T ) and synaptosomal ( S ) GAD2 and gephyrin between PAK1 KO and WT control mice ( KO T-gephyrin = 0 . 99 ± 0 . 06 , n = 7 ( 7 ) , p=0 . 797; T-GAD2 = 1 . 00 ± 0 . 10 , n = 8 ( 8 ) , p=0 . 990; S- gephyrin = 1 . 00 ± 0 . 09 , n = 7 ( 7 ) , p=0 . 986; S-GAD2 = 0 . 89 ± 0 . 07 , n = 7 ( 7 ) , p=0 . 158; t-tests ) . ( f ) Phase contrast ( upper ) and GFP ( lower ) images of cultured hippocampal neurons showing how the genotype of mixed neurons was identified based on the presence or absence of GFP; ( g ) Sample traces ( upper ) and summary graph ( lower ) showing no differences in the amplitude of responses evoked by 1 mM GABA puff ( arrows ) delivered to the cell body of the neurons ( WT = 1103 . 74 ± 93 . 43 pA , n = 18 ( 3 ) ; KO = 1086 . 86 ± 104 . 94 pA , n = 18 ( 3 ) ; p=0 . 91; t-test ) . Scale bar: 500 pA/0 . 5 s . DOI: http://dx . doi . org/10 . 7554/eLife . 14653 . 01410 . 7554/eLife . 14653 . 015Figure 4—figure supplement 1 . Normal GABAergic neurons and synapses in the cortex . DOI: http://dx . doi . org/10 . 7554/eLife . 14653 . 01510 . 7554/eLife . 14653 . 016Figure 5 . GABAergic transmission is independent of postsynaptic actin cytoskeleton . ( a–f ) Sample traces of sIPSCs and summary graphs showing neither the actin polymerization inhibitor cytochalasin D ( a–c ) nor Rac1 inhibitor NSC23766 ( d–f ) had an effect on the amplitude ( b: DMSO = 12 . 09 ± 0 . 65pA , n = 6 ( 3 ) ; Cyto-D = 11 . 82 ± 0 . 54 pA , n = 6 ( 3 ) ; p=0 . 757; e: Ctrl = 11 . 21 ± 0 . 26 pA , n = 6 ( 3 ) ; NSC = 11 . 57 ± 1 . 29 pA , n = 7 ( 3 ) ; p=0 . 806; t-tests ) or frequency ( c: DMSO = 0 . 79 ± 0 . 25 Hz , n = 6 ( 3 ) ; Cyto-D = 0 . 57 ± 0 . 24 Hz , n = 6 ( 3 ) ; p=0 . 538; f: Ctrl = 2 . 90 ± 0 . 51 Hz , n = 6 ( 3 ) ; NSC = 3 . 01 ± 0 . 48 Hz , n = 7 ( 3 ) ; p=0 . 867; t-tests ) . Scale bar: 20 pA/1 s . ( g–i ) Sample traces of sEPSCs and summary graphs showing actin polymerization inhibitor cytochalasin D had no effect on the amplitude ( h: Ctrl = 13 . 48 ± 0 . 91 pA , n = 8 ( 3 ) ; NSC = 15 . 48 ± 1 . 33 pA , n=6 ( 3 ) ; p=0 . 248; t-test ) , but significantly increased the frequency of eEPSCs ( i: Ctrl = 0 . 47 ± 0 . 09 Hz , n = 8 ( 3 ) ; NSC = 0 . 95 ± 0 . 17 Hz , n = 6 ( 3 ) ; p=0 . 034; t-test ) . ( j , k ) Representative traces and summary graph of evoked EPSC ( eEPSC ) showing that the Rac1 inhibitor NSC23766 blocked DHPG-induced LTD ( Ctrl = 52 . 04 ± 7 . 33% , n = 5 ( 5 ) ; NSC = 90 . 03 ± 5 . 21% , n=5 ( 5 ) ; p=0 . 027; t-test ) . Scale bar: 25 pA/25 ms . DOI: http://dx . doi . org/10 . 7554/eLife . 14653 . 016 Endocannabinoids ( eCB ) are known to be generated and secreted from postsynaptic pyramidal neurons to act as a retrograde messenger to inhibit GABA release ( Katona and Freund , 2012; Piomelli , 2003 ) . While the tonic secretion of eCB affects basal synaptic transmission , its phasic secretion induced by postsynaptic depolarization regulates synaptic plasticity ( Katona and Freund , 2012; Morena et al . , 2016; Piomelli , 2003 ) . An enhanced tonic signaling would reduce the probability of GABA release , and thus decrease IPSC frequency similar to what we observed in neurons of PAK1 KO mice or in WT neurons loaded with PAK1 inhibitors . Thus , we hypothesized that disruption of PAK1 would enhance tonic eCB signaling . To test this hypothesis , we first examined the effect of AM251 , a CB1 receptor antagonist and inverse agonist . In WT , bath application of AM251 ( 5 μM ) increased eIPSCs to approximately 150% of the baseline response ( Figure 6a ) , reflecting disinhibition of GABA release by blocking tonically active CB1 receptors ( Neu et al . , 2007 ) . Remarkably , in PAK1 KO mice , AM251 enhanced eIPSC amplitudes to 250% of the baseline response ( Figure 6a ) . Acute inhibition of postsynaptic PAK1 by IPA3 ( 10 μM ) or by the PAK1 inhibitory peptide ( 20 μg/ml ) prior to the AM251 application produced similar results as shown in PAK1 KO mice ( Figure 6b , c ) . These findings indicate that tonic eCB signaling is restricted by PAK1 and that disruption of PAK1 causes a robust increase in tonic eCB effect . Consistent with this , an increase in tonic eCB signaling could explain why the frequency of sIPSC/mIPSC is reduced in PAK1 KO mice as shown in Figure 2 and 3 . To further investigate whether the enhanced eCB signaling in the KO mice is due to changes in CB receptors and/or their downstream signaling processes in the presynaptic terminal , we examined the protein level of CB1 receptors , but found no differences in either total brain lysates ( Figure 6d ) or synaptosomal fractions ( Figure 6e ) . In addition , bath application of the CB1 receptor agonist WIN ( 5 μM ) , designed to maximally activate the receptor , depressed synaptic responses to the same degree in both WT and PAK1 KO mice ( Figure 6f ) , suggesting that CB1 receptors and their downstream events are intact in the KO mice . It is important to note that DMSO ( the dissolvent for many of the pharmacological agents in this study ) had no effect on eIPSC ( Figure 6—figure supplement 1 ) . 10 . 7554/eLife . 14653 . 017Figure 6 . PAK1 disruption enhances endocannabinoid signaling . ( a ) Sample traces of eIPSC and averaged data showing that bath application of the CB1 receptor antagonist AM251 potentiated the amplitude of eIPSC significantly more in PAK1 KO compared to WT control ( genotype: F ( 1 , 11 ) = 7 . 51 , p=0 . 02; time: F ( 3 , 43 ) = 26 . 833 , *p <0 . 001; repeated measures two-way ANOVA [also see Figure 6—source data 1]; at 21-30 min post AM251 application: WT = 154 . 77 ± 20 . 57% , n = 7 ( 5 ) ; KO = 261 . 41 ± 35 . 00% , n = 6 ( 4 ) , p=0 . 02; t-test ) . ( b ) Sample traces of eIPSC and averaged data showing that postsynaptic infusion of the PAK1 inhibitor IPA3 ( for 30 min ) was sufficient to enhance the subsequent potentiating effect of AM251 on eIPSC ( genotype: F ( 1 , 11 ) = 4 . 919 , p=0 . 049; time: F ( 3 , 33 ) = 15 , p<0 . 001; repeated measures two-way ANOVA [also see Figure 6—source data 2]; at 21-30 min post AM251 application: Ctrl = 129 . 17 ± 7 . 20% , n = 6 ( 5 ) ; IPA3 = 227 . 26 ± 38 . 26% , n = 7 ( 5 ) ; p=0 . 024; t-test ) . Baseline responses ( -10-0 min ) shown here were taken 30 min after whole-cell break-in . ( c ) Sample traces of eIPSC and averaged data showing that postsynaptic infusion of the PAK1 inhibitory peptide ( for 30 min ) was sufficient to enhance the subsequent potentiating effect of AM251 on eIPSC ( genotype: F ( 1 , 11 ) = 10 . 254 , p=0 . 008; time: F ( 3 , 33 ) = 70 . 824 , p<0 . 001; repeated measures two-way ANOVA [also see Figure 6—source data 3]; at 21–30 min post AM251 application: Ctrl = 181 . 89 ± 15 . 94% , n = 7 ( 5 ) ; peptide = 295 . 71 ± 27 . 40% , n = 6 ( 4 ) ; p=0 . 001; t-test ) . Baseline responses ( -10-0 min ) shown here were taken 30 min after whole-cell break-in . ( d , e ) Western blots of hippocampal lysate and summary graphs showing that both total ( d ) and synaptosomal ( e ) CB1R protein levels were unaltered in PAK1 KO mice ( KO total = 1 . 15 ± 0 . 20 , n = 7 ( 7 ) ; p=0 . 468 normalized and compared to the WT; KO synaptosomal = 0 . 94 ± 0 . 11 , n = 8 ( 8 ) ; p=0 . 567 normalized and compared to the WT; t-tests ) . ( f ) Sample traces of eIPSC and averaged data showing that bath application of the CB1 receptor agonist WIN depressed eIPSC to the same degree in PAK1 KO and WT control ( genotype: F ( 1 , 8 ) = 0 . 008 , p=0 . 932; time: F ( 3 , 24 ) = 31 . 556 , p<0 . 001; repeated measures two-way ANOVA [also see Figure 6—source data 4]; at 21-30 min post WIN application: WT = 60 . 78 ± 10 . 78% , n = 5 ( 5 ) ; KO = 60 . 88 ± 6 . 73% , n = 5 ( 5 ) ; p=0 . 977; t-test ) . All scale bars: 40 pA/25 ms . DOI: http://dx . doi . org/10 . 7554/eLife . 14653 . 01710 . 7554/eLife . 14653 . 018Figure 6—source data 1 . Statistical data summary for Figure 6a: Effect of AM251 on eIPSC in WT and PAK1 KO using repeated measures two-way ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 14653 . 01810 . 7554/eLife . 14653 . 019Figure 6—source data 2 . Statistical data summary for Figure 6b: Effect of AM251 on eIPSC in the presence or absence of IPA3 using repeated measures two-way ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 14653 . 01910 . 7554/eLife . 14653 . 020Figure 6—source data 3 . Statistical data summary for Figure 6c: Effect of AM251 on eIPSC in the presence or absence of PAK1 inhibitory peptide using repeated measures two-way ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 14653 . 02010 . 7554/eLife . 14653 . 021Figure 6—source data 4 . Statistical data summary for Figure 6f: Effect of WIN on eIPSC in WT and PAK1 KO using repeated measures two-way ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 14653 . 02110 . 7554/eLife . 14653 . 022Figure 6—figure supplement 1 . The lack of effect of DMSO on eIPSCs . DOI: http://dx . doi . org/10 . 7554/eLife . 14653 . 02210 . 7554/eLife . 14653 . 023Figure 6—figure supplement 1—source data 1 . Statistical data summary for Figure 6—figure supplement 1: Lack of effect of DMSO on eIPSC using repeated measures two-way ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 14653 . 023 The increased effect of AM251 in the absence of any changes in the level of CB1 receptors or their activation strength suggests that the levels of the eCB molecules might be elevated by PAK1 disruption . To test this possibility directly , we measured the tissue level of AEA and 2-AG in the hippocampus . Interestingly , tissue levels of AEA ( Figure 7a ) , but not 2-AG ( Figure 7b ) , were significantly elevated in PAK1 KO mice . These data are consistent with the belief that AEA mediates tonic actions of the eCB system ( Di et al . , 2013; Kim and Alger , 2010; Tabatadze et al . , 2015 ) , whereas 2-AG mediates essentially all forms of phasic eCB signaling in the CNS ( Katona and Freund , 2012; Morena et al . , 2016 ) . More so , as AEA mediates tonic inhibition of GABA release within the hippocampus ( Di et al . , 2013; Lee et al . , 2015; Tabatadze et al . , 2015 ) , these data indicate that PAK1 disruption results in an increase in AEA production and a consequential reduction in tonic GABA release . 10 . 7554/eLife . 14653 . 024Figure 7 . Elevated AEA and reduced COX-2 in PAK1 KO mice . ( a ) Summary graph showing a significant increase in hippocampal tissue AEA in PAK1 KO compared to WT control ( WT = 5 . 82 ± 0 . 44 pmol/g , n = 11 ( 11 ) ; KO = 7 . 64 ± 0 . 70 pmol/g , n = 13 ( 13 ) , p=0 . 046; t-test ) . ( b ) . Summary graph showing no differences in hippocampal tissue 2-AG between PAK1 KO and WT control ( WT = 12 . 46 ± 1 . 73 nmol/g , n = 11 ( 11 ) ; KO = 12 . 30 ± 0 . 82 nmol/g , n = 13 ( 13 ) , p=0 . 932; t-test ) . ( c ) Sample traces of eIPSC and averaged data showing reduced DSI in PAK1 KO compared to WT control ( time to reach 50% of baseline response T50%: WT = 15 . 06 ± 2 . 06s , n = 9 ( 5 ) ; KO = 6 . 98 ± 0 . 99s , n = 9 ( 3 ) ; p=0 . 003; t-test ) . ( d ) Western blots of hippocampal lysate and summary graphs showing no differences in total MGL protein levels between PAK1 KO and WT control ( KO = 0 . 98 ± 0 . 10 , n = 6 ( 6 ) , p=0 . 854 normalized and compared to WT; t-test ) . ( e ) Western blots of hippocampal synaptosomal protein fraction and summary graph showing no difference in the amount of MGL protein between PAK1 KO and WT control ( KO = 1 . 01 ± 0 . 09 , n = 7 ( 7 ) ; p=0 . 253 normalized and compared to WT; t-test ) . ( f ) Western blots of hippocampal lysate and summary graph showing no differences in total DGL protein levels between PAK1 KO and WT control ( KO = 1 . 09 ± 0 . 15 , n = 6 ( 6 ) , p=0 . 583 normalized and compared to WT; t-test ) . ( g ) Western blots of hippocampal synaptosomal protein fraction and summary graph showing no difference in the amount of DGL protein between PAK1 KO and WT control ( KO = 0 . 96 ± 0 . 13 , n = 9 ( 9 ) ; p=0 . 738 normalized and compared to WT; t-test ) . ( h ) Western blots of hippocampal lysate and summary graph showing no differences in total FAAH protein levels between PAK1 KO and WT control ( KO = 0 . 87 ± 0 . 11 , n = 5 ( 5 ) , p=0 . 247 normalized and compared to WT; t-test ) . ( i ) Western blots of hippocampal synaptosomal protein fraction and summary graph showing no difference in the amount of FAAH protein between PAK1 KO and WT control ( KO = 1 . 01 ± 0 . 11 , n = 6 ( 6 ) ; p=0 . 937 normalized and compared to WT; t-test ) . ( j ) Western blots of hippocampal lysate and summary graphs showing no differences in total COX-2 protein levels between PAK1 KO and WT control ( KO = 0 . 89 ± 0 . 09 , n = 7 ( 7 ) , p=0 . 263 normalized and compared to WT; t-test ) . ( k ) Western blots of hippocampal synaptosomal protein fraction and summary graph showing reduced COX-2 protein in PAK1 KO compared to WT control ( KO = 0 . 54 ± 0 . 09 , n = 10 ( 10 ) ; p<0 . 001 normalized and compared to WT; t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14653 . 02410 . 7554/eLife . 14653 . 025Figure 7—figure supplement 1 . Reduced synaptosomal COX-2 in PAK1 KO hippocampus . DOI: http://dx . doi . org/10 . 7554/eLife . 14653 . 025 Quite interestingly , a recent report has indicated that AEA and 2-AG signaling within the hippocampus may compete with each other , such that elevations in AEA signaling dampen 2-AG regulation of GABAergic transmission , through an AEA-TRPV1 mediated mechanism ( Lee et al . , 2015 ) . To investigate whether this interaction occurs in PAK1 KO mice , we analyzed phasic suppression of inhibition induced by postsynaptic depolarization ( DSI ) and found that the amplitude of DSI was significantly reduced in the KO compared to WT mice ( Figure 7c ) , supporting the idea that elevated AEA competes with 2-AG , resulting in reduced phasic 2-AG signaling . Together , these data create a compelling argument that disruption of PAK1 selectively augments tonic AEA signaling to dampen constitutive synaptic GABA transmission . To determine the mechanism by which PAK1 modulates the levels of eCBs , we examined how its deletion impacted a series of enzymes known to be involved in eCB metabolism . Consistent with the fact that there were no changes in 2-AG , the enzymes involved in the generation and metabolism of 2-AG , including monoacylglycerol lipase ( MGL ) and diacylglycerol lipase ( DGL ) , were not altered in PAK1 KO mice ( Figure 7d–g ) . Surprisingly , however , both total and synaptosomal protein levels of the enzyme primarily involved in AEA metabolism , fatty acid amide hydrolase ( FAAH ) , were also not altered in PAK1 KO mice ( Figure 7h , i ) . As several reports have identified that COX-2 can be an important regulator of AEA signaling ( Glaser and Kaczocha , 2010; Hermanson et al . , 2013 ) , we examined if PAK1 deletion could impact AEA signaling through a COX-2 mediated mechanism . Notably , we found that although the total protein level of COX-2 was not altered ( Figure 7j ) , its synaptosomal fraction was significantly reduced in the KO brain ( Figure 7k , Figure 7—figure supplement 1 ) , suggesting that PAK1 is specifically important for COX-2 expression at the synapse and/or during synaptic activity . Together these results suggest that the enhanced eCB signaling , due to the disruption of PAK1 , might be mediated by reduced COX-2 and subsequently elevated AEA . To directly test if the reduced COX-2 expression is responsible for the elevated effect of the eCB signaling on GABA transmission in PAK1 KO mice , we tested the effect of the COX-2 inhibitor Nimesulide on eIPSC . As shown in Figure 8a , application of Nimesulide ( 30 μM ) in WT neurons depressed IPSCs to approximately 45% of the baseline response , presumably due to increased eCB production , but this Nimesulide-induced depression was significantly reduced ( to 70% of the baseline response ) in PAK1 KO neurons , consistent with an already reduced COX-2 level in the KO mice . Following Nimesulide application and after the depressed responses became stabilized , we then treated the neurons with the CB1 receptor antagonist AM251 ( 5 μM ) . As shown in Figure 8b , both WT and PAK1 KO neurons now showed equally enhanced responses ( 250% of the baseline response ) that were similar to PAK1 KO neurons treated with AM251 alone as shown in Figure 6a . Therefore , inhibition of COX-2 in WT neurons recapitulated the effect of PAK1 disruption . 10 . 7554/eLife . 14653 . 026Figure 8 . COX-2 inhibition recapitulates the effect of PAK1 disruption . ( a ) Sample traces and averaged data of eIPSCs showing that bath application of the COX-2 inhibitor Nim depressed eIPSCs in WT , but this depression was significantly reduced in PAK1 KO neurons ( genotype: F ( 1 , 12 ) = 7 . 639 , p=0 . 017; time: F ( 2 , 24 ) = 87 . 676 , **p<0 . 001; repeated measures two-way ANOVA [also see Figure 8—source data 1]; at 11-20 min post Nim application: WT = 48 . 26 ± 6 . 42% , n = 7 ( 3 ) ; KO = 71 . 51 ± 4 . 75% , n = 7 ( 3 ) ; p=0 . 013; t-test ) . Scale bar: 30 pA/25 ms . ( b ) Sample traces and averaged data showing that following the Nim treatment , AM251 potentiated eIPSCs to the same degree in WT and PAK1 KO neurons ( genotype: F ( 1 , 9 ) = 0 . 044 , p=0 . 839; time: F ( 3 , 27 ) = 15 . 222; *p<0 . 001; repeated measures two-way ANOVA [also see Figure 8—source data 2]; at 21-30 min post AM251 application: WT = 308 . 58 ± 55 . 37% , n = 5 ( 4 ) ; KO = 280 . 53 ± 69 . 32% , n = 6 ( 5 ) ; p=0 . 766; t-test ) . Scale bar: 60 pA/25 ms . Baseline responses ( -10-0 min ) shown here were taken 30 min after the onset of the Nim treatment . Nim was present throughout the entire experiment . ( c ) Sample traces and averaged data of eIPSCs showing that bath application of the FAAH inhibitor URB597 depressed eIPSCs in WT , but this depression was significantly reduced in PAK1 KO neurons ( genotype: F ( 1 , 13 ) = 13 . 830 , p=0 . 003; time: F ( 3 , 39 ) = 14 . 122 , *p<0 . 001; repeated measures two-way ANOVA [also see Figure 8—source data 3]; at 21-30 min post URB597 application: WT = 54 . 19 ± 7 . 59% , n = 7 ( 5 ) ; KO = 93 . 74 ± 6 . 80% , n = 8 ( 3 ) ; p=0 . 002; t-test ) . Scale bar: 70 pA/25 ms . ( d ) Sample traces and averaged data showing that following the URB597 treatment , AM251 potentiated eIPSCs to the same degree in WT and PAK1 KO neurons ( genotype: F ( 1 , 8 ) = 0 . 055 , p=0 . 821; time: F ( 3 , 24 ) = 23 . 459 , *p<0 . 001; repeated measures two-way ANOVA [also see Figure 8—source data 4]; at 21-30 min post AM251 application: WT = 246 . 61 ± 45 . 34% , n = 5 ( 3 ) ; KO = 208 . 59 ± 17 . 52% , n = 5 ( 4 ) ; p=0 . 391; t-test ) . Scale bar: 60 pA/25 ms . Baseline responses ( -10-0 min ) shown here were taken 30 min after the onset of the URB597 treatment . URB597 was present throughout the entire experiment . ( e ) Sample traces and averaged data showing no differences in eIPSC depression by the MGL inhibitor JZL184 between PAK1 KO and WT control ( genotype: F ( 1 , 16 ) = 0 . 265 , p=0 . 614; time: F ( 3 , 48 ) = 17 . 292 , p<0 . 001; repeated measures two way ANOVA [also see Figure 8—source data 5]; at 21-30 min post JZL184 application: WT = 69 . 53 ± 9 . 60% , n = 10 ( 8 ) ; KO = 64 . 99 ± 10 . 76% , n = 8 ( 6 ) ; p=0 . 757; t-test ) . Scale bar: 35 pA/25 ms . ( f ) Sample traces and averaged data showing that following JZL184 treatment , AM251 still induced eIPSC potentiation significantly more in PAK1 KO compared to WT control ( genotype: F ( 1 , 9 ) = 7 . 770 , p=0 . 021; time: F ( 3 , 27 ) = 30 . 146 , **p<0 . 001; repeated measures two-way ANOVA [also see Figure 8—source data 6]; at 21–30 min post AM251 application: WT = 159 . 76 ± 16 . 08% , n = 6 ( 6 ) ; KO = 242 . 60 ± 26 . 03% , n = 5 ( 4 ) ; p=0 . 020; t-test ) . Baseline responses ( -10-0 min ) shown here were taken 30 min after the onset of the JZL184 treatment . JZL184 was present throughout the entire experiments . Scale bar: 60 pA/25 ms . DOI: http://dx . doi . org/10 . 7554/eLife . 14653 . 02610 . 7554/eLife . 14653 . 027Figure 8—source data 1 . Statistical data summary for Figure 8a: Effect of Nimesulide on eIPSC in WT and PAK1 KO using repeated measures two-way ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 14653 . 02710 . 7554/eLife . 14653 . 028Figure 8—source data 2 . Statistical data summary for Figure 8b: Effect of AM251 on eIPSC after Nimesulide treatment in WT and PAK1 KO using repeated measures two-way ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 14653 . 02810 . 7554/eLife . 14653 . 029Figure 8—source data 3 . Statistical data summary for Figure 8c: Effect of URB597 on eIPSC in WT and PAK1 KO using repeated measures two-way ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 14653 . 02910 . 7554/eLife . 14653 . 030Figure 8—source data 4 . Statistical data summary for Figure 8d: Effect of AM251 on eIPSC after URB597 treatment in WT and PAK1 KO using repeated measures two-way ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 14653 . 03010 . 7554/eLife . 14653 . 031Figure 8—source data 5 . Statistical data summary for Figure 8e: Effect of JZL184 on eIPSC in WT and PAK1 KO using repeated measures two-way ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 14653 . 03110 . 7554/eLife . 14653 . 032Figure 8—source data 6 . Statistical data summary for Figure 8f: Effect of AM251 on eIPSC after JZL treatment in WT and PAK1 KO using repeated measures two-way ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 14653 . 032 As COX-2 only metabolizes a proportion of AEA , changes in COX-2 are consistent with the magnitude of AEA changes we documented in the hippocampus , which are significantly less dramatic than what would be seen following disruption of FAAH activity ( Cravatt et al . , 2001 ) . Finally , to provide further evidence that this increase in tonic eCB signaling in PAK1 KO mice is mediated by a selective increase in AEA , and not 2-AG , signaling , we examined the effects of specific AEA and 2-AG hydrolysis inhibitors . Similar to what was seen following COX-2 inhibition , the ability of the FAAH inhibitor URB597 ( 1 μM ) to reduce eIPSC was reduced in PAK1 KO mice relative to WT mice ( Figure 8c ) and the effect of AM251 ( 5 μM ) on eIPSC was no longer different between the two genotypes after the FAAH treatment ( Figure 8d ) , consistent with the fact that there is already an elevated AEA tone . Importantly , the effect of the MGL inhibitor JZL184 was not altered in PAK1 KO mice; that is , bath application of JZL184 ( 5 μM ) depressed eIPSC to the same degree in both WT and KO neurons ( Figure 8e ) and had no effects on the degree of disinhibition by AM251 ( 5 μM ) ( Figure 8f ) . Therefore , we conclude that the increased inhibition of GABA transmission by eCB signaling in PAK1 KO mice is caused by reduced COX-2 expression and consequential elevation in AEA , but not 2-AG , signaling at GABA synapses in the CA1 region of the hippocampus . To further elucidate why disruption of PAK1 and reduced COX-2 affected inhibitory , but not excitatory synaptic transmission , we performed immunostaining experiments using cultured hippocampal neurons to determine the subcellular distribution of PAK1 and COX-2 , and how their synaptic localization was affected by PAK1 disruption . First , we showed that in WT neurons , PAK1 was colocalized with both PSD-95 ( excitatory synaptic marker , Figure 9a ) and gephyrin ( GABAergic synaptic marker , Figure 9b ) , suggesting that PAK1 is expressed at both excitatory and inhibitory synapses . In addition , we showed that PAK1 was colocalized with COX-2 ( Figure 9c ) . Next , we showed that a small portion of COX-2 was colocalized with PSD-95 ( Figure 9d ) and this colocalization was not altered in PAK1 KO neurons ( Figure 9e–h ) , suggesting that PAK1 disruption does not affect COX-2 localization at the excitatory synapse . Finally , we showed that a much larger portion of COX-2 was colocalized with gephyrin ( Figure 9i ) , and importantly , the level of this colocalization was significantly reduced in PAK1 KO compared to WT neurons ( Figure 9i–m ) . The total protein levels of COX-2 ( Figure 9f , k ) , PSD-95 ( Figure 9g ) and gephyrin ( Figure 9l ) were unaltered in PAK1 KO neurons . These results together suggest that PAK1 disruption specifically impairs COX-2 localization at GBABergic synapses , providing a mechanism for PAK1-COX2 signaling to specifically regulate inhibitory synaptic transmission . 10 . 7554/eLife . 14653 . 033Figure 9 . Reduced COX-2 localization at GABAergic synapses in PAK1 KO neurons . ( a–c ) Cultured hippocampal neurons costained for PAK1 and the excitatory marker PSD-95 ( a ) , the GABAergic marker gephyrin ( b ) or COX-2 ( c ) showing PAK1 colocalization with PSD-95 , gephyrin and COX-2 . ( dh ) Cultured hippocampal neurons costained for COX-2 and PSD9-5 in WT ( d ) and PAK1 KO neurons ( e ) and summary graphs ( fh ) showing no differences between genotypes in total COX-2 ( f , WT = 100 ± 9 . 11 , n = 21 ( 3 ) ; KO = 119 . 08 ± 7 . 50 , n = 16 ( 3 ) ; p=0 . 131; t-test ) , total PSD-95 ( g , WT = 100 ± 7 . 33 , n = 21 ( 3 ) ; KO = 100 . 03 ± 5 . 68 , n = 16 ( 3 ) ; p=0 . 997; t-test ) and COX-2 colocalized with PSD-95 ( h , WT = 12 . 55 ± 0 . 26% , n = 21 ( 3 ) ; KO = 12 . 72 ± 0 . 35% , n = 16 ( 3 ) ; p=0 . 700; t-test ) . ( im ) Cultured hippocampal neurons costained for COX-2 and gephyrin in WT ( i ) and PAK1 KO neurons ( j ) and summary graphs ( km ) showing no changes in total COX-2 ( k , WT = 100 ± 4 . 90 , n = 17 ( 3 ) ; KO = 102 . 09 ± 6 . 14 , n = 15 ( 3 ) ; p=0 . 792; t-test ) or total gephyrin ( l , WT = 100 ± 5 . 31 , n = 17 ( 3 ) ; KO = 94 . 29 ± 4 . 91 , n = 15 ( 3 ) ; p=0 . 432; t-test ) , but reduced COX-2 colocalized with gephyrin ( m , WT = 80 . 35 ± 1 . 14 , n = 17 ( 3 ) ; KO = 75 . 32 ± 1 . 47 , n = 15 ( 3 ) ; *p=0 . 008; t-test ) . Scale bars: 10 μm for whole neuron images and 5 μm for the enlarged dendritic fragments . Arrows indicate colocalization and arrowheads for no colocalization . DOI: http://dx . doi . org/10 . 7554/eLife . 14653 . 033 Although PAK1 is known to be important in the regulation of spine properties such as spine actin and morphology , its effect on excitatory synaptic transmission is minimal ( Asrar et al . , 2009 ) . In fact , basal excitatory synaptic transmission is not altered in PAK1 KO mice , possibly due to the existence of other functionally redundant PAKs ( e . g . PAK2 and PAK3 ) ( Bokoch , 2003; Huang et al . , 2011; Kelly and Chernoff , 2012; Meng et al . , 2005 ) . In contrast , disruptions of PAK1 , either by genetic deletion or pharmacological blockade , dramatically reduces GABAergic transmission , indicating that the primary role of PAK1 is to facilitate inhibitory but not excitatory synaptic function . This selective effect of PAK1 disruption is rather surprising because it has been shown previously that expression of a dominant negative form ( i . e . the autoinhibitory domain ) of PAK1 affects the excitatory synaptic morphology ( Hayashi et al . , 2004 ) . One possible explanation is that the expression of this mutant PAK1 may affect other members of the PAK family , including PAK3 , which is also highly expressed in the brain ( Meng et al . , 2005 ) . Indeed , the double KO mice lacking both PAK1 and PAK3 display severe deficits in spines and the actin cytoskeleton ( Huang et al . , 2011 ) . These results together suggest that while PAK1 and PAK3 are functionally redundant at the excitatory synapses , PAK1 is a unique regulator at the inhibitory synapses , which may not be readily replaced by other members of the PAK family . Therefore , the ability of the PAK1 inhibitors in ameliorating the deficits associated with models of fragile X syndrome and neurofibromatosis ( Dolan et al . , 2013; Hayashi et al . , 2007; Molosh et al . , 2014 ) may be mediated by its effect on the GABA function . Consistent with this notion , inhibitory synaptic transmission is commonly altered in these brain disorders ( Cui et al . , 2008; Olmos-Serrano et al . , 2010; Radhu et al . , 2015 ) . In PAK1 KO mice , the frequency , but not the amplitude of mIPSCs and sIPSCs is reduced . The reduction in the frequency is not caused by changes in the number of inhibitory neurons or synapses because none of these parameters are altered in the KO mice . The lack of changes in synapse number is also consistent with the observations that the frequency reduction can be rapidly induced ( within mins ) by infusion of the PAK1 inhibitors , a perturbation that is not likely to cause any changes in the number of neurons or synapses . Our data also indicate that although the frequency reduction is induced by postsynaptic inhibition of PAK1 , it is independent of postsynaptic GABA receptors or the actin cytoskeleton . Therefore , postsynaptic PAK1 affects GABA transmission through a retrograde messenger to modulate neurotransmitter release . Several lines of evidence support that eCBs are the retrograde messenger to mediate such an effect here . First , the effect the CB1 receptor antagonist AM251 is greatly enhanced by the loss of PAK1; application of AM251 potentiates eIPSC three times greater in PAK1 KO or IPA3 treated neurons than in WT neurons . The simplest interpretation of this result is that loss of PAK1 results in increased tonic eCB signaling and therefore a greater inhibition of GABA release . Thus , blocking eCB inhibition causes a greater potentiation of GABA release and eIPSC in the absence of PAK1 . There are at least two possibilities by which the eCB signaling can be enhanced: an increase in the signaling pool of eCB molecules and/or altered signaling processes triggered by activation of the CB1 receptors . Because neither the expression level of CB1 receptors nor the effect of a CB1 receptor agonist are enhanced , a straightforward interpretation would be that eCB secretion is elevated in the absence of PAK1 . Consistent with this , biochemical analysis of hippocampi from PAK1 KO mice indicate a significant increase in the tissue level of the eCB AEA , but not 2-AG . While 2-AG has been found to mediate the majority of phasic eCB processes , such as DSI and LTD , AEA is believed to mediate the tonic signaling actions of the eCB system ( Katona and Freund , 2012; Morena et al . , 2016 ) . As such , the selective increase in AEA is consistent with the fact that tonic , but not phasic , eCB signaling is enhanced by disruption of PAK1 . In fact , a recent report has suggested that elevations in AEA signaling reduce the ability of 2-AG signaling to modulate GABAergic transmission in the hippocampus ( Lee et al . , 2015 ) , and again consistent with this finding , our data demonstrate that DSI , a measure of activity-dependent , phasic eCB signaling that is mediated by 2-AG , is reduced in PAK1 KO mice . Finally , PAK1 KO mice exhibit no alteration to the synaptic response of a 2-AG hydrolysis inhibitor , but do demonstrate reduced responses to a AEA-hydrolysis inhibitor , which is to be expected if PAK1 deletion selectively increases AEA , and not 2-AG , signaling . Taken together , these results suggest that PAK1 normally restricts tonic AEA level to enhance GABA release and maintains sufficient inhibitory transmission . The disruption of PAK1 could increase AEA signaling through a variety of mechanisms , although if AEA synthesis were increased by PAK1 deletion , then the effects of an AEA hydrolysis inhibitor would be expected to be amplified , not impaired , because the inhibitor would lock all of the excess AEA within the synapse . If PAK1 disruption reduced AEA hydrolysis , however , one would expect an occlusion to the addition of an AEA-hydrolysis inhibitor . As the latter , and not the former , is what the current data indicate , this leads us to investigate how PAK1 was modulating AEA clearance , not synthesis . FAAH is the primary enzyme responsible for AEA hydrolysis , however our biochemical analysis did not reveal any difference between total or synaptic expression of FAAH protein in PAK1 KO mice . A growing body of work has identified that aside from the canonical metabolism of AEA by FAAH , COX-2 represents an additional mechanism of AEA clearance ( Glaser and Kaczocha , 2010; Hermanson et al . , 2013 ) . Notably , the impact of COX-2 inhibition on AEA levels ( Hermanson et al . , 2013 ) is roughly comparable to the increase we documented herein in PAK1 KO mice . Consistent with this , we found that PAK1 KO mice did exhibit reductions in COX-2 protein . It is important to emphasize that only synaptic , but not total COX-2 is affected in PAK1 KO mice , implying that PAK1 is particularly important for COX-2 regulation at the synapse . Furthermore , although COX-2 is expressed at both excitatory and inhibitory synapses , the effect of PAK1 deletion on COX-2 appears to be specific to inhibitory synapses , which may contribute to the specific effect of PAK1-COX-2 signaling on inhibitory synaptic transmission . The mechanism by which PAK1 regulates COX-2 localization is unknown , but it is possible that local regulation of COX-2 , both at the level of protein synthesis and/or trafficking , could be targeted by PAK1 . With respect to neurodevelopmental disorders , such as autism , these data could have significant implications for both the therapeutic potential of PAK1 inhibitors or agents that enhance AEA signaling . For example , PAK1 deletion has been shown to normalize synaptic and behavioral deficits in the Neurofibromatosis model of autism ( Molosh et al . , 2014 ) . Specifically , deletion of PAK1 resulted in a reduction of mIPSC in the amygdala ( similar to what was found within the hippocampus within the current study ) that was associated with an improvement in social behaviors ( Molosh et al . , 2014 ) . In line with this , elevating AEA signaling within the amygdala ( which has been shown to dampen GABA release in the amygdala [Azad et al . , 2004] ) has been shown to improve social interaction and increase social behavior ( Trezza et al . , 2012 ) . More so , a recent report has indicated that the neuroligin-3 mutation related to autism causes a disruption in tonic eCB signaling within the hippocampus ( Földy et al . , 2013 ) , resulting in an increase in GABA release and a shift in the E/I balance of the hippocampus that is the exact opposite of what was produced by PAK1 disruption . Finally , PAK1 disruption has also been shown to normalize behavioral changes in the Fragile X model of autism ( Dolan et al . , 2013; Hayashi et al . , 2007 ) , which parallels recent behavioral work similarly demonstrating that administration of AEA , but not 2-AG , hydrolysis inhibitors , can normalize some of the same behavioral deficits seen in Fragile X mice ( Qin et al . , 2015 ) . Taken together , while speculative at this time , these data collectively create a compelling picture that impairments in tonic AEA signaling may relate to the pathology of neurodevelopmental disorders , such as autism , and that inhibition of PAK1 may exert its potentially beneficial effects by enhancing tonic AEA signaling . Future research will need to investigate this mechanism more directly , but these data establish a framework to approach this question . More so , as PAK1 has been associated with other disease states ( Gilman et al . , 2011; Kelly and Chernoff , 2012; Kumar et al . , 2006; Ma et al . , 2012 ) , the relationship between PAK1 , COX-2 and AEA signaling could prove to be highly relevant for a wide array of pathological processes , given that COX-2 and AEA have similarly been implicated in the etiology of these disease processes ( Hermanson et al . , 2014 ) . In summary , we have identified a novel process by which PAK1 regulates the eCB system and inhibitory synaptic function ( Figure 1 ) . Given that PAK1 is involved in both normal physiological and pathological processes as discussed above , that range from cancers , allergies to mental disorders , our results provide a new mechanism and treatment scheme by which PAK operates in these various systems , and opens the door to a mechanism-driven therapeutic approach which targets the interaction of these systems . 10 . 7554/eLife . 14653 . 034Figure 10 . A hypothetical model . In wild type neurons , constitutively active PAK1 is required for maintaining a sufficient level of synaptic COX-2 to keep AEA low , thus less suppression of GABA release and normal inhibitory transmission . In the absence of PAK1 , synaptic COX-2 is reduced , which leads to accumulation of AEA , increased suppression of GABA release and impaired inhibitory transmission and E/I balance . DOI: http://dx . doi . org/10 . 7554/eLife . 14653 . 034 The generation and initial characterization of PAK1 KO mice were described previously ( Asrar et al . , 2009; Huang et al . , 2011 ) . All the mice used in this study were PAK1 KO ( Pak1-/- ) offspring and their wild type littermaze ( Pak1+/+ ) derived PAK1 heterozygous ( Pak1+/- ) breeders to minimize the effect of genetic or environmental factors . The age of the animals in all experiments was 25–32 days . The following PCR primers were used for PAK1 mouse genotyping: ( Pak1+F: 5’-CTGAGGGAAGAGACTGCAGAG-3’ , Pak1+R: 5’-AGGCAGAGGTTTGGAGCCGTG-3’; Pak1-F: 5’-CTGAGGGAAGAGACTGCAGAG-3’ , Pak1-R:5’-GGGGGAACTTCCTGACTAGG-3’ ) . The absence of PAK1 in the PAK1 KO mice was confirmed by Western blot analysis . The mice were housed under a standard 12/12 light/dark cycle condition . All the procedures used for this study were approved by the Animal care committees at the Hospital for Sick Children , Canada and Southeast University , China . AM251 , cytochalasin D ( Cyto-D ) , JZL184 , Nimesulide , NSC23766 ( NSC ) and URB597 were from Selleck; D-2-amino-5-phosphonovalerate ( D-APV ) and Glycyl-H 1152 dihydrochloride ( GH1152 ) were from Tocris; all other drugs ( NBQX disodium salt hydrate , ( R ) - ( + ) -WIN 55 , 212–2 mesylate salt [WIN] , IPA3 , TTX , and picrotoxin ) were from Sigma . The procedures for the preparation and recording of hippocampal slices were described previously ( Meng et al . , 2005; Meng et al . , 2002; Zhou et al . , 2011 ) . Briefly , mouse brains were quickly dissected and transferred to ice-cold artificial cerebrospinal fluid ( ACSF ) saturated with 95% O2/5% CO2 and sliced to 360 μm sagittal slices . The slices were recovered at 32°C for at least 2 hrs before a single slice was transferred to the recording chamber . ACSF contained ( in mM ) : 120 NaCl , 3 . 0 KCl , 1 . 0 NaH2PO4 , 26 NaHCO3 , 11 D-glucose , 2 . 0 CaCl2 , and 1 . 2 MgSO4 . To obtain evoked synaptic responses , the stimulation electrode was placed near the stratum pyramidal layer of the CA1 area to execute 0 . 1Hz stimulation . Whole cell recordings were performed under the voltage clamp mode with a holding potential of −70 mV except in those to construct I/V curves of synaptic currents . Series resistance was monitored by a −3 mV step throughout the entire experiment of whole cell access and if it fluctuated more than 20% , the data were excluded from the analysis . For the E/I ratio experiments , slices were first perfused by ACSF to record total postsynaptic current ( PSC ) for 10 min , then 10 μM NBQX/50 μM APV were added to specifically record inhibitory postsynaptic current ( IPSC ) , and finally 100 μM picrotoxin was added to verify the inhibitory response . The E/I ratio was calculated as ( PSC-IPSC ) /IPSC . Spontaneous IPSC ( sIPSC ) and miniature IPSC ( mIPSC ) were recorded by including 10 μM NBQX plus 50 μM APV with or without 1 μM TTX in ACSF respectively . Spontaneous EPSC ( sEPSC ) and miniature EPSC ( mEPSC ) were recorded with 100 μM picrotoxin with or without 1 μM TTX respectively . For the E/I ratio and s/mIPSC/EPSC ( except sIPSC with NSC ) experiments , whole-cell recordings were made using glass pipettes ( 3–5 MΩ ) filled with an intracellular solution containing ( in mM ) : 130 CsMeSO4 , 5 NaCl , 1 MgCl2 , 0 . 05 EGTA , 10 HEPES , 1 Mg-ATP , 0 . 3 Na3-GTP , and 5 QX-314 ( pH 7 . 25 ) ( 280–300 mOsm ) , and for all other measurements , the recordings were made with an intracellular solution containing ( in mM ) : 110 K-gluconate , 25 KCl , 10 Na2-creatine phosphate , 0 . 2 EGTA , 10 HEPES , 2 Mg-ATP and 0 . 3 Na3-GTP ( pH 7 . 25 ) ( 280–300 mOsm ) ( Hashimotodani et al . , 2007; Huang and Woolley , 2012 ) . Some of the experiments were repeated in a high Cl intracellular solution containing ( in mM ) : 130 CsCl , 5 NaCl , 10 HEPES , 0 . 5 EGTA , 5 QX314 , 4 Mg-ATP and 0 . 3 Na3-GTP ( pH 7 . 2–7 . 4 ) ( 280 mOsm ) and similar results were obtained , but the data were not included in this study . For whole cell infusion of chemical inhibitors ( e . g . peptide , IPA3 , and NSC23766 ) , the stock solutions or control vehicle were added to the intracellular solution right before the start of the experiments . The sequence of the PAK1 inhibitory peptides are as follows: active peptide: KKEKERPEISLPSDFEHT; ctrl peptide: GPPARNPRSPVQPPP ( final concentration at 20 μg/ml , Genscript ) . For the DSI experiments , the stimulation frequency was increased to 0 . 33 Hz and the depolarization was from −70 mV to 0 mV lasting 5 s . For the FAAH inhibitor URB597 experiments , the summary graphs ( Figure 8c , d ) represented pooled data from both females and males ( the ratio of females vs males was balanced between WT and PAK1 KO mice ) . Because the effect of URB597 is more prominent in female rats ( Tabatadze et al . , 2015 ) , we also repeated the URB597 experiment shown in Figure 8c in female mice only and the result was the same as the pooled data ( data not shown ) . Synaptic depression in response to sustained high frequency stimulation was induced by 5 Hz lasting 3 min . All data acquisition and analysis were done with pCLAMP and MiniAnalysis programs . All evoked data were normalized to the average of the baseline response . In order to accurately compare the WT and PAK1 KO neurons , we used a co-culture system where we plated the WT and KO neurons on the same coverslips to keep the culture conditions and other procedures identical between genotypes . To accomplish this , we crossed Pak1+/- mice to 57BL/6-Tg ( CAG-EGFP ) to obtain four genotypes: Pak1+/+/EGFP+ , Pak1+/+/EGFP- , Pak1-/-/EGFP+ and Pak1-/-/EGFP- . Hippocampi from Pak1+/+/EGFP+ and Pak1-/-/EGFP- pups or from Pak1+/+/EGFP- and Pak1-/-/EGFP+ were mixed and plated on the same coverslips and the genotype of the neurons was identified based on the presence or absence of EGFP . The procedures for hippocampal culture and recordings were described previously ( Meng et al . , 2002; Zhou et al . , 2011 ) . Briefly The hippocampi from two pups with suitable genotypes ( as described above ) from the same litter were dissected and subjected to trypsinization ( 0 . 25% at 37°C , 15–20 min ) , centrifugation ( 1200 g , 3 min ) and resuspension in maintenance medium containing Neurobasal A , 0 . 5 mM GlutaMax , B27 and 1% penicilin , before being placed on 24-well plate with poly-D-lysine coated glass coverslips . The maintenance medium was half replaced by fresh medium every 4 days . At 12-15DIV coverslips were transferred to a recording chamber containing ( in mM ) : 120 NaCl , 3 KCl , 25 HEPES , 25 Glucose , 1 . 2 MgCl2 , and 2 . 0 CaCl2 ( pH 7 . 2–7 . 4 ) ( 280 mOsm ) , and whole cell recordings were made as described above . GABA currents were evoked by 1 mM GABA puff ( 100 ms ) delivered to the cell body through a glass electrode using the pneumatic picopump PV830 ( WPI ) . Standard methods for extraction and analysis of protein lysates were followed ( Meng et al . , 2002; Zhou et al . , 2011 ) . Briefly , the brain tissues were dissected quickly in ice-cold 0 . 1 M PBS and transferred to a homogenizer containing ice-cold RIPA lysis buffer ( Beyotime ) with 0 . 5% protease inhibitor cocktail ( Roche ) and lysed for 45 min at 4°C . Debris was excluded by centrifugation at 15 , 000 g for 10 min ( 4°C ) . For synaptosomal fractions , the protein lysate was first processed by the synaptic protein extraction reagents ( Thermo ) , followed by centrifugations at 1200 g for 10 min to collect the supernatant and additional centrifugations at 15 , 000 g for 20 min to collect the pellet to be resuspended in RIPA lysis buffer . Proteins were separated on a SDS-PAGE ployacrylamide gel and electrotransfered to a PVDF filter . Filters were then blocked with 5% dry milk TBST ( 20 mM Tris base , 9% NaCl , 1% Tween-20 , pH 7 . 6 ) and incubated overnight at 4°C with appropriate primary antibodies in TBST . After washing and incubation with appropriate secondary antibodies , filters were developed using enhanced chemiluminescence ( Thermo ) method of detection and analyzed using the AlphaEaseFC software as per manufacturer’s instruction . Protein loading was further controlled by normalizing each tested protein with actin , α/β-tubulin or GAPDH immunoreactivity on the same blot . Primary antibodies included: anti-PAK1 ( 1:1000 , CST , rabbit ) , anti-GAD2 ( 1:1000 , CST , rabbit ) , anti-COX-2 ( 1:3000 , CST , rabbit ) , anti-Actin ( 1:2000 , CST , rabbit ) , anti-DGLα ( 1:1000 , CST , rabbit ) and anti-α/β tubulin ( 1:3000 , CST , rabbit ) , anti-CB1R ( 1:1000 , Proteintech , rabbit ) , anti-gephyrin ( 1:1000; BD , mouse ) , anti-GAPDH ( 1:1000; Bioworld , rabbit ) , anti-MGL ( 1:1000; Proteintech , rabbit ) , and anti-FAAH ( 1:1000; Proteintech , rabbit ) . n represents the number of independent experiments ( i . e . samples from separate mice and tested independently on Western blots ) . Brain regions underwent a lipid extraction process as previously described ( Qi et al . , 2015 ) . In brief , tissue samples were weighed and placed in borosilicate glass culture tubes containing 2 ml of acetonitrile with 5 pmol of [2H8] AEA and 5 nmol of [2H8] 2-AG for extraction . These samples were homogenized with a glass rod , sonicated for 30 min , incubated overnight at -20°C to precipitate proteins , then centrifuged at 1500 g for 5 min to remove particulates . Supernatants were removed to a new glass culture tube and evaporated to dryness under N2 gas , re-suspended in 300 µl of methanol to recapture any lipids adhering to the tube and re-dried again under N2 gas . The final lipid extracts were suspended in 200 µl of methanol and stored at -80°C until analysis . AEA and 2-AG contents within lipid extracts were determined using isotope-dilution , liquid chromatography-tandem mass spectrometry ( LC-MS/MS ) as previously described ( Qi et al . , 2015 ) . n in the summary graphs of Figure 7a , b represents the number of mice . The procedure for brain processing and immunohistochemistry were described previously ( Meng et al . , 2005; Meng et al . , 2002 ) . Briefly , mice were anesthetized by 10% Chloral hydrate , subjected to cardiac perfusion with 0 . 1 M phosphate-buffered saline ( PBS ) , followed by 4% paraformaldehyde ( PFA , in PBS ) . The brain was then dissected and transferred to 4% PFA for additional 24 hrs , and then to 30% sucrose solution till it was saturated . The brain was enbeded in Tissue-Tek OCT . compound and frozen by liquid nitrogen before being sliced to 25 μm coronal crystat sections ( Leica CM1950 ) . The brain sections were transferred to a glass slide coated with poly-D-lysine for immunostaining . Sections were permeabilized by 0 . 25% TritonX-100 in 0 . 1 M PBS ( PBT ) for 30 min , blocked with 10% fetal bovine serum ( FBS ) for 1 hr , and incubated with primary antibodies overnight at 4°C , and then appropriate secondary antibodies 2 hrs at 37°C . Primary antibodies used included: anti-GABA ( 1:200 , Sigma , rabbit ) , anti-VGAT ( 1:200 , CST , rabbit ) , and anti-gephyrin ( 1:200 , BD , mouse ) . Cell nucleus was marked with 4 , 6-diamidino-2-phenylindole ( DAPI , Cayman Chemical ) . The stained coverslips were mounted using DAKO mounting medium for image collections . Confocal images were obtained on Zeiss LSM 700 . For each section , approximately 400 μm width×1200 μm depth of equivalent cortical and hippocampal areas were analyzed to estimate the number of GABAgeric neurons and synapses . For cortical superficial layers , an area of 400 μm width×320 μm depth per section was used . Measurements were performed using Zeiss AimImage Browser software . n represents the number of brain sections and the number of mice for each genotype was no less than 3 . Hippocampal low-density neuronal cultures were prepared from postnatal day 1 of PAK1 KO and WT littermates as described above ( Meng et al . , 2002; Zhou et al . , 2011 ) . At 17–18 DIV , culture medium was quickly replaced with ice-cold 4% paraformaldehyde + 4% sucrose for 20 min and permeabilized with 0 . 25% TritonX-100 for additional 20 min . Cells were then blocked with 3% donkey serum and 3% BSA in PBS for 1 hr , incubated with primary antibodies overnight at 4°C followed by appropriate secondary antibodies for 1 hr at room temperature . After extensive washing with PBS , coverslips were mounted using ProLong Antifade mounting medium ( Invitrogen ) for image collections . The primary antibodies ( 1:750 dilution ) used for immunostaining were anti-PAK1 ( CST , rabbit ) , anti-COX-2 ( CST , rabbit ) , anti-gephyrin ( BD , mouse ) , anti-PSD-95 ( Millipore , mouse ) and anti-COX-2 ( Santa Cruz , mouse ) . Secondary antibodies ( 1:1000 dilution ) were: donkey anti-mouse IgG ( H+L ) Alexa Fluor 488 ( Invitrogen ) and donkey anti-rabbit IgG ( H+L ) Alexa Fluor 546 ( Invitrogen ) . Confocal images were obtained on Zeiss LSM 700 at 2048 × 2048 pixels using Zeiss 63× ( NA 1 . 4 ) objective under the same settings and configurations within each experiment . ImageJ ( NIH ) software was used for measurements of total fluorescence intensity and puncta staining . 10x30 μm sections of primary dendrites were randomly taken for puncta ( with an area of greater than 0 . 1 μm2 ) counting . All images were analyzed by experimenters blind to the treatment or genotype of the images . For each experiment , 3 independent cultures from three animals were used for analysis . All the averaged data were reported as mean ± SEM and statistically evaluated by one-way ANOVA , two-way ANOVA or repeated measures two-way ANOVA , wherever appropriate , followed by post-hoc t-tests . p<0 . 05 was considered to be significant and indicated with p<0 . 05 , p<0 . 01 or *p<0 . 001 in the summary graphs . n represented the number of cells/slices or independent experiments and was used for calculating the degree of freedom . The number of animals was also indicated by the number in the bracket following each n . The statistical data for key summary graphs were provided in the Source data .	Brain cells communicate by sending chemical signals that activate or excite neighbouring cells . However , too much signalling can be harmful . As such the brain has systems in place to inhibit brain signals , and healthy brain activity relies striking a proper balance between excitation and inhibition . In some brain mental health conditions , like autism or schizophrenia , the balance is skewed which has an impact on the brain’s activity . A chemical produced by brain cells called endocannabinoid helps maintain the appropriate balance in brain excitation and inhibition . Endocannabinoid is similar to a chemical found in cannabis , but little is known about how it works and which proteins interact with endocannabinoid . A family of proteins called p21-activated kinases ( PAKs ) has been implicated in autism and other disorders like Huntingtin disease and Alzheimer disease , but it is not fully understood how these proteins interact with endocannabinoid . Now , Xia , Zhou et al . show that one member of this protein family called PAK1 plays a key role in controlling endocannabinoid activity . The experiments showed that mice genetically engineered to lack the PAK1 protein have higher levels of endocannabinoids and , as a consequence , the chemical signals that inhibit brain cells are affected more . The experiments also revealed that PAK1 does not interact directly with endocannabinoids . Instead PAK1 boosts levels of another protein called COX-2 and reduces levels of a molecule called anandamide , which together restrict endocannabinoid’s inhibitory effects . Scientists are currently interested in developing drugs that target the endocannabinoids and their regulators in the brain as a way to treat anxiety , pain and sleep problems . Drugs that block PAK1 are already being studied . Future studies are needed to determine if such PAK1-targeting drugs could be useful for restoring excitatory and inhibitory balance in brain diseases or for treating other diseases involving the PAK proteins .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	p21-activated kinase 1 restricts tonic endocannabinoid signaling in the hippocampus
Pore-blocking toxins inhibit voltage-dependent K+ channels ( Kv channels ) by plugging the ion-conduction pathway . We have solved the crystal structure of paddle chimera , a Kv channel in complex with charybdotoxin ( CTX ) , a pore-blocking toxin . The toxin binds to the extracellular pore entryway without producing discernable alteration of the selectivity filter structure and is oriented to project its Lys27 into the pore . The most extracellular K+ binding site ( S1 ) is devoid of K+ electron-density when wild-type CTX is bound , but K+ density is present to some extent in a Lys27Met mutant . In crystals with Cs+ replacing K+ , S1 electron-density is present even in the presence of Lys27 , a finding compatible with the differential effects of Cs+ vs K+ on CTX affinity for the channel . Together , these results show that CTX binds to a K+ channel in a lock and key manner and interacts directly with conducting ions inside the selectivity filter . Poisonous animals such as tarantula spiders , green mamba snakes , or the deathstalker scorpion rely on their venom for efficient defense and precapture strategies . These venoms are stored in dedicated glands and are rapidly delivered through specialized apparatus via subcutaneous , intramuscular , or intravenous routes . What is the underlying cause of toxicity of these venoms ? The principal toxic components in all of them are peptidic in nature . These venoms usually contain libraries of hundreds of peptide-based toxins that together encompass a high degree of stereochemical diversity ( Han et al . , 2008; Liang , 2008; Rodriguez de la Vega et al . , 2010 ) . Only a small fraction of these molecules , however , have been pharmacologically characterized thus far . The targets of these toxins are typically a variety of ion channels—voltage-gated Na+ ( Nav ) , K+ ( Kv ) , and Ca2+ ( Cav ) channels , and cell-surface ‘receptor’ ion channels , such as the nicotinic acetylcholine ( Ach ) receptor ( Billen et al . , 2008; King et al . , 2008; Mouhat et al . , 2008; Kasheverov et al . , 2009 ) . The remarkable molecular diversity of these toxins is borne out by the fact that multiple different toxins can target different components of the same ion channel/receptor . However , the end result is alteration of the normal physiology of the ion channel/receptor , thereby eliciting the desired reaction of the venom . Potassium channels ( K+ channels ) , a large and diverse class of ion channels , are targets of a large number of toxins that have been characterized up to now ( Carbone et al . , 1982; Miller et al . , 1985; Galvez et al . , 1990; Garcia et al . , 1994; Swartz and MacKinnon , 1995 ) . Most K+ channels are tetrameric in architecture—four pore domains together form an ion-conduction pathway through the membrane ( Figure 1A; MacKinnon , 1991; Doyle et al . , 1998 ) . In addition , in the voltage-gated family of K+ channels ( Kv channels ) , each channel monomer has attached onto the N-terminal end of the pore domain a transmembrane voltage sensor domain that senses the transmembrane voltage difference ( Figure 1A; Papazian et al . , 1987; Jiang et al . , 2003 ) . Kv channels are targeted by toxins primarily at two distinct sites—pore-blocking toxins that bind at the extracellular mouth of pore domains and gating-modifier toxins that bind to voltage sensor domains ( Figure 1A; MacKinnon et al . , 1990; Goldstein et al . , 1994; Swartz and MacKinnon , 1997 ) . 10 . 7554/eLife . 00594 . 003Figure 1 . Structures of the channel used in this study and of representative scorpion toxins , including the one used in this study . ( A ) Side view showing two diagonal subunits of the paddle chimera channel in complex with the auxiliary β-subunit shown in ribbon trace ( PDB ID 2R9R; Long et al . , 2007 ) . The pore domain of paddle chimera is colored green , and the voltage sensor domain and the linker between the voltage sensor and the pore are colored in yellow . The cytoplasmic T1 domain and the auxiliary β-subunit are shown in blue . The K+ ions in the selectivity filter are shown as cyan spheres . The area corresponding to the membrane is shaded in light gray . The channel forming α-subunit is indicated . Each α-subunit forms a complex with a β-subunit and four such α- and β-heterodimers make up the tetramer . Note that because the voltage sensors arrange around the pore domains in a domain-swapped fashion , the voltage sensor domains and the pore domains shown in the figure belong to different molecules of the tetrameric channel . Sites on the channel for binding the pore-blocking toxins and gating-modifier toxins are shown with arrows . ( B ) The lowest energy NMR structures of representative scorpion toxins with activity on Kv channels—charybdotoxin ( CTX; PDB ID 2CRD; Bontems et al . , 1991 ) , Agitoxin2 ( AgTx2; PDB ID 1AGT; Krezel et al . , 1995 ) and Noxiustoxin ( PDB ID 1SXM; Dauplais et al . , 1995 ) are shown in blue ribbon trace . A critical lysine ( Lys27 for CTX and AgTx2 and Lys28 for Noxiustoxin ) that is conserved in this family of toxins and the conserved cysteines are shown in ball and stick rendition and are colored by atoms . DOI: http://dx . doi . org/10 . 7554/eLife . 00594 . 003 Scorpion venom specifically has been an abundant source of pore-blocking toxins for K+ channels . These are small peptides , typically ranging from 30 to 40 residues in length , held together by three or four disulfide bonds in a rigid architecture ( Figure 1B; Bontems et al . , 1991; Johnson and Sugg , 1992; Fernandez et al . , 1994 ) . Pore-blocking toxins have profoundly impacted research in the K+ channel field primarily in two ways . First , they have enabled purification of specific novel K+ channels such as the BK channel , a Ca2+ and voltage-gated K+ channel ( Garcia et al . , 1997 ) . Second , they provided our first knowledge about channel subunit stoichiometry and the shape of the extracellular K+ pore entryway at a time when no three-dimensional structure was available for any ion channel ( MacKinnon , 1991; Goldstein et al . , 1994; Gross et al . , 1994; Stampe et al . , 1994; Hidalgo and MacKinnon , 1995; Naranjo and Miller , 1996; Ranganathan et al . , 1996 ) . Charybdotoxin ( CTX; Figure 1B ) , a pore-blocking toxin for K+ channels , is a 37-residue peptide isolated from the venom of the scorpion Leiurus quinquestriatus ( Miller et al . , 1985 ) . Early experiments with CTX inhibition of the BK channel revealed that CTX binds to the extracellular surface of the channel with a 1:1 channel:toxin stoichiometry , that both the open and closed states of the channel are competent for toxin binding , and that electrostatic interactions play an important role in enhancing the toxin’s affinity ( Anderson et al . , 1988 ) . Furthermore , CTX affinity was found to be voltage dependent , a property later shown to result from the destabilization of the toxin-channel complex by permeant ions entering from the intracellular side ( an effect called ‘trans-enhanced dissociation’; MacKinnon and Miller , 1988 ) . Ions that were unable to traverse the ion conduction pathway did not elicit trans-enhanced dissociation . These observations led to a hypothesis that CTX physically occludes the ion-conduction pathway , and in doing so brings a positive charge on CTX close to a K+ ion-binding site near the extracellular side ( MacKinnon and Miller , 1988 ) . The positive charge was later identified as Lys27 , a residue that is conserved in all members of the CTX-like toxin family ( Figure 2A; Park and Miller , 1992; Goldstein and Miller , 1993 ) . Studies with other members of the CTX toxin family , most extensively , Agitoxin2 ( AgTx2 ) , supported the conclusion that they function in a manner similar to CTX ( Garcia et al . , 1994; Krezel et al . , 1995; Hidalgo and MacKinnon , 1995; Ranganathan et al . , 1996 ) . Most notably , conservation of the toxin shape and the functionally important lysine suggested that they all bind with a similar orientation on the K+ channel and inhibit through a common mechanism , whereby a lysine amino group functions as a K+ ion mimic to block the pore ( Miller , 1995; Figure 1B ) . 10 . 7554/eLife . 00594 . 004Figure 2 . Sequence alignments of toxins and pore regions of K+ channels . ( A ) Sequence alignments of five toxins belonging to the CTX family of K+ channel toxins . The conserved cysteines that form disulfide bonds are shown in blue and the conserved lysine that competes with K+ is shown in red . ( B ) Sequence alignment of the pore regions of selected members of the Shaker family of channels and KcsA–paddle chimera , rat Kv2 . 1 ( GI:24418849 ) , rat Kv1 . 2 ( GI:1235594 ) , human Kv1 . 1 ( GI:119395748 ) , human Kv 1 . 3 ( GI:88758565 ) , human Kv1 . 5 ( GI:25952087 ) , Shaker Kv ( GI:13432103 ) , and KcsA ( GI:61226909 ) . The sequence of the selectivity filter is shown in red . DOI: http://dx . doi . org/10 . 7554/eLife . 00594 . 004 Double-mutant cycle studies between toxins ( CTX and AgTx2 ) and the Shaker K+ channel provided numerous pairwise restraints for mapping the extracellular-facing pore surface ( Goldstein et al . , 1994; Gross et al . , 1994; Stampe et al . , 1994; Hidalgo and MacKinnon , 1995; Naranjo and Miller , 1996; Ranganathan et al . , 1996 ) . NMR-derived models using the KcsA K+ channel also provided valuable structural data ( Takeuchi et al . , 2003; Yu et al . , 2005 ) . However , models derived from the double-mutant cycle and NMR data were largely silent as to the influence of toxin on the conducting ions . Here , we have used x-ray crystallography to determine the structure of a complex between CTX and the paddle chimera , a mutant of the Kv1 . 2 K+ channel from rat brain , with particular focus on the influence of toxin on the selectivity filter structure and distribution of ions in the pore ( Figure 2B; Alabi et al . , 2007; Long et al . , 2007 ) . Electrophysiological studies of paddle chimera in planar lipid bilayers had revealed that CTX inhibits paddle chimera with high affinity ( ∼20 nM Kd; Tao and MacKinnon , 2008 ) . We crystallized the complex of paddle chimera with CTX by mixing together separately purified preparations of the channel and the toxin , and setting up cocrystallization trials . The highest resolution data were obtained from the complex of paddle chimera with the selenomethionine derivative of CTX . We used this dataset to solve the structure of the toxin complex of paddle chimera to 2 . 5 Å resolution . The architecture of the paddle chimera channel typifies the family of eukaryotic Kv channels such as Shaker , with four pore domains together forming the ion-conduction pathway through the membrane and four voltage sensor domains surrounding the pore ( Figure 1A; Long et al . , 2007 ) . The voltage-sensors are linked to the cytoplasmic T1 domains that form a cytosolic tetrameric interface . Each channel-forming α-subunit is associated with an accessory β-subunit on the cytoplasmic side . The toxin channel complex shares the same overall architecture , with the fourfold symmetry axis of the channel tetramer coinciding with the fourfold crystallographic symmetry axis . There are two molecules of the α- and β-heterodimeric complex in the asymmetric unit ( Figure 3 ) . We refer to these as molecule A ( top ) and molecule B ( bottom ) . Consequently , the symmetry operations generate two distinct tetramers in the lattice . For the toxin-channel complex , an initial omit map without any toxin in the model clearly shows electron density corresponding to the toxin at the pore entryway ( Figure 4A ) . However , the electron density for the toxin bound to molecule A , henceforth referred to as toxin A , was clearer and so we used this map for building a model of the toxin-channel complex . 10 . 7554/eLife . 00594 . 005Figure 3 . Lattice structure of paddle chimera–CTX complex . The asymmetric unit contains two independent molecules of channel forming α-subunits each in complex with an auxiliary β-subunit ( see Figure 1A ) . They are called molecule A , shown in stick rendition in different shades of cream; and molecule B , shown in stick rendition in different shades of blue . The toxin bound to molecule A was modeled and is shown in red . The outlines of a unit cell are shown in green . DOI: http://dx . doi . org/10 . 7554/eLife . 00594 . 00510 . 7554/eLife . 00594 . 006Figure 4 . Initial placement of the CTX molecule in the toxin-channel complex and heavy atom derivatives used for subsequently improving the accuracy of placement . ( A ) Stereoview showing the pore domains from two diagonal subunits ( molecule A ) of paddle chimera ( in the toxin-channel complex ) in green α carbon trace and a toxin-omit weighted 2Fo − Fc electron density map in wire mesh at 0 . 8 σ contour level . The initial placement of the CTX molecule using the NMR structure ( PDB ID 2CRD; Bontems et al . , 1991 ) is shown in blue α carbon trace within the omit map . ( B ) Chemical structures of the heavy atom derivatives of CTX used in this study are shown schematically . The NMR structure ( PDB ID 2CRD; Bontems et al . , 1991 ) is used to depict the rest of the molecule in blue α carbon trace . The heavy atoms are shown in color , purple for iodine and copper for selenium . DOI: http://dx . doi . org/10 . 7554/eLife . 00594 . 00610 . 7554/eLife . 00594 . 007Figure 4—figure supplement 1 . Inhibition of paddle chimera by the different heavy atom derivatives of CTX used in this study . The experiments were carried out using a planar bilayer system ( see ‘Materials and methods’ ) . The current traces before addition of the CTX derivative are shown in black , and the current traces after adding 100 nM CTX derivative are shown in red . Voltage pulses used are shown schematically on the right of each panel . ( A ) Wild-type CTX . ( B ) Selenomethionine derivative of CTX . ( C ) Diselenide mutant of Cys7-Cys28 . ( D ) 4-Iodophenylalanine mutant of Tyr14 . DOI: http://dx . doi . org/10 . 7554/eLife . 00594 . 007 The consequences of an asymmetric toxin , binding to a fourfold symmetric channel , are that the toxin can bind to the tetrameric channel in four statistically distinguishable but structurally and energetically equivalent orientations . In this case , these orientations are related by the fourfold symmetry axis . Thus , the observed electron density map is a superposition of the electron densities for four such individual orientations of the toxin ( Figure 4A ) . Moreover , since in the absence of any external constraints , each orientation is populated with one-fourth occupancy , the electron density for each orientation is inherently weak . Certain secondary structural features in the map were discernible and allowed approximate placement of a CTX molecule , whose structure was determined previously using NMR ( Figures 1B , 4A; Bontems et al . , 1991 ) . It was apparent from the outset that this initial model would require additional data to achieve a reasonable level of accuracy . To obtain the additional data , we incorporated electron-dense marker atoms individually , at several sites on the toxin and purified each heavy atom-modified toxin either by peptide synthesis followed by refolding and purification or by overexpression in Escherichia coli and following literature procedures ( Park et al . , 1991 ) . We used three different heavy atom markers—replacement of a disulfide by a diselenide , 4-iodophenylalanine , and selenomethionine ( Figure 4B ) . We tested each of these derivatives in a planar bilayer system , and they efficiently blocked the paddle chimera channel ( Figure 4—figure supplement 1 ) . We then crystallized each toxin derivative with paddle chimera and collected single-wavelength anomalous diffraction data at an appropriate wavelength for each derivative . For each derivative , an anomalous difference electron density map showed four heavy-atom peaks , corresponding to the four orientations of the toxin ( Figure 5A ) . We collected datasets from crystals with each derivative at a distinct site on the toxin , a total of three datasets ( Table 1 ) . We next used our highest resolution dataset ( of all the different toxin derivative complexes ) and roughly placed the toxin in the omit electron density map ( Figure 4A ) . We used this approximate initial placement of the toxin to determine which one of the four symmetry-equivalent peaks in the anomalous difference map ( for each marker ) corresponded to which orientation of our initial toxin placement . This provided a set of three experimental constraints corresponding to the three individual heavy atom peak positions ( for three different markers; Figure 5B ) . The coordinates of the corresponding three sites from the known NMR structure of the toxin ( Bontems et al . , 1991 ) were used as three reference constraints . We then used RMSD-based superposition to minimize the sum of distances between the peaks in the map and the predicted positions on the toxin ( Figure 5C; final RMSD 1 . 6 Å ) . This procedure yielded a constrained placement of the toxin , which was subsequently refined by rigid-body , coordinate-based B-factor refinement in CNS with manual adjustments , where appropriate ( Figure 5D ) . 10 . 7554/eLife . 00594 . 008Figure 5 . Improvement of the initial placement of the toxin . ( A ) Close-up stereoview showing part of the pore domains from two diagonal subunits ( molecule A ) of paddle chimera . Peak positions from the anomalous electron density maps for the three derivatives ( see Figure 4B ) are shown and colored as follows—purple ( 4-iodophenylalanine mutant at position 14 ) , orange ( diselenide mutant of Cys7-Cys28 ) , and yellow ( SeMet mutant at position 29 ) . Note that the four peaks correspond to the four positions of the toxin . In addition , the peak on the symmetry axis in the anomalous electron density map of the 4-iodophenylalanine mutant likely derives from reinforcement of noise peaks that are very close to the symmetry axis . ( B ) Stereoview showing initial placement of CTX in the omit map using the NMR structure ( PDB ID 2CRD; Bontems et al . , 1991 ) , in blue α carbon trace ( same as in Figure 4A ) . The heavy atoms in the depicted orientation are shown as oversized spheres with the corresponding peak positions ( the closest of the four shown in Figure 5A ) in the anomalous electron density maps in wire mesh . Color-coding of the maps are the same as in Figure 5A . A dummy atom has been placed to indicate the position of each peak . ( C ) The toxin molecule is shown in purple α carbon trace , after RMSD superposition of the heavy atom positions in the structure onto experimental peak positions as illustrated by the dummy atoms in Figure 5B . The initial placement of the toxin as shown in Figure 5B is also shown in blue . ( D ) The final refined model of the toxin after crystallographic refinement ( please see text and ‘Materials and methods’ ) is shown in orange α carbon trace together with the model after initial RMSD superposition in purple and a weighted 2Fo − Fc electron density map in wire mesh at 1σ contour level . DOI: http://dx . doi . org/10 . 7554/eLife . 00594 . 00810 . 7554/eLife . 00594 . 009Figure 5—figure supplement 1 . Validation of the model for the toxin . Shown is an RMSD superposition of molecule A with the bound CTX onto molecule B . The main chain atoms of the pore domains were used for the RMSD superposition . Shown are the pore domains ( two diagonal subunits ) from molecule A ( in green ) , the pore domains ( two diagonal subunits ) from molecule B ( in blue ) , and the CTX bound to molecule A in orange α carbon trace . The side chain of Met29 of CTX is shown in stick rendition , and the sulfur is shown as an oversized sphere . The peaks from the anomalous electron density map from the complex of the selenomethionine derivative of CTX with paddle chimera are shown in wire mesh . Note that the position of the sulfur is close to the expected position of selenium in the selenomethionine derivative of CTX . In addition to the four peaks corresponding to the four orientations of CTX , there is a peak on the fourfold symmetry axis , likely resulting from the reinforcement of noise in the electron density map . DOI: http://dx . doi . org/10 . 7554/eLife . 00594 . 00910 . 7554/eLife . 00594 . 010Table 1 . Crystallographic data and refinement statistics tableDOI: http://dx . doi . org/10 . 7554/eLife . 00594 . 010Data collection DatasetPaddle chimera–SeMet CTXPaddle chimera–Lys27Met CTXPaddle chimera–CTX in CsCl Space groupP4212P4212P4212 Cell constants ( Å ) a = b = 144 . 200; c = 283 . 608a = b = 144 . 842; c = 283 . 938a = b = 145 . 434; c = 285 . 591α = β = γ = 90°α = β = γ = 90°α = β = γ = 90° SourceBNL X29BNL X29BNL X29 Wavelength ( Å ) 0 . 97911 . 0751 . 075 Resolution ( Å ) 2 . 52 . 542 . 56 Unique reflections99 , 90797 , 91398 , 762 <I>/<σI>29 . 4 ( 2 . 59–2 . 50/1 . 9 ) 18 . 4 ( 2 . 58–2 . 54/2 . 2 ) 13 ( 2 . 60–2 . 56/1 . 04 ) Redundancy6 . 9 ( 2 . 59–2 . 50/2 . 3 ) 7 . 3 ( 2 . 58–2 . 54/6 . 0 ) 7 . 1 ( 2 . 60–2 . 56/5 . 3 ) Completeness ( % ) 96 . 1 ( 2 . 59–2 . 50/71 . 1 ) 96 . 7 ( 2 . 58–2 . 54/32 . 8 ) 99 . 3 ( 2 . 60–2 . 56/87 . 6 ) Rmerge ( % ) 5 . 9 ( 2 . 59–2 . 50/39 . 8 ) 8 . 6 ( 2 . 58–2 . 54/73 . 2 ) 9 . 5 ( 2 . 60–2 . 56/>100 ) Model Refinement Resolution ( Å ) 50–2 . 550–2 . 5450–2 . 56 Reflections ( free set ) 96 , 658 ( 4605 ) 93 , 308 ( 4406 ) 91 , 104 ( 4292 ) Rwork/Rfree ( % ) 21 . 1/23 . 621 . 0/23 . 423 . 7/26 . 2 RMSD bond lengths ( Å ) 0 . 0060 . 0070 . 007 RMSD bond angles ( ° ) 1 . 091 . 2491 . 299 Mean B-factor ( Å2 ) 73 . 0975 . 0772 . 11 Ramachandran plot Allowed ( % ) 99 . 599 . 499 . 4 Disallowed ( % ) 0 . 50 . 60 . 6*Numbers in parentheses represent the resolution range of the highest resolution shell followed by the value of the parameter for the highest resolution shell . As noted previously , the density for toxin A in the omit map is better defined than for toxin B , and thus we chose toxin A density for initial placement and for building and refining the model of the toxin . However , when we superimposed the toxin-bound channel A onto channel B using the pore domains of channels A and B for the superposition , we observed a very reasonable model for toxin B , which agrees well with the omit electron density map for toxin B and places the side-chain of Met29 very near to one of the four experimentally observed heavy atom peaks for toxin B in the selenomethionine dataset ( Figure 5—figure supplement 1 ) . Since channels A and B are independent molecules in the asymmetric unit and only the heavy atom peak positions for toxin A were used for the initial placement of the toxin , this provided further validation of our model and boosted our confidence in the placement of the toxin . We did not include a model for toxin B in our final model since the overall electron density is not as well defined as for toxin A and building toxin B caused a small increase in Rfree . Superposition of the channel in the toxin complex onto the channel in the toxin-free structure ( Long et al . , 2007 ) shows that the channel undergoes no discernible structural changes ( Figure 6A , F ) ; RMSD 0 . 33 Å , residues 321–414 , main chain atoms , molecule A; RMSD 0 . 16 Å , residues 321–414 , main chain atoms , molecule B . This is consistent with the idea that the toxin fits into the mouth of the channel in a lock and key manner . The oblate-shaped toxin binds asymmetrically to the mouth of the pore domain of the channel such that the wider end is closer to the symmetry axis of the channel than the tapered end ( Figure 6B ) . The helical part of the toxin molecule faces away from the channel and the edge of the toxin formed by the residues 25–29 on a β-strand faces toward the channel . The inherent architecture of this class of toxins is such that the three disulfide bonds that hold the folded toxin together are the main buried components in the structure and most of the side chains are displayed on the surface of the toxin . These side chains are in a position to engage into a number of different kinds of interactions with the channel molecule ( Figures 7A , B ) . Closer to the fourfold symmetry axis of the channel , the aromatic ring of Tyr36 is positioned to pack simultaneously against Asp375 and Val377 of one subunit and Met29 is able to pack against Asp375 of an adjacent subunit ( Figure 7A , B ) . Closer to the periphery , the side chain of Arg25 is within close proximity of Gln353 , and the peptide backbone near Thr8-Thr9 is held against Gln353 of another subunit ( Figure 7A , B ) . There are also residues that should be involved in long-range electrostatic interactions , that is , Arg25 is within 5 . 5 Å of Asp359 . The guanidium headgroup of Arg25 could also , in principle , engage in electrostatic interactions with an ordered lipid that is present in the structure . Although the density of the lipid headgroup is not clear in this case , it is likely that the headgroup will be placed close enough for the Arg25 to make electrostatic contact with it . Arg34 is another residue that approaches the channel closely enough to make electrostatic interactions , that is , with the carbonyl oxygen of Asp375 . Arg34 is also within H-bonding distance of Gln353 and within long-range electrostatic contact of Asp375 . Asn30 is another residue that appears to approach close to Asp375 to enable a weak H-bonding interaction . 10 . 7554/eLife . 00594 . 011Figure 6 . Structure of the toxin-channel complex . ( A ) Side view showing pore domains from two diagonal subunits from paddle chimera ( yellow; PDB ID 2 R9R; Long et al . , 2007 ) and the toxin-channel complex ( green ) in α carbon trace . They have been superimposed by RMSD superposition of the main chain atoms from residues Met321-Thr414 . ( B ) The tetrameric pore domain in the toxin-channel complex is shown in green ribbon trace from an extracellular view looking into the molecule . The bound CTX is shown in surface rendition in orange . ( C ) Side view of the selectivity filter ( two diagonal subunits , molecule B ) from the paddle chimera structure shown in stick rendition with the K+ ions shown as cyan spheres . Sites S1 through S4 in the selectivity filter are labeled ( labels on left side ) in cyan . A weighted 2Fo − Fc electron density map contoured at 3σ is shown in wire mesh . ( D ) Side view of the selectivity filter ( two diagonal subunits , molecule B ) from the toxin-channel complex is shown in stick rendition with the K+ ions shown as cyan spheres . A weighted 2Fo − Fc electron density map contoured at 3σ is shown in wire mesh . ( E ) The selectivity filter from the toxin-channel complex is shown in stick rendition with the K+ ions shown as cyan spheres . Also shown is the bound CTX molecule in orange ribbon trace and the side chain of the Lys27 residue in stick rendition . Close contact between the amino headgroup and the carbonyl oxygen in the selectivity filter are shown in dotted lines . ( F ) RMSD superposed structures of the selectivity filter regions ( same regions as shown in Figures 6C , D ) of the paddle chimera ( pale green ) and the toxin-channel complex ( green ) shown in stick rendition . The superposition was done using the main chain atoms from residues Met321-Thr414 . The K+ ions in the paddle chimera structure are shown in light blue and those in the toxin-channel complex in cyan . DOI: http://dx . doi . org/10 . 7554/eLife . 00594 . 01110 . 7554/eLife . 00594 . 012Figure 6—figure supplement 1 . Electron density for the side chain of Lys27 of CTX in the paddle chimera–CTX complex . Shown in stereo is a close-up view of the top part of the selectivity filters of two diagonal subunits of paddle chimera in the paddle chimera–CTX complex ( molecule A ) in stick rendition . The K+ ions in the selectivity filter are shown as cyan spheres . Also shown are the four symmetry-related orientations of the toxin molecule in purple , orange , teal , and green in α carbon traces . Only the lower parts of the toxin that are close to the channel are visible here . The side chains of Lys27 of CTX for the four corresponding orientations are shown in orange stick rendition . Also shown in wire mesh is a simulated annealing composite omit map at 1σ contour level , contoured on the atoms of the side chains of the four orientations of Lys27 . DOI: http://dx . doi . org/10 . 7554/eLife . 00594 . 01210 . 7554/eLife . 00594 . 013Figure 7 . Interactions between the bound CTX and the channel . ( A ) Stereoview showing the CTX receptor region of paddle chimera from the toxin-channel complex shown in stick rendition together with the top parts of the selectivity filter . The bound CTX is shown in orange ribbon trace . Side chains of selected residues of the toxin are shown in purple and are labeled . ( B ) Shows a view orthogonal to that in ( A ) . Consequently the receptor regions shown in ( B ) are from the two diagonal subunits of paddle chimera that are not shown in ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00594 . 013 The most striking aspect of the toxin-channel complex concerns the distribution of K+ ions in the selectivity filter of the channel . In the structure of the paddle chimera , as well as other high-resolution K+ channel structures , the selectivity filter contains four distinct ion-binding sites or positions , S1 through S4 , S1 being the most extracellular ( Figure 6C; Zhou et al . , 2001; Long et al . , 2007; Nishida et al . , 2007 ) . From analyses of high-resolution diffraction data on KcsA , the prototypical K+ channel pore , it was inferred that during conduction , these ion-binding sites are occupied alternately in what are referred to as 1 , 3 and 2 , 4 configurations ( Morais-Cabral et al . , 2001 ) . In accordance , the occupancy of each site was experimentally determined to be roughly 0 . 5 ( Zhou and MacKinnon , 2003 ) . In the toxin-channel complex , only sites S2 through S4 have discernible electron density ( Figure 6D ) . The top ion-binding site appears empty . It is important to note here that this observation holds strictly true for both channel molecules in the asymmetric unit . We have several datasets for all the different derivative toxin complexes , and this observation holds true for all of them as well , in both channel molecules A and B , in the asymmetric unit . Why is the distribution of ions dramatically different in the toxin-bound complex ? The toxin binds at the mouth of the pore positioned in such a manner to project the side chain of Lys27 straight into the pore , allowing the amino group to approach the top of site S1 ( Figure 6E and Figure 6—figure supplement 1 ) . Thus , Lys27 is within the range to make hydrogen-bonding interactions with all four carbonyl oxygen atoms that would otherwise constitute the top-half layer of coordinating ligands for a K+ at site S1 ( Figure 6E; Zhou et al . , 2001 ) . We suspect therefore two reasons why K+ is disfavored at site S1 . First , there is electrostatic repulsion from the closely placed positively charged amino group , and second , carbonyl oxygen atoms that would otherwise constitute half of the coordination are not fully available for coordination with a K+ at site S1 . The structure thus offers a simple rationale for the altered distribution of ions in the selectivity filter of the toxin-bound channel . The altered ion distribution informs us that the toxin interacts with ions inside the pore . This interaction is compatible with the electrophysiological observation that intracellular ions destabilize extracellular toxin—the trans-enhanced dissociation effect ( MacKinnon and Miller , 1988; Park and Miller , 1992 ) . To further correlate the ion distribution in the crystal with trans-enhanced dissociation in electrophysiology experiments , we crystallized the channel with a mutant toxin , Lys27Met . Miller and coworkers had shown that Lys27 mutants reduce toxin affinity and abolish trans-enhanced dissociation ( Park and Miller , 1992; Goldstein et al . , 1994 ) . Lys27Met CTX indeed inhibits the paddle chimera channel with reduced affinity ( ∼630 nM ) but still forms a complex at the high concentrations of channel ( ∼20 μM ) and toxin ( 60–80 μM ) present in crystallization trials ( Figure 8A , B ) . Electron density is present at S1 in an ion omit map , compatible with the presence of an ion , although the density at S1 is weaker relative to the ion densities at S2–S4 , as if the occupancy at S1 is reduced ( Figure 8C ) . Thus , it appears that the ability of toxin to interact with ions in the pore ( and thus alter the ion distribution ) is directly connected to the ability of ions from the intracellular solution to destabilize toxin on the extracellular side through interaction with Lys27 . A simple mechanistic explanation could be that K+ and toxin—via Lys27—compete for stabilizing interactions at S1 in the selectivity filter . 10 . 7554/eLife . 00594 . 014Figure 8 . Effects of Lys27Met mutation of CTX and Cs+ separately on toxin-channel interactions . ( A ) CTX inhibition . The fraction of unblocked current ( I/Imax , mean ± SEM or range of mean; n = 2–4 ) is graphed as a function of CTX concentration ( in nM ) and fit to the equation I/Imax = 0 . 1 + 0 . 9 × ( 1 + [CTX]/Kd ) −1 . Voltage pulses: holding −110 mV , depolarized to +110 mV , followed by a step back to −110 mV . Paddle chimera–wild-type CTX with KCl on both sides is shown in red , paddle chimera–wild-type CTX with CsCl on both sides is shown in blue and paddle chimera–Lys27Met CTX with KCl on both sides is shown in black . The modified equation accounts for approximately 10% current that is due largely to the channels facing the other side and are not blocked by toxin . ( B ) The pore domains from two diagonal subunits of paddle chimera ( molecule A ) in the Lys27MetCTX–paddle chimera complex are shown in green α carbon trace and a toxin-omit weighted 2Fo − Fc electron density map is shown in wire mesh at 1σ contour level . ( C ) Side view of the selectivity filter ( two diagonal subunits; molecule B ) of the Lys27Met CTX–paddle chimera complex shown in stick rendition with the K+ ions shown as cyan spheres . An ion-omit weighted 2Fo − Fc electron density map is shown in wire mesh at 3 . 2σ contour level . ( D ) The pore domains from two diagonal subunits of paddle chimera ( molecule A ) in CTX–paddle chimera complex in CsCl , are shown in green α carbon trace and a toxin-omit weighted 2Fo − Fc electron density map is shown in wire mesh at 0 . 8σ contour level . ( E ) Side view of the selectivity filter ( two diagonal subunits; molecule B ) of the CTX–paddle chimera complex in CsCl shown in stick rendition with the Cs+ ions shown as yellow spheres . A weighted 2Fo − Fc electron density map is shown in wire mesh at 3σ contour level . DOI: http://dx . doi . org/10 . 7554/eLife . 00594 . 014 The above mechanistic proposal provides motive to wonder what happens when K+ is replaced with Cs+ , because Cs+ in crystal structures of K+ channels binds at only three sites and with unusually high occupancy at S1 ( Zhou and MacKinnon , 2003 ) . A CTX complex with the paddle chimera channel in the presence of Cs+ shows that toxin is bound but that Cs+ adopts its expected distribution ( i . e . , similar to its distribution in KcsA in the absence of toxin ) with an ion at S1 ( Figure 8D , E ) . It is difficult to tell where in the electron density map the amino group of Lys27 resides , but it is clear that Cs+ competes effectively for site S1 , despite the presence of CTX . The mechanism of competition put forth above predicts that CTX should not bind with high affinity in the presence of Cs+ . As shown , this prediction holds: CTX inhibits with a nearly 10-fold reduced affinity in the presence of Cs+ compared to K+ ( Figure 8A ) . We report here the first x-ray structure of a K+ channel bound to a toxin . Many of the original studies on pore-blocking toxins for K+ channels were carried out with eukaryotic voltage-gated K+ channels , such as Shaker and Kv1 . 3 , closely related in sequence to the mutant version of the eukaryotic Kv1 . 2 channel that we employed in our structural studies ( MacKinnon and Miller , 1989; MacKinnon , 1991; Goldstein and Miller , 1992; Stampe et al . , 1992 , 1994; Goldstein et al . , 1994; Gross et al . , 1994; Aiyar et al . , 1995 , 1996; Hidalgo and MacKinnon , 1995; Gross and MacKinnon , 1996; Naini and Miller , 1996; Naranjo and Miller , 1996; Ranganathan et al . , 1996; MacKinnon et al . , 1998 ) . Our first major finding is that we do not see discernible changes in the structure of the channel between the toxin-bound and the toxin-free paddle chimera structures ( Figure 6A , F ) . This , we note , is in contrast to the solid-state NMR ‘structure’ of a KcsA mutant with kaliotoxin , where such rearrangements in the channel were proposed ( Lange et al . , 2006 ) . In this NMR study , chemical shift changes induced by toxin were interpreted as resulting from dihedral angle changes ( i . e . , structural ) ; however , such chemical shift changes could have other origins ( i . e . , electrostatic ) . Moreover , no channel-toxin distance restraints were included in the determination of this solid-state NMR ‘structure’ ( Lange et al . , 2006 ) . In the crystal structure presented here , the good precomplex complementarity between the shape of the channel entryway and the shape of the toxin ( i . e . , the absence of channel structural change upon toxin binding ) suggests a possible explanation for two prominent features of pore-blocking toxins . First , toxins can bind with relatively high affinity to their target channels because binding free energy is not ‘spent’ bringing about a protein conformational change . And second , single mutations in the ‘toxin receptor’ region of the channel can drastically alter the affinity for the channel by disrupting the good fit ( Goldstein et al . , 1994; Garcia et al . , 1997 ) . The relatively static architecture of the K+ channel pore entryway undoubtedly reflects the requirement of a well-ordered selectivity filter structure to select K+ ions . In hindsight , this static toxin receptor on Kv channels lends credence to the idea that was originally proposed for using the pore-blocking toxins as ‘molecular slide calipers’ for gauging , at the resolution of mutagenesis , the shape of the protein surface on the extracellular part of the pore domain ( Goldstein et al . , 1994; Stampe et al . , 1994; Hidalgo and MacKinnon , 1995; Ranganathan et al . , 1996 ) . A large body of data has emerged from these studies on mutagenesis-based electrophysiological measurements of toxin-channel interactions in the Kv channel family , mainly using the toxin-channel pairs Shaker–CTX , Shaker–AgTx2 , and to a lesser extent , Kv1 . 3–CTX ( Goldstein et al . , 1994; Stocker and Miller , 1994; Hidalgo and MacKinnon , 1995; Aiyar et al . , 1995; Naini and Miller , 1996; Naranjo and Miller , 1996; Ranganathan et al . , 1996; Rauer et al . , 2000 ) . We have mapped these data onto our structure . Among these measurements , the data obtained with mutant cycle analyses are more reliable in gauging residue proximities on either side of the toxin-channel interface . The mapping has been done in the following manner: for the channel , we have used a sequence alignment ( Figure 2B ) to match the corresponding residue in Shaker or Kv1 . 3 onto paddle chimera . Data from the studies using CTX are shown in Figure 9A and those using AgTx2 in Figure 9B . For AgTx2 , we used the known structure of AgTx2 and the guidelines in Krezel et al . ( 1995 ) to superimpose AgTx2 onto CTX in the structure of the paddle chimera–CTX complex . A few data points were derived from lysine scanning , which alters the length of the wild-type residue appreciably . For the lysine mutations , for the sake of representation , the corresponding residue in paddle chimera was mutated in silico ( in Coot ) to lysine , and the rotamer of lysine with the least distance between the corresponding toxin residue was chosen . Our structure is overall in excellent agreement with not only the CTX data but also with the AgTx2 data . 10 . 7554/eLife . 00594 . 015Figure 9 . Mapping of toxin-channel interactions reported in literature on the structure of paddle chimera–CTX complex . ( A ) Shown is a stereoview of the bound CTX in surface rendition in the CTX–paddle chimera structure from an intracellular perspective viewing down the fourfold symmetry axis of the channel ( channel not shown ) . Residues in the channel that have been reported in the literature to be proximal to the toxin are represented as green spheres taking the coordinates of the closest atom from the structure of the CTX–paddle chimera complex ( see text ) . They are connected to the corresponding residues in the toxin with a black line . The corresponding residues in CTX are indicated in orange . Note that certain residues on the channel have been reported to be proximal to multiple residues on the toxin and thus they are represented more than once in the map . This figure shows data derived with CTX . ( B ) Shown are the residues in AgTx2 reported in the literature to be proximal to the channel , in the same format as ( A ) . In order to depict AgTx2 , it was superimposed on CTX in the paddle chimera–CTX complex using the published NMR structure of AgTx2 ( PDB ID 1AGT; Krezel et al . , 1995 ) and using the guidelines in Krezel et al . ( 1995 ) . AgTx2 is shown in surface rendition and the residues on the channel proximal to AgTx2 are shown as blue spheres . The corresponding residues in AgTx2 are indicated in red . DOI: http://dx . doi . org/10 . 7554/eLife . 00594 . 015 We wish to discuss two cases of proximity deduced by the mutant cycle analyses and how the structure reveals atomic insights into them . From analyzing mutants of AgTx2 and Shaker , MacKinnon and Ranganathan derived a coupling energy of >3 kT for the Gly10Val ( AgTx2 ) –Phe425Gly ( Shaker ) pair and ∼1 . 5 kT for the Gly10Val ( AgTx2 ) –Thr449Cys ( Shaker ) pair ( Ranganathan et al . , 1996 ) . Phe425 and Thr449 in Shaker map onto Gln353 and Val377 in paddle chimera ( Figure 2B ) . An inspection of the structure reveals clearly that Gly10 is close to Gln353 ( Figure 10A ) , which is consistent with the high coupling energy . Intriguingly , Val377 is not within the first layer of surrounding residues contacted by Gly10 . However , Val377 is within close proximity to Gln353 such that a mutation of Gly10 to valine would incur a clash of Gln353 with Val377 ( Figure 10A ) . It is worthwhile noting that pairwise mutant cycle analysis per se does not distinguish between direct interactions and such interactions mediated by a third residue . However , the relative magnitude of the coupling energies , in hindsight , is consistent with such a mode of interaction . 10 . 7554/eLife . 00594 . 016Figure 10 . Mutant cycle data from Ranganathan et al . ( 1996 ) mapped onto the structure of paddle chimera–CTX complex . The toxin molecule shown in purple is AgTx2 , which was superimposed onto CTX in the structure of the paddle chimera–CTX complex using the NMR structure of AgTx2 ( PDB ID 1AGT; Krezel et al . , 1995 ) . Guidelines in Krezel et al . ( 1995 ) were used for the superposition . ( A ) Close-up view of AgTx2 showing Glycine 10 in ball and stick rendition and the rest of the toxin in α carbon trace . The channel subunit that is most proximal to Gly10 is shown in green α carbon trace . Shown in ball and stick are residues Gln353 and Val377 in paddle chimera , that correspond to Phe425 and Thr449 , respectively , in Shaker using a sequence alignment ( Figure 2B ) . Ranganathan and MacKinnon reported that Gly10 is coupled to Phe425 and Thr449 in Shaker . ( B ) Close-up view of AgTx2 showing the side-chain of Lys27 in ball and stick rendition and the rest of the toxin in α carbon trace . The nearby regions of the selectivity filter from all four subunits of the channel are shown in green α carbon trace . Shown in ball and stick is Tyr373 in all four subunits of paddle chimera , which map onto Tyr445 in Shaker using a sequence alignment ( Figure 2B ) . Ranganthan and MacKinnon reported Tyr445 to be coupled to Lys27 , and this was dependent on the concentration of K+ . DOI: http://dx . doi . org/10 . 7554/eLife . 00594 . 016 Another case worth highlighting is the coupling energy , ∼1 . 5 kT , between the Lys27Met ( AgTx2 ) –Tyr445Phe ( Shaker ) pair . Tyr445 in Shaker maps to Tyr373 in paddle chimera ( Figure 2B; Ranganathan et al . , 1996 ) . Since mutation of Tyr to Phe incurs only a loss of the hydroxyl group , one likely interpretation of this would have been that Lys27 contacts the hydroxyl group of Tyr445 . Intriguingly , Lys27 only contacts the backbone carbonyl oxygen of Tyr445 , a point farthest from the hydroxyl group ( Figure 10B ) . However , since this is a tightly packed part of the protein structure with the hydroxyl group being a part of an interaction network , mutation of the hydroxyl is felt at the backbone carbonyl by the toxin . MacKinnon and Ranganathan also found that the interaction between these two residues is dependent on the K+ concentration . This is also consistent with Lys27 inserting into the selectivity filter and displacing K+ from site S1 . The structure also shows the chemical rationale behind an experimental observation that is decades old concerning the role of a highly conserved lysine in voltage-dependent block by CTX and other pore-blocking toxins in this family . In studies with the BK channel , MacKinnon and Miller ( 1988 ) first observed that permeant ions coming through the channel from the intracellular side enhance the dissociation of a toxin bound to its receptor on the extracellular side . This trans-enhanced dissociation effect of permeant ions on CTX block of K+ channels was subsequently confirmed for Shaker as well ( Goldstein and Miller , 1993 ) . In order to explain this phenomenon , MacKinnon and Miller put forward the hypothesis that the toxin binds at a site close to the ion permeation pathway and places a positive charge close to one of the ion-binding sites in the channel . Additionally , it was also observed that this voltage-dependent and permeant ion–dependent block was nearly completely abolished when Lys27 was mutated to a neutral Asn or Gln residue ( Park and Miller , 1992 ) . Mutation of no other residue on the toxin had the same effect . This implied that the electrostatic interaction between the K+ in the ion-binding site and the toxin was wholly mediated by this single Lys residue , and the structure of the toxin-paddle chimera complex shows exactly why that is so . In the structure of the WT CTX-K+ complex , the top ion-binding site S1 has no density corresponding to K+ ions ( Figure 6C ) and the positively charged amino group is positioned to form contacts with S1 instead ( i . e . , making hydrogen-bonding interactions with the carbonyl oxygen atoms comprising the top layer of coordinating ligands at S1; Figure 6D ) . From the structure , it is clear that only the side chain of Lys27 approaches close enough to the S1 site to exert such an effect ( Figure 6—figure supplement 1 ) . It is worth noting here that the long linear alkyl chain and the tetrahedral disposition of hydrogens around the terminal nitrogen of Lysine make it uniquely and chemically suited for making maximal contacts with the top layer of coordinating carbonyl oxygen atoms at S1 . Because this region is buried within the protein , this is likely to be a highly stabilizing interaction . Even a conservative mutation of Lys27 to Arginine causes ∼1000-fold destabilization of the toxin-channel complex in Shaker ( Goldstein et al . , 1994 ) . The CTX family of K+ channel toxins has remarkable diversity in sequence and binding affinities to specific channel subtypes . It is interesting to note , in light of the present structure , the basic mechanistic principles underlying the mode of action of these toxins . In a strikingly simple but effective strategy , the toxins target a functional aspect common to all the K+ channels—ion conduction . At the structural level , this is executed by presenting the amino group of Lys27 , a highly conserved residue ( Figure 2A and Figure 6—figure supplement 1 ) , to the top ion-binding site in the selectivity filter . This acts as a tethered surrogate cation and effectively plugs the ion-conduction pathway . How is this surrogate cation brought to this site in the first place ? The structures of these miniproteins are held together by rigid disulfide bonds , that are highly conserved in this and other families of small peptide-based toxins . This provides a relatively rigid scaffold . Through evolutionary sampling of intervening less-conserved residues , individual toxins have gained the ability to engage in specific interactions by long-range electrostatic interactions and a few subtype-specific close contacts , thereby ensuring efficient channel block . This is a remarkable example of combinatorial diversity in nature with the constraints of a rigid scaffold and a conserved mechanism to target one of the most important classes of sensory molecules in biology . Scorpion toxins have evolved to fit like a lock and key into the pore entryway of potassium channels , and disrupt ion conduction through presentation of a lysine amino group that competes with potassium in the selectivity filter .	The deadly toxins produced by many creatures , including spiders , snakes , and scorpions , work by blocking the ion channels that are essential for the normal operation of many different types of cells . Ion channels are proteins and , as their name suggests , they allow ions—usually sodium , potassium , or calcium ions—to move in and out of cells . They are especially important for cells that generate or respond to electrical signals , such as neurons and the cells in heart muscle . Ion channels are located in the lipid membranes that surround all cells , and the ions enter or leave the cell via a pore that runs through the channel protein . They can be opened and closed ( or ‘gated’ ) in different ways: some ion channels open and close in response to voltages , whereas others are gated by biomolecules , such as neurotransmitters , that bind to them . Now , Banerjee et al . have used x-ray crystallography to study the structure of the complex that is formed when charybdotoxin ( CTX ) , a toxin that is found in scorpion venom , blocks a voltage-gated potassium channel . Previous studies have shown that CTX binds to the channel on the extracellular side of the pore . Banerjee et al . show that the toxin fits into the entrance to the channel like a key into a lock , which means the toxin is preformed to fit the shape of the channel . The potassium ion channel is made up of four subunits , and the pore contains four ion-binding sites that form a ‘selectivity filter’: it is this filter that ensures that only potassium ions can pass through the channel when it is open . When CTX binds to the channel , a lysine residue poised at a critical position on the toxin is so close to the outermost ion-binding site that it prevents potassium ions binding to the site . The structure determined by Banerjee et al . explains many previous findings , including the fact that ions entering the pore from inside the cell can disrupt the binding between the toxin and the ion channel protein . It remains to be seen if the toxins that target the pore of other types of ion channels work in the same way .	[ "Abstract", "Introduction", "Results", "Discussion" ]	[ "structural", "biology", "and", "molecular", "biophysics", "neuroscience" ]	2013	Structure of a pore-blocking toxin in complex with a eukaryotic voltage-dependent K+ channel
Soluble N-ethylmaleimide-sensitive factor attachment protein receptors ( SNAREs ) are evolutionarily conserved machines that couple their folding/assembly to membrane fusion . However , it is unclear how these processes are regulated and function . To determine these mechanisms , we characterized the folding energy and kinetics of four representative SNARE complexes at a single-molecule level using high-resolution optical tweezers . We found that all SNARE complexes assemble by the same step-wise zippering mechanism: slow N-terminal domain ( NTD ) association , a pause in a force-dependent half-zippered intermediate , and fast C-terminal domain ( CTD ) zippering . The energy release from CTD zippering differs for yeast ( 13 kBT ) and neuronal SNARE complexes ( 27 kBT ) , and is concentrated at the C-terminal part of CTD zippering . Thus , SNARE complexes share a conserved zippering pathway and polarized energy release to efficiently drive membrane fusion , but generate different amounts of zippering energy to regulate fusion kinetics . Soluble N-ethylmaleimide-sensitive factor attachment protein receptor ( SNARE ) -mediated membrane fusion is ubiquitous in eukaryotes and underlies numerous basic processes in humans , including neurotransmission , hormone secretion , and antibody production ( Sollner et al . , 1993; Sudhof and Rothman , 2009; Wickner , 2010; Jahn and Fasshauer , 2012 ) . Malfunction of fusion has been associated with many important diseases such as neurological disorders and diabetes ( Burre et al . , 2010; Stockli et al . , 2011 ) . Consistent with their diverse functions and dysfunctions , these intracellular membrane fusion processes exhibit distinct kinetics and regulation ( Kasai et al . , 2012 ) . For example , fusion of synaptic vesicles occurs within 0 . 2 ms in response to the arrival of an action potential ( Sabatini and Regehr , 1996 ) , whereas vacuole fusion in yeast is constitutive and lasts minutes ( Wickner , 2010 ) . Although these diverse processes have long been identified , it is not fully understood how SNAREs specialize in membrane fusion and become adapted to and regulated for various fusion speeds . SNAREs constitute a large family of proteins with highly conserved modular structures ( Fasshauer et al . , 1998 ) , including 38 SNARE proteins in humans . Each SNARE protein contains one or two defining SNARE motifs of around 60 amino acids in eight heptad repeats ( Figure 1A ) . The motif is often connected to a C-terminal transmembrane domain via a short linker domain ( LD of ∼10 a . a . ) . Complementary SNAREs are anchored to transport vesicles ( v-SNAREs ) and their targeted membranes ( t-SNAREs ) in disordered or partially disordered conformations . Their specific interactions lead to coupled folding and assembly into a stable parallel four-helix bundle , drawing the two membranes into close proximity for fusion ( Sollner et al . , 1993; Sudhof and Rothman , 2009; Gao et al . , 2012 ) . In the core of each SNARE bundle are 15 layers of hydrophobic amino acids and one middle layer of ionic amino acids . The ionic layer is formed by three glutamine residues ( Q ) and one arginine residue ( R ) from each of the SNARE motifs categorized as Qa , Qb , Qc , and R SNAREs ( Fasshauer et al . , 1998; Figure 1B ) . Crystal structures show that the four-helix bundle structures are highly conserved in different SNARE complexes ( Sutton et al . , 1998; Zwilling et al . , 2007; Stein et al . , 2009 ) , which can be aligned to the angstrom level ( Strop et al . , 2008 ) . 10 . 7554/eLife . 03348 . 003Figure 1 . Chimeric SNARE construct and experimental setup used to study functional assembly of single SNARE complexes using dual-trap high-resolution optical tweezers . ( A ) Modular parallel four-helix bundle structure of an assembled neuronal SNARE complex mediating membrane fusion . The SNARE complex contains different functional domains: an N-terminal domain ( NTD ) , an ionic layer , a C-terminal domain ( CTD ) , a linker domain ( LD ) , two transmembrane domains , and other domains not shown here . ( B ) Diagram showing the chimeric SNARE construct and the experimental setup . Each SNARE complex contains one SNARE motif from the four highly conserved Qa , Qb , Qc , and R SNARE families . These motifs are joined into one protein through spacer sequences ( dashed lines ) to facilitate the single-molecule manipulation experiment . The same color coding for different SNARE proteins is used throughout this work . See Figure 1—figure supplement 1 for complete sequences of the chimeric SNAREs and Figure 1—figure supplements 2–4 for minimal effects of the spacer sequences on the folding energy and kinetics of the SNARE complexes . DOI: http://dx . doi . org/10 . 7554/eLife . 03348 . 00310 . 7554/eLife . 03348 . 004Figure 1—figure supplement 1 . Amino acid sequences of the chimeric SNARE protein constructs used for the single-molecule manipulation study of SNARE assembly . The four sets of SNARE proteins are neuronal SNAREs of rat syntaxin ( SX ) 1A , rat VAMP2 , and mouse SNAP-25B , GLUT4 SNAREs of rat syntaxin 4A , rat SNAP-23 , and rat VAMP2 , endosomal SNAREs of rat syntaxin 13 , mouse Vti1a , human Stx6 , and mouse VAMP4 , and yeast SNAREs of Sso1 , Sec9 , and Snc2 . SNARE sequences within and between different rows are aligned in their SNARE motifs , in which amino acids in the 15 hydrophobic layers and the ionic layer are shown in red and green , respectively , with layer numbers ( from −7 to +8 ) indicated above the sequences . The amino acids in the SNARE linker domains ( LDs ) are shown in bold . In yeast R SNARE Snc2 , the amino acids deleted in the LD- and C-terminal domain ( CTD ) -truncated constructs ( Figure 2—figure supplement 1 ) are underlined by blue dashed lines . All natural cysteine residues were mutated to serine ( underlined and in italic ) . The pulling sites ( cysteine and lysine ) are indicated in blue . The spacer sequences ( Sp1–Sp3 ) are derived from an unstructured sequence found in a POU transcription factor . The Avi-tag as a substrate for biotinylation ( on lysine ) is underlined by a black line . DOI: http://dx . doi . org/10 . 7554/eLife . 03348 . 00410 . 7554/eLife . 03348 . 005Figure 1—figure supplement 2 . The chimeric neuronal SNARE protein correctly folds into an expected four-helix SNARE bundle . Circular dichroism spectrum of the chimeric SNARE protein . The presence of two local minima at 208 nm and 222 nm indicates a high content of alpha-helical structure in the protein . Based on the CD spectrum , we estimated that about 50% of the amino acids in the protein are in an alpha-helical configuration , consistent with a fully folded SNARE four-helix bundle in the chimeric protein . DOI: http://dx . doi . org/10 . 7554/eLife . 03348 . 00510 . 7554/eLife . 03348 . 006Figure 1—figure supplement 3 . The chimeric neuronal SNARE protein folds into a homogenous SNARE four-helix bundle with an expected molecular weight . Gel filtration profile of the chimeric protein . The purified chimeric SNARE protein was eluted from the column ( Superdex 200 10/300 GL ) in one major peak corresponding to a molecular weight of 57 kDa , consistent with a folded SNARE four-helix bundle . DOI: http://dx . doi . org/10 . 7554/eLife . 03348 . 00610 . 7554/eLife . 03348 . 007Figure 1—figure supplement 4 . The chimeric t-SNARE protein supports lipid mixing between liposomes . ( A ) Schematic representation of the chimeric t-SNARE and the v-SNARE used for the lipid-mixing assay . The chimeric t-SNARE protein is lipidated through a cysteine introduced to the C-terminus of syntaxin . ( B ) Comparison of membrane fusion activity mediated by the chimeric t-SNARE ( blue ) and the transmembrane domain truncated wild-type t-SNARE complex ( black ) in the fluorescence de-quenching assay . Both t-SNAREs were anchored to membrane through a unique cysteine at the C-termini of syntaxin . The fusion depends on formation of trans-SNARE complexes , as pre-incubation of t-SNARE proteoliposomes with the cytoplasmic domain of VAMP2 ( CDV ) inhibits membrane fusion ( green ) . The reduced fusion activity of the chimeric t-SNARE compared to the wild-type t-SNARE was caused by the spacer sequence between the C-terminus of syntaxin and the cysteine residue ( McNew et al . , 2000; Figure 1—figure supplement 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03348 . 007 The conserved sequences of SNAREs and their similar initial and final conformations implicate a conserved pathway of SNARE folding/assembly . However , the kinetics and energetics of SNARE folding have not been well characterized . It is notoriously difficult to study SNARE assembly using traditional ensemble-based experimental approaches due to the many states and pathways involved in the folding process , especially misassembled states ( Weninger et al . , 2003; Pobbati et al . , 2006 ) . In addition , functional SNARE assembly occurs in the presence of the opposing force imposed by membranes , which has a great impact on the kinetics and regulation of SNARE assembly ( Sudhof and Rothman , 2009; Gao et al . , 2012 ) . Although studies of SNARE assembly are facilitated by the use of soluble SNAREs isolated from membranes , the lack of an essential force load may complicate data interpretation regarding functional SNARE assembly . For example , whereas complexin can suspend assembly of trans-SNAREs in a partially zippered state ( Kummel et al . , 2011; Li et al . , 2011; Malsam et al . , 2012 ) , it cannot do so for isolated SNAREs ( Chen et al . , 2002 ) . Thus , new methods are required to better elucidate SNARE assembly . Recently , we have applied high-resolution optical tweezers to quantitatively characterize the energetics and kinetics of neuronal SNARE folding for the first time ( Gao et al . , 2012 ) . This single-molecule manipulation method allows measurement of the folding energy and kinetics of macromolecules under equilibrium conditions ( Liphardt et al . , 2001 ) . Furthermore , the external force applied to the SNARE complex mimics the opposing force from membranes ( Li et al . , 2007; Liu et al . , 2009; Min et al . , 2013 ) . Using this single-molecule method , we proved the long-standing hypothesis that neuronal SNAREs assemble by a zippering mechanism and discovered a half-zippered SNARE intermediate that plays a crucial role in the synchronized , calcium-triggered synaptic vesicle fusion ( Gao et al . , 2012 ) . Step-wise SNARE zippering is initiated by slow association between N-terminal domains ( NTDs ) of t- and v-SNAREs . SNARE assembly then pauses in the half-zippered state in a force-dependent manner . Finally , the C-terminal domain ( CTD ) of the v-SNARE ( VAMP2 or synaptobrevin ) rapidly zippers along the pre-structured t-SNARE template to drive fast membrane fusion . It is unknown whether other SNAREs assemble by the same zippering mechanism . Furthermore , it is not clear how SNARE assembly is adapted to efficient and versatile membrane fusion . It has been proposed that SNAREs generally assemble in an all-or-none manner without any partially folded intermediates ( Jahn and Fasshauer , 2012; Kasai et al . , 2012 ) . It is argued that assembly of neuronal SNARE complexes occurs in a large energy gradient , and thus cannot be stopped halfway to form any partially assembled intermediates ( Jahn and Fasshauer , 2012 ) . However , despite its fast speed , downhill SNARE assembly would be poorly coupled to membrane fusion , resulting in low energy efficiency of the SNARE engine . In contrast , many molecular engines tested at a single-molecule level have nearly 100% energy efficiency ( Bustamante et al . , 2004 ) . Based on the first law of thermodynamics , a mechanochemical process has 100% energy efficiency only when the process is reversible . Therefore , to maximize their energy efficiency , SNAREs are expected to fold in a relatively smooth energy landscape ( Onuchic and Wolynes , 2004 ) in the presence of the membrane load . This requires a close match between the energy landscape of SNARE assembly and the energy profile of membrane interactions . The energy opposing membrane fusion includes contributions from the long-ranged entropic membrane undulation , membrane deformation , and electrostatic interactions , and the short-ranged membrane dehydration and van der Waals interactions ( Leckband and Israelachvili , 2001 ) . The strong short-ranged repulsion is the largest energy barrier for fusion and takes place within a few nanometers of membrane separation , thus constituting a hard core for fusion . To break this hard core , a SNARE complex is required to focus its folding energy on the membrane proximal C-terminus . Therefore , analogous to a car engine , an efficient SNARE engine is expected to change gears to meet increasing resistance as SNAREs fold towards membranes . However , it remains unclear whether such a gear-changing mechanism exists in SNARE assembly . To address the above questions , we measured the folding energy and kinetics of four representative SNARE complexes at a single-molecule level , using high-resolution optical tweezers and a new chimeric SNARE design ( Figure 1 ) . These complexes mediate highly regulated exocytosis of neurotransmitters in pre-synaptic neurons ( neuronal SNAREs: syntaxin 1 , SNAP-25B , and VAMP2 or synaptobrevin ) ( Sollner et al . , 1993 ) and translocation of glucose transporter type 4 ( GLUT4 ) in adipocytes or muscle cells ( GLUT4 SNAREs: syntaxin 4 , SNAP-23 , and VAMP2 ) ( Bai et al . , 2007; Stockli et al . , 2011 ) . The complexes also affect constitutive fusion of endocytic vesicles to early endosome in mammals ( endosomal SNAREs: syntaxin 13 , Vti1A , syntaxin 6 , and VAMP4 ) ( Zwilling et al . , 2007 ) and fusion of post-Golgi vesicles to plasma membranes in yeast ( yeast SNAREs: Sso1 , Sec9 , and Snc2 ) ( Strop et al . , 2008 ) . All four of these SNARE complexes were chosen for our study because they represent SNAREs in diverse evolutionary species , have different degrees of regulation , and mediate fusion with a speed ranging from 0 . 2 ms to 20 min ( Kasai et al . , 2012 ) . In addition , the crystal structures of neuronal , endosomal , and yeast SNARE complexes are available ( Sutton et al . , 1998; Zwilling et al . , 2007; Strop et al . , 2008; Stein et al . , 2009 ) , which facilitates derivation of their various assembly intermediates from our single-molecule measurements ( Gao et al . , 2012 ) , allowing us to compare the folding pathways and energy landscapes of different SNARE complexes . Our results show that all four SNARE complexes assemble via the same zippering mechanism in three sequential steps: slow NTD association , fast CTD zippering , and finally rapid LD zippering . However , the CTD zippering energy of different SNARE complexes varies greatly and is highly concentrated at the C-terminus . To facilitate protein preparation and single-molecule experiments , we constructed new chimeric SNARE proteins in which three or four cognate SNARE proteins were joined into one polypeptide with the addition of two or three spacer sequences ( Figure 1 ) . Individual cytoplasmic SNARE sequences were truncated and regions that directly participate in SNARE complex formation were kept ( Figure 1—figure supplement 1 ) . To minimize their perturbation on the structure and dynamics of SNARE complexes , the spacer sequences were chosen to be unstructured and of proper length . Each chimeric SNARE protein consisted of a unique cysteine at the C-terminus of Qa SNARE and an Avi-tag at the C-terminus of R SNARE used to pull the single SNARE complex ( Figure 1B ) . We first examined the structural and functional integrity of the chimeric SNARE complexes . For this purpose , the recombinant proteins were purified and biotinylated in vitro . The expected helical bundles that formed were confirmed by circular dichroism spectra and gel filtration profiles ( Figure 1—figure supplements 2 and 3 ) . To test the function of the SNARE protein , we similarly made a chimeric neuronal t-SNARE protein and tested its ability to mediate lipid mixing with full-length VAMP2 ( Figure 1—figure supplement 4 ) . We found that the t-SNARE protein was as fusogenic as the wild-type cytoplasmic t-SNARE complex that is covalently linked to the membrane ( McNew et al . , 2000 ) . This result suggests that the spacer sequence between syntaxin and SNAP-25 does not significantly interfere with SNARE assembly and membrane fusion . Furthermore , the chimeric neuronal SNARE complex reveals folding energy and kinetics ( see below ) consistent with our recent reports based on a different SNARE construct in which syntaxin and VAMP2 were cross-linked at their N-termini by a disulfide bond ( Gao et al . , 2012 ) . Taken together , the chimeric SNARE proteins mimic their corresponding SNARE complexes and can be used to facilitate the study of SNARE assembly at a single-molecule level . We refer to these proteins as SNARE complexes . The SNARE complexes were cross-linked to a 2260 bp DNA handle ( Cecconi et al . , 2005 ) and tethered to two polystyrene beads held in two optical traps of high-resolution optical tweezers ( Moffitt et al . , 2006; Sirinakis et al . , 2011; Figure 1B ) . Single SNARE complexes were pulled from the C-termini of Qa and R SNAREs by moving one trap relative to another at a constant speed , typically 10 nm/s . The tension and extension of the protein-DNA tether were recorded at 10 kHz and used to derive protein folding energy and kinetics . When pulled to a force up to 25 pN , all four SNARE-DNA tethers extended continuously in some force ranges , but discontinuously in other ranges ( Figure 2A , B ) . The continuous extension increase was mainly caused by stretching of the semi-flexible DNA handle while the SNARE complex remained in the same folding state . The resultant force-extension curves ( FECs ) could generally be fit by the worm-like chain model of the DNA and polypeptide ( Marko and Siggia , 1995 ) . In contrast , abrupt extension changes resulted from cooperative protein transitions between different states ( Figure 2C ) . The FECs show that all four SNARE complexes sequentially unfolded via two reversible transitions and one or two irreversible unfolding steps . Compared to the FECs reported for the neuronal SNARE complex ( Gao et al . , 2012 ) and confirmed by the detailed analysis described below , the second reversible transition ( between state 2 and state 3 ) and the first irreversible transition ( between state 3 and state 4 ) resulted from folding/unfolding transitions of CTD and NTD , respectively . Both transitions are energetically or kinetically distinct for each of the four SNARE complexes , as is demonstrated by non-overlapping distributions of the characteristic forces or different lifetimes associated with these transitions ( Figure 3 ) . In particular , NTD is mechanically more stable than CTD and unfolded generally after 10–105 CTD folding and unfolding transitions under our experimental conditions ( Figure 2B ) . 10 . 7554/eLife . 03348 . 008Figure 2 . Four representative SNARE complexes assemble or disassemble via common intermediates and pathways . ( A ) Force-extension curves ( FECs ) of the neuronal , GLUT4 , endosomal , and yeast SNARE complexes . FECs were obtained by pulling the complexes ( black ) or relaxing them ( gray ) . The reversible C-terminal domain ( CTD ) and linker domain ( LD ) folding/unfolding transitions are marked by blue solid and dashed ovals , respectively , whereas irreversible unfolding of the partially zippered SNARE complex is indicated by a red arrow . Continuous FEC regions can be fit by the worm-like chain model and represent different SNARE states numbered as in ( C ) . Below ∼6 pN , deviation of some fits from the measured FECs corresponding to the unfolded complex ( state 5 ) may be caused by intramolecular interactions or partial refolding of the complex . The t-SNARE state can be identified from some FECs , with a transient one ( ∼20 ms ) marked by a cyan rectangle . ( B ) Time-dependent extension , force , and trap separation corresponding to the CTD and N-terminal domain ( NTD ) transition region in the FEC of the GLUT4 SNARE complex in A ( marked by two magenta dots ) . In the upper and middle panels , the positions of different SNARE folding states are indicated by red dashed lines . About 90 CTD transitions occurred before NTD unzipping and reaching ∼18 . 6 pN equilibrium force ( indicated by a red arrow in the middle panel ) . Here the equilibrium force for a two-state protein folding/unfolding process is defined as the average state forces ( marked by dashed lines ) under which the folded and the unfolded states are equally populated . Note that most NTD unzipping took place in the CTD-unfolded state ( state 3 ) . In the middle panel , the first CTD and the first NTD unzipping events during the pulling process are indicated by green dots and their time and force differences indicated . The time and force distributions are shown in Figure 3B , C . In the bottom panel , the separation between two optical traps was increasing at a speed of 10 nm/s to slowly pull the single SNARE complex . ( C ) Different SNARE assembly states partly derived from model-fitting of FECs shown in ( A ) . Gray arrows indicate the pulling direction . Data associated with all FECs shown in this work were mean-filtered using 5 ms time window . See more FECs and their associated features in Figure 2—figure supplements 1–4 . DOI: http://dx . doi . org/10 . 7554/eLife . 03348 . 00810 . 7554/eLife . 03348 . 009Figure 2—figure supplement 1 . Distinct linker domain and C-terminal domain transitions . Force-extension curves ( FECs ) of yeast SNARE complexes containing the v-SNARE Snc2 truncated in the linker domain ( LD ) region or in both the LD region and part of the C-terminal domain ( CTD ) region ( Figure 1—figure supplement 1 ) . The CTD transition in the LD-truncated complex is marked by a solid blue oval . The cooperative SNARE reassembly is indicated by a black arrow . DOI: http://dx . doi . org/10 . 7554/eLife . 03348 . 00910 . 7554/eLife . 03348 . 010Figure 2—figure supplement 2 . The neuronal t-SNARE complex as a transient unfolding intermediate of the half-zippered SNARE complex . Force-extension curves ( FECs ) of a single neuronal SNARE complex corresponding to three pulling cycles . The inset shows close-up views of the unfolding process of the half-zippered SNARE complex in extension-time trajectories . The transient t-SNARE complex appeared in the second pulling cycle ( indicated by a cyan rectangle ) , but was not discernible in the other two pulling cycles . Note that the C-terminal domain ( CTD ) transition seen in the second pulling cycle was significantly slower than those in the other cycles . DOI: http://dx . doi . org/10 . 7554/eLife . 03348 . 01010 . 7554/eLife . 03348 . 011Figure 2—figure supplement 3 . The GLUT4 t-SNARE complex as a transient unfolding intermediate of the half-zippered SNARE complex . Force-extension curves ( FECs ) of the same GLUT4 SNARE complex corresponding to three pulling cycles . The overlapping FECs ( ‘All’ ) and individual FECs from successive pulling cycles ( #1–#3 ) were shifted along the x-axis for better comparison . The cooperative SNARE reassembly is indicated by a black arrow . DOI: http://dx . doi . org/10 . 7554/eLife . 03348 . 01110 . 7554/eLife . 03348 . 012Figure 2—figure supplement 4 . The yeast t-SNARE complex is a stable unfolding intermediate of the half-zippered SNARE complex . Force-extension curves ( FECs ) of a single yeast SNARE complex corresponding to two successive pulling cycles and their best fits by the worm-like chain model in the continuous phases . The cooperative SNARE reassembly is indicated by black arrows . The FECs corresponding to the completely unfolded SNARE complex are shown in gray . DOI: http://dx . doi . org/10 . 7554/eLife . 03348 . 01210 . 7554/eLife . 03348 . 013Figure 3 . Distinct transition kinetics and stabilities of SNARE C-terminal domain and N-terminal domain . ( A ) Histogram distributions of the C-terminal domain ( CTD ) equilibrium force and the N-terminal domain ( NTD ) unzipping force for different SNARE complexes . The vertical axis shows the percentage of the event number in each bin . The average CTD equilibrium force ( f1/2 ) or NTD unzipping force ( funzip ) scored on the total numbers of transition events ( NT ) and single SNARE complexes ( Nm ) are indicated , with the number in parenthesis designating the standard deviation of the mean . ( B , C ) Histogram distributions of the force and time differences of the first NTD and CTD unzipping events ( Figure 2B ) . The average force difference ( Δfunzip ) or time difference ( Δtunzip ) is indicated . The distinct CTD and NTD transition kinetics are revealed by non-overlapping force distributions for neuronal , endosomal , and yeast SNARE complexes or significant force and time differences associated with the first unzipping events of CTD and NTD of the GLUT4 SNARE complex . Note that optical tweezers have a force measurement accuracy of 10% absolute forces between different single molecules and of <0 . 1 pN relative forces within same single molecules ( Moffitt et al . , 2006 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03348 . 013 After the last irreversible unfolding event , the FECs obtained by pulling proteins to higher forces ( >25 pN ) did not show any additional discontinuous extension changes ( Figure 2A ) , indicating that the SNARE complexes had been completely unfolded . When relaxed , the SNARE complex remained unfolded until the force was dropped to ∼4 pN , leading to a large hysteresis in the FECs . Further relaxation of the complex to lower forces led to FECs overlapping with those of the FECs in the pulling phase , often with small and sudden extension drops manifesting cooperative reassembly of SNARE complexes ( Figure 4 ) . Additional cycles of pulling and relaxation generally revealed overlapping FECs , suggesting that the SNARE complexes could fully reassemble into nearly identical structures under our experimental conditions . 10 . 7554/eLife . 03348 . 014Figure 4 . Overlapping force-extension curves obtained by repeatedly pulling a single neuronal or GLUT4 SNARE complex , revealing robust and common step-wise SNARE assembly and disassembly . The overlapping force-extension curves ( FECs ) ( designated by ‘All’ ) are shifted along the x-axis to reveal individual FECs corresponding to different pulling cycles ( numbered ) . The cooperative reassembly events are indicated by black arrows . The neuronal SNARE-DNA tether broke in the third pulling cycle of the neuronal SNARE complex at the maximum pulling force . The GLUT4 SNARE complex unfolded at 2 . 5 pN force ( red arrow ) in the last pulling cycle , indicating that the complex was not properly assembled at the end of the fifth pulling cycle . Note that heterogeneity in SNARE zippering was observed , a phenomenon also seen in many single-molecule experiments ( Lu et al . , 1998; Sirinakis et al . , 2011 ) . The heterogeneity is manifested by changes in the rate and/or the equilibrium force of the C-terminal domain ( CTD ) transition detected in different pulling cycles of the same chimeric SNARE protein . For the single neuronal SNARE protein shown here , both equilibrium force and rate of the CTD transition are slightly lower in the first pulling cycle than in the following two cycles . More heterogeneity can be seen in Figure 2—figure supplements 2 and 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 03348 . 014 The reversible SNARE transitions could be better observed under approximately constant forces ( Figure 5A , Figure 6A ) . In this case , the time-dependent extension change represented spontaneous folding/unfolding transition of the protein under tension due to thermal fluctuations . Both transitions in each of the four SNARE complexes were binary , as indicated by the two peaks in the histogram distributions of extension ( Figure 5B , Figure 6B ) . The transitions remained cooperative at all forces tested , but were shifted to unfolded states at higher forces . Furthermore , the four SNARE complexes had similar average extension changes for both transitions ( Table 1 ) , implying that the same SNARE domains were involved in the observed transitions . Taken together , the results from experiments in variable and constant forces suggest that all four SNARE complexes follow similar pathways to assembly or disassemble via at least two intermediates . 10 . 7554/eLife . 03348 . 015Figure 5 . Comparison of the two-state C-terminal domain transitions of four SNARE complexes . ( A ) Force-dependent extension-time trajectories under approximately constant forces ( f ) revealing the unfolding probability ( p ) of C-terminal domain ( CTD ) as indicated . The idealized two-state transitions ( red lines ) were calculated based on a hidden Markov model ( HMM ) . ( B ) Histogram distributions of the extensions shown in A ( symbols ) and their best fits with double-Gaussian functions ( lines ) . For best comparison , the distributions for each SNARE complex were shifted along the x-axis to align them at the same average position of the unfolded CTD state . Distributions at different forces are color-coded as the corresponding extension traces in A . All the extension-time trajectories shown in this work were mean-filtered using a 1 ms time window . ( C ) CTD unfolding probabilities of four SNARE complexes . ( D ) The corresponding folding rates ( hollow symbols ) and unfolding rates ( solid symbols ) of CTD transitions . The best-fit unfolding probability ( solid line ) , folding rate ( dashed line ) , and unfolding rate ( solid line ) were obtained by non-linear least-squares fitting using a simplified energy-landscape model of SNARE assembly ( ‘Materials and methods’ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03348 . 01510 . 7554/eLife . 03348 . 016Figure 6 . Comparison of the two-state linker domain transitions of four SNARE complexes . ( A ) Extension-time trajectories ( black lines ) and their best hidden Markov model ( HMM ) fits ( red lines ) showing fast binary transitions of linker domains ( LDs ) under constant forces . The force ( f ) and unfolding probability ( p ) are indicated . ( B ) Extension histogram distributions corresponding to the trajectories in A ( symbols ) and their best fits with double-Gaussian functions ( red lines ) . ( C ) Force-dependent unfolding probability and transition rates ( symbols ) and their best fits ( solid or dashed lines ) of GLUT LD . Similar data corresponding to other SNARE complexes are shown in Figure 6—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 03348 . 01610 . 7554/eLife . 03348 . 017Figure 6—figure supplement 1 . Folding energy and kinetics of SNARE linker domains . Unfolding probabilities ( top panel ) and folding rates ( open symbols in bottom panel ) or unfolding rates ( solid symbols ) of linker domains ( LDs ) in the yeast , endosomal , and neuronal SNARE complexes . The corresponding best fits based on the energy landscape model ( ‘Materials and methods’ ) are shown as solid or dashed lines . DOI: http://dx . doi . org/10 . 7554/eLife . 03348 . 01710 . 7554/eLife . 03348 . 018Figure 6—figure supplement 2 . Minor effect of the spacer sequences in the chimeric SNARE proteins on the folding energy of SNARE complexes . Predicted structures of the folded and unfolded states in linker domain ( LD ) transition and the accompanying extension change of the spacer sequence connecting syntaxin and SNAP-25 ( red dashed line ) . The apparent LD folding energy measured by optical tweezers ( ΔGapp ) contains true LD folding energy ( ΔGLD ) and the free energy change of the spacer sequence due to its extension change ( ΔGsp ) , that is , ( 6 ) ΔGapp=ΔGLD+ΔGsp . To correct for the true LD folding energy , we estimated the spacer energy change using a worm-like chain model . In the LD folded state , the spacer sequence is 65 a . a . long or 23 . 7 nm in contour length and has an extension of 11 . 8 nm , or the distance between point 1 ( SNAP-25 R8 ) and point 2 ( syntaxin R260 ) . Thus , the ratio ( r ) of the extension to the contour length ( l ) is 0 . 5 . Using Equation 3 , we calculated the free energy of the spacer sequence in the folded LD state to be 9 kBT . In the unfolded LD state , the spacer sequence ( including the unfolded LD sequence in syntaxin ) becomes 73 a . a . long or 26 . 6 nm in contour length , with an extension of 10 . 6 nm ( the distance between point 1 and point 3 at syntaxin A254 ) . The free energy of the spacer in the unfolded LD state was similarly calculated to be 5 . 9 kBT . Thus , the energy change of the spacer sequence upon LD folding is ΔGsp = 3 . 1 kBT . The apparent energy change derived from the singe-molecule measurement is ΔGapp = −6 . 6 kBT . Thus the LD folding energy ΔGLD = ΔGapp − ΔGsp = −9 . 7 kBT . The two spacer sequences in the chimeric neuronal SNARE protein have no effect on the C-terminal domain ( CTD ) transition , because their extensions do not change in this process . The spacer sequence connecting SNAP-25 and VAMP2 ( not shown ) slightly decreases N-terminal domain ( NTD ) association energy by 4 . 3 kBT . However , this reduced NTD stability does not alter its much greater lifetime than the CTD . DOI: http://dx . doi . org/10 . 7554/eLife . 03348 . 01810 . 7554/eLife . 03348 . 019Table 1 . Average equilibrium force , extension change , folding energy , and folding energy barrier and position associated with C-terminal domain and linker domain transitions of the four different SNARE complexesDOI: http://dx . doi . org/10 . 7554/eLife . 03348 . 019SNARE complexC-terminal domainLinker domainForce ( pN ) Extension change ( nm ) Folding energy ( kBT ) Transition state energy* ( kBT ) Transition state position† ( a . a . ) Force ( pN ) Extension change ( nm ) Folding energy ( kBT ) Transition state energy* ( kBT ) Transition state position† ( a . a . ) Neuron16 . 2 ( 0 . 9 ) 7 . 2 ( 1 . 2 ) −27 ( 4 . 7 ) −5 . 5 ( 1 . 5 ) 17 ( 3 ) 8 ( 1 ) 4 . 7 ( 0 . 5 ) −9 . 7 ( 1 . 6 ) 5 . 5 ( 1 . 5 ) 31 ( 1 ) GLUT418 . 5 ( 1 . 8 ) 6 . 0 ( 0 . 9 ) −23 ( 4 . 1 ) −0 . 8 ( 1 . 0 ) 11 ( 2 ) 8 . 6 ( 0 . 9 ) 5 . 6 ( 1 . 1 ) −12 ( 2 . 7 ) 2 ( 1 . 0 ) 30 ( 1 ) Endosome11 . 9 ( 0 . 9 ) 6 . 9 ( 0 . 4 ) −16 ( 1 . 5 ) 2 . 1 ( 1 . 4 ) 12 ( 2 ) 6 . 3 ( 1 . 2 ) 5 . 1 ( 1 . 8 ) −6 . 1 ( 2 . 4 ) 4 . 9 ( 1 . 5 ) 32 ( 2 ) Yeast10 . 1 ( 1 . 4 ) 5 . 8 ( 0 . 8 ) −13 ( 2 . 5 ) 3 . 2 ( 1 . 5 ) 13 ( 2 ) 6 . 0 ( 1 . 6 ) 5 . 1 ( 1 . 2 ) −5 . 7 ( 2 . 0 ) 3 . 6 ( 2 . 0 ) 32 ( 2 ) *Here , negative energy indicates downhill protein folding ( Yang and Gruebele , 2003 ) . †The number of the amino acids in the R SNARE C-terminal to the ionic layer ( chosen as 0 ) . The equilibrium force and extension change were determined at an unfolding probability of 0 . 5 for the two-state processes . The standard deviations of the averages are shown in parenthesis . The equilibrium force distribution , the number of transitions , and the number of single molecules scored for C-terminal domain ( CTD ) transitions are shown in Figure 3 . For parameters related to linker domain ( LD ) transitions , a total of 18 , 35 , 11 , and 24 LD transitions in single neuronal , GLUT4 , endosomal , and yeast SNARE complexes were scored , respectively . To derive the structures of the intermediates observed in our experiments , we fit the continuous regions of the FECs using a quantitative model of the protein-DNA conjugate previously reported ( Gao et al . , 2012; Xi et al . , 2012 ) . In this model , the extension of the structured portion of the SNARE complex is force-independent , but varies as the SNARE complex changes its conformation ( ‘Materials and methods’ ) . The model generally fit the measured FECs and extension changes obtained at constant forces well ( Figure 2A ) . Extensive analysis revealed two common intermediates for the four SNARE complexes: the LD-unfolded state and the half-zippered state ( Figure 2C ) . In the former , the SNARE LD was unfolded , while its CTD remained approximately intact . In the latter , the C-terminal half of the R SNARE was unzipped , whereas three Q SNARE motifs remained intact ( Kummel et al . , 2011; Gao et al . , 2012; Li et al . , 2014 ) . Specifically , neuronal , GLUT4 , endosomal , and yeast SNARE complexes in the half-zippered state had their R SNAREs unzipped to −1 , +3 , +1 , and +3 amino acids relative to the ionic layer , respectively , where the positive sign designates the C-terminal amino acids . The standard deviation of all positions was less than three amino acids ( Table 1 ) . To further confirm the derived structures of the intermediate states , we truncated the LD or the CTD of the v-SNARE Snc2 in the yeast SNARE complex and repeated the pulling experiment . We found that LD truncation eliminated the LD but not the CTD transition , while the CTD truncation abolished both transitions ( Figure 2—figure supplement 1 ) . These results support the inferred structures for the intermediate states . Finally , both CTD and LD folded more rapidly than similar coiled-coil proteins ( Xi et al . , 2012 ) , with their transition rates greater than 50 s−1 , even at the equilibrium forces ( Figure 5 , Figure 6 ) . Further unzipping of the half-zippered states of all four SNARE complexes became irreversible and they remained unfolded for over 50 s under the slow relaxation conditions in our experiment ( Figure 2A ) , indicating a large energy barrier for SNARE NTD association . Close inspection of the FECs showed that a fraction of half-zippered SNARE complexes , that is , 10% , 50% , and 30% for neuronal , GLUT4 , and endosomal SNAREs , respectively , unfolded via a transient intermediate with a typical lifetime of less than 50 ms ( Figure 2A , B , Figure 2—figure supplements 2 and 3 ) . The yeast SNARE complex is special , because this additional intermediate appeared in 85% of the unfolding transitions of the half-zippered complex and generally lasted for more than 5 s ( Figure 2—figure supplement 4 ) . For all SNARE complexes , these intermediate states are located at an extension approximately halfway between the half-zippered states and the fully unfolded states , indicating their similar structures . Based on their relative extension positions , the intermediate states are estimated to be t-SNARE or Q SNARE complexes with ordered NTDs but disordered CTDs ( Figure 2C ) . In conclusion , all four SNARE complexes assemble or disassemble via three common intermediate states: the t-SNARE state , the half-zippered state , and the LD-unfolded state , and along similar folding pathways and kinetics , particularly slow NTD association and fast CTD and LD zippering . The biggest difference between the four SNARE complexes lay in their CTD equilibrium forces ( Figure 3A , Figure 5C ) , indicating different CTD folding energies . To quantify CTD folding energy and kinetics , we measured CTD transitions at different constant forces in their corresponding force ranges ( Figure 5A ) . We analyzed each extension-time series using a two-state hidden Markov model ( HMM ) and determined the positions of the folded and unfolded CTD states and their corresponding fluctuations , the unfolding probability , and transition rates ( Gao et al . , 2012; Xi et al . , 2012 ) . The HMM-based analyses yielded idealized state transitions and extension histogram distributions that closely matched the corresponding experimental measurements ( Figure 5A , B ) . The unfolding probability rises with the force increase in a sigmoidal manner ( Figure 5C ) . The folding rate or unfolding rate decreases or increases approximately exponentially upon a force increase in the narrow force range tested ( Bustamante et al . , 2004; Figure 5D ) . Both observations suggest a two-state CTD transition and the existence of a single major energy barrier corresponding to the transition state for the folding/unfolding process . The position of the transition state relative to the folded or unfolded state can be determined from the force-dependent transition rates . We adopted a simplified energy landscape model to derive the energy and rate of SNARE folding at zero force ( ‘Materials and methods’ ) ( Gao et al . , 2012; Xi et al . , 2012 ) . Non-linear least-squares fitting of the model matched the experimental data well ( Figure 5C , D ) , which revealed the free energy of the folded state and the transition state and their relative positions ( Table 1 ) . The CTD folding energy of neuronal , GLUT4 , endosomal , and yeast SNARE complexes were −27 ( ±5; SD throughout the text ) kBT , −23 ( ±4 ) kBT , −16 ( ±2 ) kBT , and −13 ( ±3 ) kBT , respectively . The CTD folding energy and the equilibrium rate ( ∼100 s−1 ) of the neuronal SNARE complex were very close to the energy ( 28 ± 3 kBT ) and the rate ( ∼160 s−1 ) reported earlier ( Gao et al . , 2012 ) , indicating that the spacer sequences in the chimeric construct used here have minimal effect on the folding energy and kinetics of the SNARE complex . The binary CTD transition manifested the existence of an energy barrier and its associated transition state for CTD folding and unfolding in the presence of the external force . When extrapolated to zero force , the CTD folding energy barrier became minimal for endosomal and yeast SNARE complexes or disappears for neuronal and GLUT4 SNARE complexes ( Table 1 ) . In both scenarios , free energy of the transition states can still be defined ( Gao et al . , 2012; Xi et al . , 2012 ) . The transition states of four SNARE complexes are located between the third and sixth hydrophobic layers . The energy and position of the transition state is important for characterizing the energy landscape of SNARE folding described later in the text . The relatively small folding energy barrier suggests that the rate of SNARE-mediated fusion is not limited by the intrinsic rate of CTD folding ( at zero force ) , and that the stability of the half-zippered state is strongly force-dependent . Any partially zippered trans-SNARE complexes involved in vesicle docking and priming are likely in strained states imposed by the membranes and regulatory proteins ( Guzman et al . , 2010; Jahn and Fasshauer , 2012 ) . The folding and unfolding transitions of the LD in four SNARE complexes were similar . They were reversible , binary , and fast ( Figure 6 , Figure 6—figure supplement 1 ) . Furthermore , the LD transitions occurred in narrow force ranges ( 6–8 . 6 pN ) , in contrast to the CTD transition ( 10 . 1–18 . 5 pN ) ( Table 1 ) . Although the LD of the endosomal SNARE complex forms a four-helix bundle ( Zwilling et al . , 2007; Figure 1—figure supplement 1 ) , rather than a two-stranded coiled coil as in the other three SNARE complexes , it has similar LD transition kinetics , associated extension change , and lower equilibrium force than its CTD . This observation corroborates the conclusion that LD is a domain distinct from CTD , even in the endosomal SNARE complex . The energy and kinetics of LD zippering at zero force was obtained in a way similar to CTD , as previously described ( Gao et al . , 2012 ) . For neuronal SNARE complexes , the new chimeric construct led to an equilibrium force of 8 ( ±1 ) pN for LD transition , compared to 12 ( ±2 ) pN previously measured for the same transition . Correcting for the minor effect of the spacer sequence added between syntaxin and SNAP-25 ( Figure 6—figure supplement 2 ) , we obtained LD zippering energy of −10 ( ±2 ) kBT for the neuronal SNARE complex , consistent with our previous measurement of −8 ( ±2 ) kBT . Similarly , we derived the zippering energy of LDs and their associated energy barriers for the other three SNARE complexes ( Table 1 ) . In the four SNARE complexes , LD zippering outputs less energy than CTD zippering . Thus , CTD zippering serves as the major power stoke for membrane fusion ( Walter et al . , 2010 ) . Our above analysis revealed a simplified folding energy landscape of each SNARE complex ( Figure 7 ) . To illustrate how such an energy landscape is adapted to stage-wise membrane fusion ( Figure 7A ) , we calculated the energy landscape of SNARE assembly in the presence of membranes using a neuronal SNARE complex as an example ( Gao et al . , 2012 ) . The interaction energy between membranes containing lipid-anchored t- and v-SNAREs has been measured by the surface forces apparatus ( SFA ) ( Li et al . , 2007 ) . The interaction as a function of membrane separation contains two exponentially decaying components with decay constants of 2 . 5 nm ( d1 ) and 6 nm ( d2 ) . The short-ranged component represents membrane repulsion just before fusion , including membrane dehydration ( Leckband and Israelachvili , 2001 ) , and the long-range component results from the steric repulsion between unfolded or partially unfolded t- and v-SNAREs before their association . 10 . 7554/eLife . 03348 . 020Figure 7 . Energy landscape of SNARE zippering perfectly meets the needs of membrane fusion . ( A ) Cartoons of different assembly states of the trans-SNARE complex corresponding to the points indicated in B . ( B ) Free energy of membrane fusion per SNARE complex ( red line ) , a single loaded trans-SNARE complex ( blue ) , or a single unloaded SNARE complex ( black ) as a function of the distance between two membrane surfaces . The experimental and alternative energy landscapes are plotted in solid and dashed lines , respectively . For the unloaded SNARE complex , the free energy at each membrane distance represents the energy of the SNARE complex in the same zippering state as the trans-SNARE complex at that distance . The experimental energy landscape of the unloaded SNARE complex was derived from the measured energy at characteristic points ( marked by circles ) through interpolation . ( C ) Repulsive force between two membranes opposing their fusion . ( D ) Extension of all the unfolded amino acids in Qa and R SNAREs under membrane tension ( gray in the left axis ) or the folded portion of the SNARE complex ( black ) and zippering stage of the amino acids ( A . A . ) in R SNARE ( red line in the right red axis ) . The amino acid number indicates the position of the amino acid relative to the ionic layer ( 0 ) . The amino acids with negative numbers are in N-terminal domain ( NTD ) and those with positive numbers in C-terminal domain ( CTD ) . At each amino acid number in the right axis , the R SNARE motif has assembled from its N-terminus ( A . A . at −24 ) to the amino acid with this number . Note that NTD association is accompanied by a relatively small change in membrane distance compared to CTD zippering , because the extension decrease due to NTD folding is largely canceled by the extension increase of the folded t-SNARE . DOI: http://dx . doi . org/10 . 7554/eLife . 03348 . 02010 . 7554/eLife . 03348 . 021Figure 7—figure supplement 1 . Estimation of the average forces generated by zippering of the N-terminal and C-terminal CTD of neuronal SNARE complex . We estimated the C-terminal domain ( CTD ) zippering forces based on a simplified energy landscape model for a two-state process . The energy landscape is characterized by the experimentally measured energy of the unfolded state and the transition state and their associated positions in terms of extension . The folded state is chosen as a reference 0 here for energy and extension . We chose the average extension change upon CTD unzipping ( 7 . 2 nm , Table 1 ) as the length unit . Thus , the x-axis is the normalized extension and the energy shown in y-axis has a unit of 7 . 2 pN × nm . Based on our measurement , two-thirds of the folding energy is released upon zippering of the C-terminal one-third of the CTD , which leads to the energy landscape shown by the solid black line . The folded state , the transition state , and the unfolded state are located at positions of 0 , 1/3 , and 1 in the unit of total extension change of CTD transition . In this unit , the energy of the unfolded state is equal to the equilibrium force measured for CTD transition ( ∼16 pN ) . The average forces of the C-terminal one-third and the N-terminal two-thirds of CTD can be calculated based on the slopes of the corresponding energy changes , yielding 32 pN and 8 pN , respectively . The average force does not depend on the detailed energy landscape between the transition state and the folded or the unfolded state . Assuming an energy profile of E ( x ) between extension position 0 and a , the average force in this region is defined as〈f〉≡1a∫0a ( dEdx ) dx=Eaa , where f ( x ) ≡dEdx is the local force at position x and E ( 0 ) =0 , E ( a ) =Ea . Thus the average force in a region can be simply determined by the slope of the line going through the energy points at the two ends of the region , regardless of the detailed energy profile in the region . DOI: http://dx . doi . org/10 . 7554/eLife . 03348 . 021 We chose the membrane interaction energy ( V ) per SNARE complex versus the distance between two membrane surfaces at the sites of SNARE attachment ( d ) as ( 1 ) V ( d ) =Em1+α[exp ( −d−dcd1 ) +α exp ( −d−dcd2 ) ] , d≥dc , where Em determines the energy barrier for membrane fusion per SNARE complex when membranes are brought to the minimal distance allowed by the molecular dimension of the fully folded SNARE four-helix bundle ( dc = 1 nm ) ( Sutton et al . , 1998; Stein et al . , 2009; Figure 7B ) . Below this critical distance , membrane fusion occurs irreversibly . The amplitude ratio of the two exponential components ( α ) was set to 0 . 5 based on the SFA measurement ( Li et al . , 2007 ) . The energy barrier for membrane fusion ( N × Em ) and the exact number of SNARE complexes required for fusion ( N ) are under much discussion ( Karatekin et al . , 2010; Mohrmann et al . , 2010; van den Bogaart et al . , 2010 ) . To bypass these uncertainties , we chose Em ≈ 50 kBT per SNARE complex , consistent with our measured folding energy per neuronal SNARE complex and other estimations ( Li et al . , 2007; van den Bogaart et al . , 2010 ) . The role of the LD in membrane fusion is not clear . Evidence suggests that LDs bind membranes and constitute parts of membrane anchors with SNARE transmembrane domains ( Li et al . , 2007; Ellena et al . , 2009; Borisovska et al . , 2012 ) . To simplify our calculations , we did not explicitly consider extension and energy contributions from LDs but assumed instead that membrane fusion occurs when CTD is fully zippered . We computed the energy landscape of the loaded SNARE complex as the sum of the energy of the unloaded SNAREs , the entropic energy of the stretched and unfolded SNARE polypeptides calculated based on Equation 3 in ‘Materials and methods’ , and the membrane interaction energy . At each SNARE zippering stage , an equilibrium membrane distance was calculated by equating the SNARE pulling force to the membrane repulsive force ( Figure 7C , D ) . The calculated SNARE energy was plotted in Figure 7B as a function of the membrane distance . SNARE NTD association was initiated at the very N-termini of t- and v-SNAREs at a large distance of 12 . 5 nm ( Figure 7B–D ) . The association was accompanied by coil-to-helix propagation of the partially disordered t-SNARE towards its C-terminus ( Li et al . , 2014; Figure 7A , state ii ) . Further NTD zippering led to the half-zippered trans-SNARE complex at 9 . 7 nm ( state iii ) ( Bharat et al . , 2014 ) . Thus , NTD association occurs in a narrow distance range of 9 . 7–12 . 5 nm , where the membrane repulsive force is small ( 2–5 pN ) . Formation of the half-zippered state in the presence of membranes leads to a net energy release of ∼26 kBT ( the energy difference between state i and state iii ) ( Gao et al . , 2012 ) , which can be used to dock or prime vesicles and prevent dissociation of the half-zippered trans-SNARE complex . In contrast to NTD association , CTD zippering from the half-zippered state was directly and tightly coupled to membrane fusion ( Figure 7B–D ) . The membrane-loaded half-zippered SNARE complex had an energy of ∼34 kBT relative to its folded state , which consists of ∼27 kBT CTD folding energy and ∼7 kBT entropic energy stored in the stretched VAMP2 CTD . CTD zippering drew two membranes from 9 . 7 nm to 1 nm for fusion against a large average force . As a result , CTD zippering of the trans-SNARE complex reduced the energy barrier of membrane fusion ( Figure 7A , state iv ) to 7 . 4 kBT per SNARE complex ( the energy difference between state iv and state iii ) , consistent with a fusion rate of ∼600 s−1 , where we assumed a maximum fusion rate of 106 s−1 in the absence of any energy barrier ( Yang and Gruebele , 2003; Gao et al . , 2012 ) . Correspondingly , the metastable half-zippered SNARE complex in the absence of a force load was stabilized by the short-ranged membrane opposing force ( Figure 7B ) together with regulatory proteins , such as complexin ( Sudhof and Rothman , 2009; Gao et al . , 2012; Min et al . , 2013 ) . In contrast , the optical trapping force used in our experiments was long-range , with a typical force constant of 0 . 1 pN/nm , compared to an average force constant of 7 . 1 pN/nm for the membrane force opposing CTD assembly . As a result , the half-zippered SNARE state was typically short-lived ( <0 . 2 ms ) upon reassembly of the SNARE complex at the low forces favoring NTD association ( Figure 4 ) , because the trapping force opposing CTD zippering remained small immediately after NTD was zippered . Thus , the membrane's repulsive force was an integral component of SNARE assembly and regulation . The exponential increase in the membrane repulsive force below 5 nm ( from 17 pN to 60 pN , Figure 7C ) required an increasing force output as CTD zippers toward its C-terminus . SNARE zippering indeed met this requirement by producing a high force in this region . Here , the magnitude of local force generated by SNARE zippering was equivalent to the slope of the energy landscape of the unloaded SNARE complex with respect to extension , which also represents the energy density ( defined as the folding energy per unit length of R SNARE polypeptide chain zippered ) distributed along the SNARE bundle . While zippering of the first two-thirds of CTD generated an average force of 8 pN , zippering of the last one-third of CTD produces an average force of up to 32 pN ( Figure 7—figure supplement 1 ) , well suited to counteracting the short-ranged membrane opposing force . The position where CTD changed its energy density ( Figure 7B , at 5 nm ) was intriguing , because it was close to the energy barrier of the trans-SNARE complex . This position overlap is no accident , because any force applied to the SNARE complex tilts the CTD zippering energy landscape of the unloaded SNARE complex toward the unfolded state ( Bustamante et al . , 2004 ) , which tends to make the density-changing point an energy barrier ( for example , see Figure S9 in Xi et al . , 2012 ) . It is this energy barrier that results in the binary CTD transition and the polarized CTD energy distribution . To further corroborate the essential role of the polarized CTD energy distribution in membrane fusion , we calculated the energy landscape of the trans-SNARE complex based on an alternative energy landscape of SNARE zippering ( Figure 7B , black dashed line ) . In this alternative landscape , the energy of CTD is enriched at its N-terminus , rather than its C-terminus , but with the same total CTD zippering energy . The half-zippered trans-SNARE complex is now greatly stabilized by membranes ( at 8 nm of the blue dashed line ) , but nearly unable to fuse them , because the energy barrier for fusion increases to 24 kBT ( the energy difference between states at membrane distances of 1 nm and 8 nm ) compared to 7 . 4 kBT for the wild-type SNARE complex . In this case , significant energy from CTD zippering is not transmitted to membranes , but dissipated as heat . In addition , no additional energy barrier appears before fusion . Similarly , SNAREs alone with this alternative folding energy landscape are expected to zipper or unzip in a continuous manner in response to the force exerted by optical tweezers , in contrast to the observed cooperative two-state manner . Various parameters for the membrane interaction energy have been reported for different model membranes ( Leckband and Israelachvili , 2001 ) . To test how variation in membrane properties may change the requirement for the polarized energy distribution , we repeated our above calculations by changing the parameters in Equation 1 with L1 and α ranging from 1 nm to 3 nm and from 0 to 0 . 5 , respectively . Although energy landscapes of the loaded SNARE complex quantitatively change with these parameters , the energy landscape with a C-terminal polarized energy distribution always led to a much lower energy barrier for fusion than the alternative energy landscapes with an N-terminal polarized energy distribution . Thus , a C-terminal polarized energy distribution is a general requirement for efficient membrane fusion . Taken together , our observed binary CTD transition indicates that the polarized CTD energy distribution is essential for efficient membrane fusion . In contrast , continuous and progressive SNARE assembly leads to poor coupling to membrane fusion . The identification of half-zippered intermediates in all four representative SNARE complexes suggests that SNARE complexes follow a common zippering mechanism to drive membrane fusion ( Hanson et al . , 1997 ) . This observation is not consistent with alternative mechanisms by which the entire SNARE four-helix bundle assembles in an all-or-none manner ( Jahn and Fasshauer , 2012; Kasai et al . , 2012 ) or in a continuous layer-by-layer manner . Instead , our data reveal two distinct cooperative assemblies of the NTD and CTD in the SNARE complex , which clarifies the detailed zippering kinetics . The presence of a partially zippered SNARE complex has been supported by many experiments ( Xu et al . , 1999; Schwartz and Merz , 2009; Walter et al . , 2010; Diao et al . , 2012; Gao et al . , 2012; Min et al . , 2013 ) . Our work further demonstrates that the partially zippered complex is a half-zippered complex intrinsic to a SNARE complex and functionally important for membrane fusion . Mutations and truncations that alter the structure of the half-zippered neuronal SNARE complex and/or its folding energy and kinetics abolish membrane fusion ( Ma L , Gao Y , Yang G , and Zhang YL , manuscript in preparation ) . Thus , such a half-zippered SNARE complex is required for fast and regulated synaptic vesicle fusion ( Kummel et al . , 2011; Li et al . , 2011 , 2014 ) . Our finding suggests that the half-zippered structure is more ancient in SNARE evolution than any regulators that target this structure , and may have more conserved function in membrane fusion than previously thought . We propose that the step-wise assembly enables reversible folding of SNARE complexes , as shown in our calculations . The step-wise and reversible assembly enhances not only the coupling between SNARE zippering and membrane fusion , but also the specificity of SNARE pairing . Furthermore , the half-zippered SNARE complexes may be the target of their cognate Sec1p/Munc18 ( SM ) -family proteins essential for SNARE-mediated membrane fusion ( Shen et al . , 2007; Sudhof and Rothman , 2009; Jorgacevski et al . , 2011; Ma et al . , 2013 ) . For the neuronal SNARE complex , we and others suggested that assembly of the NTD and the CTD has distinct functions: while NTD assembly is responsible for vesicle docking and priming , CTD zippering directly drives membrane fusion ( Walter et al . , 2010; Gao et al . , 2012 ) . Fusion of GLUT4-storage vesicles ( GSVs ) with the plasma membrane mediated by the GLUT4 SNARE complex appears to be very similar to fusion of synaptic vesicles , including distinct stages of vesicle docking , priming , fusion ( Stockli et al . , 2011 ) , and close CTD zippering energy . However , the docked GSVs take about 1 min to fuse after insulin triggering ( Bai et al . , 2007 ) . In this case , the observed fusion rate is probably limited by the slow NTD association , but not the fast CTD zippering . Thus , insulin may mainly regulate steps upstream of NTD association . The CTD zippering energy puts a strong constraint on the detailed mechanism of membrane fusion , including on the number of SNARE complexes required for fusion ( N ) . If the total CTD zippering energy of N trans-SNARE complexes is used to lower the energy barrier of membrane fusion ( Eb ) ( Montecucco et al . , 2005; Mohrmann et al . , 2010 ) , the fusion rate ( k ) should be an exponential function of the total zippering energy , that is , k = k0 × exp ( N × ECTD − Eb ) , where ECTD is the CTD zippering energy per SNARE complex and k0 a pre-constant . The large difference of CTD zippering energy between either endosomal or yeast SNARE complex and neuronal SNARE complex ( 12 or 14 kBT ) suggests that more endosomal or yeast SNARE complexes than neuronal SNARE complexes may be required to mediate fusion ( Mohrmann et al . , 2010; Wickner , 2010; Shi et al . , 2012 ) . SNARE-mediated membrane fusion in vivo involves fixed numbers of SNARE complexes characteristic of different fusion processes , likely controlled by regulatory proteins ( Montecucco et al . , 2005 ) . This result is in contrast with reconstituted SNARE-mediated membrane fusion in vitro , in which the SNARE number is probably not controlled ( Karatekin et al . , 2010; Shi et al . , 2012; van den Bogaart et al . , 2010 ) . As a result , liposome–liposome fusion mediated by the four different SNARE complexes alone exhibit similar fusion rates ( Shen et al . , 2007; Zwilling et al . , 2007; Yu et al . , 2013 ) . Taken together , the difference in CTD zippering energy contributes to the large variation in the fusion rate mediated by SNARE complexes and indicates a different number of SNARE complexes required for different fusion processes in vivo . A common feature of our derived energy landscapes for all tested SNARE complexes is their polarized energy distribution , with much higher energy density at the C-terminus of CTD . This distribution allows SNAREs to increase their force output as CTD zippering draws two membranes into close proximity . Thus , as specialized engines for membrane fusion , SNAREs contain a built-in automatic transmission system that adjusts their force output to accommodate the large force change required for membrane fusion . This system ensures efficient and tight coupling between SNARE zippering and membrane fusion . A mismatch between the force output from SNAREs and the load from membranes would inevitably lead to dissipation of the SNARE zippering energy into heat , reducing the efficiency or the rate of membrane fusion . Thus , membrane fusion requires SNAREs to ‘save the best for last’ . How does the SNARE complex focus its zippering energy to the C-terminus ? Li et al . ( 2014 ) have recently shown that the association of the N-terminal half of VAMP2 to the N-terminal t-SNARE triggers folding of the t-SNARE C-terminal domain ( Figure 7A , state ii ) . The ordered t-SNARE then serves as a template for fast and energetic VAMP2 zippering ( Gao et al . , 2012 ) . Furthermore , tight association between v- and t-SNAREs near the C-terminus of CTD is achieved by key amino acids in that region , including the highly conserved phenylalanine residue shared by the v-SNAREs in all four SNARE complexes ( Figure 1—figure supplement 1 ) . Substitution of the phenylalanine residue with alanine abolishes the binary CTD transition in vitro ( Ma L , Gao Y , Yang G , and Zhang YL , manuscript in preparation ) and exocytosis ( Walter et al . , 2010 ) . These observations strongly suggest that the polarized energy distribution of SNARE complexes is essential for membrane fusion . In summary , as the molecular machine for membrane fusion , SNARE proteins share a working mechanism conserved from yeast to humans . They couple their step-wise folding/assembly to membrane fusion through a distinct half-zippered state . SNAREs contain a built-in transmission system that produces the highest forces at the very C-termini required for efficient membrane fusion . This unified mechanism provides a basis for dissecting the diverse functions of SNAREs in more detail . Amino acid sequences of the SNARE constructs used in our study are listed in Figure 1—figure supplement 1 . The corresponding genes were codon-optimized , synthesized , subcloned into the protein expression pET-SUMO vector , and expressed in BL21 ( DE3 ) Escherichia coli cells as previously described ( Gao et al . , 2012 ) . The proteins were purified using Ni-NTA resin ( GE Healthcare Biosciences , Pittsburgh , PA ) and biotinylated using biotin ligase ( Avidity , Aurora , CO ) . The tweezers were home-built and located in an acoustically isolated room with controlled temperature and air flow as previously described ( Moffitt et al . , 2006; Sirinakis et al . , 2011 ) . The machine was operated remotely through a computer interface written in LabVIEW ( National Instruments , Austin , TX ) . The force and displacement measured by optical tweezers were calibrated by Brownian motion of polystyrene beads in optical traps before each single-molecule experiment . The beads were trapped in aqueous buffer in a microfluidic channel 0 . 2 mm in thickness , which was formed by sandwiching two coverslips with parafilm ( Zhang et al . , 2012 ) . The cysteine-containing SNARE complex was reduced by TCEP or DTT , treated with dithiodipyridine ( DTDP ) , mixed with the thiol-containing DNA handle in a typical 20:1 protein:DNA molar ratio , and cross-linked to the DNA handle overnight . An aliquot of the protein-DNA conjugate was mixed with anti-digoxigenin-coated beads and injected into the microfluidic channel . One DNA-bound bead was caught by one optical trap , brought close to a streptavidin-coated bead held in another optical trap , and formed a single SNARE-DNA tether . The SNARE complex was then pulled at a uniform trap separation speed or held at an approximately constant force or trap separation . The single-molecule folding experiment was performed at room temperature ( 22°C ) in phosphate-buffered saline . An oxygen scavenging system was added to prevent photo-damage of the SNARE-DNA tether ( Gao et al . , 2012 ) . Methods of data analysis are described in detail elsewhere ( Gao et al . , 2012; Xi et al . , 2012 ) and are summarized here . The observed extension and energy changes contain contributions from the structured and unstructured parts of the SNARE protein as well as the DNA handle . The extension of the structured SNAREs was derived from the crystal structure of the SNARE complex and was assumed to be force-independent . The extensions of the unstructured polypeptide and the DNA handle were determined by the worm-like chain model ( Marko and Siggia , 1995; Smith et al . , 1996 ) . Specifically , the extension ( x ) of a worm-like chain is related to the stretching force ( F ) and the contour length ( l ) by the Marko-Siggia formula ( 2 ) F ( r ) =kBTP[14 ( 1−r ) 2+r−14] , where r=x/l , P is the persistence length of the polypeptide ( 0 . 6 nm ) or DNA ( 30–50 nm ) , and kBT = 4 . 1 pN × nm the product of the Boltzmann constant and the room temperature . Extending a worm-like chain decreases its entropy . The associated energy increase can be obtained by integrating the force in Equation 2 with respect to the extension , yielding ( 3 ) E ( l , r ) =kBTPl4 ( 1−r ) ( 3r2−2r3 ) . For the two-state transitions of LD and CTD , we determined the unfolding probability , transition rates , and average state extensions and forces at different trap separations based on the measured extension and force trajectories using a two-state hidden Markov model . Then , we constructed a force-dependent energy landscape model that relates these experimental measurements to model parameters , including free energy of the folded state and transition state and their associated positions . Finally , we fit this model to the experimental data and determined the folding energy , folding energy barrier , and their associated structures . The unfolding energy ( ΔG ) of a protein can be measured based on the mechanical work to reversibly unfold the protein , that is , ( 4 ) ΔG=f1/2×ΔX−E ( Δl , Δx/Δl ) , where f1/2 is the measured equilibrium force , ΔX the corresponding extension change , and E the entropic energy of the unfolded polypeptide . Δx and Δl are the extension change and the contour length change , respectively , of the unfolded polypeptide associated with protein unfolding . However , neither Δx nor Δl in Equation 4 is directly measurable and both are related to ΔX in a model-dependent manner ( see Equations 12 and 13 in Gao et al . , 2012 ) . In addition , our experiments were not performed under exactly constant force , but constant trap separation for maximum spatiotemporal resolution ( Sirinakis et al . , 2012 ) . As a result , a SNARE domain in a two-state transition experiences slightly different average forces in the folded state ( f1 ) and the unfolded state ( f2 ) ( Figure 2B ) . Nevertheless , the average of the two state forces f= ( f1+f2 ) /2 remains constant , which we have simply referred to as force ( Gao et al . , 2011; Figures 5 and 6 ) . Therefore , we constructed a detailed energy landscape model to quantitatively account for the correlation between protein structural transitions and the observed extension changes . We chose the contour length of the unfolded polypeptide directly pulled by optical traps , which is 0 . 365 nm per amino acid , as a reaction coordinate to describe the extension change and the energy landscape associated with SNARE folding and unfolding . Different structural models were used for LD and CTD transitions: in LD transition , the two helices in Qa and R SNAREs fold and unfold symmetrically , whereas in CTD transition , R SNARE folds and unfolds along the pre-structured t-SNARE template ( Kummel et al . , 2011; Gao et al . , 2012; Li et al . , 2014 ) . Using these structural models , we could fit the calculated extension to the measured FEC to determine the contour length parameter associated with each state . The total energy of the single-molecule system additionally includes the harmonic potential energy of two beads in optical traps . The folding energy of the structured part of the SNARE complex as a function of the contour length gives the folding energy landscape of SNARE folding . In our data analysis , this energy landscape was characterized by the free energy of the folded state and the transition state and their associated positions in the reaction coordinate , all relative to the unfolded state . Both energy and positions were chosen as model parameters first to calculate the total system energy and the extension of the SNARE-DNA tether . Then these calculations were used to further compute the opening probability based on the Boltzmann distribution and the folding and unfolding rates based on Kramer's theory , as well as the extension change , for the transition of each SNARE domain . These values from model predictions were fit against the corresponding experimental data by the non-linear least-squares method , which yielded the best-fit model parameters , including the energy of the folded state and the transition state . In our NTD structural model , we incorporated a detailed mechanism for t-SNARE folding induced by NTD association ( Li et al . , 2014 ) . Because the kinetics of this coupled binding and folding process is unclear , we assumed that t-SNARE is gradually structured as VAMP2 starts to zipper from its N-terminus , and becomes fully structured when two-thirds of VAMP2 NTD has been zippered ( Figure 7D ) . This forms a structure corresponding to the transition state of NTD association . Further zippering stabilizes NTD and forms the half-zippered state . Combined with the structural model for CTD transition , a complete model for assembly of the SNARE four-helix bundle was defined . This model also established the structure and the extension of the folded SNARE complex h as a function of the contour length l . The membrane distance d was determined by equating the SNARE pulling force to the membrane repulsive force , that is , ( 5 ) F ( x ) =−V′ ( d ) , where x=d−h−l3/5p2/5/2 is the effective extension of the unfolded polypeptide with contour length l and V′ the derivative of the membrane interaction energy ( Figure 7D ) . The last term in the effective extension expression ( l3/5p2/5/2 ) corrects for the residual extension of the unfolded polypeptide in the absence of external force when one end of the polypeptide is attached to the membrane , which is estimated to be half of the Flory radius of a semi-flexible chain ( Li et al . , 2007 ) . Solving this non-linear equation at different contour lengths , we obtained the membrane distance d at any SNARE zippering stage . The energy landscapes of trans-SNAREs are the total energy of SNAREs ( including the folding energy and the elastic energy of unfolded polypeptide ) and membrane interaction energy as a function of the membrane distance ( Figure 7B ) . The calculations were performed using Matlab codes that are available as source codes .	Many processes in living things need molecules to be transported within , or between , cells . For example , damaged or waste molecules are transported within a cell to structures that can break the molecules down , while nerve impulses are transmitted from one neuron to the next via the release of signaling molecules . Cells—and the compartments within cells—are surrounded by membranes that act as barriers to certain molecules . Vesicles are small , membrane-enclosed packages that are used to transport molecules between different membranes; and in order to release its cargo , a vesicle must fuse with its target membrane . To fuse like this , the forces that act to push membranes away from one another need to be overcome . Proteins called SNARES , which are embedded in both membranes , are the molecular engines that power the fusion process . Once the SNARE proteins from the vesicle and the target membrane bind , they assemble into a more compact complex that pulls the two membranes close together and allows fusion to take place . The final shape of an assembled SNARE complex is essentially the same for all SNARE complexes; however , it is not known whether all of these complexes fold using the same method . Now Zorman et al . have used optical tweezers—an instrument that uses a highly focused laser beam to hold and manipulate microscopic objects—to observe the folding and unfolding of four different types of SNARE complex . All four SNARE complexes followed the same step-by-step process: the leading ends of the SNARE proteins slowly bound to each other; the process paused; then the rest of the proteins rapidly ‘zippered’ together . Zorman et al . revealed that , although the steps in the processes were the same , the energy released in the last step was different when different complexes assembled . This suggests that the energy released by the ‘zippering’ of different SNARE proteins is optimized to match the required speed of different membrane fusion events . Furthermore , Zorman et al . propose that the reason why the majority of energy is released in the later stages of complex assembly is because this is when the repulsion between the two membranes is strongest . The discoveries of Zorman et al . will now aid future efforts aimed at understanding better how the numerous other proteins that interact with SNARE proteins regulate the process of membrane fusion .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2014	Common intermediates and kinetics, but different energetics, in the assembly of SNARE proteins
To investigate the phenomic and genomic traits that allow green algae to survive in deserts , we characterized a ubiquitous species , Chloroidium sp . UTEX 3007 , which we isolated from multiple locations in the United Arab Emirates ( UAE ) . Metabolomic analyses of Chloroidium sp . UTEX 3007 indicated that the alga accumulates a broad range of carbon sources , including several desiccation tolerance-promoting sugars and unusually large stores of palmitate . Growth assays revealed capacities to grow in salinities from zero to 60 g/L and to grow heterotrophically on >40 distinct carbon sources . Assembly and annotation of genomic reads yielded a 52 . 5 Mbp genome with 8153 functionally annotated genes . Comparison with other sequenced green algae revealed unique protein families involved in osmotic stress tolerance and saccharide metabolism that support phenomic studies . Our results reveal the robust and flexible biology utilized by a green alga to successfully inhabit a desert coastline . Green algae play important ecological roles as primary biomass producers and are emerging as viable sources of commercial compounds in the food , fuel , and pharmaceutical industries . However , relatively few species of green algae have been characterized in depth at the genomic and metabolomic levels ( Koussa et al . , 2014; Salehi-Ashtiani et al . , 2015; Chaiboonchoe et al . , 2016 ) . Also , recent genomics studies lack accompanying phenotype studies that could provide valuable context ( Bochner , 2009; Chaiboonchoe et al . , 2014 ) . Analyses of this scope are necessary to better understand the ecology and physiology of microscopic algae ( microphytes ) , both at the local and global scales , and to optimize their cultivation and yield of bioproducts for industrial applications ( Abdrabu et al . , 2016 ) . Currently , species with superior growth characteristics for large-scale cultivation remain understudied , under-developed , and under-exploited ( Fu et al . , 2016 ) . Our study focuses on the green alga – formerly identified as Chloroidium sp . DN1 and accessioned at the Culture Collection of Algae at the University of Texas at Austin ( UTEX ) as Chloroidium sp . UTEX 3007 , which we recently isolated in a screen for lipid-producing algae ( Sharma et al . , 2015 ) . Its oleogenic properties , robust growth , and capacity to survive in a wide range of environmental conditions prompted us to carry out whole-genome sequencing and phenomic analyses . Genomic ( Dataset 1 ) and phenomic ( Dataset 2 ) datasets are available online at Dryad ( Nelson et al . , 2017 ) . We discovered that this species accumulates palmitic acid that , in conjunction with other traits , may enable its survival in a desert climate . The mechanisms employed by desert extremophiles such as Chloroidium sp . UTEX 3007 to maintain cellular integrity despite the oxidative insults of a desert climate may yield insight into the biology of a key player in a desert ecosystem and may also provide resources for the production of highly thermo-oxido-stable oils for human use . One of the most important thermo-oxido-stable oils for human civilization is produced by the oil palm tree , Elaeis guineensisis ( Barcelos et al . , 2015 ) . Palm oil’s distinguishing characteristics stem from its uniquely high concentrations of palmitic acid , which , due to its carbon chain length and lack of reactive double bonds , remains stable in conditions that destroy other longer and more unsaturated fatty acids . However , cultivation of trees for palm oil production has led to extensive destruction of high-biodiversity rainforests and wetlands that threatens to expand around the globe ( Abrams et al . , 2016 ) . In addition to the loss of unique and diverse flora and fauna ( Labrière et al . , 2015; Wich et al . , 2014 ) , clear-cutting these regions for palm oil plantations has devastated major carbon sinks ( Yue et al . , 2015 ) and caused intensive production of smoke pollution that affects millions of people in densely populated areas ( Bhardwaj et al . , 2016; Vadrevu et al . , 2014 ) . Thus , an alternative source of palmitic acid could be valuable in terms of elevated product output , environmental preservation and a reduction of smoke pollution . Chloroidium sp . UTEX 3007 reproduces via unequal autospores ( 2–8 cells/division ) ; cells vary greatly in size from 1 to 12 μm in diameter ( mean = 6 μm; Figure 2a ) and exhibit ovoid morphology , parietal chloroplasts , and accumulation of prominent intracellular lipid deposits ( Figures 1 and 2 ) . Stationary phase is reached after ~2 weeks of growth in F/2 media ( Figure 2 ) . Chloroidium sp . UTEX 3007 showed robust growth in open pond simulators ( OPSs ) , which generate a sinusoidal approximation of daily light based on geographical information system ( GIS ) data on daily light/dark cycles to simulate the growth conditions of an outdoor large-scale algae growth operation ( Figure 2—figure supplement 1 ) ( Lucker et al . , 2014; Tamburic et al . , 2014 ) . We note that , because OPSs tend to underestimate growth rates compared to actual open ponds ( Lucker et al . , 2014; Tamburic et al . , 2014 ) , the actual productivity in an open pond may be higher . Under simulated parameters , growth occurs at 0–60 g/L NaCl and culture maturity ( i . e . , the transition from log to stationary phase ) is reached after about two weeks ( Figure 2b ) . 10 . 7554/eLife . 25783 . 004Figure 2 . Cell size , growth , and lipid accumulation in Chloroidium sp . UTEX 3007 . ( a ) Cell size distribution of Chloroidium sp . UTEX 3007 in late log phase ( 14 days ) . Cell size analysis was performed using a Cellometer Auto M10 from Nexcelcom Bioscience ( Lawrence , MA , USA ) on 2 × 20 ul from each liquid algal sample . As autospores typically generate six progeny , the distribution was fitted with the sixth order polynomial equation: ( a + bx + cx2 + dx3 + ex4 + fx5 + gx6 ) . ( b ) Growth of Chloroidium sp . UTEX 3007 cultures in tris-minimal media supplemented with 0 , 0 . 3 , 0 . 7 . and 1M NaCl at 20°C and 400 umol photons m−m s−s of full-spectrum light ) . Cells were grown in a Multi-Cultivator MC 1000 by Photon Systems Instruments ( Drasov , Czech Republic . See Figure 2—figure supplement 2 for chlorophyll/cell count curve and Figure 2—figure supplement 1 for a growth curve in the open pond simulators ( OPSs ) . ( c ) Fluorescence intensity of cells ( RFUs=relative fluorescence units ) measured with a BD FACSAria III flow cytometer ( BD Biosciences , San Jose , CA ) at mid-log phase ( time=day 1 , one week after inoculation ) and one week into stationary phase ( time=day 14 ) with and without staining ( indicated as ( + ) or ( - ) , respectively , from the lipophilic dye BODIPY 505/515 . Whiskers indicate range , box edges indicate standard deviation and the box centerline indicates the median fluorescence intensity reading . DOI: http://dx . doi . org/10 . 7554/eLife . 25783 . 00410 . 7554/eLife . 25783 . 005Figure 2—source data 1 . Cell diameter measurements . DOI: http://dx . doi . org/10 . 7554/eLife . 25783 . 00510 . 7554/eLife . 25783 . 006Figure 2—source data 2 . Cell concentration time course measurements . DOI: http://dx . doi . org/10 . 7554/eLife . 25783 . 00610 . 7554/eLife . 25783 . 007Figure 2—source data 3 . Flow cytometry measurements . DOI: http://dx . doi . org/10 . 7554/eLife . 25783 . 00710 . 7554/eLife . 25783 . 008Figure 2—figure supplement 1 . OPS growth curve . DOI: http://dx . doi . org/10 . 7554/eLife . 25783 . 00810 . 7554/eLife . 25783 . 009Figure 2—figure supplement 2 . Chlorophyll/cell count curve . DOI: http://dx . doi . org/10 . 7554/eLife . 25783 . 009 Upon entering stationary phase , lipid accumulation commences and causes an increase in size , weight ( 2-6x , ~100–200 pg/cell dry weight ) , and relative fluorescence of lipophilic dye stained cells ( Figures 1 and 2 ) . After reaching stationary phase , Chloroidium sp . UTEX 3007 cultures remain stable and continue to increase biomass at a reduced rate ( Figure 2b ) . Our results indicate that the slowly accruing biomass in post-stationary phase is concurrent with an accumulation of triacylglycerols ( TAGs ) composed of mostly palmitic acid side-chains ( Figure 3a ) . The presence of palmitic acid instead of other longer and more unsaturated fatty acids is expected to increase fitness at higher temperatures because palmitic acid is more thermostable . 10 . 7554/eLife . 25783 . 010Figure 3 . Lipid composition of Chloroidium sp . UTEX 3007 observed with HPLC/MS and GC-FID , and comparison with other photosynthetic , oleagenic species . ( a ) Chloroidium sp . UTEX 3007 fatty acid content estimated by extracting total lipid , creating methyl esters , and running the esters on a GC-FID ( whiskers = range , box boundaries = 1 SD , center box line = mean [8 replicates] ) . ( b ) Comparison of fatty acid profile of Chloroidium sp . UTEX 3007 with that of oil palm ( Elaeis guineensis ) ( Barcelos et al . , 2015 ) and several other algal isolates ( Lang et al . , 2011 ) . Asterisks mark the presence of palmitic acid in Chloroidium sp . UTEX 3007 . ( c ) Base peak chromatograms ( BPCs ) from stationary phase Chloroidium sp . UTEX 3007 and Chlamydomonas reinhardtii extracts run in positive mode on an Agilent LC-MS QToF 6538 ( Agilent , Santa Clara , CA , USA ) using an acetonitrile/ammonium formate/isopropanol gradient . Cultures were grown for three weeks in F/2 media with 0 g/L NaCl ( freshwater media ) . Triacylglycerols ( TAGs ) can be seen in abundance in Chloroidium sp . UTEX 3007 while Chlamydomonas reinhardtii contained a higher ratio of polar lipids ( Dataset 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25783 . 01010 . 7554/eLife . 25783 . 011Figure 3—source data 1 . GC-FID results for major fatty acid species in Chloroidium sp . UTEX 3007 . DOI: http://dx . doi . org/10 . 7554/eLife . 25783 . 01110 . 7554/eLife . 25783 . 012Figure 3—source data 2 . Fatty acid profiles of Chloroidium sp . UTEX 3007 , Elaeis guineensis ( Barcelos et al . , 2015 ) , and several other algal isolates ( Lang et al . , 2011 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25783 . 01210 . 7554/eLife . 25783 . 013Figure 3—source data 3 . HPLC-MS base peak chromatograms ( BPCs ) for Chloroidium sp . UTEX 3007 and Chlamydomonas reinhardtii extracts . DOI: http://dx . doi . org/10 . 7554/eLife . 25783 . 013 We performed gas ( GC-FID for fatty acids and GC-MS for polar primary metabolites ) and liquid chromatography ( UHPLC/Q-TOF-MS/MS ) on cellular extracts of Chloroidium sp . UTEX 3007 to characterize its intracellular metabolites . Fatty acids were extracted for gas chromatography with flame-ionization detection ( GC-FID ) analysis at one week after the onset of stationary phase . Fatty acid methyl esters were created to compare their abundance with a fatty acid methyl ester ( FAME ) standard library . Comparison with an internal standard ( C15:0 ) revealed that fatty acids had accumulated to 78 . 2 ± 7 . 7% ( from 8 replicates ) , and ~41 . 8% of total fatty acids consisted of palmitic acid at the time of harvest ( Figure 3a ) . The palmitic acid content of Chloroidium sp . UTEX 3007 was found to be higher than many other known algae ( Lang et al . , 2011 ) roughly equivalent to that of palm oil from Elaeis guineensis ( Barcelos et al . , 2015 ) ( Figure 3b ) . We performed UHPLC/MS-QToF to identify intact and distinct lipid species in Chloroidium sp . UTEX 3007 and Chlamydomonas reinhardtii ( CC-503 ) ( Figure 3c ) . Microwave-assisted methanol whole-cell extracts were filtered and used for LCMS analyses ( Dataset 2 ) . Chlamydomonas reinhardtii contained more membrane-type lipids including sphingolipids and monogalactosyl diacylglycerides ( MGDGs ) , which indicated that the accumulation of neutral lipids had not yet commenced under these conditions . Chloroidium sp . UTEX 3007 contained a larger fraction of triacylglycerols at the conditions tested ( 3 weeks at 25°C , 400 μmol photons/m2/s ) . In stationary phase Cloroidium sp . UTEX 3007 cells , we observed a marked accumulation of mono- , di- , and triacylglycerides . These larger lipid molecules formed six notable peaks at the end of the chromatograms where the concentration of isopropanol in the elution solvent was between 70–90% . Chloroidium sp . UTEX 3007 cultures grown for the same duration , under the same conditions , and having the same chlorophyll a content as Chlamydomonas reinhardtii contained a dramatically higher percentage of triacylglycerols in the final extracts ( Figure 3c ) . To define the spectrum of nutrients and metabolites that support heterotrophic growth , we exposed the strain to hundreds of chemical compounds in phenotype microarray plates ( Figure 4a , Dataset 2 ) . This method has been used successfully to identify new compounds that support heterotrophic growth in the model green alga Chlamydomonas reinhardtii ( Chaiboonchoe et al . , 2014 ) . Chloroidium sp . UTEX 3007 grew on more than 40 different carbon sources , including several desiccation-promoting sugars including trehalose , sorbitol , raffinose , and palatinose ( Dataset 2 ) . 10 . 7554/eLife . 25783 . 014Figure 4 . Metabolic profiling of Chloroidium sp . UTEX 3007 . ( a ) Biolog phenotype microarrays were run using an Omnilog instrument ( Biolog Inc . , Hayward , USA ) as previously described ( Chaiboonchoe et al . , 2014 ) . In total , 380 substrate utilization assays for carbon sources ( PM01 and PM02 plates ) , 95 substrate utilization assays for nitrogen sources ( PM03 plate ) , 59 nutrient utilization assays for phosphorus sources , and 35 nutrient utilization assays for sulfur sources ( PM04 plate ) , along with peptide nitrogen sources ( PM06-08 plates ) were performed ( Dataset 2 ) . All microplates were incubated at 25°C for up to 8 days , and the dye color change ( in the form of absorbance ) was read with the Omnilog system every 15 min . As the Omnilog instrument does not provide a source of continuous light during incubation , the algae are assumed to be carrying out heterotrophic respiration . In addition , a marked increase in the chlorophyll a content and total cell count was confirmed for wells with suggested growth . Kinetic curves were plotted from the raw data in the form of heatmaps , and statistical analysis was carried out to visualize the metabolic properties and generate Omnilog values . Heatmap density correlates to Omnilog-registered color density . In addition to the dye color change , Omnilog also registers color change resulting from the accumulation of other pigments , including chlorophyll a . Thus , growth is displayed as cumulative color change density . ( b ) Extraction and analysis by gas chromatography coupled with mass spectrometry was performed as described in Lisec et al . ( 2006 ) . The color scale corresponds to chromatogram peak areas reported in the GC-MS results in Dataset 2 . Significant increase of diverse carbon compounds was observed in Chloroidium sp . UTEX 3007 ( Cm ) as compared to Chlamydomonas reinhardtii ( Cre ) , and vice versa for nitrogen compounds . These differences may reflect acclimatization to their respective habitats and lifestyles . DOI: http://dx . doi . org/10 . 7554/eLife . 25783 . 01410 . 7554/eLife . 25783 . 015Figure 4—source data 1 . Phenotype microarray results for plates PM1 , PM2 , and PM3 . DOI: http://dx . doi . org/10 . 7554/eLife . 25783 . 01510 . 7554/eLife . 25783 . 016Figure 4—source data 2 . GC-MS results for Chloroidium sp . UTEX 3007 ( Cm ) and Chlamydomonas reinhardtii ( Cre ) intracellular polar metabolites . DOI: http://dx . doi . org/10 . 7554/eLife . 25783 . 016 Although Chloroidium species have been documented to accumulate the pentose ribitol , to date , no green alga has been shown to use a pentose sugar ( for example arabitol and lyxose , which Chloroidium sp . UTEX 3007 can assimilate ) for heterotrophic growth . Thus , our experiments show that some green algae have a wider range of sugar utilization than previously recognized . Because Chloroidium sp . UTEX 3007 was able to use common plant polysaccharides and sugars , including cellulose , fructose , and fucose for heterotrophic growth ( Figure 4 ) , our nutrient assays suggest a herbivoric capacity for the Chloroidium species . Strikingly , growth on L-glucose was also observed in our assays . In contrast to Chlamydomonas reinhardtii , which can only use acetate as a carbon source for heterotrophic growth , Chloroidium sp . UTEX 3007 lacked the capacity to assimilate extracellular acetate for growth . We found that a different profile of nitrogen sources supports growth in Chloroidium sp . UTEX 3007 in comparison with Chlamydomonas reinhardtii . Although Chlamydomonas reinhardtii can use a broad range of nitrogen sources , including several D-amino acids ( Chaiboonchoe et al . , 2014 ) , major nitrogen sources promoting growth of Chloroidium sp . UTEX 3007 were limited to ammonium , nitrate , urea , agmatine , ornithine , and glutamine ( Figure 4a ) . These differences in nutrient assimilatory capacities might reflect differences in the habitats of the two algae , such as the composition of the soil or other species in their immediate vicinity . A desert environment may offer only limited nitrogen sources , while the fertile soil of typical Chlamydomonas reinhardtii habitats , that is , the temperate climate zone , often contains a wide variety of nitrogen sources ( Harris , 2009 ) . In addition to GC-FID and HPLC-MS , we also performed metabolite profiling using GC-MS to examine other simple carbon and nitrogen compounds in Chloroidium sp . UTEX 3007 and Chlamydomonas reinhardtii . While Chloroidium sp . UTEX 3007 accumulated a wide assortment of sugars , a much broader range of nitrogen metabolites was accumulated by Chlamydomonas reinhardtii ( Figure 4b ) . These results complement the phenotype microarray results . Carbon compounds accumulated by Chloroidium sp . UTEX 3007 include the desiccation resistance-promoting sugars arabitol , ribitol , and trehalose ( Figure 4b ) . Based on previous findings , ( Santacruz-Calvo et al . , 2013; Watanabe et al . , 2016 ) , these sugars and lipids are likely to be involved in osmotic stabilization in C . sp . UTEX 3007 . We sequenced and annotated the genome of Chloroidium sp . UTEX 3007 at high depth ( ~200 x ) with PCR-free Illumina reads to yield a 52 . 5 Mbp genome ( N50 = 148 kbps ) . Although we did not perform classical chromosome quantification , we estimate that Chloroidium sp . UTEX 3007 has approximately 16 nuclear chromosomes based on synteny analyses performed with Coccomyxa subellipsoidea C-169 ( 20 chromosomes ) and Chlorella variabilis NC64A ( 12 chromosomes ) . Chlorella-type ( G3T3A ) repeat telomeres were discovered on nuclear contigs . Protein-coding sequences , comprising 40 . 0% of the genome , contained 9455 distinct Pfam domains ( Dataset 1 , within 8155 genes ) as predicted with Pfam-A . hmm ( v31 . 0 ) . Gene predictions using an Arabidopsis thaliana hidden Markov model ( HMM ) in SNAP ( Korf , 2004 ) were used for downstream analyses ( Dataset 1 ) . Sequence similarity analysis of the predicted proteome to known species revealed matches at or below E-values <10−10 with proteins from species of Coccomyxa and Chlorella ( Dataset 1 ) . We hypothesize that Chloroidium sp . UTEX 3007 is epiphytic in nature , and that such an ecological niche is responsible for the broad carbon assimilatory behavior we observed in both the ( 1 ) phenotype microarrays and ( 2 ) reconstructed genetic pathways from genomic data . The retention of these sugar-inclusive metabolic pathways may provide survival advantages in a desert climate due to the osmotic stability and desiccation tolerance conferred by some of the produced sugars . From our functional annotation of Chloroidium sp . UTEX 3007’s genome , we reconstructed a genome-scale metabolic pathway model by mapping predicted genes to metabolic network nodes reconstructed from pathways in the BioCyc database ( https://biocyc . org/ ) . The model can be found in Dataset 1 . Our metabolic model highlights known pathways of potential lipid biosynthesis in Chloroidium sp . UTEX 3007 . This analysis also revealed unusual lipid biosynthesis mechanisms including TAG biosynthesis from membrane lipids instead of an acyl-CoA pool . This has previously been documented in some yeast and plant species that use phospholipids as acyl donors and diacylglycerol as the acceptor ( Cases et al . , 2001; Dahlqvist et al . , 2000 ) . To investigate possible pathways in Chloroidium sp . UTEX 3007 that could synthesize TAGs from polar membrane lipids , we examined enzymes with phospholipid-cleaving or phospholipid-translocating activities . Our BLASTP/BLAST2GO analyses highlighted an abundance of phospholipid-translocating ATPases that might be involved in this process ( Dataset 1 ) . From our protein family domain analysis , we detected lecithin retinol acyltransferase ( LRAT ) and phospholipase D ( PLD ) domain-containing enzymes in Chloroidium sp . UTEX 3007 ( Figure 5 ) that are also candidate proteins for rapidly modifying membrane lipids ( Dataset 1 ) . Membrane-bound LRAT enzymes are involved in the transfer of palmitoyl groups to and from a variety of biomolecules ( Furuyoshi et al . , 1993; Golczak and Palczewski , 2010 ) , and several phospholipases , including those from the PLD family , cleave phosphate groups from membrane lipids . One predicted PLD in Chloroidium sp . UTEX 3007 contains several C2 calcium-binding domains , indicating that it acts on membrane phospholipids to produce membrane-bound phosphatidic acid ( PA ) and cytosolic choline ( Rahier et al . , 2016 ) . PA has noted roles in environmental stress response ( Peppino Margutti et al . , 2017 ) , and its liberation has been shown to create ‘supersized’ lipid droplets ( Fei et al . , 2011 ) . We hypothesize that phospholipases acting on membrane lipids could play particularly important roles in osmotic stress resistance and lipid accumulation in Chloroidium sp . UTEX 3007 as they represent a key branch point between polar and non-polar lipid species . 10 . 7554/eLife . 25783 . 017Figure 5 . Phospholipase D ( PLD; EC number: 3 . 1 . 1 . 4 ) domain-containing gene in Chloroidium sp . UTEX 3007 . Functional protein family domains in a Chloroidium sp . UTEX 3007 predicted protein with high similarity to PLD proteins from closely related organisms . PLDs are members of the phospholipase superfamily and produce phosphatidic acid as a main product . Phosphatidic acid is involved in signaling , membrane curvature , and is rapidly converted to diacylglycerol . The Chloroidium sp . UTEX 3007 PLD contains several domains with calcium-binding , amidation , and phosphorylation sites ( annotated as C2 ( 2 ) , A ( 2 ) , or P ( 11 ) domains ) . C2 domains act to target proteins to cell membranes and allow phosphatases to de-phosphorylate membrane lipids without removing them from the membrane . DOI: http://dx . doi . org/10 . 7554/eLife . 25783 . 01710 . 7554/eLife . 25783 . 018Figure 5—source data 1 . Locations , confidence scores , and accession numbers for PLD and C2 Pfam domains in Chloroidium sp . UTEX 3007 . DOI: http://dx . doi . org/10 . 7554/eLife . 25783 . 018 We sought to compare and contrast Chloroidium sp . UTEX 3007’s genome with other species from diverse clades within the green algae . Our approach compares the euryhaline Chloroidium sp . UTEX 3007 with salt- and fresh-water-dwelling species and highlights Pfam domains that are unique to Chloroidium sp . UTEX 3007 . The genomes of Chlorella variabilis NC64A ( Blanc et al . , 2010 ) , Micromonas pusilla ( Worden et al . , 2009 ) , Ostreococcus tauri ( Blanc-Mathieu et al . , 2014 ) , and Coccomyxa subellipsoidea ( Blanc et al . , 2012 ) were downloaded from the National Center for Biotechnology Information ( NCBI ) , and the Chlamydomonas reinhardtii genome ( Merchant et al . , 2007 ) was downloaded from Phytozome ( v5 . 5; www . phytozome . net ) . De novo gene prediction using an A . thaliana HMM in SNAP yielded a set of peptide predictions ( Dataset 1 ) that served as a base for HMM alignment with Pfam-A ( v31 . 0 , ( http://hmmer . org ) ) . Figure 6 provides a graphical representation of unique and shared Pfam domains among the six algal species in an interactive Venn diagram ( Figure 5—source data 1 ) using InteractiVenn ( http://www . interactivenn . net/ ) ( Heberle et al . , 2015 ) . Chlorella sp . NC64A contained the lowest number of unique Pfam domains ( 174 ) , while , surprisingly , Ostreococcus tauri , known for its minimal genome , had the highest number of unique Pfams ( 427 ) . 10 . 7554/eLife . 25783 . 019Figure 6 . Protein family ( Pfam ) domains in Chloroidium sp . UTEX 3007 compared with algae from other clades . Well-curated assemblies from genomes of algae from four other clades were downloaded from NCBI [assemblies - Micromonas pusilla , Ostreococcus taurii , and Coccomyxa subellipsoidea] . The Chlamydomonas reinhardtii genome was downloaded from Phytozome ( v5 . 5 ) ( Merchant et al . , 2007 ) . De novo gene prediction , including exon-intron structural modeling , yielded a set of peptide predictions that served as a base for HMM alignment with Pfam-A ( v31 . 0 , ( http://hmmer . org ) . ) . ( a ) Shared and unique Pfam sets from representative species of major green algae clades . Pfam lists can be viewed by selecting the numbers displayed for each shared or unique value in Figure 6—source data 3 . ( b ) Legend for ( a ) including additional metrics describing the algal assemblies used for de-novo coding sequence and Pfam predictions . DOI: http://dx . doi . org/10 . 7554/eLife . 25783 . 01910 . 7554/eLife . 25783 . 020Figure 6—source data 1 . Predicted Pfam designations for each species in Figure 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 25783 . 02010 . 7554/eLife . 25783 . 021Figure 6—source data 2 . Table with QUAST results used in ( b ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25783 . 02110 . 7554/eLife . 25783 . 022Figure 6—source data 3 . Interactive Venn diagram that can be viewed at interactivenn . net to obtain Pfam sets for numbers displayed in Figure 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 25783 . 022 In comparison with other algae , we found 235 Pfam domains that were unique to Chloroidium sp . UTEX 3007 and may play roles in its inhabitation of a desert region . Selected , unique , Pfam entries with relevance to phenotypes documented in this manuscript and unique to Chloroidium sp . UTEX 3007 are shown in Table 1 . For example , the detection of a NST1 domain suggests the presence of proteins involved in cellular response to hypo-osmolarity ( Leng and Song , 2016 ) . Similarly , the presence of a proline-rich domain is associated with thermotolerance ( Cvikrová et al . , 2012; Khan et al . , 2013 ) . In all other cases for the abiotic stress column , oxidative stress resistance , either direct or metal-mediated , is suggested . Because mechanisms to deal with oxidative stress from a variety of sources ( sun , salt , drought , heat ) are necessary to survive in a desert climate , gene products encoded with these domains may play a key role in the successful colonization of a desert region , and its various sub-habitats , by Chloroidium sp . UTEX 3007 . 10 . 7554/eLife . 25783 . 023Table 1 . Protein families ( Pfams ) with roles in ( a ) abiotic stress resistance and ( b ) saccharide metabolism unique to Chloroidium among the green algae explored . Pfam domains were predicted using HMMsearch against the Pfam-A ( v31 . 0 , http://hmmer . org ) database . The i-Evalue or the ‘independent E-value’ , signifies the E-value that the query would have received if it were the only domain , irrespective of homology with any other entries in the database thereby providing a stringent confidence metrics for the hit ( http://hmmer . org ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25783 . 023DescriptionKeyi-Evalue ( a ) Iron-containing redox enzymeHaem_oxygenas_27 . 50E-05Salt tolerance down-regulatorNST10 . 00012Proline-richPro-rich0 . 00015Peroxisome biogenesis factorPEX-2N0 . 003OsmC-like proteinOsmC0 . 0066NA , K-Atpase interacting proteinNKAIN0 . 003MetallothioneinMetallothio_20 . 0074Mercuric transport proteinMerT0 . 0054 ( b ) Beta-galactosidase jelly roll domainBetaGal_dom4_50 . 013Cryptococcal mannosyltransferaseCAP59_mtransfer3 . 50E-15Carbohydrate binding domain CBM49CBM490 . 0085Carbohydrate binding domain ( family 25 ) CBM_250 . 0023Carbohydrate binding domainCBM_4_90 . 00036Cellulose biosynthesis protein BcsNCBP_BcsN0 . 0018Glycosyl hydrolase family 70Glyco_hydro_700 . 00054Glycosyl hydrolase family 9 ( heptosyltransferase ) Glyco_transf_90 . 0025Carbohydrate esterase/acetylesteraseSASA2 . 00E-06Activator of aromatic catabolismXylR_N0 . 011 BLAST2GO analysis revealed that Chloroidium sp . UTEX 3007 contains a disproportionately high number of genes involved in redox chemistry and metal binding ( Dataset 1 ) Exploring these genes further with functional and phylogenetic analyses highlighted a cluster of genes ( CDS1/2/3 ) with homologs in desiccation-associated proteins from Deinococcus species ( Figure 7 , Figure 7—figure supplement 1 ) . 10 . 7554/eLife . 25783 . 024Figure 7 . Chloroidium sp . UTEX 3007 genes exhibiting homology with desiccation-responsive Deinococcus genes . ( a ) Gene models for the putative manganese catalases ( CDS1/2/3 ) . Functional annotation of homologs from ( CDS1/2/3 ) is limited: 96 of the top 100 BLAST hits of CDS3 in the non-redundant database ( nr v . 2 . 31 , NCBI ) are annotated as hypothetical or predicted proteins . Of the other four , three are characterized as desiccation related ( from Auxenochlorella protothecoides , Deinococcus gobiensis I-0 , and Salinisphaera shabanensis E1L3A ) . ( b ) Experimental structures of representative Mn catalases ( Antonyuk et al . , 2000; Barynin et al . , 2001; Bihani et al . , 2013 ) showing the homohexamers ( left ) and isolated monomers ( right ) . The latter are colored according to structural motifs as given . ( c ) Left: Homology model of CDS3 , colored according to linear structure shown below ( the additional presumptive loop is indicated by the red arrow and the predicted Mn coordinating side chains are shown as sticks ) . Right: Zoom in of same homology model aligned with the metal coordinating residues of Anabaena PCC 7120 Mn catalase ( pdb 4R42 ) ; highlighting the spatial alignment of side chains ( homology model side chains shown as thick sticks , 4R42 as thin sticks ) . ( d ) Sequence alignment of CDS1-3 shown with the predicted secondary structure . The presumptive helical regions are also aligned with the corresponding sequences from the known Mn catalase structures . Canonical Mn coordinating residues are in red and other positions with sequence identity between CDS1-3 and at two least of the known catalases are in bold . DOI: http://dx . doi . org/10 . 7554/eLife . 25783 . 02410 . 7554/eLife . 25783 . 025Figure 7—source data 1 . Models of CDS1-3 in protein data bank ( PDB ) format . DOI: http://dx . doi . org/10 . 7554/eLife . 25783 . 02510 . 7554/eLife . 25783 . 026Figure 7—source data 2 . Amino acid residue alignment of CDS1-3 . DOI: http://dx . doi . org/10 . 7554/eLife . 25783 . 02610 . 7554/eLife . 25783 . 027Figure 7—figure supplement 1 . Alignment of CDS3 to a Deinococcus globiensis desiccation-related protein ( Yuan et al . , 2012 ) . Metal-binding motif residues are highlighted . DOI: http://dx . doi . org/10 . 7554/eLife . 25783 . 027 The proteins from ( CDS1/2/3 ) contain ferritin-like domains and may be involved in neutralizing reactive oxygen species ( ROS ) via manganese redox cycles ( Daly , 2009; Fredrickson et al . , 2008 ) . Consistent with the primary sequence analysis , homology-based tertiary structure modeling confirmed that these proteins contain the metal binding motifs and extended C-terminal tails characteristic of manganese ( Mn ) catalases ( Whittaker , 2012 ) ( Figure 7 ) . As intracellular iron can become toxic in the face of overwhelming oxidative stress from changing environments , the utilization of manganese as a replacement would offer a protective advantage in the face of severe oxidative stress ( Daly , 2009; Fredrickson et al . , 2008 ) . Halo-tolerant algae are in demand for the cost-effective commercial production of a variety of compounds ( Chokshi et al . , 2015 ) . Euryhaline algae such as Chloroidium sp . UTEX 3007 allow algae cultivators to use local water sources including lakes , inland seas , or oceans . As fresh water is a scarce resource , the ability to produce commercial products photosynthetically by mass culture of algae using alternative water sources , including salt water , could lead to a reduced environmental footprint in the mass culture of algae . We found that Chloroidium sp . UTEX 3007 accumulated high levels of palmitic acid and various sugars ( Figures 1–4 ) . As growth and cell cycle progression involves the global synthesis and breakdown of an array of proteins and signaling molecules , the energy that the cell would usually spend in cell cycle progression and for biomass production during log phase might funnel directly into the accumulation of various carbon compounds in stationary phase . Chlorophytes are known to assume cystic morphology to survive desert conditions , and completely desiccated algae have been shown to survive until they reach nutrient-rich environments ( Abe et al . , 2014 ) . The retention of chlorophyll a well into stationary phase indicates that Chloroidium sp . UTEX 3007 cells may transition into a carbon storage state wherein cells photosynthetically fix carbon as resting autospores . We propose that Chloroidium sp . UTEX 3007 accumulates stable hydrocarbons such as palmitic acid to survive the extended periods of heat , drought and nutrient depletion that are characteristic of desert regions . As an example of the potential importance of palmitic acid in thermotolerance , a mutant of Arabidopsis deficient in palmitic acid desaturation had an increased optimal growth temperature range ( Kunst et al . , 1989 ) . Likewise , the various sugars that accumulated in Chloroidium sp . UTEX 3007 have been shown to promote desiccation tolerance in a wide variety of species ( Bradbury , 2001; Breeuwer et al . , 2003; Petersen et al . , 2008; Petitjean et al . , 2015; Santacruz-Calvo et al . , 2013; Watanabe et al . , 2016 ) . Uncommon sugars are known to be involved in the stress responses of green algae , but their constituents vary widely between even closely related species ( Gustavs et al . , 2010 ) . For example , a study focusing on sugars in aeroterrestrial algae found sorbitol in the Prasiola clade , ribitol in the Elliptochloris and Watanabea clades , and erythritol in Apatococcus lobatus sap ( Darienko et al . , 2015; Gustavs et al . , 2010 ) , the expanded carbon source utilization capacity we observed in our phenotype microarray experiments supports the sequence data in placing this isolate within the herbivoric/saprophytic/aeroterrestrial Watanabea clade of green algae ( Darienko et al . , 2010 ) . The proximity of Chloroidium sp . UTEX 3007 isolation sites with date palm trees ( Phoenix dactylifera ) and mangrove species ( Avicennia marina ) indicates that these desert- and saline-hardy trees may be hosts for Chloroidium sp . UTEX 3007 . An anomaly in the natural world , heterotrophic growth on L-glucose could further indicate a close relationship with plants . Of the very few species documented to use L-glucose for growth , two species , one bacterium and one fungal species , live endo-/epi-phytic lifestyles ( Sasajima and Sinskey , 1979; Shimizu et al . , 2012 ) . L-glucose biosynthesis in a plant system was characterized in the 1970's ( Barber , 1971 ) . Barber concluded that L-glucose biosynthesis proceeds via an epimeration of the D-mannose moiety of guanosine 5’-diphosphate D-mannose to the L-galactose of β-L-galactose 1-phosphate . Our data suggest that the uptake of L-glucose may proceed through a mannose-nucleotide intermediate ( Lunn et al . , 2014 ) . The genome of Chloroidium sp . UTEX 3007 revealed many potential cellular mechanisms for dealing with osmotic stress ( Figure 3 , Table 1 ) . Due to the high temperatures encountered in its natural habitat , the presence of longer unsaturated fatty acids may be detrimental to membrane stability and greater quantities of stable fatty acids such as palmitic acid may be required . As unsaturated lipids are at greater risk for peroxidation upon salt stress ( Shu et al . , 2015 ) , an increase in saturated membrane lipids can also be protective against rapid shifts in salinity . Our observations suggest that proteins involved in salt stress response can translate both into heat stress tolerance and increased lipid accumulation due to their release of PA from membrane lipids . As increased PA has been shown to cause ‘supersized’ lipid droplets ( Fei et al . , 2011 ) , and salt stress promotes lipid storage ( Wang et al . , 2016 ) , the accumulation of lipids in response to salt stress may be occurring through the action of PLD . Several studies have found that the overexpression of transgenic PLD confers salt and drought tolerance in plants , however a trans gene from a euryhaline organism has not yet been tested ( Ji et al . , 2017; Wang et al . , 2014; Yu et al . , 2015 ) . As such , the PLD from the euryhaline Chloroidium might present an attractive target for biotechnology efforts . The stores of palmitate we observed may play other roles than energy storage or resistance to heat and salt stress . A pool of palmitic acid may be necessary for dynamic palmitoylation of membrane-docked ion pumps . Palmitoylation , mediated by LRAT , can increase targeted regional hydrophobicity and is an expected requisite for a massive influx of ion channels into a cell membrane . However , although at least three subunits of the human cardiac sodium pump are regulated by palmitoylation , the full functional outcome of these palmitoylation events is still not well characterized ( Howie et al . , 2013 ) . The increased targeting of ion pumps into Chloroidium sp . UTEX 3007 membranes by palmitoylation could also assist in surviving rapid salinity shifts and hypersaline conditions . Chloroidium sp . UTEX 3007 appeared to have an especially sophisticated network of genes involved in metal metabolism . As metal-containing enzymes have important roles in oxidative stress reduction and in protein folding in the face of severe environmental conditions , these genes may promote survival in a desert region . Daly and colleagues have proposed that Mn redox cycles are a critical mechanism by which desiccation and radiation tolerant organisms avoid damage caused by reactive oxygen species ( ROS ) , which can be generated by spurious Fe-catalysis ( Daly , 2009; Fredrickson et al . , 2008 ) . An intermediate protective step to prevent ROS damage is the decomposition of hydrogen peroxide , as accumulated hydrogen peroxide can react with free iron via Fenton chemistry to generate extremely reactive hydroxyl radicals ( Daly , 2009; Fredrickson et al . , 2008 ) . The radiation tolerance of organisms shows a positive correlation with a higher intracellular Mn to Fe ratio . Moreover , Mn catalase overexpression provides oxidative protection in cyanobacteria ( Fenton , 1894 ) , and homologs of these genes are upregulated in response to desiccation stress in other Chlorella strains ( Gao et al . , 2014 ) . Thus , the cumulative evidence suggests that Mn catalases in Chloroidium sp . UTEX 3007 and other species could be used to mitigate ROS and cellular damage from environmental stressors . The focus of our study , Chloroidium sp . UTEX 3007 , grows robustly in a broad spectrum of conditions and accumulates palmitic acid . Thus , it may serve as an alternative source for this valuable fatty acid . The demand for palmitic acid has caused public pressure for the increased cultivation of oil palm trees ( Abrams et al . , 2016 ) . However , the traditional cultivation and harvest of palm oil involves environmentally damaging practices that destroy high-biodiversity rainforests and produce large volumes of smoke pollution ( Newbold et al . , 2015 ) . Although Southeast Asia is now the primary source of palm oil , new plantations are rapidly growing in Africa and Central America at the expense of rainforests that are no less diverse or unique than those of Asia ( Barcelos et al . , 2015 ) . Recent studies have shown that transformation of native rainforest land causes approximately 50% reduction in species diversity , density , and biomass in animal communities ( Yue et al . , 2015 ) . This is a critical global issue , as the pace of deforestation has accelerated dramatically in recent years ( Newbold et al . , 2015 ) . The development of an alternative method for producing palm oil substitutes is therefore highly valuable , but thus far no viable option has been found . We envision that our characterization of Chloroidium sp . UTEX 3007 and its associated genome and phenome information can be used to develop resources for the sustainable production of palm oil alternatives . Figure supplements and source data can be found online at Dryad ( Nelson et al . , 2017 ) . Chloroidium sp . UTEX 3007 was sequenced with the Illumina HiSeq 2500 ( Illumina , San Diego , USA ) as described previously ( Sharma et al . , 2015 ) . We performed an assembly with the CLC Genomics Workbench assembler ( v8 . 5 , CLC Genomics , Qiagen , Aarhus , Denmark ) with a kmer length of 45 , word size of 22 . After assembly , reads were mapped back to contigs and contigs were updated according to the mapped reads ( 95% sequence identity required ) . The insert size for paired genomic reads ( 2 × 100 bp ) was ~720 bp on average and reached a maximum of 1200 bp . The reads assembled into 710 scaffolds with N50 of 150 . 8 kbp and a total length of 52 . 5 Mbp . For mapping , 335 , 693 , 730 reads were obtained of which 323 , 780 , 836 ( 98 . 6% ) reads were matched to the final assembly ( Dataset 1 ) . Using an Arabidopsis thaliana HMM in SNAP ( Korf , 2004 ) to predict gene models , we predicted 42 , 000 individual transcripts and their translated protein amino acid sequences . Kingdom-specific HMM libraries have been found to increase the sensitivity and accuracy of genome annotations ( Alam et al . , 2007 ) , and we found that using the A . thaliana HMM as a chlorophyte-specific HMM yielded the best gene models ( in terms of intact Pfam domains ) for Chloroidium sp . UTEX 3007 as well as algae from several other lineages . The assemblies of ther algae used in the manuscript for comparative analyses ( Figure 6 and Table 1 ) were analyzed with QUAST and the results are presented in Figure 6 . The full sets of resulting annotations are available as transcripts and proteins in fasta ( . fa ) format and GFF ( . gff ) annotation files ( Dataset 1 ) . All assemblies used in this manuscript were analyzed by us using QUAST ( Gurevich et al . , 2013 ) . We used the predicted genes/transcripts/proteins from SNAP ( default parameters ) for functional annotation , metabolic network reconstruction , and comparative analyses . The predicted proteins were analyzed for Pfam domains ( Rodrigues et al . , 2015 ) using HHMER ( Sinha and Lynn , 2014 ) for protein family analysis ( E-value <0 . 01 , Pfam-A . hmm ( v31 . 0 , current ) [Finn et al . , 2014] ) and analyzed for similarity to known proteins using BLASTP ( v2 . 2 . 31 ) to create reference tables for annotated gene functions ( Dataset 1 ) . A model of the Chloroidium sp . UTEX 3007 metabolic network was reconstructed from functional gene annotation and EC ( enzyme code ) evidence from BLASTP ( Dataset 1 ) searches using the BLAST2GO v1 . 1 . 0 functional annotation tool ( https://www . blast2go . com/ ) ( Jones et al . , 2014 ) , which includes profile-based searches , to obtain model component genes ( Dataset 1 ) . The curated list of genes was processed using Pathway Tools v19 . 0 ( Karp et al . , 20092010 ) software . The pathologic function of Pathway Tools generated the genome-scale metabolic pathway model by matching our EC assignments with known enzyme/reaction references of MetaCYC ( Caspi et al . , 2014 ) . Enzymes that were deleted or modified by the enzyme commission were manually updated . A total of 1400 proteins associated with 1448 genes , of which , 441 were associated with more than one EC number . The final model has 229 pathways , 1651 enzymatic reactions , 17 transport reactions , 1464 polypeptides , 22 transporters , and 1245 compounds . PathoLogic assigned pathways include the following classes: activation/inactivation/interconversion ( 3 ) , biosynthesis ( 165 ) , degradation/utilization/assimilation ( 72 ) , detoxification ( 3 ) , generation of precursor metabolites and energy ( 21 ) , metabolic clusters ( 5 ) , and superpathways ( 30 ) . We assigned 19 transport reactions , in which 15 are transport-energized by ATP hydrolysis . The number of genes , enzymes , and metabolites are comparable to a well-curated and verified reconstruction of the Chlamydomonas reinhardtii metabolic network , iRC1080 ( Chang et al . , 2011 ) and its recent expansion , iBD1106 ( Chaiboonchoe et al . , 2014 ) , which includes 1106 genes and 1959 individual metabolites . Pathway classes and corresponding reactions that were shared between the Chlamydomona s reinhardtii iRC1086 Cobra model and the Chloroidium sp . UTEX 3007 model ( CF-SNAP ) were: activation/inactivation/interconversion ( 3/0/2 ) , superpathway ( 7 ) , generation of precursor metabolites and energy ( 18 ) , degradation/utilization/assimilation ( 28 ) , biosynthesis ( 963 ) , palmitate biosynthesis ( 53 ) and detoxification ( 5 ) . The pathways/reactions unique to the Chloroidium sp . UTEX 3007 model consists of: biosynthesis ( 796 ) , generation of precursor metabolites and energy ( 18 ) , degradation/utilization/assimilation ( 28 ) , biosynthesis ( 462 ) , palmitate biosynthesis ( 31 ) , detoxification ( 6 ) . The Pfam database categorizes the CDS1-3 protein sequences ( Figure 7 ) in the Ferritin_2 family ( Finn et al . , 2014 ) . The structural core of ferritins and ferritin-like proteins is 4-helix bundle ( with the pattern: helix A – hairpin- helix B – long crossover loop – helix C- hairpin – helix D ) ( Andrews , 2010 ) . The center of the helical bundle often contains a di-metal binding site and is frequently an active site for various forms of redox chemistry . For example , in canonical ferritins , this is a di-iron site that oxidizes Fe2+ to Fe3+ for mineral iron storage . Examination of the sequence indicated the proteins contain metal-binding motifs ( helix A[E ( 6 ) Y] , helix B[ExxH] , helix C[E ( 6 ) Y] , helix D [ExxH] ) similar to those characteristic of two ferritin-like protein classes: erythrins and manganese catalases ( Andrews , 2010 ) . Our homology-based tertiary structure modeling was performed using ModWeb ( Eswar et al . , 2003 ) . The helix A/B and helix C/D loops ( residues 57–84 and 109–152 , respectively ) were optimized with Rosetta3 kinematic loop closure in centroid mode ( command line flag: -loops:remodel perturb_kic , loops were built from randomly selected cut-points ) ( Leaver-Fay et al . , 2011; Mandell et al . , 2009 ) . The lowest energy structure was repacked in full atom fixed backbone design mode ( with -ex1 -ex2aro flags ) , Thus , the CDS3 model spans residues 33–202 of the 290 total residues . Crystal structures of three Mn catalases , from Thermus thermophilus , Lactobacillus plantarum , and Anabaena PCC 7120 have been solved ( Antonyuk et al . , 2000; Barynin et al . , 2001; Bihani et al . , 2013 ) . They are globular homohexamers and contain canonical ferritin four-helix bundles with notable features including a short ( ~20 ) residue N-terminal segment before Helix A; a quite long ( ~40–60 residue ) cross-over loop between helices B and C ( as a comparison the crossover in a pseudo-nitzschia diatom ferritin is 20 residues [Marchetti et al . , 2009] , and a long C-terminal tail ( of 50–100 residues ) that wraps around a neighboring monomer in the homohexamer complex . Notably , the CDS1-3 sequences contain both a predicted N-terminal segment ( of 10 to 33 residues ) before the assumed start of helix A and a C-terminal tail of ~70–100 residues . The predicted crossover loop in the algal sequences has a mid length ( ~40 residues ) between those found in ferritins and the structurally characterized Mn catalases . In the known Mn catalase structures , the long crossover loops constitute much of the contact surface between chains in the homohexamer , and we speculate that the additional presumptive loop between helices A and B in the algal proteins may participate in similar inter-chain binding interactions . We observed cells with an Olympus BX53 microscope using a UPlanFL N 100x/1 . 30 Oil Ph 5 UIS two objective ( Olympus Corporation , Tokyo , Japan ) using brightfield display or fluorescence . An X-cite series 120 Q fluorescent illumination source ( Lumen Dynamics , Ontario , Canada ) was used to visualize Bodipy-505/515 stained cells . Bodipy 505/515 absorbs at 488 nm and emits at 567 nm . Lipid staining was performed as follows: 1 µl of 10 mg/mL Nile-red or Bodipy-505/515 and Triton X-100 ( 0 . 1% ) was added to 100 µl of cells ( concentration of appx . 1 × 106 ) and allowed to incubate at room temperature for one hour . Cell size analysis was performed using a Cellometer Auto M10 from Nexcelcom Bioscience ( Lawrence , MA , USA ) . 2 × 20 µl from each liquid algal sample was analyzed for the live/dead count , total cell count , mean diameter ( μm ) , viability ( % ) , and live cell concentration . Chloroidium sp . UTEX 3007 was maintained as single colony isogenic monocultures repeatedly throughout our experiments . Algal cultures were grown under constant illumination ( 400 umol photons m−2 s−1 ) at 25°C . Media used for algal culture were modified F/2 media ( Tamburic et al . , 2014 ) enriched with nitrate ( 10 mM KNO3 ) and magnesium ( 2 mM MgSO4 ) . Saltwater medium was made using Ao Reef Salt ( Ao Aqua Medic , Bissendorf , Germany ) to 40 g/L unless indicated otherwise . Growth was measured using the bioreactor growth chamber Multi-Cultivator MC 1000 and MC 1000-OD by Photon Systems Instruments ( Drasov , Czech Republic ) . For artificial , open pond , simulation , we grew algae in outdoor pond simulator bioreactors ( PBR-101 , Phenometerics Inc . , East Lancing , MI , USA ) . The open pond simulators were approximately cylinder photobioreactors ( PBRs ) and were set up with a working volume of 400 ml and a light depth in the culture of 14 . 0 cm . The PBRs were injected with air enriched with 2 . 0–3 . 0% CO2 at a flow rate of 0 . 20 L/min . We maintained temperature and pH at 25 ± 1°C and 7 . 0 ± 0 . 2°C , respectively . Stirring was set at a constant rate of 200 RPM using a 28 . 6 mm stir bar . The cultures were grown under a light-dark cycle of 16:8 using a sinusoidal approximation of daily light with a peak intensity of 2000 μmol photons m−2 s−1 . Cells were prepared as described above and per Phenometerics Inc . ’s instructions included in the bioreactor setup guide . Briefly , cells were inoculated from solid agar media into liquid media for 24 hr in illuminated flasks on a shaker in the growth chambers described above . After inoculation of 10 mL of pre-prepared ( concentration of about 5 × 107 cells/mL ) into 70 mL of media in each bioreactor tube ( 80 mL total liquid volume in eight tubes/bioreactor setup ) , optical density ( OD ) readings were recorded at 680 nm every hour for 8–12 days . Unless otherwise noted , bioreactors were maintained at 25°C and were illuminated at 400 μmol photons m−2 s−1 . Cell counting was performed using a Cellometer Auto M10 from Nexcelcom Bioscience ( Lawrence , MA , USA ) . 2 × 20 ul from each liquid algal sample was analyzed for the live/dead count , total cell count , mean diameter ( μm ) , viability ( % ) , and live cell concentration . Cell concentrations for growth curves were calculated by constructing a standard curve correlating to absorbance at 680 nm , which is roughly proportional to chlorophyll a concentrations , with measured cell concentrations , R > 0 . 99 . Cells were harvested at one and three weeks of growth in PBRs to obtain cells in mid-log and stationary phases , respectively . Staining was performed according to the microscopy procedure ( above ) . Forward scatter ( FSC ) signal , assumed to be proportional to cell size or cell volume , and side scatter ( SSC ) signal , related to the complexity of the cell ( Ramsey et al . , 2016 ) , were collected and plotted against fluorescence from BODIPY 505/515 in the FITC-A channel . Phenotyping was done using standard Biolog assay plates and using the Omnilog instrument ( Biolog Inc . , Hayward , USA ) as previously described ( Chaiboonchoe et al . , 2014 ) . In total , 380 substrate utilization assays for carbon sources ( PM01 and PM02 ) , 95 substrate utilization assays for nitrogen sources ( PM03 ) , 59 nutrient utilization assays for phosphorus sources , and 35 nutrient utilization assays for sulfur sources ( PM04 ) , along with peptide nitrogen sources ( PM06-08 ) were performed ( Dataset 2-biolog ) . A defined tris-acetate-phosphate ( TAP ) medium ( Gorman and Levine , 1965 ) containing 0 . 1% tetrazolium violet dye ‘D’ ( Biolog , Hayward , CA , USA ) was used for the PM tests . The carbon , nitrogen , phosphorus , or sulfur component of the media was omitted from the defined medium when applied to the respective PM microplates that tested for each of those sources . Cells were grown in new tris-minimal media to mid-log phase , then spun down at 2000 g for 10 min , and then re-suspended in fresh media to a final concentration of 1 × 106 cells/mL before inoculation into Biolog’s 96-well plates . A 100 μL aliquot of cell-containing media was inoculated into each well before the plates were inserted into the Omnilog system . A final concentration of 400 μL/mL Timentin ( GlaxoSmithKline , Brentford , UK ) was used to inhibit bacterial growth in all plates . Bacterial contamination was monitored by streaking cells on yeast extract/peptone plates and performing gram stains before and after Biolog assays . All microplates were incubated at 30°C for up to 8 days , and the dye color change ( in the form of absorbance ) was read with the Omnilog system every 15 min . As the Omnilog instrument does not provide a source of continuous light during incubation , the algae are assumed to be carrying out heterotrophic respiration . The Biolog Phenotype Microarray ( PM ) data analysis was conducted using an Omnilog Phenotype Microarray ( OPM ) software package that runs within the R software environment ( Vaas et al . , 2013 , 2012 ) . The raw kinetic data were exported as csv files to the OPM package , and then the biological information was added as metadata ( e . g . strain designation , growth media , temperature , etc . ) . For the extraction , 50 mg algal material ( dry weight ) was used by adding 1 ml of 1 N HCl/methanol solution ( Sigma-Aldrich , Darmstadt , Germany ) . An internal standard ( 100 µl of FA15:0 , pentadecanoic acid ) was added to each sample before incubation at 80°C in a water bath for 30 min . After cooling to room temperature , 1 ml of 0 . 9% NaCl and 1 ml of 100% hexane were added to each vial . Vials were shaken for 5 s and centrifuged for 4 min at 1000 rpm . The upper FAME-containing hexane phase was transferred to a new glass vial , where it was concentrated under a stream of N2 . Finally , FAMEs were dissolved in hexane and filled into GC glass vials . The details of the GC-FID method are as follows: injector temperature of 250°C; helium carrier gas; head pressure 25 cm/s ( 11 . 8 psi ) ; GC column , J and W DB23 ( Agilent , Santa Clara , CA , USA ) , 30 m 9 0 . 25 mm 9 0 . 25 lm; detector temperature 250°C; detector gas H240 ml/min , air 450 ml/min , He make-up gas 30 ml/min . FAME peaks were identified by comparing their retention time and equivalent chain length on standard FAME . We used standards of fatty acid methyl esters ( FAME ) from Supelco 37 Component FAME Mix ( CRM47885 , SIGMA-ALDRICH , Darmstadt , Germany ) and Supelco PUFA No . 3 , from Menhaden Oil ( 47085 U SIGMA-ALDRICH , Darmstadt , Germany ) . Source data can be found in Dataset 2 . Extraction and analysis by gas chromatography coupled with mass spectrometry was performed using the same equipment set up and exact same protocol as described in Lisec et al . ( 2006 ) . Briefly , frozen ground material was homogenized in 300 μL of methanol at 70°C for 15 min and 200 μL of chloroform followed by 300 μL of water were added . The polar fraction was dried under vacuum , and the residue was derivatized for 120 min at 37°C ( in 40 µl of 20 mg ml-1 methoxyamine hydrochloride in pyridine ) followed by a 30 min treatment at 37°C with 70 µl of MSTFA . An autosampler Gerstel Multi-Purpose system ( Gerstel GmbH and Co . KG , Mülheim an der Ruhr , Germany ) was used to inject the samples to a chromatograph coupled to a time-of-flight mass spectrometer ( GC-MS ) system ( Leco Pegasus HT TOF-MS ( LECO Corporation , St . Joseph , MI , USA ) ) . Helium was used as carrier gas at a constant flow rate of 2 ml/s and gas chromatography was performed on a 30 m DB-35 column . The injection temperature was 230°C and the transfer line and ion source were set to 250°C . The initial temperature of the oven ( 85°C ) increased at a rate of 15 °C/min up to a final temperature of 360°C . After a solvent delay of 180 s mass spectra were recorded at 20 scans s-1 with m/z 70–600 scanning range . Chromatograms and mass spectra were evaluated by using Chroma TOF 4 . 5 ( Leco ) and TagFinder 4 . 2 software . Source data can be found in Dataset 2 . For the extraction , cultured algae was scraped from agar plates and placed into 5 mL methanol and vortexed . Algal-methanol solutions were microwaved on high power ( 1150 watts , 2450 Mhz ) in a Samsung ME732K microwave with a triple distribution system ( Samsung , Seoul , South Korea ) . Solutions were microwaved until boiling five times and then vortexed . Extracts were filtered with a 2 μm filter ( Millipore ( Merck Millipore , Billeric , MA , USA ) and maintained in the dark at 4°C . The HPLC/MS method was developed in-house and is based on a comprehensive shotgun lipidomic technique from the Castro-Perez laboratory ( Castro-Perez et al . , 2010 ) . Four µl of the extract from each sample was injected into a reverse-phase C18 column heated to 50°C . A quaternary pump maintained a 300 µl/min flow rate of the solvent composite over the sample . The starting solvent was 36% water with 30 mM ammonium formate , 36% acetonitrile , and 28% isopropanol which approached 90% isopropanol and 10% acetonitrile in a semi-linear gradient over 18 min . The end stream was diverted to a quadrupole time-of-flight mass spectrometer in positive mode with accurate mass profiling enabled that was tuned within 1 hr before the experiments ( reference compounds were 121 . 050873 and 922 . 009798 m/z ) ( Agilent LCMS QToF 6538 ( Agilent , Santa Clara , CA , USA ) ) . The data were processed using Agilent’s software and XCMS with METLIN ( Sana et al . , 2008; Smith et al . , 2005 ) . Source data can be found in Dataset 2 .	Single-celled green algae , also known as green microalgae , play an important role for the world’s ecosystems , in part , because they can harness energy from sunlight to produce carbon-rich compounds . Microalgae are also important for biotechnology and people have harnessed them to make food , fuel and medicines . Green microalgae live in many types of habitats from streams to oceans , and they can also be found on the land , including in deserts . Like plants that live in the desert , these microalgae have likely evolved specific traits that allow them to live in these hot and dry regions . Yet , fewer scientists have studied microalgae compared to land plants , and until now it was not well understood how microalgae could survive in the desert . Nelson et al . analyzed green microalgae from different locations around the United Arab Emirates and found that one microalga , known as Chloroidium , is one of the most dominant algae in this area . This included samples from beaches , mangroves , desert oases , buildings and public fresh water sources . Chloroidium has a unique set of genes and proteins and grew particularly well in freshwater and saltwater . Rather than just harnessing sunlight , the microalgae were able to consume over 40 different varieties of carbon sources to produce energy . The microalgae also accumulated oily molecules with a similar composition to palm oil , which may help this species to survive in desert regions . A next step will be to develop biotechnological assets based on the information obtained . In the future , microalgae could be used to make an oil that represents an alternative to palm oil; this would reduce the demand for palm tree plantations , which pose a major threat to the natural environment . Moreover , understanding how microalgae can colonize a desert region will help us to understand the effects of climate change in the region .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "plant", "biology", "genetics", "and", "genomics" ]	2017	The genome and phenome of the green alga Chloroidium sp. UTEX 3007 reveal adaptive traits for desert acclimatization
Evolutionary origin of muscle is a central question when discussing mesoderm evolution . Developmental mechanisms underlying somatic muscle development have mostly been studied in vertebrates and fly where multiple signals and hierarchic genetic regulatory cascades selectively specify myoblasts from a pool of naive mesodermal progenitors . However , due to the increased organismic complexity and distant phylogenetic position of the two systems , a general mechanistic understanding of myogenesis is still lacking . In this study , we propose a gene regulatory network ( GRN ) model that promotes myogenesis in the sea urchin embryo , an early branching deuterostome . A fibroblast growth factor signaling and four Forkhead transcription factors consist the central part of our model and appear to orchestrate the myogenic process . The topological properties of the network reveal dense gene interwiring and a multilevel transcriptional regulation of conserved and novel myogenic genes . Finally , the comparison of the myogenic network architecture among different animal groups highlights the evolutionary plasticity of developmental GRNs . Muscle cells are present in most animals with the characteristic of containing protein filaments that slide and produce contractions . In bilaterians , muscles develop usually from mesodermal cells in a process called myogenesis , during which the naive mesodermal progenitors are selectively specified as myoblasts and later differentiate into muscle cells . In vertebrate embryos , the mesoderm is subdivided into several regions from which different muscle types originated . For example , the axial skeletal muscles originate in the segmented paraxial mesoderm called somites , whereas the cardiac muscles develop from the lateral plate mesoderm . The subdivisions of different mesoderm regions require differential expression of the Forkhead ( Fox ) family of transcription factors , such as FoxC1 and FoxC2 for the paraxial mesoderm ( Wilm et al . , 2004 ) and FoxF1 for the lateral plate mesoderm ( Mahlapuu et al . , 2001 ) . Analyses of molecular mechanisms underlying myogenesis in various parts of the somites further lead to a series of complex gene regulatory networks ( GRNs ) ( Bryson-Richardson and Currie , 2008; Yokoyama and Asahara , 2011; Bentzinger et al . , 2012 ) . These networks are composed of multiple signals that drive the expression of the genes encoding basic helix-loop-helix ( bHLH ) domain-containing myogenic regulatory factors , which include myogenic differentiation 1 ( MyoD ) , myogenic factor 5 ( Myf5 ) , Myf6 , and myogenin ( Myog ) . Other factors , including the myocyte enhancer binding factor 2 ( MEF2 ) family of MADS-box proteins , transcription factors Pitx2 , Pitx3 , sine oculis-related homeobox ( six ) family members and their cofactors eyes-absent homologs ( Eya ) , are also involved in specification and migration of the myogenic precursor cells ( Molkentin and Olson , 1996; Yokoyama and Asahara , 2011 ) . These myogenic transcriptional regulators then in turn initiate the transcription of muscle differentiation markers , such as myosin heavy chain proteins ( MHC ) , that determine the morphological and functional identity of the differentiated muscle cells ( Bryson-Richardson and Currie , 2008; Bentzinger et al . , 2012 ) . In addition to the transcription factors mentioned above , myogenesis in different parts of the vertebrate somites depends on several signals including Notch , fibroblast growth factor ( FGF ) , Wnts , bone morphogenetic factor 4 ( BMP4 ) , and Sonic hedgehog ( Shh ) that are secreted from adjacent tissues , such as neural tube , notochord , dorsal , and lateral ectoderms ( Marcelle et al . , 1997; Vasyutina et al . , 2007; Delfini et al . , 2009 ) . Among these signals , FGF-signaling pathway is of particular interest because it has a crucial and conserved role in mesoderm patterning and muscle development in various animal models . In vertebrates , FGF induces the expression of myogenic genes in the primary myotome , controls the timing of the epithelial–mesenchymal transition of the dermomyotome , which triggers the emergence of muscle progenitors ( Delfini et al . , 2009 ) and positively regulates muscle differentiation ( Marics et al . , 2002; Groves et al . , 2005 ) . In flies , an FGFR ortholog ( Heartless ) -mediated pathway is essential for cell migration and fate induction of the visceral mesoderm , heart , and the somatic muscle lineages ( Beiman et al . , 1996 ) . In ascidians , FGF signaling is required for the specification of the heart progenitor cells and secondary muscle development ( Beh et al . , 2007; Tokuoka et al . , 2007 ) . In the nematode Caenorhabditis elegans , the crosstalk between FGF and Wnt signals is involved in larval sex myoblast specification and migration ( Burdine et al . , 1998; Lo et al . , 2008 ) . In the amphioxus Brachiostoma lanceolatum , FGF signaling is necessary for the formation of the anterior somites that will generate the musculature ( Bertrand et al . , 2011 ) and in the hemichordate Saccoglossus kowalevskii , FGF signaling is necessary for all types of mesoderm development ( Green et al . , 2013 ) . The overall genetic regulatory cascade that drives myogenesis appears to be highly conserved between ecdysozoans ( Drosophila melanogaster and C . elegans ) and vertebrates ( Ciglar and Furlong , 2009 ) . However , due to the high complexity that characterizes the vertebrate organogenesis and the distant phylogenetic position between ecdysozoans and vertebrates , the logics behind the genomic regulatory interactions that drive muscle development remain unclear . Therefore , elucidating the properties of the key genetic interconnections that orchestrate myogenesis in a larger set of model organisms is necessary to reveal the core myogenic circuits . Moreover , given the fact that the common origin of musculature is a highly debated topic ( Seipel and Schmid , 2005; Burton , 2008; Steinmetz et al . , 2012 ) , a larger interspecies comparison of the transcriptional myogenic networks would provide new insights into the evolution of muscle and the kernel driven hypothesis . Echinoderms occupy a key phylogenetic position since they are early branching deuterostomes , therefore , are more closely related to vertebrates compared to the other two most studied invertebrate model systems ( D . melanogaster and C . elegans ) . Sea urchin is a powerful model system for studying regulatory events that take place during development due to their advantageous properties for GRN analysis ( Oliveri and Davidson , 2004 ) . GRNs provide causal explanations of the molecular interactions occurring during dynamic developmental processes such as cell specification and differentiation ( Davidson et al . , 2002; Ben-Tabou de-Leon and Davidson , 2006 ) . In the last decade , the extensive amount of data collected by perturbation analyses has led to the assembly of the largest so far known endomesodermal GRN ( Davidson et al . , 2002; Peter and Davidson , 2011; Materna et al . , 2013 ) representing most of the transcriptional interactions that take place during endomesoderm formation in the sea urchin embryo . These properties make echinoderms an excellent model to study molecular developmental mechanisms and address aspects in evolution of organogenesis . Sea urchin larvae possess a muscular apparatus that surrounds their esophagus and produces contractile force ( Burke , 1981; Burke and Alvarez , 1988 ) . These muscle cells originate from mesoderm and adapt the myogenic fate by segregating from the other three non-skeletogenic mesodermal ( NSM ) lineages at the early gastrula stage ( Ruffins and Ettensohn , 1996 ) . The first appearance of myoblasts is seen at the late gastrula stage as a few cells in the oral vegetal domain of each coelomic sac at the tip of the archenteron and they express a number of muscle-specific transcription regulators ( Andrikou et al . , 2013 ) . Few hours later , at the prism stage , these cells extend pseudopods towards the midline of the esophagus , increase in number and diameter , fuse to each other , and finally form the circumesophagael contractile bands ( Burke and Alvarez , 1988 ) . Despite the existence of large amount of data concerning the mechanisms patterning the early endomesoderm segregation , little is known about the regulatory landscape of the later NSM specification . Parts of the regulatory events that underlie the specification of the three NSM mesodermal lineages ( blastocoelar , pigment , and coelomic pouch cell lineages ) were recently revealed ( Luo and Su , 2012; Ransick and Davidson , 2012; Materna et al . , 2013; Solek et al . , 2013 ) , and only scattered information regarding the molecular interplay that establish the muscle lineage is available in the literature . In particular , transcription factors such as Twist ( Wu et al . , 2008 ) and a sea urchin lineage-specific Fox family factor , FoxY ( Materna et al . , 2013 ) , and signaling pathways such as Delta/Notch ( D/N ) ( Sherwood and McClay , 1999; Sweet et al . , 2002 ) and Hedgehog ( Hh ) ( Walton et al . , 2009; Warner et al . , 2014 ) seem to be involved in muscle development , but the molecular mechanisms remain poorly understood . In our previous work , we identified several homologues of myogenic regulators in the sea urchin embryo and revealed the regulatory state of the myoblasts and their precursor cells ( Andrikou et al . , 2013 ) . Further functional analyses are needed to confirm the myogenic function of these transcription factors and to unravel the regulatory architecture of the muscle GRN . This study provides an explanatory mechanism on how the sea urchin muscle lineage is specified using a perturbation approach accompanied by a combination of temporal and spatial gene expression analyses . We show that FGF signaling is necessary for specifying naive NSM cells to myoblast precursors at the very early gastrula stage . The four Fox family factors FoxY , FoxC , FoxF , and FoxL1 ( Tu et al . , 2006 ) constitute the central part of our myogenic GRN model and are activated sequentially . Other conserved key factors that occupy different hierarchical levels in the myogenic GRN are members of the T-box ( Tbx6 ) ( Howard-Ashby et al . , 2006 ) , bHLH ( MyoD2 ) ( Andrikou et al . , 2013 ) , SRY ( Sox ) ( Howard-Ashby et al . , 2006 ) , Scratch ( ScratchX ) ( Materna et al . , 2013 ) , and Six ( Six1/2 ) ( Poustka et al . , 2007 ) family genes . These findings imply an overall high level of functional conservation of both key myogenic transcriptional regulators and signaling components but not of the myogenic GRN architecture per se . Also , they explain in a rational way the logics and properties of the regulatory interactions that drive myogenesis in the sea urchin embryo . As shown in our previous work ( Andrikou et al . , 2013 ) , the myogenic lineage seems to segregate as early as the very early gastrula stage ( 30 hr post fertilization ) . The relative position of these cells is not precise at the earlier stages but they seem to be located at the oral/lateral periphery of the vegetal plate , at the border between the blastocoelar and pigment cell precursors ( Figure 1—figure supplement 1 [Ruffins and Ettensohn , 1996] ) . To reveal the cause of myoblast emergence at the early gastrula stage , we searched for putative candidate signaling components that are expressed at that time in the myoblast precursors . As elsewhere demonstrated , a mesodermal Delta/Notch signal activates FoxY , which is required for the specification of the two mesoderm derivatives , coelomic pouches and muscles ( Materna and Davidson , 2012; Materna et al . , 2013 ) . The segregation of these two derivatives should thus specifically rely on another signal that triggers myoblast specification at that particular developmental time . There are one FGF ligand , FGFA ( Figure 1—figure supplement 2 ) , and two FGF Receptors ( FGFRs ) , FGFR1 , and FGFR2 ( Figure 1—figure supplement 3 ) , annotated in the sea urchin genome ( McCoon et al . , 1996; Lapraz et al . , 2006 ) . A previous study showed that inhibition of FGFR2 affects morphogenesis of the embryonic skeleton ( Rottinger et al . , 2008 ) . We therefore focused on characterizing the expression patterns of genes encoding FGFR1 and FGFA ligand in the sea urchin Strongylocentrotus purpuratus . The expression profile of both genes matches with the one already described in another sea urchin species , Paracentrotus lividus ( Rottinger et al . , 2008 ) . To better resolve the expression patterns of these two genes relatively to the myoblast precursors , double fluorescent in situ hybridizations ( FISHs ) with FoxC , the earliest myoblast precursor marker ( Andrikou et al . , 2013 ) , were performed . At the very early gastrula stage , FGFR1 was expressed in all FoxC-positive cells ( Figure 1A ) whilst at the late gastrula stage ( 48 hr ) , only a number of FGFR1 transcripts overlap with FoxC-positive cells at the tip of the archenteron ( Figure 1—figure supplement 4 ) . Moreover , FGFA transcripts were observed in the ventrolateral ectodermal regions in the vicinity of FoxC-positive cells ( Figure 1B ) . The spatial and temporal expression of FGFR1 and FGFA suggests a putative role of FGF signaling in sea urchin myogenesis . 10 . 7554/eLife . 07343 . 003Figure 1 . Expression analysis of genes encoding sea urchin FGF-signaling components and FoxC by double FISH . FGFR1 and FGF were stained in green and FoxC in red at the very early gastrula stage ( 30–32 hr ) . Nuclei were labeled blue with DAPI . Yellow circles indicated by yellow arrowheads show cells co-expressing the analyzed genes . Panels A and B are stacks of merged confocal Z sections of all three channels , while separate channels over DAPI are presented in the other panels . Insets in panels A–A″ show representative single confocal sections to confirm that the two genes are indeed expressed in the same cell . Embryos in A–A″ are seen in a lateral view along the animal-top/vegetal-down axis . Embryos in B–B″ are displayed in a vegetal view . fv , frontal view; vv , vegetal view; o , oral , ab , aboral . The position of the putative unspecified myoblast precursors is indicated in Figure 1—figure supplement 1 . Phylogenetic analyses of sea urchin fibroblast growth factor ( FGF ) and FGFR protein sequences are reported in Figure 1—figure supplements 2 , 3 , respectively . A co-expression analysis of FGFR1 and FoxC at late gastrula stage ( 48 hr ) is reported in Figure 1—figure supplement 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 07343 . 00310 . 7554/eLife . 07343 . 004Figure 1—figure supplement 1 . Three-color FISH of Gcm , Ese , and FoxA . Co-staining of Gcm as a marker for aboral pigment cell precursors , Ese for oral blastocoelar cell precursors , and FoxA for endoderm at 26 hr . Gcm transcript was stained in light green , Ese in red , and FoxA in cyan . Nuclei were labeled blue with DAPI . Yellow circles indicate the SMs , and white circles indicated by white arrows show single cells that do not express any of the analyzed genes . In 85% of the embryos analyzed , approximately two triple-negative cells were observed that might represent myoblast precursors . All embryos are in a vegetal view . Panels A , B , and C are merged confocal stacks , while panels A′–A‴ depict separate channels over DAPI . Panel D is a schematic representation of the vegetal surface of a sea urchin mesenchyme blastula orientated along the oral right/aboral left ( O/Ab ) axis , seen from the vegetal pole . The different NSM domains identified by distinct regulatory signatures at the vegetal plate are shown in different colors as indicated in the legend . Primary mesenchyme cells are not visible as they have already ingressed into the blastocoel at this stage . DOI: http://dx . doi . org/10 . 7554/eLife . 07343 . 00410 . 7554/eLife . 07343 . 005Figure 1—figure supplement 2 . Phylogenetic analysis of the sea urchin FGFA protein sequence . ( A ) The protein domain structure of the SpFGFA contains a signal peptide and a FGF core domain . ( B ) Phylogenetic analysis of FGF proteins . The tree was constructed by the neighbor-joining method based on the multiple alignments of the FGF core domains from various organisms . Bootstrap values over 50% are shown at the branch points . Sea urchin FGFs ( Sp-FgfA and Pl-FgfA ) do not group with the seven vertebrate FGF families from A to G . The scale indicates the % amino acid difference with Poisson correction . GenBank accession numbers for FGFs are: acorn worm SkFgf20 like , NM_001171225; amphioxus AmphiFgf1/2 , EU606032; AmphiFgf8/17/18 , EU606035; AmphiFgf9/16/20 , EU606036; ascidian CiFgf8/17/18 , NM_001032476; CiFgf11/12/13/14 , NM_001032561; Caenorhabditis elegans CeEGL17 , NM_075706; Drosophila melanogaster DmPYR , AY55396; DmTHS , NM_136857; human HsFgf19 , NM_005117; mouse MmFgf1 , NM_010197; MmFgf2 , NM_008006; MmFgf3 , Y00848; MmFgf4 , NM_010202; MmFgf5 , NM_010203; MmFgf6 , NM_010204; MmFgf7 , NM_008008; MmFgf8 , NM_010205; MmFgf9 , U33535; MmFgf10 , NM_008002; MmFgf11 , NM_010198; MmFgf12 , NM_183064; MmFgf13 , NM_010200; MmFgf14 , NM_010201; MmFgf15 , NM_008003; MmFgf16 , AB049219; MmFgf17 , NM_008004; MmFgf18 , NM_008005; MmFgf20 , AB049218; MmFgf21 , NM_020013; MmFgf22 , NM_023304; MmFgf23 , NM_022657; sea urchin Sp-FgfA , HQ107979; Pl-FgfA , EF157978 . DOI: http://dx . doi . org/10 . 7554/eLife . 07343 . 00510 . 7554/eLife . 07343 . 006Figure 1—figure supplement 3 . Phylogenetic analysis of the sea urchin FGFR protein sequences . ( A ) The protein domain structures of SpFGFR1 and SpFGFR2 . The Fibronectin III domain ( FN3 ) , three Ig domains ( Ig1-3 ) , acid box ( AB ) , transmembrane region ( TM ) , and Tyrosine kinase domain are indicated . ( B ) For the phylogenetic analysis , the tyrosine kinase domains of FGFRs were aligned , and the tree was built in the same manner as the FGF tree . SpFGFR1 and SpFGFR2 are not orthologous to the human FGFR1 and FGFR2 , respectively . GenBank accession numbers for FGFRs are: human HsFgfR1 , AB208919; HsFgfR2 , NM_000141; HsFgfR3 , NM_000142; HsFgfR4 , AY892920; ascidian CiFgfR , NM_001044355; C . elegans CeEGL-15 , NM_077441; D . melanogaster DmBreathless , NM_168577; DmHeartless , NM_169784; sea urchin SpFgfR1 , NM_214537; SpFgfR2 , JF499690 . DOI: http://dx . doi . org/10 . 7554/eLife . 07343 . 00610 . 7554/eLife . 07343 . 007Figure 1—figure supplement 4 . Coexpression analysis of FGFR1 and FoxC by double FISH . Relative spatial expression domains of FGFR1 and FoxC at late gastrula stage ( 48 hr ) . Image is a stack of merged confocal Z sections in all channels . Inset shows representative single confocal section of the tip of the archenteron . Color code of channel association to each gene is shown in each panel . Nuclei are stained blue with DAPI . Embryo is seen in a frontal view . DOI: http://dx . doi . org/10 . 7554/eLife . 07343 . 007 To test whether FGF signaling through FGFR1 is involved in sea urchin myogenesis , we perturbed FGF-signaling pathway by using ( a ) SU5402 , a FGFR inhibitor ( Mohammadi et al . , 1997 ) ; ( b ) U0126 , a MEK inhibitor ( Favata et al . , 1998 ) ; ( c ) an antisense morpholino oligonucleotide ( MO ) targeted to FGFR1; ( d ) a dominant negative form ( Dn ) of FGFR1 . A summary of the phenotypes observed after inhibitor treatments is in Figure 2—figure supplement 1A . 70% of the 300 pluteus larvae treated with SU5402 from 26 hr showed an abnormal elongated archenteron missing the pyloric and possibly the anal sphincter constrictions , although the cardiac sphincter was still formed ( Figure 2A–D ) suggesting that the three distinct sphincters are differentially regulated ( Annunziata and Arnone , 2014 ) . Moreover , all larvae had shorter spicules and their triradiate skeleton was never fully shaped possibly due to the inhibition of FGFR2 , which is specifically expressed in the skeletogenic primary mesenchyme cells ( PMCs ) that produce the larval skeleton ( Rottinger et al . , 2008 ) . Finally , the coelomic pouches were not formed and the circumesophagael muscle fibers were completely absent when tested for MHC gene expression by FISH at the prism stage ( Figure 2D ) . To reveal the downstream cascade of the FGF signaling , we used the MEK inhibitor U0126 . In one of our previous studies , we showed that MAPK/ERK activation is required for muscle formation ( Fernandez-Serra et al . , 2004 ) . We repeated the experiment by treating the embryos with U0126 at 26 hr , as we did with the SU5402 inhibitor . The treated larvae were tested again for the presence of muscle fibers by examining the MHC protein level using immunostaining . Again , the level of MHC protein was greatly reduced in the treated larvae ( Figure 2—figure supplement 1B , C ) . These findings show an involvement of FGF signaling through MAPK/ERK pathway in muscle development . 10 . 7554/eLife . 07343 . 008Figure 2 . Perturbation of the FGF pathway . To analyze the phenotype of FGF perturbation , bright-field images were taken with differential interference contrast ( DIC ) . Effects on muscle formation were also tested by detection of MHC expression by fluorescent in situ hybridization ( FISH ) or of myosin heavy chain ( MHC ) protein localization by immunostaining on pluteus larvae ( 72 hr ) . The ciliary band and gut internal cilia were stained with an anti-acetylated tubulin antibody ( AcT ) . Panels ( A–D ) show the effect of SU5402 in the formation of the coelomic pouches ( B ) and MHC expression ( D ) . Panels ( E–H ) show the effect of anti-FGFR1 translation morpholino oligonucleotide ( MO ) in the formation of the coelomic pouches , MHC protein localization , and gut morphology . Two representative phenotype embryos , both with impaired muscles while differing for gut sphincter formation , are reported in F ( normal gut , 70% of cases ) and H ( reduced sphincters , 30% of cases ) . Panels ( E , G , I , and J ) show the effect in MHC protein localization caused by injection of FGFR1 dominant negative RNA ( DnRNA ) ( J ) . Panels ( A , C ) show control embryos treated with DMSO . Panel ( G ) shows a larva injected with a fluo-control MO and panels ( E , I ) show control uninjected larvae ( for MO injection controls see also ‘Materials and methods’ and Figure 2—figure supplement 2 ) . The inset in panel H is a magnified view of the cilia at the apical organ . Pictures in C , D , and G–J are stacks of merged confocal Z sections . MHC was stained in red and acetylated tubulin in green . Nuclei were labeled blue with DAPI . Spicules are seen in DIC analysis as reflecting polarized light objects . All embryos are seen in frontal view except the ones in panels E , F , and I that are seen in lateral view with the oral side on the right ( fv , frontal view; lv , lateral view ) . White arrows indicate the position of cardiac sphincters , whilst yellow and red arrows show , where present , the pyloric and anal sphincters , respectively . Black lines indicate pigment cells ( pc ) . White lines indicate muscle fibers ( mf ) . The asterisks indicate the absence of coelomic pouches ( cp ) . A summary of SU5402 and U0126 treatments as well as MHC protein expression analysis after MEK pathway perturbation is reported in Figure 2—figure supplement 1 . Control MO experiments are reported in Figure 2—figure supplement 2 . Co-expression analysis of genes encoding putative MAPK effectors and FoxC as well as P-Elk protein detection is reported in Figure 2—figure supplement 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 07343 . 00810 . 7554/eLife . 07343 . 009Figure 2—figure supplement 1 . Summary of SU5402 and U0126 treatments and MHC protein detection by immunostaining after MEK pathway perturbation . ( A ) Scheme summarizing the drug treatments performed and the morphological phenotypes observed . ( B , C ) MHC protein localization was tested by immunostaining in ( B ) control pluteus larva and ( C ) pluteus larva treated with the MEK inhibitor U0126 as indicated in ( A ) by the green line . The ciliary band and gut internal cilia were stained with an anti-acetylated tubulin antibody . MHC was stained in red and acetylated tubulin in green . Nuclei were labeled blue with DAPI . All embryos are seen in lateral view with the oral side on the right . DOI: http://dx . doi . org/10 . 7554/eLife . 07343 . 00910 . 7554/eLife . 07343 . 010Figure 2—figure supplement 2 . Control experiments for MOs . Panels A- D show control uninjected embryos at very early gastrula ( A ) , late gastrula ( B ) , and pluteus larva ( C , D ) . Panels ( E–G ) show embryos injected with fluo-control MO at very early gastrula ( E ) , late gastrula ( F , H ) , and pluteus larva ( G , I ) . Panels J and K show the effect in the translation of FGFR1-GFP fusion protein expression after FGFR1 MO injection ( K–K′ ) compared to controls injected with FGFR1-GFP mRNA only ( J–J′ ) . A–C , E–G , J and K are bright-field images taken with DIC , E′–G′ , J′ and K′ are fluorescent images , whilst D , H , and I are stacks of merged confocal Z sections . In panels D , H , and I , MHC was immunostained in red , acetylated tubulin in green , and nuclei were labeled blue with DAPI . Embryos in panels D and I are the same shown in Figure 2I and Figure 4A , respectively . Embryos in panels A–E , G , and I–K are seen in lateral view while those in panels B , F , and H are in frontal view . fv , frontal view; lv , lateral view; cp , coelomic pouches; pc , pigment cells; sp , spicules; mo , mouth; st , stomach; in , intestine . DOI: http://dx . doi . org/10 . 7554/eLife . 07343 . 01010 . 7554/eLife . 07343 . 011Figure 2—figure supplement 3 . Immunostaining of P-Elk and expression analysis of genes encoding putative MAPK effectors and FoxC by double FISH . ( A , B ) Localization of P-Elk protein by immunostaining . ( C–F ) Spatial expression domains of ( C , D ) Erg and ( E , F ) Ets1/2 with respect to FoxC by double FISH . Panels A , B are DIC images , and panels C–F are stacks of merged confocal Z sections . Insets in panels E , F show representative single confocal sections . FoxC is red , Erg is green , and Ets1/2 is cyan . Nuclei were stained blue with DAPI . Embryos in A and C panels are viewed from the vegetal pole while all the others are seen in frontal view . DOI: http://dx . doi . org/10 . 7554/eLife . 07343 . 011 To reinforce the role of FGF signaling mediated through FGFR1 during myogenesis , we performed gene knockdown experiments by injecting anti-translation MO against FGFR1 at a concentration of 500 μM . The MO toxicity and efficacy were examined by using a control MO and a GFP fusion construct , respectively ( Figure 2—figure supplement 2 ) . Although in 90% of the morphant larvae , a malformation of coelomic pouches ( Figure 2F ) and a severe down-regulation of the MHC protein level were evident ( Figure 2G , H ) , two different phenotypes concerning the compartmentalization of the archenteron were observed . 70% of the larvae displayed a fully elongated gut , with the cardiac and anal sphincters well formed ( Figure 2F ) , while in the rest cases both sphincters were absent or partly formed ( Figure 2H ) . In addition , an abnormal extension of cilia in the apical ectoderm was observed ( Figure 2H insert ) that provides a putative additional role of FGFR1 in patterning the apical organ ( Paola Oliveri , personal communication ) consistent with the expression of FGFR1 in the apical domain ( Figure 1A ) . An evident reduction of the cilia in the gut lumen is also observed ( Figure 2H ) , which may be related to the FGFR1 expression seen in the endoderm during gastrulation ( Figure 1—figure supplement 4 ) . Moreover , the larval skeleton and the formation of pigment cells were not affected , confirming that FGFR1 is not involved either in PMC patterning and subsequent skeleton formation or in pigment cell development ( Figure 2F ) . An even more severe phenotype was obtained by injecting the embryos with the mRNA encoding FGFR1 Dn with the cardiac and anal sphincters being absent in 60% of the cases ( Figure 2J ) . As expected , MHC levels were greatly reduced whilst the same abnormal extension of cilia in the apical ectoderm and the reduction of the cilia in the gut lumen were observed . Taking together , these experiments not only demonstrate clearly an essential role of FGF signaling through FGFR1 in sea urchin myogenesis but also suggest its possible involvement in the formation of the ciliated gut epithelium and the ciliary ectoderm . We then tested the expression of the myoblast marker genes FoxY , FoxC , and FoxF in the FGF signaling-perturbed embryos at the very early gastrula ( 28 hr ) , mid gastrula ( 36 hr ) , and late gastrula ( 48 hr ) stages , respectively . While FoxY expression did not change when FGF signaling was perturbed ( Figure 3A–D ) , a severe impact was observed in FoxC and FoxF transcript levels in all treated embryos ( Figure 3E–L ) . To check whether the specification of other oral NSM cell lineages were affected as well by the FGFR1 perturbation , we also tested the expression of Ese , a marker of the blastocoelar cell lineage , and found it unaffected ( Figure 3A , B ) . These results further support the involvement of FGF signaling in myogenesis and demonstrate that FoxC and FoxF factors are acting downstream of it . 10 . 7554/eLife . 07343 . 012Figure 3 . Spatial analysis of gene expression after FGF pathway perturbation by FISH . FoxY ( A–D ) FoxC ( E–H ) , FoxF ( I–L ) , MHC ( I , J ) , and Ese ( A , B ) transcript localization tested by FISH in control embryos ( A , C , E , G , I , K ) and in embryos treated with SU5402 ( B , F , J ) or injected with FGFR1 MO ( D , H , L ) ( for MO injection controls see also ‘Materials and methods’ and Figure 2—figure supplement 2 ) . Panels A , B , I , and J show double FISH . FoxY was stained in green , FoxC and FoxF in red , MHC in cyan , and Ese in magenta . Nuclei were labeled blue with DAPI . Each picture is a stack of merged confocal Z sections . Yellow circles indicated by yellow arrowheads show cells co-expressing the analyzed genes . The orientation of the larvae is reported for each panel: fv , frontal view; av , animal view; lv , lateral view . DOI: http://dx . doi . org/10 . 7554/eLife . 07343 . 012 ETS family transcription factors are known effectors of the MAPK-signaling pathway . In order to understand whether ETS-domain-containing transcriptional regulators are likely to be the downstream effector of the MAPK pathway leading to FoxC and FoxF transcription in presumptive myoblasts , we searched for ETS family factors that are significantly expressed in the NSM at the very early gastrula stage . From the 11 members of the ETS gene family that are present in the sea urchin genome , only 3 are expressed in the NSM at 30 hr; Elk , Erg , and Ets1/2 ( Rizzo et al . , 2006 ) . We performed an immunostaining experiment on the phosphorylated active form of Elk ( P-Elk ) and double FISH of FoxC and Erg or Ets1/2 . The spatial localization of P-Elk protein seemed to be excluded from the myogenic cells , when comparing it to the expression pattern of FoxC ( Figure 2—figure supplement 3A–C ) , suggesting that P-Elk is probably not the candidate effector of MAPK involved in myoblast specification . Erg expression was also not coincided with FoxC transcripts at 30 hr ( Figure 2—figure supplement 3C , D ) , thus , indicating that this factor is also not the downstream effector . On the other hand , Ets1/2 expression showed a clear co-localization with FoxC transcripts at around 30 hr ( Figure 2—figure supplement 3E , F ) that perfectly corresponds to the onset of myogenesis , suggesting a putative role of Ets1/2 as one of the terminal effectors of MAPK pathway acting during myoblast specification . Further perturbation experiments are needed to confirm this hypothesis . Given the interesting temporal and spatial expression profile of FoxY , FoxC , FoxL1 , and FoxF regulators during myogenesis and the decrease in FoxC and FoxF expression observed after perturbing FGF signaling , translation-blocking MOs specific for all four Fox factors were designed and tested at a concentration of 200 μM . The injected embryos were fixed at the larval stage for immunostaining to test MHC protein levels . In 90% of the FoxY morphants , the archenteron was fully elongated with signs of incomplete differentiation and the coelomic pouches were lacking , as was shown previously ( Song and Wessel , 2012; Materna et al . , 2013 ) . The MHC level was also severely reduced ( Figure 4A , B ) . 80% of the FoxC morphants showed a similar morphological alteration; the archenteron was fully invaginated and partially compartmentalized , missing completely the coelomic pouches and MHC protein ( Figure 4C ) . In 90% of the FoxF morphant larvae , a fully tripartite archenteron was present with disorganized coelomic pouches and a great decrease in the MHC level ( Figure 4D ) . 95% of the FoxL1 MO-injected larvae possessed a differentiated archenteron with disrupted coelomic pouches and reduced MHC protein level ( Figure 4E ) . The phenotypes caused by FoxC , FoxF , and FoxL1 MOs were confirmed by injecting a second translation-blocking MO targeting to different sequences ( Figure 4—figure supplement 1 ) . These observations reinforce the crucial role of Fox family factors in sea urchin myogenesis . 10 . 7554/eLife . 07343 . 013Figure 4 . MHC protein detected by immunostaining after perturbation of putative myogenic regulators . MHC protein localization was tested by immunostaining in fluo-control MO-injected pluteus larvae ( 72 hr ) ( A ) and in embryos of the same age injected with MOs against FoxY ( B ) , FoxC ( C ) , FoxF ( D ) , FoxL1 ( E ) , MyoD2 ( F ) , Six1/2N ( G ) , and Tbx6 ( H ) ( for MO injection controls see also Materials and methods and Figure 2—figure supplement 2 ) . The ciliary band and gut internal cilia were stained by immunohistochemistry with an anti-acetylated tubulin antibody . Each picture is a stack of merged confocal Z sections with MHC in red and acetylated tubulin in green . Nuclei were labeled blue with DAPI . All embryos are seen in lateral view with the oral side on the right . White arrows indicate the position of cardiac sphincters . White lines indicate muscle fibers ( mf ) . Below each panel , statistics of muscle fiber phenotype observed are reported as normal ( 6–7 mf ) , mild ( 4–5 mf ) , or strong ( 0–2 mf ) . A co-expression analysis of Six1/2 and FoxC is reported in Figure 4—figure supplement 1 . Analysis of the temporal expression profile of two distinct Six1/2 isoforms and visualization of pigmentation after perturbing Six1/2N isoform are reported in Figure 4—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 07343 . 01310 . 7554/eLife . 07343 . 014Figure 4—figure supplement 1 . Control experiments for MOs . Circumesophagael muscles were tested by phalloidin staining in fluo-control MO-injected pluteus larvae ( 72 hr ) ( A , A′ ) and in embryos of the same age injected with MOs against FoxC ( B , B′ ) , FoxF ( C , C′ ) , and FoxL1 ( D , D′ ) at different concentrations . Each picture is a stack of merged confocal Z sections . Phalloidin is seen in green and nuclei are labeled blue with Hoechst . All embryos are seen in lateral view with the oral side on the left . Below each panel , statistics of muscle fiber phenotype observed are reported . DOI: http://dx . doi . org/10 . 7554/eLife . 07343 . 01410 . 7554/eLife . 07343 . 015Figure 4—figure supplement 2 . Co-expression analysis of Six1/2 and FoxC by double FISH . Relative spatial expression domains of Six1/2 and FoxC at the mid gastrula stage ( 42 hr ) . Image is a stack of merged confocal Z sections in all channels . Inset shows representative single confocal section of the tip of the archenteron . Color code of channel association to each gene is shown in each panel . Nuclei are stained blue with DAPI . Embryo is seen in a frontal view . DOI: http://dx . doi . org/10 . 7554/eLife . 07343 . 01510 . 7554/eLife . 07343 . 016Figure 4—figure supplement 3 . The two Six1/2 isoforms . ( A ) Upstream sequence of the of the Six1/2 gene . Two ATGs are shown in red . The upstream one , highlighted in bold , corresponds to the first ATG in the long isoform ( Six1/2N ) that is probably generated by an alternative transcription start . The downstream ATG indeed corresponds to the first ATG of the short isoform in which transcription starts a few nucleotides upstream of it ( Andrew Ransick , personal communication ) . Highlighted in different colors show the regions where the different set of qPCR primers were designed: the ones used to amplify the upstream sequence belonging to Six1/2N isoform only , are in yellow , while the ones used to amplify part of the homeobox domain ( in bold ) , common to both isoforms , are highlighted in olive green . The target sequence used to design the MO against the long isoform Six1/2N is highlighted in cyan . ( B ) Temporal expression profiles of Six1/2 distinct isoforms during sea urchin embryogenesis . Graph shows the number of transcripts per embryo during embryogenesis revealed by qPCR . Six1/2HD represents the sum of the number of transcripts of the two isoforms , while Six1/2N shows only the number of transcripts for Six1/2N . The columns represent average of various measurements , and the error bars are standard deviations . ( C , D ) Bright-field images were taken with DIC of ( C ) control uninjected larva and ( D ) Six1/2N morphant ( 72 hr ) for visualizing the effect on pigmentation . DOI: http://dx . doi . org/10 . 7554/eLife . 07343 . 01610 . 7554/eLife . 07343 . 017Figure 4—figure supplement 4 . Control experiments for MOs . Circumesophagael muscles were tested by phalloidin staining in fluo-control MO-injected pluteus larvae ( 72 hr ) ( A , A′ in Figure 4—figure supplement 1 ) and in embryos of the same age injected with MOs against Six1/2 ( A , A′ ) and Tbx6 ( B , B′ ) at a concentration of 100 μM . Each picture is a stack of merged confocal Z sections . Phalloidin is seen in green and nuclei are labeled blue with Hoechst . All embryos are seen in lateral view with the oral side on the left . Below each panel , statistics of muscle fiber phenotype observed are reported . DOI: http://dx . doi . org/10 . 7554/eLife . 07343 . 017 MOs blocking the translation of three mesodermal markers Tbx6 , MyoD2 , and Six1/2 that are known to have conserved roles in muscle development were included in the study . MyoD2 and Tbx6 are both parts of the molecular fingerprint of the myoblasts at the late gastrula stage , which makes them good candidates as myogenic regulators ( Andrikou et al . , 2013 ) . Six1/2 is known to be a part of the regulatory state of the aboral mesoderm , which gives rise to two other mesodermal derivatives ( pigment cells and coelomic pouches ) ( Poustka et al . , 2007 ) . At the mid gastrula stage , Six1/2 expression partially overlapped with that of FoxC in the oral mesodermal domain ( Figure 4—figure supplement 2 ) . This transient expression seems to be correlated with the transcription of a late isoform whose peak of expression was seen at 42 hr ( Figure 4—figure supplement 3A , B ) . Therefore , a MO blocking the translation of the late isoform was designed and tested . 85% of the MyoD2 morphants showed an elongated archenteron , and in most of the cases with reduced sphincter constrictions . The coelomic pouches were also not formed properly and the MHC level was dramatically reduced ( Figure 4F ) . In 80% of the Six1/2 morphants , the archenteron was elongated but not differentiated , the coelomic pouches were absent , and the MHC level was decreased ( Figure 4G ) . Interestingly , unlike what previously shown by using an MO against the early isoform of Six1/2 , which negatively affects pigmentation ( Ransick and Davidson , 2012 ) , embryos injected with the MO targeted to the late Six1/2 isoform were fully pigmented ( Figure 4—figure supplement 3C , D ) . Finally , 95% of Tbx6 morphants had a milder phenotype , with no signs of malformed coelomic pouches and only a partial disrupted muscle fiber assembly as shown by MHC immunostaining ( Figure 4H ) . Similar phenotypes were observed when injecting a second MO against Six1/2 and Tbx6 ( Figure 4—figure supplement 4 ) . The above experiments demonstrate an important role of MyoD2 and the late isoform of Six1/2 as well as a limited role of Tbx6 in sea urchin muscle development , thus , suggesting that Tbx6 acts in a redundant manner with other factors , a known property of T-box factors ( Gentsch et al . , 2013 ) . To test this hypothesis , double knockdown assays should be conducted . In order to identify the myogenic gene core set and unravel in detail the inter-regulatory mechanisms that take place during myogenesis , we tested the expression of genes encoding selected mesodermal factors and signaling components in the morphants by FISH ( Figure 5 ) , quantitative PCR ( qPCR ) ( asterisks in Figure 6 ) , and NanoString analysis ( Figure 6—source data 1 , 2 ) . We selected three different developmental stages for our analyses that correspond to three distinct steps of myogenesis: early gastrula ( 30–35 hr ) as myoblast specification stage , a broad mid gastrula ( 36–44 hr ) , and late gastrula ( 45–48 hr ) stage as intermediate and later steps of myogenesis , respectively . For NanoString analyses , we used a code set containing probes for most known NSM factors expressed at these developmental stages and a number of signaling components . The MO effects on transcript levels analyzed using the three methods were mostly consistent . For quantitative analyses using QPCR and Nanostring ( Figure 6 ) , changes in the transcript level were considered significant if the effect between control and MO injected embryos is more than twofolds . Epistatic interactions were evaluated when the effects were observed in at least two significant data points . 10 . 7554/eLife . 07343 . 018Figure 5 . Spatial analysis of gene expression after MO perturbation of selected putative myogenic regulators by FISH . FoxC , FoxY , FoxF , MHC , and nanos transcripts were detected by FISH in fluo-control MO injected embryos ( A , C , E , G , I ) and in embryos injected with MOs against FoxY ( B , D , F ) , FoxC ( H ) , and FoxF ( J ) . All images are obtained as stacks of merged confocal Z sections . Panels G , H show double FISH . In panel G , single channels over DAPI are shown as insets . FoxY was stained in green , FoxC and FoxF in red , MHC in cyan , and Nanos in magenta . Nuclei were labeled blue with DAPI . Yellow circles indicated by yellow arrowheads show cells co-expressing the analyzed genes . The orientation of the embryos is indicated in each panel: fv , frontal view; lv , lateral view . DOI: http://dx . doi . org/10 . 7554/eLife . 07343 . 01810 . 7554/eLife . 07343 . 019Figure 6 . Effects of FoxY , FoxC , FoxF , FoxL1 , MyoD2 , Six1/2 , and Tbx6 perturbations on transcript levels of selected mesodermal genes at 44 hr and 48 hr . Each diamond represents a single measurement of three independent biological experiments . Fold differences were calculated between experiments and control counts using the quantitative data obtained from the NanoString nCounter . Onefold change represents no change; ≥ 2 indicates increased expression level significantly ( blue labels ) ; ≤ 0 . 5 indicates decreased expression level significantly ( red labels ) . Asterisks indicate perturbation effects as measured in independent biological experiments by qPCR . NanoString and qPCR perturbation data normalized against controls are provided in Figure 6—source data 1 , and raw NanoString data are provided in Figure 6—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 07343 . 01910 . 7554/eLife . 07343 . 020Figure 6—source data 1 . Perturbation data derived from NanoString and qPCR analysis showing fold differences of gene expression in MO-injected embryos after normalization against controls . DOI: http://dx . doi . org/10 . 7554/eLife . 07343 . 02010 . 7554/eLife . 07343 . 021Figure 6—source data 2 . Raw data derived from NanoString analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 07343 . 021 At the early gastrula stage , FoxY and FoxC expression was strongly reduced in FoxY MO-injected embryos ( Figure 5A–D ) , suggesting that FoxY positively regulates itself and FoxC . This positive regulation continued to the mid and late gastrula stages and FoxY also positive regulated the other two Fox factors , FoxF and FoxL1 ( Figure 6 ) . The expression of Nanos , a germ cell marker ( Juliano et al . , 2006 ) that is part of the molecular signature of the myoblasts precursors at that developmental time point ( Andrikou et al . , 2013 ) , was also downregulated in FoxY morphants ( Figure 5E , F and Figure 6 ) . FoxC , FoxF , and FoxL1 were downstream of FoxC at the mid gastrula stage . Expression of FoxF and FoxL1 required FoxC input , whereas FoxC negatively regulated itself ( Figure 5G , H and Figure 6 ) . At the late gastrula stage , FoxY expression domain was expanded in the FoxF morphants to include the oral vegetal region of the archenteron tip , where myogenesis takes place ( Figure 5I , J ) . In addition to the inter-regulatory properties of the Fox factors , at the mid gastrula stage , FoxY also positively regulated Pitx2 ( Hibino et al . , 2006 ) , MyoR , a gene expressed in sea urchin mesoderm but not in the myoblasts ( Andrikou et al . , 2013 ) , Six1/2 and ScratchX . FoxC also activated Tbx6 and repressed Not and Dachshund ( Dachs ) , an aboral NSM gene in the sea urchin embryo ( Luo and Su , 2012 ) . Six1/2 appeared to activate FoxL1 , MyoR , and Pitx2 , as previously suggested ( Hibino et al . , 2006 ) . At the late gastrula stage , FoxY continued to activate the aforementioned downstream factors although its effect on Six1/2 was diminished and activated in addition SoxE gene . FoxL1 gave positive inputs to FoxF , MyoR , ScratchX , and Pitx2 and repressed Dachs , FoxY , and Not . Tbx6 repressed FoxY , Dach , and Scl and activated Pitx2 and MyoR . Positive inputs on Tbx6 gene from FoxC and on MyoR gene from Six1/2 remained at the late gastrula stage . These perturbation analyses revealed complex positive and negative regulatory interactions among the four Fox factors and other transcriptional regulators , and the myogenic GRN models were formulated based on these results . This detailed perturbation analysis coupled with the available high-resolution transcriptional profiling during sea urchin embryogenesis ( Materna et al . , 2010 ) led to the construction of a GRN model that orchestrates sea urchin myoblast specification . Based on the examined developmental stages and regulatory states , three GRN diagrams have been illustrated using the BioTapestry software ( www . biotapestry . org ) , one for the very early and two for the mid and late gastrula stages . The three proposed GRN models are summarized in Figure 7 . 10 . 7554/eLife . 07343 . 022Figure 7 . Schematic representation and view from all nuclei of the NSM regulatory interactions in early , mid , and late sea urchin gastrulae . On the left side , three developmental stages of the sea urchin embryo are schematized: ( A ) early , ( B ) mid , and ( C ) late gastrula stage . On the right side , the genetic interactions found within this study are summarized . Different colors are used for each domain showing exclusive regulatory states: oral animal non-skeletogenic mesodermal ( NSM ) ( OR AN NSM ) , salmon pink; NSM , blue; aboral NSM ( AB NSM ) , lavender; small micromere derivatives ( SM ) , green; myogenic domain ( M ) , light red; endoderm ( ENDO ) , yellow-green; oral ectoderm ( OR ECTO ) , light gray . Genes are presented as horizontal thick lines and their names are reported below the thick lines . The wiring among the genes is shown with solid lines , although none of them has been demonstrated to be direct . Arrows represent positive regulation , bars represent repression , and white bullets , together with the dashed lines , indicate signaling events . Genes that are expressed in more than one domain , for which the putative inputs were revealed by NanoString but not validated by spatial expression analysis , are shown on a shaded background . The asterisk in A relates to the fact that we did not demonstrate which FGF factor signals to FGFR1 . A co-expression analysis of several genes included in the gene regulatory network ( GRN ) diagrams is reported in Figure 7—figure supplement 1 . Numbers associated to inputs indicate the evidence for all interactions reported and are listed in Figure 7—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 07343 . 02210 . 7554/eLife . 07343 . 023Figure 7—source data 1 . Evidence for all inputs reported in Figure 7 . DOI: http://dx . doi . org/10 . 7554/eLife . 07343 . 02310 . 7554/eLife . 07343 . 024Figure 7—figure supplement 1 . Co-expression analysis of genes encoding mesodermal factors by double FISH . Relative spatial expression domains of ( A ) SoxE and FoxC at the mid gastrula stage ( 40 hr ) , ( B ) Not and FoxC at the mid gastrula stage ( 44 hr ) , and ( C ) Pitx2 and FoxC at the late gastrula stage ( 48 hr ) . Each picture is a stack of merged confocal Z sections in all channels . Inset in panel A shows a representative single confocal section of the tip of the archenteron . Color code of channel association to each gene is shown in each panel . Nuclei are stained blue with DAPI . All embryos are in a frontal view . DOI: http://dx . doi . org/10 . 7554/eLife . 07343 . 024 At the very early gastrula stage ( 28–32 hr; Figure 7A ) , FGFA is produced in the ventrolateral ectoderm and is received by FGFR1 , which is expressed in all myoblast precursor cells . FGF signaling induces myoblast specification through the MAPK/ERK pathway and a downstream effector ( indicated as a circle in Figure 7A ) , possibly Ets1/2 , and results in the activation of FoxC transcription . FoxC activation needs an additional positive input from FoxY . Since FoxY expression is excluded from the other two NSM lineages ( pigment and blastocoelar cells ) ( Andrikou et al . , 2013 ) , it is reasonable to consider the existence of a Non-Skeletogenic Mesoderm Repressor ( NSMR ) that inhibits FoxY transcription in these cell populations . The existence of such a repressor has also been postulated by Materna and Davidson ( 2012 ) . Here , we propose that in the myoblast precursors this repressive action is blocked through a double-negative gate , due to the existence of another repressor ( X ) . At the mid gastrula stage ( 40–44 hr; Figure 7B ) , FoxY establishes the myoblast regulatory state by activating a variety of transcriptional regulators including all four Fox factors . FoxC is a broad connected hub gene and gives positive inputs to a number of factors that compose the muscle gene battery such as FoxF , FoxL1 , and Tbx6 . FoxL1 needs also an additional input from Six1/2 . Finally , SoxE is also part of the molecular identity of the myoblasts at the mid gastrula stage since it appears to largely co-express with FoxC ( Figure 7—figure supplement 1A ) ; however , the upstream factor ( s ) controlling SoxE expression remain unknown . At the late gastrula stage ( 44–48 hr; Figure 7C ) , the expression of FoxY , Nanos , Six1/2 , and SoxE clears from the myogenic territory and confines in other NSM domains ( small micromere [SM] and aboral NSM [AB NSM domain] ) . This occurs due to the repressive functions of FoxF and Tbx6 on FoxY , which results in a subsequent loss of Six1/2 , SoxE , and Nanos expression . Similarly , Not ( whose expression domain relative to that of FoxC is reported in Figure 7—figure supplement 1B ) , as well as Scl and Dach receive negative inputs from FoxL1 and Tbx6 , respectively , that together prevent their expression in the myogenic domain . The initiation of MHC transcription marks the terminal differentiation state ( Andrikou et al . , 2013 ) . The analyses of the perturbation data provided additional information concerning the regulatory interactions seen in the other three NSM domains . As elsewhere stated ( Materna et al . , 2013 ) , FoxY is at the top of the hierarchy of the NSM GRN . Tbx6 and FoxL1 positively regulate Pitx2 ( whose expression domain relative to that of FoxC is reported in Figure 7—figure supplement 1C ) and together with Six1/2 activate MyoR . However , the fact that Tbx6 and FoxL1 are not expressed in the AB NSM ( Andrikou et al . , 2013 ) lead us to the conclusion that their inputs on MyoR are indirect . FoxL1 may also activate Pitx2 in the oral ectoderm ( OR ECT ) . Finally , in the SM lineage , FoxY is upstream of Nanos , FoxC , and Pitx2 . In this study , we propose that the muscle progenitors originate from a pool of unspecified cells located at the oral/lateral periphery of the vegetal plate at the mesenchyme blastula stage and they adapt the myogenic fate by receiving an inductive signal at the moment of gastrulation . This would further imply that all developmental decisions regarding the separation of all four NSM regulatory states ( pigment , blastocoelar , muscle , and coelomic pouch cells ) take place during the interval between the blastula and the very early gastrula stage . A key finding of this work was the recruitment of FGF signal in myoblast specification . FGF-signaling cascade is reported to trigger the emergence and/or promote the expansion of muscle lineage progenitors in different organisms and probably possessed an ancestral role in mesoderm patterning ( at least in deuterostomes ) , along with other developmental processes . Genes encoding FGF and FGFR were already present in the last common ancestor of all metazoans , but the origin of the FGF/FGFR couple appears to be an innovation specific to the Eumetazoa , potentially linked to the increase of animal complexity ( Bertrand et al . , 2014 ) . The acquisition of diverse roles and functions of FGF signaling occurred with the expansion of FGF and FGFR families during evolution ( Itoh and Ornitz , 2011 ) . For example , vertebrates are known to have the largest number of FGF-signaling components due to the two whole-genome duplication events ( 22 ligands and four receptors ) . FGF signals are involved in various developmental contexts ( e . g . , somitogenesis , cancer , gastrulation , metabolism , neural induction , etc ) ( Dorey and Amaya , 2010; Oki et al . , 2010; Naiche et al . , 2011; Ko et al . , 2014; Neugebauer and Yost , 2014; Sandhu et al . , 2014 ) . Similar situations have been observed in Drosophila where the moderate expansions of both FGFR ( breathless and heartless ) and FGF ( Pyramus [Pyr] , Thisbe [Ths] , and Branchless [Bnl] ) families resulted in different biological functions such as cell–cell interactions during mesoderm layer formation , caudal visceral muscle formation , tracheal morphogenesis , and glia differentiation ( Muha and Muller , 2013 ) . In echinoderms , the two paralogues FGFR1 and FGFR2 are involved in two distinct developmental processes: myoblast specification ( this study ) and PMC migration ( Rottinger et al . , 2008 ) , respectively . Diverse functions of FGF signaling seen among different animal taxa suggest that this signaling system is redeployed at different levels of GRN hierarchy and acquire new functions in developmental and physiological processes . Moreover , the hierarchical position of FGF signaling seems to depend on the level of complexity of the developmental process placing it as a ‘checkpoint’ in a non-conserved way regarding its downstream targets , a known property of GRN ‘plug-in’ devices defined previously ( Davidson , 2010 ) . The functional importance of being attachable onto any kind of circuitry reflects the need of developmental GRNs to adopt factors that act as turn on/off apparatuses , resulting in change of the undefined cell regulatory state to their specific cell lineage . An interesting finding of this analysis is the hierarchical organization of the regulatory interactions in the network topology . FoxY , a sea urchin-specific Fox family factor , lies on the top of the GRN architecture and is the key upstream regulator of sea urchin myogenesis and coelomic pouch formation . As shown elsewhere , FoxY is a direct target of the D/N signal , produced in the adjacent NSM . D/N signaling is known to be effective in cells being in direct contact with the ligand-producing cells ( Wang , 2011 ) . Therefore , the positive autoregulation of FoxY maintains its own expression during gastrulation when cells invaginate into the blastocoel and may be away from the signaling source . FoxY then promotes the expression of downstream positive regulators necessary for the execution of the myogenic lineage fate . The next tier of the myogenic GRN includes FoxC , SoxE , ScratchX , and Six1/2 factors and may be regarded as the core level of GRN . FoxC initially needs both positive inputs from FGF signaling and FoxY to be activated but then acts as an autorepressor in order to stabilize its number of transcripts . All factors belonging to this group are highly interconnected and provide multiple inputs to their downstream targets . On the same level of the GRN , immediately adjacent to the core , stands the intermediate layer of the network composed of the FoxF and FoxL1 transcriptional effectors . Finally , the last level of the network includes the differentiation driver and repressor Tbx6 that triggers the terminal myoblast differentiation . Tbx6 seems to cooperate redundantly with other factors in an OR logic ( Davidson , 2010 ) , which also explains the mild phenotype seen in the Tbx6 perturbations . The differentiation drivers , such as MyoD ( Andrikou et al . , 2013 ) , are seen in the bottom of the GRN and probably activate the differentiation gene battery set , composed of a number of structural genes ( e . g . , MHC ) that account for the specific function of the muscle cell . This highly multilevel transcriptional regulation seen in myogenesis can be interpreted as an example of ‘correlated evolution’ where the increase of complex processes is accompanied by the expansion of the relevant regulatory systems whilst in less complex systems a shallower regulatory system can be provided . Another important outcome that arose from this analysis is the sequential inter-regulatory mechanism observed among the four Fox family members FoxY , FoxC , FoxF , and FoxL1 . The Fox family of transcription factors is an ancient gene family and it has been proposed that its evolutionary origin occurred in a clade of unicellular organisms ( Baldauf , 1999 ) . This family has expanded over time through multiple duplication events , and sometimes through gene loss , and resulted in over 40 members in vertebrates , grouped in 23 subclasses ( Hannenhalli and Kaestner , 2009 ) . Among them , FoxC , FoxF , and FoxL subfamilies are of particular interest because they are clustered in most metazoan genomes and usually involved in mesoderm specification/differentiation processes ( Mazet et al . , 2006; Shimeld et al . , 2010 ) . Moreover , a linked activation or an overlapping expression has been reported in some cases , as in vertebrates , where FoxC specifies the dorsal mesoderm and derivatives , while FoxF patterns the lateral mesoderm and derivatives ( Beh et al . , 2007; Zinzen et al . , 2009; Amin et al . , 2010; Fritzenwanker et al . , 2014 ) . In the sea urchin , where a cluster of FoxC , FoxF , and FoxL1 genes is present ( Andrikou et al . , 2013 ) , we witness a similar regulatory logic where the different Fox factors pattern in an overlapping fashion in different compartments of the coelomic pouches . This cluster was probably expressed initially in developing mesodermal tissues and further evolved in regulating the specification and compartmentalization of mesodermal derivatives . Moreover , our findings highlight the importance of the hierarchical position of the Fox family factors in the GRN . The expression patterns and sequential activation of FoxC , FoxF , and FoxL1 genes in time reflect the linkage properties of the retained cluster: among the three genes , FoxC is the first to be turned on in the myogenic lineage and is necessary for the activation of the downstream factors FoxF and FoxL1 , which are inter-regulated . The outcome of such complex inter-wiring GRN may contribute to the establishment of a more robust output , able to mask putative perturbations of single nodes , as recently proposed ( Macneil and Walhout , 2011 ) . This study provides additional insights into understanding the logic of the exclusive mechanisms that occur in the sea urchin embryo during myoblast specification ( Davidson , 2009 ) . By 24 hr post fertilization , the separation of the different NSM regulatory states is defined . The vegetal plate of the embryo consists of four distinct NSM cell populations: the aboral part that occupies about two-thirds of the total cells will differentiate into the pigment cells; the oral part that consists of the remaining one-third of the total cells will become the blastocoelar cells; the four SMs in the center are primordial germ cells; approximately 2–3 cells in the oral/lateral domain of the periphery of the vegetal plate will give rise to the myoblast precursors . As showed elsewhere , the specification of the distinct NSM domains depends on D/N and Nodal signals ( Sherwood and McClay , 1999; Ransick and Davidson , 2006; Duboc et al . , 2010 ) . At 9 hr post fertilization , Delta ligand produced by the skeletogenic mesoderm activates a D/N cascade , which subsequently initiates Gcm and GataE transcription in all NSM cells ( Ransick and Davidson , 2006; Lee et al . , 2007 ) . Later , at 24 hr , the NSM GRN becomes regionalized into distinct oral and aboral NSM GRNs in consequence of Nodal signaling through the immediate expression of a Nodal target , Not , in the oral NSM . This causes the repression of the aboral GRN in the oral NSM cells , and the aboral NSM cells still express Gcm and GataE as part of the regulatory state of the pigment cell lineage ( Materna et al . , 2013 ) . In the oral NSM , the expression of a new suite of regulatory genes is taking place that belong to the blastocoelar lineage GRN ( Solek et al . , 2013 ) . Our study showed that in the periphery of the oral/lateral NSM , the appearance of 2–3 unspecified cells is evident . These cells will soon get specified , at 28/30 hr , by receiving an inductive FGF signaling at the moment of gastrulation . FGF signal reception together with the double-negative gate caused by the repressive action of factor X on NSMR leads to the transcriptional activation of FoxY and FoxC expression . Positive inputs from FoxY and FoxC genes into downstream effectors promote the recruitment of the myogenic genes and lock down the myoblast regulatory state . Genes such as Scl , Dachs , Pitx2 , and Not that belong to different NSM regulatory states are excluded from the myogenic domain leaving only the muscle GRN operating . Finally , spatial repression circuits generate regulatory transitions in the expression of key genes such as FoxY , Nanos , SoxE , and Six1/2 , which now are established only in the other NSM GRNs . The sea urchin myogenic GRN is a nice example of how dynamic developmental processes can be encoded in the genome and shows clearly that understanding in depth the wiring properties of a developmental GRN model can provide a comprehensive view on the relationship between the regulatory architecture and gene expression dynamics . The nature of the evolutionary alterations that arise from regulatory changes depends on the hierarchical positions of these changes within a GRN . One of the most striking findings of this study concerns the paradox that genes are constantly re-used in the same context , but are rewired in different networks . Despite the conserved regulatory modules found in the system , the myogenic GRN structure has diverged extensively among animal groups . As a consequence , the level of the functional importance of the homologous transcriptional regulators ( e . g . , Six1/2 in sea urchin and Ceh-34 in C . elegans or MyoD in vertebrates and Nautilus in Drosophila ) ( this study , [Olson and Klein , 1994; Balagopalan et al . , 2001; Amin et al . , 2009] ) , as reflected from the position within the GRN , is often diversified . New , lineage-specific genes are recruited ( FoxY in sea urchin ) , and ‘master regulatory genes’ either have been lost completely ( e . g . , the absence of Pax3/7 ortholog in the sea urchin ) or lose their hierarchical position within the network ( e . g . , Nautilus in Drosophila and MyoD in vertebrates ) , with the level of complexity to be reflected in the wiring density and in the GRN organization . However , the fact that the same factors are used over and over again in such different animal systems indicates that the modular components are somehow required for keeping their myogenic activity during evolutionary time . It seems that as transcription factor families expanded and functionally diversified during evolution , the ancestral myogenic function may have been preserved in a more distant family member , rather than the homologous gene , providing the system with several regulatory alternatives , and explaining the high degree of evolutionary plasticity of developmental GRN architecture ( Andrikou and Arnone , 2015 ) . In conclusion , this large scale GRN analysis demonstrated a necessary hierarchical role for a large number of transcriptional regulators in muscle development and explained in a rational way the core gene network that is orchestrating the specification of the myogenic lineage . Moreover , it revealed the key signaling events involved in the activation of the muscle gene battery and underlined their crucial role in transforming an unspecified cell into a specific cell type with a characteristic molecular signature . Finally , this study reinforces the importance of GRN-based approach in understanding in detail complex developmental processes by assessing the causality of the regulatory mechanisms that accompany each step of the process . Adult S . purpuratus were obtained from Patrick Leahy ( Kerckhoff Marine Laboratory , California Institute of Technology , Pasadena , CA , USA ) and housed in circulating seawater aquaria in the Stazione Zoologica Anton Dohrn of Naples . Spawning was induced by vigorous shaking of animals or by intracoelomic injection of 0 . 5 M KCl . Embryos were cultured at 15°C in Millipore filtered Mediterranean seawater diluted 9:10 ( Vol:Vol ) in deionized H2O . FoxC , FoxF , FoxL1 , MyoD2 , Tbx6 , FGFR1 , and Six1/2 antisense MOs were obtained from Gene Tools ( Pilomath , OR , USA ) and injected at different concentrations in the presence of 0 . 12 M KCl . Various MO concentrations were tested and the lowest that enabled the observation of a phenotype was used for the experiments ( 500 μM for the FGFR1 MO and 200 μM for FoxC , FoxF , FoxL1 , MyoD2 , Tbx6 , and Six1/2 ) . Second MOs for FoxC , FoxF , FoxL1 , Six1/2 , and Tbx6 were used to confirm the morphant phenotypes ( Figure 4—figure supplements 1 , 4 ) . As a control experiment , a Standard Morpholino Control oligo end modified with 3′-Carboxyfluorescein ( control-fluo MO , Gene Tools ) was injected in parallel at the same concentration as the corresponding experiments ( Figure 2—figure supplement 2 ) . Embryos injected with FoxY , FoxC , FoxF , FoxL1 , MyoD2 , Tbx6 , and Six1/2 MOs displayed a normal gross morphology , similar to uninjected or fluo-control MO injected embryos up to the pluteus stage , except for the effects on the coelomic pouches , and in the case of FoxC MO , the apical organ , thus , suggesting confined effects of these MOs in the expression domains of the corresponding targeted genes . To test FGFR1 MO efficacy , embryos were injected with mRNA containing the MO target sequence fused to the 5′ of the gfp-coding sequence ( 500 ng/μl ) with or without the MO ( Figure 2—figure supplement 2 ) . FoxY MO was kindly provided by Stefan Materna ( Caltech , USA ) . MO sequences used in this study are listed in Supplementary file 1 . SU5402 was dissolved in DMSO and added to a final concentration of 20 μM at 26 hr , 28 hr , 30 hr , or 36 hr up to the collection time . Higher concentrations than this were lethal to the embryos soon after the addition of the drug and addition of the drug after 30 hr did not show any effect . U0126 was dissolved in DMSO and added to a final concentration of 10 μM at 24 hr as reported previously ( Fernandez-Serra et al . , 2004 ) . A corresponding volume of DMSO was added as controls . A table summarizing the drug treatments and the observed phenotypes is seen in Figure 2—figure supplement 1 . The primers used to amplify FGFA from embryonic cDNA were designed based on the gene models . 5′ and 3′ sequences were extended by the FirstChoice RLM-RACE Kit ( Ambion , Austin , TX , United States ) . The complete mRNA sequences were deposited into GenBank ( Sp-FGFA , HQ107979 ) . FGFR1 was amplified based on the published sequence ( U17164 ) . Both primer sets are in Supplementary file 1 . To construct the dominant-negative form of FGFR1 , a DNA fragment containing the signal peptide , the extracellular , and transmembrane domains was amplified by PCR ( forward primer: 5′-CGGGATCCATGAGTCTGCCGCGTTGTCC-3′ , reverse primer: 5′-CCATCGATTGTCTCGAGGGAACTCCCAC-3′ ) and cloned into the pCS2+MT vector . In situ RNA probe sequences for FoxY , FoxC , FoxF , Ese , Nanos , Six1/2 , SoxE , MHC , and Gcm are as previously published ( FoxY: [Ransick et al . , 2002]; FoxC , FoxF: [Tu et al . , 2006]; Ese: [Rizzo et al . , 2006]; Nanos: [Juliano et al . , 2006]; Gcm: [Ransick et al . , 2002]; Six1/2 , SoxE , and MHC: [Andrikou et al . , 2013] ) . Labeled probes were transcribed from linearized DNA using digoxigenin-11-UTP or fluorescein-12-UTP ( Roche , Indianapolis , IN , USA ) , or labeled with DNP ( Mirus , Madison , WI , USA ) following kit instructions . For whole-Mount In situ hybridization ( WMISH ) with single probe , we followed the protocol outlined in ( Minokawa et al . , 2004 ) . Double FISH and immunohistochemistry coupled to FISH were performed as described in ( Andrikou et al . , 2013 ) . For triple FISH , the third signal was developed using a 488 fluorophore-conjugated tyramide ( Invitrogen ) . Embryos were imaged with a Zeiss Axio Imager M1 . FISHs were imaged with a Zeiss 510Meta confocal microscope . Embryos were collected by gentle centrifugation and fixed in 2% paraformaldehyde in PBS for 15 min , washed 3 times in PBST , and incubated in 4% sheep serum and 1 mg/ml BSA in PBST for 30 min . Embryos were then incubated with a primary antibody ( anti-SpMHC , rabbit polyclonal antibody , diluted 1:600 , PRIMM , Italy or a commercially available anti-P-Elk1 [Serine 383] , mouse monoclonal antibody , dilution 1:100 , Santa Cruz Biotechnology , Santa Cruz , USA ) overnight at 4°C , washed 4 times in PBST , and followed by another incubation in 4% sheep serum and 1 mg/ml BSA in PBST for 30 min . Similarly , embryos were then incubated with a secondary antibody ( anti rabbit-AlexaFluor 555 , Invitrogen or anti mouse-HRP ) diluted 1:1000 for 1 hr in RT , washed 4 times in PBST , and imaged with a Zeiss 510Meta confocal microscope . Total RNA was isolated from cultures of various embryonic stages , approximately 100 embryos per replica . The RNA was extracted with RNAquous ( Ambion ) . The samples were treated with DNase I ( Ambion ) to remove DNA contamination as described by the manufacturer . First-strand cDNA was synthesized from total RNA using the VILO kit ( Invitrogen ) according to the manufacturer's protocol . Expression levels were quantified using the NanoString nCounter ( NanoString , UCL , London ) with a custom-designed probe set of 40 genes ( Supplementary file 1 ) . Samples were processed according to manufacturers' instructions and data processed as described previously ( Materna and Davidson , 2012 ) . Thresholds of 2- and 0 . 5-fold differences were chosen as significant changes ( Materna and Davidson , 2012 ) . Some data points were supplemented with results from quantitative real-time PCR ( qPCR ) analyses . qPCR was conducted as described ( Rast et al . , 2000 ) , using the ViiA 7 REAL TIME PCR detection system and SYBR green chemistry ( Applied Biosystems , Foster City , CA , USA ) . The primer sequences used are included in Supplementary file 1 . ddCt values were calculated between experiment and control embryos and converted to fold differences to be comparable to the NanoString data . Fold changes were calculated using poly-ubiquitin as a reference , and a threshold of twofold difference was chosen as a significant change ( Materna and Oliveri , 2008 ) . Normalized perturbation data against control are reported in Figure 6—source data 1 , and raw data are reported in Figure 6—source data 2 . The signal peptide cleavage site of FGFA was predicted by SignalP 3 . 0 ( http://www . cbs . dtu . dk/services/SignalP/ ) . The core sequences of FGFs from different organisms were aligned using the Clustal W program , and the alignment was confirmed manually . After removing gaps , the verified alignments were used to construct phylogenetic trees with the MacVector software based on the neighbor-joining method . Bootstrap support values were calculated by 1000 pseudoreplications . All phylogenetic trees were illustrated with the FigTree program ( http://tree . bio . ed . ac . uk/software/figtree/ ) . The phylogenetic tree for FGFRs was constructed in the same manner as the FGF tree but was based on the alignments of the tyrosine kinase domains .	Muscles , bones , and blood vessels all develop from a tissue called the mesoderm , which forms early on in the development of an embryo . Networks of genes control which parts of the mesoderm transform into different cell types . The gene networks that control the development of muscle cells from the mesoderm have so far been investigated in flies and several species of animals with backbones . However , these species are complex , which makes it difficult to work out the general principles that control muscle cell development . Sea urchins are often studied in developmental biology as they have many of the same genes as more complex animals , but are much simpler and easier to study in the laboratory . Andrikou et al . therefore investigated the ‘gene regulatory network’ that controls muscle development in sea urchins . This revealed that proteins called Forkhead transcription factors and a process called FGF signaling are crucial for controlling muscle development in sea urchins . These are also important factors for developing muscles in other animals . Andrikou et al . then produced models that show the interactions between the genes that control muscle formation at three different stages of embryonic development . These models reveal several important features of the muscle development gene regulatory network . For example , the network is robust: if one gene fails , the network is connected in a way that allows it to still make muscle . This also allows the network to adapt and evolve without losing the ability to perform any of its existing roles . Comparing the gene regulatory network that controls muscle development in sea urchins with the networks found in other animals showed that many of the same genes are used across different species , but are connected into different network structures . Investigating the similarities and differences of the regulatory networks in different species could help us to understand how muscles have evolved and could ultimately lead to a better understanding of the causes of developmental diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "evolutionary", "biology" ]	2015	Logics and properties of a genetic regulatory program that drives embryonic muscle development in an echinoderm
Retinal prostheses are promising tools for recovering visual functions in blind patients but , unfortunately , with still poor gains in visual acuity . Improving their resolution is thus a key challenge that warrants understanding its origin through appropriate animal models . Here , we provide a systematic comparison between visual and prosthetic activations of the rat primary visual cortex ( V1 ) . We established a precise V1 mapping as a functional benchmark to demonstrate that sub-retinal implants activate V1 at the appropriate position , scalable to a wide range of visual luminance , but with an aspect-ratio and an extent much larger than expected . Such distorted activation profile can be accounted for by the existence of two sources of diffusion , passive diffusion and activation of ganglion cells’ axons en passant . Reverse-engineered electrical pulses based on impedance spectroscopy is the only solution we tested that decreases the extent and aspect-ratio , providing a promising solution for clinical applications . Blindness affects 45 million people around the world with an increase of 1 to 2 million people each year ( Resnikoff et al . , 2004 ) . The two main retinal pathologies are age-related macular degeneration ( AMD , Ambati and Fowler , 2012; Finger et al . , 2011 ) and Retinitis Pigmentosa ( RP , Bocquet et al . , 2013 ) . Although the genetic alterations and the mechanisms subtending photoreceptor death are well described , and therapeutic strategies are under clinical trials ( Ferrari et al . , 2013; Talcott et al . , 2011 ) , retinal degeneration inexorably leads to blindness ( Tsujikawa et al . , 2008 ) . In this perspective , retinal prostheses provide a promising solution that remains to date a unique alternative for the patients . Restoring some visual perception using implants has been already achieved ( Humayun et al . , 2012; Shepherd et al . , 2013; Zrenner et al . , 2010 ) but still offers insufficient gains in visual acuity ( Humayun et al . , 2012; Zrenner et al . , 2010 ) ; ( Nanduri et al . , 2012; Rizzo , 2003 ) . Despite having fundamentally different designs and operating modes , the two main models of prostheses proposed to the patients with RP ( Argus II epiretinal prosthesis and the subretinal alpha IMS microphotodiode array ) restore some visual function although with a spatial resolution ( Ahuja and Behrend , 2013; Humayun et al . , 2012 ) ; ( Stingl et al . , 2013 ) that does not allow for the recognition of faces or autonomous locomotion . Improving the performances of such implants is thus a key strategic issue for further developments . This problem is actually a general issue shared by other sensory prosthesis , such as the cochlear implants ( Kral et al . , 1998 ) . This latter field , well in advance compared to retinal implants , has already demonstrated the importance of developing animal models for better understanding of the underlying physiological processes ( Fallon and Shepherd , 2009; Miller et al . , 2000 ) . To improve the resolution and efficiency of retinal prosthesis , it is therefore necessary to launch appropriate models that allow probing precisely and quantitatively the functional impact of prosthetic activation . Pioneering animal studies have proposed to use cortical recordings to explore the efficiency of various patterns of retinal electrical stimulation ( Chowdhury et al . , 2008 ) , including current steering methods ( Jepson et al . , 2014a; Matteucci et al . , 2013 ) or the effect of the return electrode configuration ( Cicione et al . , 2012; Matteucci et al . , 2013; Wong et al . , 2009 ) and studied the temporal aspect of prosthetic vision ( Elfar et al . , 2009; Fransen et al . , 2014; Jepson et al . , 2014a , 2013 , 2014b; Nadig , 1999; Schanze et al . , 2003; Sekirnjak et al . , 2008; Wilms et al . , 2003 ) . However , most were not designed to characterize and calibrate the functional activation of the visual system ( Chowdhury et al . , 2008; Eger et al . , 2005; Mandel et al . , 2013; Nadig , 1999; Schanze et al . , 2003; Walter et al . , 2005; Wong et al . , 2009 ) nor probed the functional impact of implants through systematic comparison with visual activation ( Chowdhury et al . , 2008; Cicione et al . , 2012; Eger et al . , 2005; Fransen et al . , 2014; Mandel et al . , 2013; Matteucci et al . , 2013; Schanze et al . , 2003; Wong et al . , 2009 ) . Hence , none of these studies allowed to fully address the question of understanding and controlling the functional impact of retinal prosthesis . To address this issue , we developed an acute animal model to quantitatively assess the functional impact of retinal prostheses by comparing the downstream activation of the visual system in response to visual versus artificial stimuli , using intrinsic optical imaging of the primary visual cortex ( V1 ) . To infer the hypothetic visual counterparts induced by electrical stimulation , we first established a quantitative mapping of the cartographic organization of the rat visual system . So far only Gias and colleagues ( Gias et al . , 2004 ) have provided a retinotopic description of the rat visual cortex using optical imaging . Here , we generalized this mapping to visual parameters that are the most important for prosthetic vision ( position , size and intensity ) . Using this cartographic benchmark , we demonstrate that prosthetic stimulation generates a functional activation that occurs at the expected retinotopic location and amplitude . However , the aspect ratio and the extent of the activation are significantly larger than expected . This can be explained through a simple model with two sources of diffusion: an electrical passive diffusion and the activation of axons en passant from ganglion cells . To control the extent of cortical activation , we tested various patterns of electrical stimulation and showed that only reverse engineering of the electrical pulses to inject the desired electrical stimulation ( Dupont et al . , 2013; Pham et al . , 2013 ) allowed focalizing the activation . This result provides a promising perspective that could be easily implemented for improving the visual acuity of already implanted patients . 10 . 7554/eLife . 12687 . 003Figure 1 . Experiment design . ( A ) Schematic view of the experimental setup with the camera and the visual pathway from the retina to V1 activated with normal visual stimuli ( left ) or with sub-retinal electrical stimulation using a MEA ( right ) . Retinal ganglion cells’ ( RGCs ) axons leaving the retina and projecting to V1 via the LGN are schematized ( black when non activated; in red for direct activation; in orange for activation of en passant fibers that could occur in the electrical stimulation case ) . Blue curves on the top of V1 schematize the expected spatial profile of cortical activation ( symmetric for visual stimulation and asymmetric for electrical stimulation if axons en passant are activated ) . ( B ) Clear optical access through thinned bone over V1; scale bar: 2 mm . L: lateral; M: medial; A: anterior; P: posterior . ( C ) Image of the eye fundus with the 9 electrodes ( top left ) and the 17 electrodes ( top right ) MEA; scale bars: 500 microns . Note that the use of an additional magnifying lens induced optical artifacts ( halos of light ) . Retinal OCT B-scan ( bottom ) of an implanted animal showing the MEA and intact retina ( PE: pigmetary epithelium; : shadowing of the external reference surrounding the 17 electrodes MEA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12687 . 003 First , we investigated whether retinal prosthesis stimulation generates activation at the expected retinotopic position . To answer this question , we mapped the cortical retinotopic organization of the rat visual cortex ( Gias et al . , 2004 ) by flashing white squares in a 5x4 grid of 20° side ( Figure 2A ) . For each animal , we computed polar maps for azimuth and elevation ( Figure 2B , F for 2 different animals ) . Animals were also implanted sub-retinally with a MEA and its retinal position was identified using fundus imaging ( Figure 2C , G ) , reported in the visual ( A , E ) and cortical domains ( B , D , F , H ) , respectively . In the first animal ( Figure 2 top row ) , we stimulated the whole MEA ( wMEA ) , the size of which ( 1 mm ) was in the range of the size of the stimuli used for visual retinotopic mapping ( 20° ) ( Hughes , 1979; Palagina et al . , 2009 ) . Figure 2D right shows the cortical activation generated by such electrical stimulation at ± 150 µA ( red contour ) . For a better visualization , we provided all the maps without contour in Supplementary Figure 2—figure supplement 1 with scale bars expressed both in Z-score ( left of the colorbar ) and in DI/I ( right of the colorbar ) . Note that Z-score and DI/I measures were highly and significantly correlated ( median correlation coefficient r2 = 0 . 81 between all pixels of Z-score vs . DI/I maps with the corresponding [20-50-80] percentiles being [0 . 73--0 . 81--0 . 88]% , all pval = 1 . 40e−45 , N = 9 rats , n = 225 maps ) . A 20° visual stimulus presented at this position ( Figure 2A white square ) generated activation at a similar position ( Figure 2D left , white circle ) but with an extent ( white contour ) that was much smaller than its electrical counterpart ( red dashed line ) . We then tested single electrode stimulation ( SE , see Materials and methods ) . Figure 2E–H shows an example in another animal for two individual electrodes each stimulated at ± 200 µA . The evoked activations for the two SE ( Figure 2H middle and right maps , cyan and purple ) were about the size of the 20° visual activation ( left map , white ) and their positions within the activation generated by the corresponding visual stimulus covering their retinal positions ( Figure 2E ) . To generalize these observations , we computed the retino-cortical magnification factors by measuring the cortical distance between the centers of activation ( white circles ) elicited by visual stimuli displayed at different positions . Across 20 animals , the cortical distance between activations was plotted against the distance separating the visual stimuli ( Figure 2I ) . We observed a linear increase of cortical distance with visual distance , at a rate of 22 µm/° ( in accordance with the literature ) ( Gias et al . , 2004 ) . For all activations ( visual and electrical ) , we used this value to estimate the error ( in mm and equivalent degree of visual angle , °eq ) between the position of the activity’s center-of-mass and its expected position within the retinotopic map . Visually evoked activations provided an approximation of the inherent variability of that measure ( Figure 2J , gray dots ) . On average , the cortical error was 0 . 32 ± 0 . 22 mm for visual stimuli , 72% of the data points falling within the retinotopic representation of the 20° stimulus ( <0 . 44 mm ) . The same measure was applied to electrically evoked activations for SE and wMEA stimulations . For SE , the estimation of the position of the evoked activity was quite accurate ( error of 0 . 46 ± 0 . 27 mm ) , 55% of evoked responses falling within the retinotopic expected position . Note that these error values are actually over-estimated since in practice , the implant was not obligatory located at the exact same retinal position that corresponds to the visual stimulus . Finally , wMEA activations yielded responses with more variable and less accurate position than SE with an average error of 0 . 68 ± 0 . 47 mm , 38% of the evoked positions being within the expected representation . 10 . 7554/eLife . 12687 . 004Figure 2 . Position . The visual ( A&E ) , retinal ( C&G ) and cortical ( B , D&F , H ) expected position and size of the MEA are compared to their corresponding visual stimuli . ( A ) Schematic view of the visual field showing the 20 positions ( grid ) of the visual stimuli used for retinotopic mapping . Optic disk: asterix; MEA: colored circles; nearest visual stimulus: white square ( stands for all Figures ) . ( B ) V1 retinotopic polar map for azimuth ( left ) and elevation ( right ) . Color hue and brightness code respectively for the retinotopic position and the strength of the response . Scale bar: 2 mm . ( C ) Image of the eye fundus with the implant . ( D ) Extent and center-of-mass ( contour and circle respectively ) of V1 activations generated by visual ( white , right map ) and stimulation at ± 150 µA wMEA ( red , left map ) . The activation amplitude depicted in the colorbar is expressed both in Z-score and in DI/I ( see Materials and methods ) . The solid line indicates the activation contour of the corresponding map and the superimposed dashed line corresponds to the compared condition; scale bar: 2 mm . ( E-H ) Same as in ( A-D ) in a different animal for 2 SE stimulations ( cyan and purple; dashed circle: MEA position ) at ± 200 µA; see Figure 6A for an example at low intensity in the same animal . ( I ) Retino-cortical magnification factor for azimuth computed over 20 animals ( n = 177 displacements ) . Boxplots represent the median and interquartile range; whiskers represent ± 2 . 7σ or 99 . 3 coverage if data are normally distributed ( any points outside are considered as outliers ) . ( J ) Retinotopic based positional cortical error for visual and electrical activations ( right: mm; left: equivalent degrees of visual angle ) . Black dots correspond to visual counterparts of electrical activations . One-sided two-sample Wilcoxon rank sum test for paired data: pSE vs . wMEA=0 . 056 , nwMEA = 21 , nSE = 82 , NwMEA = 6 , NSE = 7 . Wilcoxon rank sum test for paired data: p=0 . 025 , n = 13 , N = 4 and **p=5 . 35 10–5 , n = 80 , N = 6 ( p<0 . 05 , p<0 . 01 , p<0 . 001 , n = number of sample , N = number of rats ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12687 . 00410 . 7554/eLife . 12687 . 005Figure 2—figure supplement 1 . Raw maps . Raw z-score maps without overlaid contours corresponding to a sample of figures presented in the manuscript ( see subtitles ) . Colorbars are expressed both in z-score as well as in DI/I . DOI: http://dx . doi . org/10 . 7554/eLife . 12687 . 005 In the previous examples , we observed that the electrically-evoked activations are larger than expected ( Figure 2D , H ) . We thus quantified this effect by systematically comparing the extents of the visually and artificially evoked activations ( Figure 3 ) . As shown in the example of Figure 3A , the extent of cortical activation linearly increased with visual stimulus size with an average slope of 81 µm/° when estimated at population level ( Figure 3B , N = 7 ) . In comparison , the active surface diameters of the SE and wMEA were of 0 . 05 and 0 . 65 mm ( see Materials and methods ) equivalent to 1 and 11° of visual angle ( Hughes , 1979 ) , respectively . Although the visual stimuli covered a larger region of the retina ( 20° , Figures 2D and 3C top right ) , we observed a systematic and highly significant increase of the cortical activation extent induced by wMEA stimulation in 94% of the cases ( 17/18 conditions , 10 animals; see Materials and methods and Figure 3D legend for details on statistical procedures ) , with an average increase of 2 . 9 ± 2 . 01 times the activation extent of the visual stimulation ( Figure 3D , thin gray lines ) . Please note that for this analysis , all electrical stimulations were systematically paired with the closest visual stimulus used for retinotopic mapping ( Figure 3C , bottom maps ) . In comparison , SE stimulation led to an average activation size of 0 . 9 ± 0 . 5 times their 20° visual counterparts ( Figure 3D ) . Using the visual-size tuning function ( Figure 3B ) , we extracted the equivalent visual stimulus sizes that would have evoked such cortical activations . These values reached on average 24 . 4 ± 10 . 6 and 32 . 2 ± 11 . 5° instead of 1 and 11° for the SE and wMEA configurations , respectively . Electrical stimulation hence led to extremely large diffusion of the activation , comparatively larger for SE than wMEA . Such differences could be explained by several factors including ( i ) the extent of wMEA activation being sometimes underestimated because of the limit of the imaged region of interest and ( ii ) the close proximity of the annular counter-electrode ( reference ) in wMEA constraining lateral diffusion of the current ( Cicione et al . , 2012; Pham et al . , 2013; Wong et al . , 2009 ) . Our results unambiguously demonstrate that prosthetic retinal stimulation generates too large activation to obtain the necessary independence between electrodes; an inter-electrode spacing of at least 0 . 8 to 1 . 2 mm on the retina , at low- and high-intensity respectively , would be required to yield non-overlapping cortical activations . 10 . 7554/eLife . 12687 . 006Figure 3 . Size . ( A ) V1 activation generated by visual stimuli of increasing size ( value indicated above the maps ) . Center-of-mass: white circle; extent: white contour; equivalent ellipse orientation: white cross; scale bar: 2 mm . ( B ) Extent of cortical activation as a function of visual stimulus size pooled over 7 rats ( linear fit: dashed black line ) . ( C ) Extent of cortical activations generated in 2 animals by SE ( blue ) and wMEA ( red ) stimulation ( top maps ) at a high current intensity ( ± 200 and 150 µA respectively ) and their corresponding 20° visual stimulus ( white , bottom maps ) . Centers of mass of the activation: colored circles; scale bar: 2 mm . ( D ) Size of cortical activation generated by visual ( gray , N = 20 rats ) ; SE ( blue , N = 8 rats ) and wMEA ( red , N = 10 rats ) stimulation . Wilcoxon rank sum test for paired data: p=0 . 0019 ( n = 54 , N = 8 ) , p=1 . 83 10–4 ( n = 15 , N = 10 ) . An alternative ordinate of equivalent visual counterpart is given on the right . Gray thin lines link paired electrical stimulation to visual activation . Solid horizontal black line indicates the size of a 20° visual stimulus estimated from the fit in B . Note that we could not reveal any effect of the electrode-to-counter-electrode distance on activation size in the various SE configurations ( not shown ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12687 . 006 Next we compared the shape of cortical activations to visual and prosthetic stimulations , because it may help understanding the origin of the spread . When compared to visual stimulus ( white ) , we generally observed elongated cortical activation for SE ( Figure 4A blue ) but not for wMEA ( red ) stimulation . In order to quantify this elongation , we computed the aspect ratio ( AR ) of activation contours at the level of the population ( Figure 4B , see Materials and methods ) . Using pairwise visual-prosthetic comparison , we found a highly significant ( see Figure 4B legend ) increase in the aspect ratio of the evoked activations for SE condition compared to their visual counterparts ( 1 . 66 ± 0 . 43 vs . 1 . 35 ± 0 . 19 ) but none for wMEA . These elongations could be caused by the recruitment of ganglion cells’ axons en passant leading to oriented anisotropic cortical activations ( Figure 1A right ) . However , why would SE activation yield stronger AR than wMEA ? One explanation is that the ganglion cell neurons with axons that will be activated en passant , all converging radially towards the optic disk , are all located upstream to the electrode within a 'shadow cone' with a top angle that will ( i ) decrease with the distance between the implant and the optic disk location and ( ii ) increase with the size of the stimulated retinal surface ( Figure 4—figure supplement 1 ) . In a simple functional model ( see Materials and methods ) , we combined the predictions of what should be the shape and size of retinal activations due to local and en passant activation . Isotropic 'direct' activations were modeled as a Gaussian activation around the site of stimulation ( Figure 4—figure supplement 1 left column ) for different electrode sizes ( Figure 4—figure supplement 1 , rows ) . Anisotropic en passant recruitment of ganglion cell axons was modeled as a shadow cone activation , i . e . all peripheral ganglion somata whose axons have been activated by the electrical stimulation ( Figure 4—figure supplement 1 , middle column , Figure 4—figure supplement 1A for details in the model , see Materials and methods ) . Via a weighted sum , we combined these two activations ( right column and Figure 4C insets ) with different ratios ( Figure 4—figure supplement 1B: 0 . 5 & C: 1 ) . From these predicted activations , we extracted similar parameters ( size , position , elongation ) as we did from our cortical recordings . Since our interest is to predict the effect of axon-en-passant activation on radial elongation , we used the following convention for the AR: the numerator is the length of activation along the radial axis and the denominator is the length of activation along a tangential axis ( perpendicular to the radial axis ) . This model predicted that the resulting retinal activation should indeed be more elongated along radial axis for smaller cone angle ( Figure 4C and Figure 4—figure supplement 1 , AR > 1 ) , this effect being stronger when we increase the relative contribution of en passant activation . The model further predicts that the retinal activation should become more elongated along a tangential axis for very large cone angle ( AR < 1 ) . Please note that these predictions are left unchanged if we apply a retino-cortical transformation to the simulated activations ( see Figure 4—figure supplement 2C ) . To test for these predictions , we plotted the aspect-ratio of the activation across animals and experimental conditions as a function of the retinal 'shadow cone' angle formed by the stimulated surface ( Figure 4D ) . Please note that , to account for any potential deformation of the evoked activity due to retino-cortical magnification factor ( see Figure 4—figure supplement 2A ) or physiological noise , we normalized all electrically-induced AR to their corresponding visual AR . In this plot we observed a similar decrease of the aspect ratio with the cone angle , confirming our predictions ( Figure 4C ) . Our results thus suggest that the difference we observed between wMEA and SE is explained if we make the hypothesis that part of the functional activation of the visual system comes from of axons en passant recruitment . Note that our model suggests that the isotropy observed for wMEA activations simply results from the geometrical arrangement between the electrode size and the shadow cone . We thus conclude that wMEA activations are still suffering from contamination by the activation of fibers en passant . Hence SE simulation remains the best configuration to achieve the highest performance in spatial resolution . According to our simple model , the relative contribution of axons en passant to the global prosthetic activation should be of the same order of magnitude as direct activation ( alpha = 1 , see Materials and methods: model of retinal activation ) . 10 . 7554/eLife . 12687 . 007Figure 4 . Shape . ( A ) Shape of cortical activations generated in 2 animals by SE ( blue ) and wMEA ( red ) stimulation at high current intensity ( top ) and their corresponding 20° visual stimulus ( white , bottom ) . ( B ) Aspect ratio ( AR ) of cortical activations ( Wilcoxon rank sum test for paired data , p=1 . 06 10–4 , n = 44 , N = 7 ) . ( C ) Predictions of the elongation of electrical activations as a function of the contribution of axons en passant and the distance to the optic disk . Insets correspond to a model of retinal activation due to direct isotropic activation plus passive electrical diffusion and anisotropic activation due to axons en passant recruitment for 3 different electrode sizes . The brightness codes the strength of the response . Center of the white dashed target: position of the optic disk; black circle: position and size of the MEA active surface; gray lines: 'shadow cone' angle sustained by the MEA active surface respective to the optic disk location; colored contour: size and shape of the global retinal activation for an axons en passant contribution of 1 ( alpha , see Materials and methods ) . ( D ) Elongation of electrical activations relative to their corresponding visual activations ( AR electrical/AR visual ) as a function of the 'shadow cone' angle . ( E ) Cortical radial organization of prosthetic activations . Solid segments: orientation of cortical activations; dashed segments: optimal radial orientation towards the black disk; segment crossing; geometrical center; red dot: center-of-mass of cortical activations; Dark disk: cortical position that optimized radial organization; gray disk: median position of the optic disk . The blue lines connect the center-of-mass to the geometrical center of activations . Scale bar: 0 . 5 mm . Inset: distribution of median angular deviation expected by chance compare to our observation: blue segment . ( F ) Top: centered and reoriented deviations of the center-of-mass ( blue disks ) to the geometrical center ( center of the representation ) , horizontal dashed axis corresponds to the orientation of the radial organization . Bottom: averaged , centered and reoriented SE ( with AR > and < than 1 . 6 , left and middle respectively ) and wMEA maps ( right ) . White circle: center-of-mass; additional dashed contour corresponds to a Z-score of −4 . 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 12687 . 00710 . 7554/eLife . 12687 . 008Figure 4—figure supplement 1 . Model of retinal anisotropic activation . Model of retinal anisotropic activation due to axons en passant for 3 different electrode sizes ( rows in B&C ) . ( A ) Modeling steps used to compute the activation of axons en passant ( see Materials and methods ) . The brightness codes the strength of the response , all representations are scaled between 0 and 1 . ( B ) Direct activation with passive electrical diffusion ( left column ) . Center of the white dashed lines: position of the optic disk; black circles: position and size of the MEA active surface . Anisotropic activation due to axons en passant recruitment ( middle column ) . Size and shape of the global retinal activation ( colored contours , right column ) for an axons en passant contribution of 0 . 5 ( alpha , see Materials and methods ) . ( C ) same as in A for an axons en passant contribution of 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 12687 . 00810 . 7554/eLife . 12687 . 009Figure 4—figure supplement 2 . Model of retino-cortical transformation . To check whether retino-cortical transformation can introduce further bias to our predictions ( Figure 4—figure supplement 1 ) , we implemented a retino-cortical transformation based on the magnification measured in our data ( A ) . This model simply generates a transformation using the formula RCM = 1 / ( aR+b ) , where RCM is the retino-cortical magnification factor ( mm/deg ) , R the retinal eccentricity ( deg ) , a and b constants ( along the horizontal meridian : a = 0 . 7; b = 30 & a = 0 . 4 , b = 40 for the vertical dimension ) . In the Figure A , the resulting transformation is shown for a retinal pattern with circles of different diameter and eccentricity . ( B ) We show for 2 cone angles ( blue and red ) and ratio of axon-en-passant ( same as Figure 4—figure supplement 1 ) , what such transformation does . ( C ) From these maps , we calculated the cortical activation aspect ratio , normalized to the shape of activation without axon-en-passant ( ratio = 0 not shown ) , plotted with the same convention as in Figure 4C . DOI: http://dx . doi . org/10 . 7554/eLife . 12687 . 009 Thus , the predictions that arise from this observation are: for small cone angle ( i ) the elongation of the cortical activation should be radially organized towards the representation of the optic disk and ( ii ) the activation should be anisotropic , attracted by the radial elongation towards more eccentric positions ( i . e . away from the optic disk representation ) ; lastly ( iii ) for large cone angle , cortical activation should be more elongated along a tangential axis . To test for these predictions , we pooled together all activation profiles that were sufficiently elongated , i . e . with an AR above 1 . 6 , and plotted them in cortical space ( Figure 4E ) . In this figure , each solid segment corresponds to the orientation of a cortical activation . The dark disk is the cortical position that optimized the radial arrangement of the observed orientations ( see Materials and methods ) . Observed orientations of cortical activation deviated from an optimal radial organization ( dashed lines ) to this cortical position by only 13 . 3° ( median value across N = 20 activations ) . Importantly , this cortical position is close to , and in the same direction as , the median position of the optic disk ( gray disk ) . This latter one was estimated from 16 experiments whenever its mapping was possible . However , we could not use it systematically because of imprecision in its estimation ( see Materials and methods ) . We then checked whether this result could occur by chance . For that purpose , we generated 1000 random distributions of orientations at the observed positions and , for each , looked for the cortical position that optimized radial arrangement ( as described above ) . For this position , we computed the median angular deviation between the random orientations and the optimal radial arrangement to this position . This led to a Gaussian distribution of median angular deviation expected by chance in our particular configuration , with a mean of 23 . 6 ± 4 . 9° ( Figure 4E inset ) . Our observation of 13 . 3° ( red line ) is significantly smaller than what would be expected by chance ( p=0 . 02 , see Materials and methods ) . To test for the second prediction of anisotropic activation that our model raised , we compared the center-of-mass of the activation ( Figure 4E red dot ) to the geometric center ( dashed and solid line crossing ) . Indeed if the activity is attracted away from the blind-spot , the center-of-mass must be delocalized from the geometrical center , opposite to the position of the blind-spot . We plotted the position of the center-of-mass ( red points in Figure 4E ) , and indeed observed that it was pushed away from the optic disk representation . Figure 4F present a centered , reoriented and zoomed view on all deviations of the center-of-mass ( red disk ) to the geometrical center ( center of the representation ) within the same reference frame ( horizontal dashed axis being the radial organization ) . Center-of-masses all deviated away from the optic disk with an averaged angular deviation of −168° ( opposite deviation being 180° ) . To better illustrate this deviation , we averaged all the maps under consideration in this analysis , after realignment along the radial axis ( the axis linking the activation to the BS , here the BS is to the right ) and centering ( Figure 4F bottom left ) . We can see from this averaged map that there is indeed a radially elongated and anisotropic activation opposite to the representation of the blind spot ( dotted contour represent activation higher than 4 . 5 z-score ) . For comparison , we made similar averaging on all the other maps in response to SE ( with an aspect-ratio < 1 . 6 ) and wMEA . For SE activation with low AR , the result shows an activation that is more isotropic . For MEA activation , the averaged map shows , as predicted , a slight elongation along the tangential axis . Hence , in our experimental observations , the radial arrangement and deviation of the center-of-mass of the activation behave as expected from an activation of en-passant axons in the retina . Encoding different levels of luminance is an important aspect of prosthetic vision . We thus investigated the effect of stimulus intensity . Increasing luminance non-linearly increased the cortical activation size and amplitude ( Figure 5A ) . Across the population ( N = 8 ) , we fitted the response amplitude ( Figure 5C ) with a Naka-Rushton function ( Naka and Rushton , 1966 ) for the whole population ( black dotted line ) , as well as for each animal individually ( thin gray lines ) . The luminance of semi-saturation was 9 . 6 ± 2 . 9 cd/m2 . We then compared these results with the intensity of electrical stimulation ( Figure 5D–E ) that similarly modulated the cortical activations ( Figure 5B ) . Over 35 animals ( wMEA: 10 , SE: 25 ) , we observed a gradual increase of response amplitude for both wMEA and SE electrical stimulation , well captured by Naka-Rushton fits ( Figure 5D–E , thick dotted lines ) . The intensity of semi-saturation ( Figure 5F ) was significantly lower for SE than for wMEA ( 20 . 29 ± 3 . 25 µA vs . 52 . 66 ± 17 . 35 µA , see legend for details on statistical procedure ) . Figure 5G compares the exponent of the individual fits performed on the amplitude parameter for visual , SE and wMEA stimulation . Steep transitions were observed for wMEA ( n = 5 . 58 ± 3 . 54 ) and for visual luminance ( n = 6 . 05 ± 3 . 01 ) whereas more gradual transitions were observed for SE ( n = 1 . 34 ± 0 . 6 ) which significantly differed from wMEA and visual stimuli ( see Figure 5G legend ) . As a consequence , the operating range ( see Materials and methods ) of the intensity response function in visual stimulation was on average 9 . 2 ± 4 . 8 cd/m2 , 82 . 5 ± 88 . 4 µA for wMEA and much broader for SE ( 182 . 6 ± 151 . 4 µA ) . This is further reflected in Figure 5H plotting the hypothetic equivalent luminance of electrical activations . We estimated the equivalent visual luminance that corresponds to the response amplitude evoked by electrical stimulation using the Naka-Rushton fit ( Figure 5C ) . The functional operating range induced by current intensity manipulation allowed generating an artificial activation with an equivalent luminance that could theoretically be in the range of 2–50 cd/m2 . 10 . 7554/eLife . 12687 . 010Figure 5 . Intensity . ( A ) V1 activation generated by visual stimuli of increasing luminance ( value indicated above maps ) . Center-of-mass: white circle; extent: white contour; equivalent ellipse orientation and size of activation contour: white cross; scale bar: 2 mm . ( B ) V1 activation generated by SE ( top ) and wMEA ( bottom ) in 2 different animals at high current intensity . Center-of-mass of visual and electrical activations are indicated with white and colored circles respectively and extent of electrical activation in colored contours; scale bar: 2 mm . ( C ) Amplitude of cortical activation as a function of visual stimulus luminance computed over 8 rats ( population fit: dashed black line; individual fits: gray thin lines ) . ( D ) Amplitude of cortical activations generated by wMEA stimulation at high current intensity . Population fit: dark red dashed line; individual fits: orange thin lines ( N = 10 ) . ( E ) Amplitude of cortical activations generated by SE stimulation at high current intensity . Population fit: dark blue dashed line ( N = 25 ) . Note that individual fits ( cyan thin lines ) were only performed on N=6 animals ( tested with 7 different levels of intensity ) and not on the others ( N=19 , tested with only 2 different levels: 50 and 200 µA ) . ( F ) Constant of semi-saturation ( c50 ) for visual ( gray in cd/m2 ) and electrical activations ( wMEA: red and SE: blue , in µA ) . Two-sample Wilcoxon rank sum test: p=0 . 0017 , nSE = NSE = 6 , nwMEA = NwMEA = 10 . ( G ) Exponent of the naka-rushton fits for visual ( gray ) and electrical activations ( wMEA: red and SE: blue ) . Two-sample Wilcoxon rank sum test: pSE vs . visual=6 . 66 10–4 ( nSE = NSE = 6 , nvisual = Nvisual = 8 ) ; pwMEA vs . SE=0 . 0075 ( nSE = NSE = 6 , nwMEA = NwMEA = 10 ) ; pwMEA vs . visual=0 . 896 ( nwMEA = NwMEA = 10 , nvisual = Nvisual = 8 ) . ( H ) Amplitude-based correspondence for all electrical activations to their visual counterpart ( wMEA: red and SE: blue ) as a function of current intensity level ( in equivalent cd/m2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12687 . 010 The previous results show that artificial activations of the visual system with sub-retinal implants generate stimuli of appropriate position and scalable intensity , but whose size is on average 2 . 4 and 5 . 8 times larger than expected for wMEA and SE , respectively . This clearly impairs the functional efficiency of the implants . We therefore took advantage of our experimental design to seek solutions for controlling the spatial extent of the functional activation . Several candidate parameters have been investigated for SE stimulation ( Figures 6 and 7 , respectively blue vs . green colors ) : polarity ( cathodic vs . anodic pulse first , Figure 6A–B ) , symmetry vs . asymmetry of the biphasic square pulse ( McIntyre and Grill , 2000 ) ( Figure 6C–D ) and regular pulses vs . electrical impedance spectroscopy-based ( IS ) adaptation ( Dupont et al . , 2013 ) ( Figure 7 ) . In the example shown and over the population ( Figure 6A–B ) , our results showed that the polarity of symmetric biphasic pulses did not significantly influence the extent of the cortical activation for the two intensity levels tested . However , the polarity was found to have a significant effect in asymmetric pulses delivered at 200 µA ( Figure 6C–D ) . The combination of asymmetric with anodic-first stimulation ( Figure 6C top right ) generated smaller activation of 1 . 66 ± 0 . 71 mm corresponding to 17 . 12 ± 8 . 11°eq ( N = 7 ) , smaller than the three other combinations . However , this reduction was not systematic ( 70% of the cases compared to the symmetrical pulse ) and small ( 20 . 43% ± 54 . 32 , the corresponding [20-50-80] percentiles being [34 . 78–32 . 93–63 . 89]% ) . Please note that this negative result is the outcome of thorough manipulation of key parameters of the electrical stimulation and highlight the high variability of the changes induced . Our best result at this stage is obtained by combining the polarity and the asymmetry parameters . 10 . 7554/eLife . 12687 . 011Figure 6 . Focalization . ( A ) SE activations generated by square pulses of different polarity ( anodic first: blue; cathodic first: green ) at 2 intensity levels ( 50 and 200 µA ) . Scale bar: 2 mm . ( B ) Effect of polarity and intensity on cortical extent for SE ( anodic first: blue; cathodic first: green , N = 10 ) . One-sided Wilcoxon rank sum test for paired data , p=0 . 3802 ( n = 18 , N = 10 ) and 0 . 0615 ( n = 11 , N = 10 ) for 50 and 200 µA respectively . ( C ) Individual example of asymmetrical ( green ) and symmetrical ( blue ) square pulses , same animal as in ( A ) . ( D ) Effect of square pulse asymmetry ( green ) for two polarities on cortical extent for SE ( N = 10 ) . One-sided Wilcoxon rank sum test for paired data: *pCathoSym vs . CathoAsy=0 . 0052 ( n = 13 , N = 10 ) ; pAnoAsy vs . CathoAsy=0 . 0469 ( n = 6 , N = 10 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12687 . 01110 . 7554/eLife . 12687 . 012Figure 7 . Impedance spectroscopy adaptation . ( A ) Example of electrode-tissue interface filtering on pulse shape ( at 500 mVpp ) with ( green ) and without ( blue ) IS based adaptation . Top: injected pulse; bottom: pulse shape reaching the tissue . ( B ) Individual example of IS adaptation ( green , top maps ) and reference ( blue , bottom maps ) for 2 voltage levels; scale bar: 2 mm . Note that for this protocol only , we switched from current to voltage injection . ( C ) Effect of IS adapted pulses on cortical extent for SE at 2 intensity levels ( 5 and 12 Vpp in voltage injection mode ) . IS adapted pulses: green; reference: blue , N = 4 . One-sided Wilcoxon rank sum test for paired data: p=0 . 0117 ( n = 8 , N = 4 ) , p=0 . 0078 ( n = 7 , N = 4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12687 . 01210 . 7554/eLife . 12687 . 013Figure 7—figure supplement 1 . Principle of IS adaptation . ( A ) The stimuli processing platform’s architecture integrates an Impedance Spectroscopy ( IS ) recording module; an Identification Algorithm module fitting IS data to an Electronic Equivalent Circuit ( EEC ) ; a Transfer Function Computing module and an Adapted Stimuli Shaping module computing adapted stimuli emitted by the Stimuli Generator . ( B ) Linear EEC used for this study . I: current; V: voltage; CPE: constant phase element ( capacitance of the electrical double layer formed at the interface between a metallic electrode and an ionic solution ) ; C: retinal membrane capacitance; R: resistance; bulk: subretinal fluid . ( C ) Mean duration ( top ) and frequency content ( bottom ) of reference ( left ) and IS adapted ( right ) pulse shapes recorded in the tissue ( reference pulses: cathodic first symmetrical biphasic squared pulses ) . Duration: full width at half maximum ( FWHM ) of each recorded transients for the reference pulses ( corresponding to the 3 abrupt pulse transitions ) and of the 2 sustained phases for the adapted ones . Frequency: power of the Fourier transformed computed on the pulses temporal profiles . DOI: http://dx . doi . org/10 . 7554/eLife . 12687 . 013 One general issue encountered when injecting current or voltage in a tissue , and often ignored , is that the desired pattern to be injected can be strongly distorted by the non-ohmic properties of the electrode-tissue interface ( Geddes , 1997; Pham et al . , 2013 ) . In a previous study ( Pham et al . , 2013 ) , we indeed showed that the impedance phase and magnitude of the electrode-retina interface was found to be highly capacitive , yielding strong distortion of the applied stimulus ( Figure 7A , left bottom ) and lateral diffusion of the injected pattern . Our rationale was therefore to use the characterization of the physical properties of the electrode-tissue interface to calibrate through reverse engineering methods what needs to be injected to obtain the desired pattern in the tissue . To do so , we characterized the impedance spectrogram of selected SE of each implantation ( Pham et al . , 2013 ) . Using an equivalent electronic circuit accounting for the observed non-ohmic behavior , we simulated ( Figure 7—figure supplement 1B , see Materials and methods for details on the procedure ) the adapted pattern of voltage injection ( Figure 7A , top right ) that will generate the desired pattern shape in the implanted tissue ( Dupont et al . , 2013 ) ( Figure 7A , right bottom ) . Note that , for this protocol only , we switched from current to voltage injection and that the low voltage level used here was of the same order of magnitude as the high current level used in the previous protocols . For 100% and 88% of the low and high voltage level cases respectively ( 5 Vpp and 12 Vpp , Vpp: Volts peak-to-peak ) , the adapted pattern injected through SE decreased the extent of the activation ( Figure 7B top maps & 7C ) . This significant decrease ( see Figure 7C legend ) varied on average from 36 ± 32% to 20 . 7 ± 29 . 2% at low and high voltage , respectively ( the corresponding [20-50-80] percentiles were [11 . 78–27 . 50–59 . 05] and [1 . 89–6 . 56–39 . 67]% ) . Importantly , at high voltage it was not accompanied by a significant change in response amplitude ( Wilcoxon rank sum test: p=0 . 546 , n = 8 , N = 4 ) , suggesting that this reduction of the extent was not simply explained by a change in activation strength level , allowing for an independent control of the extent ( size ) and the amplitude ( intensity ) of the evoked activity . Prosthetic activation yielded to cortical activations that were consistent with the V1 retinotopic organization . The degree of positional precision , higher in SE compared to wMEA , was of the same order of magnitude as the 20° visual stimuli . However , this measure was quite variable , as expected given the poor precision of the murine visual system ( Euler and Wässle , 1995 ) . Here we significantly extended previous investigations that have first used mesoscopic optical recordings to image the cortical features of the prosthetic activation but on a limited number of animals ( Walter et al . , 2005 ) and with no quantitative statistical analysis ( Eckhorn et al . , 2006; Walter et al . , 2005 ) . Our approach allowed providing a clear and quantitative description of the prosthetic activation . We have also shown that increasing the level of current intensity increases the amplitude of the evoked cortical response with a sigmoidal profile similar to the one observed with visual luminance . Such non-linear profiles are typically well characterized by their threshold and slope ( Naka and Rushton , 1966 ) . The threshold differences observed in wMEA and SE stimulation configurations can trivially result from difference in active electrode surface . However , our experiments unveiled that SE offers a larger operating range ( 182 vs . 82 µA for wMEA ) to cover a theoretical modulation range of about 50 cd/m2 . This result demonstrates that small electrode sizes will be more appropriate to make fine manipulation of stimulus intensity . Importantly , our results also revealed that the increasing stimulus intensity increases both the response size and the response amplitude . This observation is crucial to consider since it will seriously challenge independent manipulation of those important parameters in prosthetic vision . Our results demonstrate that the size elicited with standard parameters is approximately 2 . 4 to 5 . 8 times larger than expected for wMEA and SE , respectively . This increase can be partly explained by passive electrical diffusion in the implant-retina interface , as we previously suggest ( Pham et al . , 2013 ) , but also by the activation of en passant ganglion axons . The spatial resolution of the prosthetic retinal activation measured in our experiments , about a millimeter of retinal space , is actually comparable to what has been reported in other animals ( as estimated from Eger et al . , 2005; Schanze et al . , 2003 ) as well as in human studies ( as estimated from Ahuja and Behrend , 2013; Humayun et al . , 2012; Stingl et al . , 2013 ) and extends beyond the retinal point spread function ( Stett et al . , 2000 , 2007 ) . Are these results accentuated by the anesthetized state of the animal ? Indeed , recent experiments showed that anesthetized mice have less surround inhibition than when awake ( Vaiceliunaite et al . , 2013 ) . However , Vizuete et al ( Vizuete et al . , 2012 ) have shown that anesthesia affects not simply the inhibition but more probably the balance between excitation and inhibition ( see also Chemla and Chavane ( 2016 ) ) . Hence , predicting how activation size of the population activity will be affected by anesthesia is not straightforward . Importantly , our study was devised to make a systematic comparison between conditions done under similar level of anesthesia ( visual vs . electrical ) and the polarity of the effect ( the size increasing or decreasing ) is not expected to reverse by changing the animal state . For SE stimulation , large activations were also accompanied by an increase in the aspect ratio of the spatial profile of the cortical activation suggesting an asymmetric recruitment of the retinal tissue , for instance through activation of ganglion axons en passant ( Nowak and Bullier , 1998a , 1998b ) . However , this increase in aspect ratio was not observed with wMEA . Using a simple model , we show that such difference in aspect ratio between stimulation conditions is actually to be expected because of differences in the spatial distribution of activation of axons en passant . Our model also allowed predicting that , to account for the observed spatial profiles , axons en passant should contribute to the same proportion than direct activation of the retina . In accordance with this prediction , we found that high aspect ratio activations distribute radially around the cortical representation of the blind spot and are asymmetrical , with center-of-masses displaced away from the blind spot representation . In contrast , Fransen et al . ( Fransen et al . , 2014 ) discard the possibility of direct RGCs activation through near infrared laser stimulation of photovoltaic subretinal array because of the absence of responses in the superior colliculus using inner retina synaptic blockers . However , several factors , including the low efficiency of photovoltaic transduction and the level of injected currents used in our study , might explain the differential recruitment of RGCs cell bodies and axons observed here using V1 population recordings . 10 . 7554/eLife . 12687 . 014Figure 8 . Differential effect of SE patterns . Averaged cortical size and aspect ratio ( ± sem: black error bars ) elicited by visual stimuli , by the wMEA and by the different SE stimulation patterns delivered at high intensity levels . Gray circles: 10 and 20 deg visual stimuli ( n = 7 and 75 , respectively ) ; cyan and red squares: high intensity symmetrical stimulation patterns for SE ( n = 57 ) and wMEA ( n = 10 ) respectively ( both polarities ) ; green and cyan circles: respectively IS adapted and non-adapted pulses delivered at 5 Vpp ( n = 7 , same order of magnitude as the other pulses delivered at high current intensity ) ; green and cyan up triangles: respectively asymmetrical ( n = 7 ) and symmetrical ( n = 21 ) anodic first pulses delivered at ± 200 µA; green and cyan down triangles: respectively asymmetrical ( n = 8 ) and symmetrical ( n = 24 ) cathodic first pulses delivered at ± 200 µA . Lines link comparable conditions . The different SE patterns evoke activations that exhibit strong correlation between size and aspect ratio ( gray shaded area ) , except for IS adapted stimuli which converge towards visual responses . DOI: http://dx . doi . org/10 . 7554/eLife . 12687 . 014 Since most implantations occurred at more or less the same retinal eccentricity , our model predicts that activation size and aspect-ratio should be inversely related: the smaller the activation size ( equivalent to a smaller cone angle for a fixed eccentricity ) , the larger the expected aspect ratio . Figure 8 indeed shows that the averaged sizes observed for standard electrical stimulations are systematically inversely correlated to their aspect ratio . Remarkably , all regular manipulations of the stimulation pattern moved the activation spatial profile along that inverse relationship . In contrast , visual activation does not display such correlation , the aspect-ratio being small and independent of activation size ( Figure 8 , gray dots ) . This relationship could therefore be taken as a diagnosis of the way electrical stimulation activates the retinal circuitry with lateral isotropic diffusion and anisotropic axons en passant activation . To mimic a functional visual activation , it is therefore needed to decrease both the extent and the aspect ratio of the prosthetic activation . Charge balanced biphasic pulses , delivered in voltage or current mode injection , are of common use in human ( Ahuja and Behrend , 2013; Fujikado et al . , 2007; Humayun et al . , 2012; Klauke et al . , 2011; Nanduri et al . , 2012; Rizzo , 2003; Shepherd et al . , 2013; Stingl et al . , 2013; Wilke et al . , 2011; Zrenner , 2013; Zrenner et al . , 2010 ) or animal ( Chowdhury et al . , 2008; Cicione et al . , 2012; Eckhorn et al . , 2006; Eger et al . , 2005; Elfar et al . , 2009; Matteucci et al . , 2013; Nadig , 1999; Schanze et al . , 2003; Walter et al . , 2005; Wong et al . , 2009 ) studies of prosthetic vision . In the literature , we observed a large consensus concerning the use of cathodic pulse first for epiretinal stimulation ( Ahuja and Behrend , 2013; Eckhorn et al . , 2006; Eger et al . , 2005; Elfar et al . , 2009; Fried et al . , 2006; Nanduri et al . , 2012; Schanze et al . , 2003; Walter et al . , 2005 ) although not for subretinal , suprachoroidal and extraocular approaches ( Chowdhury et al . , 2008; Cicione et al . , 2012; Eckhorn et al . , 2006; Fujikado et al . , 2007; Matteucci et al . , 2013; Stingl et al . , 2013; Wilke et al . , 2011; Wong et al . , 2009 ) . It has been shown in vitro ( Jensen et al . , 2005 ) that the use of single cathodic pulse lowers the threshold of retinal ganglion cells ( RGCs ) , lowers the latency of inner retina mediated RGCs response and targets more specifically RGCs cell bodies when compared to anodal pulses . As far as we know , only one study systematically examined the effect the polarity of charge balanced biphasic square pulses ( Chowdhury et al . , 2008 ) . Using an extraocular device , Chowdhury et al . did not find any significant effect of the pulse polarity on the threshold of V1 electrically evoked potentials . Similarly , we did not find any effect of the polarity on the size of cortical activations . Altogether , anodic asymmetric pulses were the best among classical stimulations to restrict cortical activations , although not systematically . This result contradicts the prediction of the pioneer modeling work of McIntyre & Grill ( McIntyre and Grill , 2000 ) , which predicts more focal activation for cathodic asymmetrical pulses using a cable model . To conclude , basic manipulation of the electrical pulses did not yield a systematic reduction of the spatial extent of the activation . Furthermore , when compared to the aspect ratio ( Figure 8 ) , it turns out that all these stimulation configurations lead to the same inverse relationship suggesting that all activated the retinal network through diffusion and axons en passant . Note that other strategies , such as current steering methods , could be investigated in future in vivo studies ( Jepson et al . , 2014a ) . These authors showed in in-vitro isolated retinal preparation , that it was possible to improve the spatial resolution of the implant up to the activation of a single RGC type with a single action potential resolution ( Jepson et al . , 2013; Sekirnjak et al . , 2008 ) . To further scrutinize optimal stimulation parameters , we have also used a more 'supervised' approach by adapting the shape of the desired retinal stimulation through impedance spectroscopy to account for the filtering properties of the electrode-tissue interface ( resistive and capacitive components ) ( Dupont et al . , 2013; Pham et al . , 2013 ) . Our results show that electrical stimulation adaptation systematically and accurately decreased the activation spread while preserving their location and amplitude , when compared to their paired controls . Crucially , it was the only configuration we tested whose decrease in extent was not accompanied by an increase in aspect ratio ( Figure 8 , filled circles ) , similarly to functional visual stimulation . This strongly suggests that such stimulation reduces both the electrical diffusion extent ( although not completely ) and the axons en passant activation . What is the origin of the reduction of electrical diffusion ? This could arise from the change of the shape of the injected current , that activate less high frequencies ( Figure 7—figure supplement 1C ) that increase the spread of the electrical diffusion and the current density magnitude ( Pham et al 2013 ) . By which mechanisms axons-en-passant would be less activated ? Two mechanisms are possible . First , this could be a consequence of the reduction of the electrical diffusion . Since we are positioned subretinally , reducing the diffusion will favor activation of elements closer to the electrode ( photoreceptors , bipolar cells… ) compared to further away ( ganglion cells ) . A second mechanism would originate from the fact that the duration of each adapted pulse ( anodic or cathodic ) in the tissue is much longer ( 1 . 2 ms on average ) than the transients contained in the non-adapted pulses ( 0 . 2 ms on average ) . This can favor activation of cell bodies since axons have shorter chronaxie , generally some hundreds of microseconds , than cell bodies , generally several milliseconds ( McIntyre and Grill , 2002; Nowak and Bullier , 1998b , 1998a; Ranck , 1975; Stern et al . , 2015; Tehovnik and Slocum , 2009 , Histed et al . , 2009 ) . This effect could be further amplified by the fact that the adapted pulse inside the tissue was also biphasic , cathodic phase first and asymmetrical in amplitude . According to the work of Mc Intyre & Grill ( McIntyre and Grill , 2002 ) , these are the conditions for which we can expect the threshold for activation to increase for axons and to decrease for the soma . It is thus expected that the adapted condition will lower the contribution of fiber-en-passant , since the amplitude of the current injected in the tissue was lower in adapted than in non-adapted mode . This is therefore a very promising solution getting closer to a more natural visual activation of the retinal network . Importantly , it is noteworthy to point out that this stimulation strategy can be easily embedded in already available commercial devices using small electronic circuits . Supervised adaptation of electrical pulse based on the electrical properties of the electrode-tissue interface is therefore a promising solution for retinal , but also probably , for cochlear prosthetic stimulations and more generally for all neuro-stimulation applications . As a perspective , one may consider to combine the advantages of the different methods that have been shown , here and in other publications , to have an effect: adaptation , asymmetry of the pulse ( McIntyre and Grill , 2001 ) and the spatial arrangement of stimulating and counterelectrodes ( Jepson et al . , 2014a ) . The possible independence of the mechanisms by which these methods decrease the activation size may allow an additive gain in spatial resolution . Here we provide a clear quantitative functional description of retinal prosthetic activations by using a systematic comparison of artificial and natural activation of the rodent visual system . Using a similar approach , further investigations will have to probe the dynamical aspect of these prosthetic activation to better understand how to generate spatio-temporal activations in the visual pathway ( Elfar et al . , 2009; Fried et al . , 2006 ) that are similar to the ones observed in response to natural stimuli . A further challenge will be to drive the stimulator to generate activity closer to the retina’s natural neural code ( Nirenberg and Pandarinath , 2012 ) , which may involve taking into account higher order correlations ( Marre et al . , 2012; Marre and Botella , 2014 ) . All these strategic steps will greatly benefit from animal models such as the one used here . To conclude , our work demonstrates that pre-clinical studies are a necessary prerequisite to validate and improve the efficiency of retinal implants , equivalent to animal models that are the foundation of any drug development for human pharmacology . We thus expect that our study will pave the way for fast and significant improvements of the clinical benefits offered to implanted patients . A total of 35 Brown Norway male rats ( 1 . 5–3 months old , 230–320 g ) were anesthetized using an intraperitoneal injection of urethane . The experimental protocol was approved beforehand by the local Ethical Committee for Animal Research and all procedures complied with the French and European regulations on Animal Research ( approval n°A12/01/13 ) as well as the guidelines from the Society for Neuroscience . Instead of a complete craniotomy , the bone was thinned with a drill until a clear optical access to V1 surface was obtained . Finally , transparent silicone was applied to the remaining bone and the preparation was covered with a glass slide . We used subretinal Micro Electrode Arrays ( MEAs ) manufactured at the CEA-LETI ( Pham et al . , 2013 ) ( Grenoble , France ) . These planar MEAs of 1 and 1 . 2 mm diameter comprise respectively 9 and 17 ( 50 μm radius , 3 μm thick ) platinum ( Pt ) contacts and a large annular Pt counter-electrode ( CE ) within a polyimide flexible substrate ( Figure 1C top row ) . The electrical connection between the MEA and the stimulator ( BioMEA , CEA-LETI ) ( Charvet et al . , 2010 ) is made using an Omnetics 18-position nanominiature connector . The MEAs were implanted between the pigmentary epithelium and the outer segment photoreceptor layer ( Figure 1C bottom row ) . Eyedrops of oxybuprocaïn chlorohydrate were used to provide a local anesthesia and the pupil was dilated using atropine drops . The bulbar conjunctiva was removed on the top of the eyeball and a millimetric incision of the sclera was performed to access the subretinal space . A cannula was then carefully inserted and a controlled injection of balanced saline solution ( BSS; from B . Braun Medical Inc . ) was used to induce a retinal detachment by hydrodissection . The implants were then carefully inserted and advanced below the retina towards the desired position . Eye fundus and/or OCT imaging were then performed to check implantation quality as well as the absence of potential lesions . After each implantation a recovery period from 30 min to 1 hr was respected before starting data acquisition and the presence of visual evoked cortical activation was used to probe the functional integrity of the implanted retinal tissue . In some animals , the retina was imaged using the scanner 3D OCT-2000 ( Topcon , Tokyo , Japan ) , allowing a complete assessment of the area around the MEA . Each acquisition combined both OCT and fundus imaging . The OCT has an axial resolution of 5 μm . The superluminescent diode light source used is centered at 840 nm with a bandwidth of 50 nm adapted for retinal imaging . The focus was adjusted manually on the retina above the MEA . The analysis was performed using 3D OCT-2000 software ( Topcon , Tokyo , Japan ) and consisted in the localization of the MEA on the corresponding B-Scan cross-sections . For fundus images , the OCT was combined with a camera ( Nikon R_D90 , Nikon Imaging Japan Inc . ) that uses a white light flash with green filter , allowing color or red-free acquisitions . An additional +20 D magnifying lens was used to enlarge the field of view of the apparatus originally designed for measurements on humans . Note that the use of this lens induced optical artifacts in fundus ( Figure 1C top row , halos of light ) . Finally , the positions of the implant and of the optic disk were back projected onto the visual stimulation screen using an ophthalmoscope coupled with a laser . We imaged 55 mm cortical windows using 2 different intrinsic imaging systems , the MiCAM ULTIMA imaging ( SciMedia , 100100 pixels , 33 . 3 Hz ) and data-acquisition system as well as a Dalsa camera ( Optical Imaging Inc , 340340 pixels , 30 Hz ) controlled by the VDAQ data-acquisition system ( Optical Imaging ) . The brain was illuminated at 605 nm by 2 optic fibers . Each trial lasted 8 s and the beginning of each acquisition was triggered by the heartbeat . Five hundred milliseconds later ( frame 0 ) the visual or electrical stimulus was displayed during 1 s . The blank condition consisted on a gray screen of 0 . 5 Cd/m2 of averaged luminance for visual stimulation and no current injection for the electrical stimulation . The trials were repeated 20 to 40 times per condition with an ISI of 3 s . Stimuli were displayed monocularly at 60 Hz on a gamma corrected LCD monitor ( placed at 21 . 6 cm from the animal eye plane ) covering 100° ( W ) x 80° ( H ) of visual angle using the Elphy software ( Elphy , Unic , Paris ) . In each trial , a gray screen with an averaged luminance of 0 . 5 Cd/m2 was displayed during 500 ms ( frame 0 ) ; the visual stimulus was then presented during 1 s followed again by a gray screen of 0 . 5 Cd/m2 during 6 . 5 s . Three different protocols were then used . For retinotopic mapping , a white square ( 49 Cd/m2 ) of 20° side was displayed on a gray background ( 0 . 5 Cd/m2 ) at 20 different positions ( 5 horizontal4 vertical ) in a grid like manner ( Figure 2A ) . For the size protocol , the same visual stimuli were displayed at 4 different positions . For each of these positions , the squared visual stimuli were displayed in different sizes . Several ranges of size were used [2 6 10 14 18°] , [6 10 14 18 22°] and [3 . 2 9 . 6 16 22 . 4 28 . 8°] . In the last protocol , 20° side stimuli were also displayed at 4 different positions but with varying luminance levels ( [2 13 . 75 25 . 5 37 . 25 49 Cd/m2] ) . Electrical stimuli were designed and injected using a BioMEA stimulator ( CEA-LETI , Grenoble , France ) ( Charvet et al . , 2010 ) in current injection mode . We used charge balanced squared pulses ( 1 ms per phase ) of varying current intensity and polarity displayed at 80 Hz during 1 s . Such pulses were either injected through single electrodes ( SE , for MEAs with 9 and 17 electrodes ) or through all active electrodes ( wMEA , for MEAs with 9 electrodes only ) in monopolar mode . The diameters of the MEA active surface are of 50 and 650 µm for SE and wMEA , respectively which correspond to 1 and 11° according to Hughes ( Hughes , 1979 ) . Four different protocols were then used . In the first protocol , we tested the effect of increasing current intensity level [± 4 8 16 37 . 5 75 150 300 µA] ( cathodic pulse first ) for SE and wMEA . For the analysis , current levels ≤ 50 µA and ≥75 µA will be considered as low and high intensities , respectively . We next investigated the effect of stimulus intensity and polarity for SE only . Two current intensity levels ( ± 50 and ± 200 µA ) and both polarities ( cathodic or anodic pulse first ) were used . For SE , we also tested the effect of pulse asymmetry and polarity at ± 200 µA . Instead of having symmetrical phase of 1 ms each , the asymmetrical pulses consisted of a first long phase of 1 ms at 40 µA followed by a short phase of 200 µs at 200 µA ( asymmetrical ratio = 5 ) . In total , 4 conditions were used in this protocol , 2 symmetrical and 2 asymmetrical pulses of opposite polarities . For the last protocol in SE , we switch from current to voltage injection mode to test the effect of Impedance Spectroscopy ( IS ) adapted stimuli as described below . Here we present a voltage-mode stimulation platform with the capability to deliver a controlled voltage pattern to the stimulation targets by taking into account the electrode-tissue interface filtering properties ( Dupont et al . , 2013 ) . This procedure requires first measuring the impedance of the electrodes ( magnitude and phase ) for a wide range of frequencies ( Impedance Spectroscopy ) . These measures unveil the capacitive nature of the electrode-tissue interface ( Geddes , 1997 ) . Thus the actual current or voltage pattern injected in the tissue is strongly distorted and does not resemble the desired pattern ( Figure 7A , left column ) . A solution proposed by Dupont et al . ( Dupont et al . , 2013 ) is therefore to estimate the equivalent electrical circuit that accounts for the observed spectrum and use it to reverse engineer what needed to be used to achieve the desired injected pattern ( Figure 7A , right column ) . To achieve this , the experimental design integrates an IS recording module , an Identification Algorithm module fitting IS data to an Electronic Equivalent Circuit ( EEC ) , a Transfer Function Computing module and an Adapted Stimuli Shaping module computing adapted stimuli emitted by the Stimuli Generator module ( Figure 7—figure supplement 1A , patent CEA-LETI ) ( Dupont et al . , 2013 ) . In this study , the test bench is built with commercial devices . IS measurements are performed by injecting low AC voltage between 100 Hz and 200 kHz with a Bio-Logic potentiostat ( SP240 type ) . Data are fitted to a chosen EEC ( Figure 7—figure supplement 1B , see legend for a detailed description of the different components of the circuit ) ( McAdams and Jossinet , 2000 , 1996; Ragheb and Geddes , 1990 ) using the potentiostat associated software ( EC-Lab ) . A transfer function H linking VIMPLANT to VTISSUE is defined . To compute the adapted signal to be generated by the stimuli generator , we use the inverse transfer function H-1 estimation . However , this transfer function does not take into account the impedance’s non-linearity toward voltage levels . To address this issue , several IS measurements are performed between 1 mVpp and 4 Vpp ( Vpp: Volts peak-to-peak ) . Each set of data is fitted onto the same EEC . The abacuses 'EEC parameters versus voltage' are then used to compute the adapted stimulus using voltage dedicated transfer functions per signal harmonics . Stacks of images were stored on hard-drives for off-line analysis . The analysis was carried out with MATLAB R2014a ( Math-Works ) using the Optimization , Statistics , and Signal Processing Toolboxes . Data were first preprocessed to allow comparison between animals and conditions . To allow comparison across conditions , sessions , animals and stimuli ( visual vs electric ) we performed temporal and spatial normalizations of optical imaging signals . We dealt beforehand with pixel-wise trial to trial variability . The averaged time-course of the outer border ( 2 pixels width ) of the image area ( outside the region of interest ) was subtracted from each pixel to remove non-functional temporal noise pattern . We then performed from each trial a first temporal normalization by subtracting the mean and dividing by the standard deviation both estimated from the frame 0 . Second we averaged over trials and applied a spatial normalization . Static maps were computed by averaging 1 . 5 to 2 . 5 s after stimulus onset ( initial dip ) . Spatial z-score normalization was based on the averaged static map of the blank condition . Pixel by pixel , we subtracted from each static map the mean value of the blank and divided the outcome by the blank standard deviation over space . Significance of cortical activation was achieved by applying a threshold of −3 . 09 Z-score ( activation contour ) which corresponds to p-value of 0 . 01 ( lower activations were not considered for further analysis ) . These activation contours were computed on a smoothed version of the maps ( convolving the raw matrix with a 15x15 pixel flat matrix ) . To give a comparison with most optical imaging publications , we also provided on each map a second colorbar scale expressed in DI/I which was computed using the standard procedure as follows ( see Reynaud et al . , 2011 ) : The first step consists of dividing each image of the stack by the mean of the first frames recorded before stimulus onset . In the second step , image stacks collected during stimulated trials are subtracted to those acquired during blank trials on a frame-by-frame basis . The cortical center-of-mass was computed on the z-scored map . In order to quantify the extent and the shape of the activation contour , we computed the equivalent ellipse ( Haralock and Shapiro , 1991 ) . This provided the length and orientation of the equivalent ellipse minor and major axis as well as its equivalent diameter and geometric center . In order to compute the retinotopic polar maps we computed for each pixel the visual activation map giving its response over the 54 visual positions . We then computed the visual center-of-mass ( equivalent to a receptive field center ) . The abscissas or ordinates of each center-of-mass computed for each pixel across positions were plotted separately for azimuth and elevation respectively as polar map ( Figure 2B , the colors representing the position across the 2 cardinal dimensions and the hue representing the intensity of the response for the center-of-mass position ) . Based on this retinotopic mapping , the expected cortical position representing each visual position was computed by taking the intersection of the 2 cardinal polar maps . For each visual or electricalevoked response , we computed the positional error as the distance between the retinotopic expected position ( given by the retinoptopic position of the visual or electrical stimulus ) and the center-of-mass of the corresponding activation ( Figure 2J ) . To compute cortical magnification factor , size or amplitude response functions , linear fits and Naka Rushton functions were used . For the intensity parameter , we computed the functional operational range ( 10–90% ) for each animal individually as follow:operationalrange=C5091n− C5091n With C50 and n being the semi-saturation parameter and the exponent of the sigmoidal fit , respectively . To investigate the spatial profile and the nature of retinal activation elicited by electrical stimulation of the MEA , we compute at the optic disk location the 'shadow cone' angle that sustains the MEA active surface as follow:shadowconeangle=2arcsin ( xy ) z X being the MEA radius , y the distance between the MEA and the optic disk location ( determined form eye fundus pictures ) . For each animal and stimulation condition ( SE vs . wMEA ) , the angle values were corrected according to the real stimulated active surface diameter ( z ) . To find the cortical position that optimized the radial arrangement of the elongated SE cortical activations , we made the following calculation . For each cortical positions ( i , j ) , we calculated the median value of the differences between the angles formed by the segments that connect each cortical activation center to the cortical position in question ( i , j ) ( e . g . expected radial arrangement , dashed segments in Figure 4E ) and the observed angle of the elongated cortical activation ( solid segments in Figure 4E ) . We then looked for the cortical position for which this difference was the smallest ( i . e . the cortical position for which the observed orientation of the activations was the closest to the expected radial arrangement ) . This gave us the cortical position that optimized the radial arrangement for our specific configuration ( black disk in Figure 4E ) . We thereafter use this position rather than the median position of the optic disk ( quite close , gray disk in Figure 4E ) because we believe it is more reliable for two reasons . First , since we positioned all cortical images within the same reference frame ( aligned to the midline and to bregma ) , the cortical representation of the optic disk should be approximately located at similar location in the map . Second , the individual mapping of the optic disk would have added unnecessary noise because of its imprecision ( and for some sessions we could not achieve a proper mapping ) . Indeed , to map the optic disk representation , we estimated its position on the screen using backward projections from an ophthalmoscope . Then we put it in correspondence to the visual stimulus of our retinotopic exploration ( stimuli of 20x20° ) it was falling within . The center of the cortical activation for this stimulus gave an estimation of where the optic disk should be represented cortically . This introduces a series of imprecision ( estimation with the ophthalmoscope , backward projection , not centered with the visual stimulus , cortical activation ) that would be detrimental to our computation . To test whether the radial arrangement of the observed orientations of SE activations occurred by chance , we used a Monte-Carlo like simulation . The question is whether the difference between the predicted ( optimal radial arrangement ) and the observed angles of activation is smaller than the one expected by chance . For that purpose , we generated 1000 random distributions of orientations of cortical activation at the observed positions . For each of these random distributions , we looked for the cortical position that optimized radial arrangement with the method described above . For each specific distribution that was randomly generated , we thus obtained a value of the minimum angular deviation between the predicted and the observed orientations . We obtained likewise a distribution of the angular deviation expected by chance ( Figure 4E inset ) , over which we used the area under the curve of our observation to estimate the p-value: p=0 . 021 ( note that the distribution was also very well captured by a Gaussian fit from which we can calculate the probability of finding the observed angular deviation , and obtained a comparable z-score of 2 . 09 , p=0 . 018 ) . A functional model was designed to predict spatial features of retinal activation elicited by electrical stimulation which depicted a 2D spatial layout of the retina , the central position corresponding to the optic disk . To predict the level , the shape and size of retinal activation , we have introduced two components of diffusion of activity that extend beyond the electrode surface: an isotropic Gaussian spread ( G ) and an 'en passant' activation ( EP ) of the ganglion cells’ axons in response to a stimulation of a given size and a given location in the retinal space . Isotropic 'direct' activation was modeled as a 'flat-top' Gaussian activation around the site of stimulation ( see Figure 4—figure supplement 1 ) :IsoA ( i , j ) ={1 , ( i−i0 ) 2 + ( j−j0 ) 2<Se ( − ( i−i0 ) 2 + ( j−j0 ) 22∗ ( σ∗S ) 2 ) , ( i−i0 ) 2 + ( j−j0 ) 2≥S Where i , j are positions in space , i0 , j0 the position of the electrode , S the size of the electrode and σ the spatial isotropic diffusion extending beyond the electrode . Anisotropic en passant recruitment of ganglion cell axons was modeled as a shadow cone activation that would result from this 'direct' activation . To compute it , we multiply cone angle activation ( Cone ) with a sigmoid function that is 0 for distance between the blind spot and the electrode and 1 for distance further away ( Sigma ) and with a general attenuation for distances very far from the electrode ( Attenuation ) :EP ( i , j ) =Att∗Sig∗AngDiff{Att ( i , j ) =e ( − ( ( i−i0 ) 2 + ( j−j0 ) 2−S ) 22∗N2 ) Sig ( i , j ) =11+e−dOD− ( j0−jod ) S/4Cone ( i , j ) =e−ConeAng22∗σ2 , ifj>jcwhere dOD ( i , j ) = ( i−iod ) 2 + ( j−jod ) 2 and ConeAng= \|tan−1 ( i−iodj−jod ) \|2∗sin−1 ( Sj0−jod ) Where Att is a general attenuation of the activation centered on the position of the electrode ( N is the size of the matrix ) , Sig is a sigmoid function centered on the position of the electrode ( iod , jod are position of the optic disk , dOD is the distance to the optic disk ) , Cone is a cone-like activation modeled as an angular ( calculated as a normalized angle in ConeAng ) Gaussian activation . The detail of all these computation steps is shown in Figure 4—figure supplement 1A . We combined these two activations ( Figure 4C insets ) with different ratios ( α ) via a weighted sum ( IsoA+αEP ) and extracted from this predicted profiles similar parameters ( size , position , elongation ) as we did from our cortical recordings . In our computation , N is 400 , σ is 1 , ( iod , jod ) is ( 200 , 100 ) , ( i0 , j0 ) is ( 200 , 150 ) , S varies between 3 and 30 , and α between 0 . 1 and 1 . Figure 4—figure supplement 1B and C shows the main steps of the computation ( left column is IsoA , middle column is EP and right column is the weighted sum ) , for two ratios ( 0 . 5 and 1 , B and C ) and three electrodes size ( 5 , 7 and 20 , rows ) . To get rid of hypotheses concerning normal distribution and equal variance of data sets , as well as to be more confident in evaluating the significance of the results , we only used two-sided non-parametric Wilcoxon rank sum tests for two-sample or paired data ( p<0 . 05 , p<0 . 01 , **p<0 . 001 ) , except where otherwise stated ( one-sided ) . The p values and sample sizes of the tests are given in the figure legends . N stands for the number of sample and N for the number of rats .	One of the most common causes of blindness is a disorder called retinitis pigmentosa . In a healthy eye , the surface at the back of the eye – called the retina – contains cells called photoreceptors that detect light and convert it into electrical signals for the brain to process . In people with retinitis pigmentosa , these photoreceptor cells die off gradually , which leads to loss of vision . The only treatment available for retinitis pigmentosa is to have an artificial retina implanted into the eye . The artificial retina consists of an array of tiny electrodes , which take over from the damaged photoreceptors and generate electrical signals . The person with the implant perceives these electrical signals as bright flashes called “phosphenes” . However , the phosphenes are too large and imprecise to provide the person with vision that is good enough for tasks such as walking unaided or reading . To find out why artificial retinas produce such poor resolution , Roux et al . compared how a rat’s brain responds to either natural visual stimuli or activation of implanted an array of micro-electrodes . Both the micro-electrodes and the natural stimuli activated the same areas of the brain . However , the micro-electrodes produced larger and more elongated patterns of activation . This is because the electrical currents generated by the micro-electrodes diffused throughout the retinal tissue and activated other neurons besides those intended . To overcome this problem , Roux et al . tested different ways of stimulating the micro-electrodes in order to identify those that induce the desired patterns of brain activity . This approach – known as reverse engineering – did indeed improve the performance of the micro-electrode array . The next step is to extend these findings , which were obtained in healthy rats , to non-human primates or animal models of retinitis pigmentosa to better understand the condition in humans . In addition , combining the current approach with other existing techniques should further improve the vision that can be achieved with artificial retinas .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Probing the functional impact of sub-retinal prosthesis
The phosphoinositide 3-kinase ( PI3K ) -Akt network is tightly controlled by feedback mechanisms that regulate signal flow and ensure signal fidelity . A rapid overshoot in insulin-stimulated recruitment of Akt to the plasma membrane has previously been reported , which is indicative of negative feedback operating on acute timescales . Here , we show that Akt itself engages this negative feedback by phosphorylating insulin receptor substrate ( IRS ) 1 and 2 on a number of residues . Phosphorylation results in the depletion of plasma membrane-localised IRS1/2 , reducing the pool available for interaction with the insulin receptor . Together these events limit plasma membrane-associated PI3K and phosphatidylinositol ( 3 , 4 , 5 ) -trisphosphate ( PIP3 ) synthesis . We identified two Akt-dependent phosphorylation sites in IRS2 at S306 ( S303 in mouse ) and S577 ( S573 in mouse ) that are key drivers of this negative feedback . These findings establish a novel mechanism by which the kinase Akt acutely controls PIP3 abundance , through post-translational modification of the IRS scaffold . Signalling networks enable cells to respond to diverse environmental challenges and maintain cellular homeostasis ( Humphrey et al . , 2015b; Kholodenko et al . , 2012; Ubersax and Ferrell , 2007 ) . They comprise an array of regulatory mechanisms that ensure the most appropriate response is achieved . For example , positive feedback loops facilitate adequate signal amplification , while negative feedback loops enable rapid , tightly regulated responses to stimuli and prevent pathway hyperactivation ( Cheong and Levchenko , 2010; Tyson et al . , 2003 ) . Negative feedback also ensures that signalling pathways are resistant to external perturbations ( Stelling et al . , 2004 ) and may optimise the energetic cost of signal transduction ( Anders et al . , 2020 ) . Identifying and characterising feedback mechanisms in signal transduction pathways is pivotal since dysfunctional cell signalling frequently underlies complex diseases ( Fazakerley et al . , 2019; Mora-Garcia and Sakamoto , 1999; Nguyen and Kholodenko , 2016 ) . The phosphoinositide 3-kinase ( PI3K ) -Akt signalling network is activated by several cellular receptors and plays a central role in regulating cell growth and metabolism ( Fruman et al . , 2017 ) . Engagement of this pathway is a multi-step process involving the actions of lipid and protein kinases and phosphatases , and relies on modulation of protein-protein and protein-lipid interactions . For example , binding of insulin to the insulin receptor ( IR ) leads to receptor auto-phosphorylation , and subsequent binding of the insulin receptor substrate ( IRS ) 1 and 2 adaptors via their phosphotyrosine-binding ( PTB ) domains ( Boucher et al . , 2014 ) . IRS proteins are then tyrosine-phosphorylated and bind the Src homology 2 ( SH2 ) domain of PI3K p85 , recruiting PI3K ( consisting of p85 and p110 subunits ) to the plasma membrane ( PM ) ( Mosthaf et al . , 1996; Songyang et al . , 1993 ) . PI3K phosphorylates phosphatidylinositol ( 4 , 5 ) bisphosphate at the PM to form phosphatidylinositol ( 3 , 4 , 5 ) trisphosphate ( PIP3 ) . Phosphatase and tensin homolog ( PTEN ) catalyses the reverse reaction ( Kabuyama et al . , 1996; Lee et al . , 2018; Maehama and Dixon , 1998; Myers et al . , 1998 ) . PIP3 is a bioactive lipid that recruits effectors such as phosphoinositide-dependent protein kinase 1 ( PDPK1 ) and Akt via their pleckstrin homology ( PH ) domains , leading to their accumulation at the PM ( Currie et al . , 1999 ) . Here , Akt is phosphorylated at T309 ( in Akt2; T308 in Akt1 , T305 in Akt3 ) by PDPK1 and S474 ( in Akt2; S473 in Akt1 , S472 in Akt3 ) by mammalian target of rapamycin complex 2 ( mTORC2 ) ( Alessi et al . , 1996a; Sarbassov et al . , 2005 ) . This leads to full activation of Akt kinase activity ( Kearney et al . , 2019 ) , and substrate phosphorylation ensues . More than 100 substrates of Akt have been reported , which regulate a myriad of biological pathways ( Manning and Toker , 2017 ) . Importantly , dysregulated PI3K/Akt activity is linked to several diseases including cancer and type 2 diabetes ( Fruman et al . , 2017 ) , making tight regulation essential . Insulin-stimulated recruitment of Akt to the PM displays overshoot behaviour ( a response where the initial maxima exceeds the final steady state ) and oscillations ( Ebner et al . , 2017; Norris et al . , 2017 ) . These dynamics are indicative of acute negative feedback signals that regulate Akt activation ( Behar and Hoffmann , 2010; Cheong and Levchenko , 2010; Nyman et al . , 2012 ) . mTORC1 and S6-kinase ( S6K ) are activated downstream of Akt ( Manning and Toker , 2017 ) , and have been reported to put the ‘brakes’ on PI3K/Akt signalling through a variety of mechanisms . For example , mTORC1/S6K phosphorylates the mTORC2 component RICTOR to attenuate mTORC2 activity and hence Akt activity ( Dibble et al . , 2009; Julien et al . , 2010 ) . Furthermore , mTORC1/S6K-mediated phosphorylation of IRS1 limits PI3K/Akt activation ( Copps and White , 2012; Hançer et al . , 2014; Harrington et al . , 2004; Shah et al . , 2004; Shah and Hunter , 2006; Tzatsos , 2009; Tzatsos and Kandror , 2006; Yoneyama et al . , 2018 ) . However , it remains unclear whether these feedbacks are responsible for the overshoot and oscillatory behaviour in Akt activation . As PI3K , Akt and mTOR are targets for cancer therapeutics ( Porta et al . , 2014 ) , a deep understanding of these network topologies is important to elucidate the consequences of pathway manipulation and ultimately provide opportunities for improving drug efficacy . Here , we applied live cell imaging , biochemical assays , and iterative computational modelling to interrogate the feedback signals regulating Akt activation . Our data uncover a potent Akt-dependent , mTORC1-independent feedback mechanism . Upon activation , Akt depletes PM localised IRS1/2 to reduce its interaction with the IR . This limits PM-associated PI3K and PIP3 synthesis , constituting a strong negative feedback loop . This feedback is driven by changes in the phosphorylation of IRS1/2 on a spectrum of residues . Specifically , we identify IRS2 as a novel substrate of Akt and show that Akt-mediated phosphorylation of IRS2 at S306 ( S303 in mouse ) and S577 ( S573 in mouse ) are critical drivers of this feedback . The recruitment behaviour of Akt in response to insulin is indicative of negative feedback , however the source of this feedback is unknown . Given the complexity of feedback architectures , we developed a computational model of insulin signalling to help dissect potential origins of negative feedback in this system ( Figure 1A ) . Our model included nodes representing proximal insulin signalling ( IR , IRS/PI3K ) and accounted for various intricacies of Akt activation , such as its phosphorylation states ( at both T309 and S474 ) as well as the positive feedback loop onto mTORC2 in response to SIN1 T86 phosphorylation ( Humphrey et al . , 2013; Yang et al . , 2015 ) . The model also included the previously described negative feedback loop from mTORC1/S6K to IRS/PI3K . As mTORC1 activation requires Akt-mediated phosphorylation of AKT1S1/PRAS40 ( Sancak et al . , 2007; Wang et al . , 2007 ) , we incorporated this feedback mechanism into the model using phospho-PRAS40 ( T246 ) as a surrogate for mTORC1 activation ( Figure 1A ) . The new model was formulated using ordinary differential equations ( ODEs ) and implemented in MATLAB that mathematically represents the network interactions as a series of ODEs based on established kinetic laws ( see Materials and methods and Supplementary file 1 for detailed model descriptions ) . To train and evaluate our model , we first assessed the recruitment of TagRFP-T tagged Akt2 ( described previously Norris et al . , 2017 ) in live 3T3-L1 adipocytes using total internal reflection fluorescence microscopy ( TIRFM ) . This cell line was used as they are exquisitely insulin responsive , providing a high signal-to-noise ratio for evaluating phenotypic responses . In response to 1 nM insulin , plasma membrane ( PM ) -associated Akt displayed overshoot behaviour . Specifically , PM levels of Akt reached a maximum at 75 s post-insulin stimulation , followed by a rapid decline that reached a new steady state that was roughly half of its maximum by 10 min post-stimulation ( Figure 1B ) . In response to 100 nM insulin , the maximum recruitment of Akt was fourfold higher than 1 nM insulin , with an initial overshoot followed by a secondary increase ( Figure 1B ) , which may reflect the engagement of a positive feedback signal at this dose . A similar overshoot was observed for insulin-stimulated phosphorylation of Akt at its activating sites T309 and S474 ( Figure 1C , D ) , indicating that the overshoot in Akt recruitment is reflected in its activation . The kinetics of T309 phosphorylation were much faster than S474 phosphorylation . Interestingly the phosphorylation of Akt substrates ( AS160 , FOXO1 , GSK3 , and PRAS40 ) did not exhibit an overshoot ( Figure 1C , D ) . Although Akt singly phosphorylated at T309 is active , the delayed onset of S474 phosphorylation doubles Akt activity ( Kearney et al . , 2019 ) and likely extends the peak phosphorylation observed for these substrates . Our model , when trained with the Akt recruitment , Akt phosphorylation , and PRAS40 phosphorylation data , was able to reasonably recapitulate the experimental data ( model 1; Figure 1E ) . We next used our model to predict the effect of a 1 nM insulin stimulus in the presence of the mTORC1 inhibitor rapamycin , to block mTORC1-mediated negative feedback . The model predicted a loss of the overshoot and a 25% increase in the recruitment of Akt to the PM ( Figure 1F ) . To test this prediction , we recapitulated these conditions in 3T3-L1 adipocytes . Despite complete inhibition of mTORC1 activity by rapamycin , no change was observed in either Akt recruitment ( Figure 1G ) or phosphorylation ( Figure 1H ) . When we included this rapamycin data for model calibration , the model could not reasonably recapitulate the experimental data ( model 2; Figure 1—figure supplement 1A , B ) , further suggesting that mTORC1 is unlikely to be involved in negative feedback under these conditions . Furthermore , calibration of a model without negative feedback was unable to recapitulate the overshoot in Akt recruitment or phosphorylation ( model 3; Figure 1—figure supplement 1C ) . Together , these data point to a feedback mechanism during the early insulin response that drives the overshoot in Akt recruitment that is independent of mTORC1 . Based on the existing model architecture , we hypothesised that negative feedback from PDPK1 , mTORC2 , or Akt could give rise to an overshoot in Akt recruitment when connected to an upstream node such as IRS/PI3K or PTEN . To explore these mechanisms , we constructed six additional models that each incorporated a potential negative feedback mechanism and calibrated each with our training data ( models 4–9; Figure 2A , Figure 2—figure supplement 1A–F , Supplementary file 1 ) . This allowed us to examine possible competing feedback structure hypotheses , by performing separate model calibrations and comparing how well each of the model variants fitted the experimental data . Identifiability analysis ( Maiwald et al . , 2016; Rateitschak et al . , 2012; Raue et al . , 2009 ) demonstrated that all models were similarly unidentifiable , with at least one parameter not identifiable in each model ( Supplementary file 2 ) , which is not surprising given the size and scope of the models ( see Materials and methods for detailed description ) . Quantitative assessment of model fitting based on the objective function revealed that Akt to IRS/PI3K ( model 9 ) best matched the experimental data , followed by mTORC2 to PTEN ( model 7 ) , and then Akt to PTEN ( model 8 ) ( Figure 2B ) . Consistent with this , qualitative assessment of models 7–9 revealed greater consistency between model dynamics and experimental data compared to models 4–6 ( Figure 2—figure supplement 1A–F ) . For example , model 4 showed a significant discordance in pAkt S474 dynamics , while models 5 and 6 displayed delayed pAkt T309 accumulation in response to 1 nM insulin ( Figure 2—figure supplement 1A–C ) . Overall , since models 7–9 displayed superior quantitative and qualitative consistency with experimental data , we focused on interrogating these models hereafter . Each of the feedback loops in these three models would be driven by Akt ( Akt to IRS/PI3K , Akt to PTEN ) or partially driven by Akt ( mTORC2 to PTEN; as Akt contributes to mTORC2 activation [Humphrey et al . , 2013; Yang et al . , 2015] ) . Thus , we simulated the effect of Akt inhibition on insulin-stimulated Akt recruitment in the three models . In all cases it was predicted that Akt inhibition would eliminate the overshoot in Akt recruitment and facilitate increased recruitment , but with distinct kinetics and magnitude ( Figure 2C ) . To test these predictions experimentally , we used TIRFM to measure the recruitment of TagRFP-T-Akt2 in the presence of the Akt inhibitor GDC0068 , which inhibits Akt by binding its ATP-binding pocket ( Figure 2—figure supplement 2A ) . Strikingly , in the presence of GDC0068 , we observed a threefold increase in Akt recruitment to the PM upon insulin stimulation and loss of overshoot ( Figure 2D ) . The recruitment kinetics and magnitude were markedly similar to the model predictions with the removal of feedback from Akt to IRS/PI3K ( Figure 2C ) . These data were corroborated by assessment of endogenous Akt localisation in adipocytes using subcellular fractionation , which showed a marked potentiation of insulin-stimulated Akt2 recruitment to the PM in the presence of GDC0068 ( Figure 2E ) . This phenomenon was not specific to adipocytes , but rather a common feature of multiple cell types – HEK293E , HeLa , 3T3-L1 fibroblasts , and MCF7 cells ( Figure 2—figure supplement 2B–E ) . To ensure these effects were not GDC0068-specific , but rather a general feature of Akt inhibition , we employed an orthogonal chemical genetics approach to inhibit Akt . The W80A mutation in Akt renders it insensitive to inhibition by MK2206 , a compound that prevents Akt PM recruitment ( Wu et al . , 2010 ) . Thus , MK2206 can block endogenous Akt activation , leaving ectopic W80A Akt kinase-dead mutants to be specifically interrogated in cells ( Green et al . , 2008; Kajno et al . , 2015; Kearney et al . , 2019 ) . In the presence of MK2206 , W80A TagRFP-T-Akt2 displayed similar recruitment kinetics to WT TagRFP-T-Akt2 ( Figure 2F ) , consistent with our earlier study ( Kearney et al . , 2019 ) . We assessed insulin-stimulated recruitment of two W80A kinase-dead TagRFP-T-Akt2 mutants , which confer loss of kinase activity by differential mechanisms; by preventing Akt phosphorylation at its activating sites ( W80A-T309A-S474A; Beg et al . , 2017 ) , or by preventing ATP binding ( W80A-K181A; Cong et al . , 1997 ) . Both mutants exhibited 3 . 5-fold augmented recruitment and loss of overshoot behaviour compared to W80A Akt ( Figure 2F ) . These responses phenocopied the recruitment of WT Akt in the presence of GDC0068 ( Figure 2—figure supplement 2F ) , and again were remarkably similar to the model prediction with the removal of feedback from Akt to IRS/PI3K ( Figure 2C ) . In the absence of MK2206 , the PM recruitment of the kinase-dead mutants was still potentiated , but also exhibited an overshoot , suggesting competition with endogenous Akt ( Figure 2G ) . Taken together , these data indicate the presence of an Akt-dependent feedback signal which limits its activation at the PM , likely through the IRS/PI3K node . While phosphorylation of Akt at T309 is essential for Akt kinase activity ( Alessi et al . , 1996a; Kearney et al . , 2019 ) , the role of S474 phosphorylation remains controversial . However , in adipocytes , we have shown S474 phosphorylation is required only for maximal kinase activity ( Kearney et al . , 2019 ) . As the overshoot in Akt recruitment correlated with the kinetics of T309 phosphorylation , but not S474 phosphorylation ( Figure 2—figure supplement 2G ) , we hypothesised that feedback is rapidly engaged following Akt T309 phosphorylation , and that S474 phosphorylation is not required for feedback engagement . To test this , we simulated the effect of losing either T309 or S474 phosphorylation on insulin-stimulated Akt recruitment in our three models ( Figure 2H ) . As expected , the predictions for loss of T309 phosphorylation in each model was identical to their prediction of Akt inhibition ( Figure 2C ) . However , each model’s prediction for loss of S474 phosphorylation was markedly different ( Figure 2H ) . We tested these predictions experimentally by observing insulin-stimulated recruitment of W80A-T309A and W80A-S474A TagRFP-T-Akt2 by TIRFM ( Figure 2I ) . Loss of S474 phosphorylation resulted in an attenuated overshoot in response to 1 nM insulin but no difference in response magnitude compared to control . This recruitment profile is consistent with a subtle loss of feedback , which likely occurs because Akt is only fully active once phosphorylated at both T309 and S474 residues ( Alessi et al . , 1996a; Kearney et al . , 2019 ) . An additional twofold increase was observed with 100 nM insulin , similar to control . In response to 1 nM insulin , W80A-T309A Akt displayed a loss of overshoot and a threefold increase in PM association in comparison to control ( Figure 2I ) . This phenocopied chemical and genetic inhibition of Akt ( Figure 2D , F ) , which was expected since T309A Akt has no kinase activity ( Alessi et al . , 1996a; Kearney et al . , 2019 ) . Additionally , with a subsequent 100 nM insulin stimulus , there was only a modest increase in W80A-T309A recruitment . These response profiles were in close agreement with the predictions arising from the Akt to IRS/PI3K model ( Figure 2H ) . These combined modelling and experimental analyses support that the phosphorylation of Akt at S474 is largely redundant in activating the negative feedback . Irrespective of where the feedback was engaged , all three models assumed that Akt regulates PIP3 levels through negative feedback . To assess whether this was the case , we measured insulin-stimulated PIP3 content in adipocytes , in the presence or absence of Akt inhibition . We utilised three small-molecule inhibitors , which abolish Akt kinase activity via distinct mechanisms: by binding the ATP-binding site of Akt ( GDC0068 ) , by preventing Akt PM recruitment ( MK2206 ) , and by inhibiting PDPK1 to prevent Akt T309 phosphorylation ( GSK2334470 ) ( Figure 3—figure supplement 1 ) . Stimulation with 1 nM insulin increased PIP3 content approximately fivefold relative to basal , confirming the validity of the assay ( Figure 3A ) . Strikingly , each inhibitor further augmented insulin-stimulated PIP3 abundance approximately fivefold relative to insulin alone ( Figure 3A ) , suggesting that Akt regulates PIP3 abundance . We hypothesised that the ability of Akt to regulate PIP3 abundance would influence the recruitment of other PIP3-binding proteins , such as PDPK1 and GAB2 ( Currie et al . , 1999; Gu et al . , 2003; Yoshizaki et al . , 2007 ) . Both PDPK1-eGFP and Gab2PH-eGFP were recruited to the PM upon insulin stimulation and displayed PM recruitment overshoots ( Figure 3B , C ) . Furthermore , their degree of recruitment was markedly enhanced with GDC0068 and MK2206 ( Figure 3B–D ) , consistent with increased PIP3 in the absence of Akt activity . To directly assess the requirement of Akt activity for feedback , we examined recruitment of Gab2PH-eGFP in cells co-expressing WT , W80A , or W80A-T309A TagRFP-T-Akt2 . In the presence of MK2206 , the Akt constructs behaved as expected ( Figure 3E ) , while MK2206-dependent hyper-recruitment of Gab2PH-eGFP was rescued by co-expression of W80A Akt , but not WT or W80A-T309A Akt ( Figure 3F ) . These data suggest that Akt regulates PIP3 abundance and this alters the localisation of PIP3-binding proteins . Elevated PIP3 abundance following Akt inhibition likely resulted from its increased production by PI3K , or suppressed breakdown by phosphatases such as PTEN . So far , the Akt to IRS/PI3K model best recapitulated the experimental data; however to further interrogate this model , we tested whether Akt controls PIP3 degradation . To this end , we simulated the effect of PI3K inhibition on the rate of Akt dissociation from the PM across the three models . Both the Akt to PTEN and mTORC2 to PTEN models predicted a slower rate of Akt dissociation in the absence of Akt activity , due to impaired PIP3 degradation by PTEN ( Figure 4A ) . In contrast , the Akt to IRS/PI3K model predicted no difference in Akt PM dissociation ( Figure 4A ) . To test this experimentally , we stimulated adipocytes expressing W80A ( active ) or W80A-T309A ( kinase-dead ) TagRFP-T-Akt2 with insulin , and then the PI3K inhibitor wortmannin , once PM Akt had achieved a steady-state concentration . There was no difference in their rate of PM dissociation , consistent with the Akt to IRS/PI3K negative feedback model ( Figure 4B ) . We next conducted the same experiment using PDPK1-eGFP , in the presence or absence of GDC0068 or MK2206 . Regardless of whether Akt was inhibited , there was no difference in the rate of PDPK1 PM dissociation ( Figure 4C ) . To corroborate these experiments , we used adipocytes expressing constitutively active PI3K ( p110* ) , as well as W80A ( active ) or W80A-T309A ( kinase-dead ) TagRFP-T-Akt2 . Expression of p110* ( Hu et al . , 1995 ) allowed a constant rate of PIP3 production , while the rate of PIP3 degradation was unaltered . When stimulated with MK2206 , there was no increase in W80A-T309A Akt recruitment , indicating no loss of feedback ( Figure 4D ) . Furthermore , there was no difference in the PM dissociation rate between W80A and W80A-T309A Akt ( Figure 4D ) . We next measured TagRFP-T-Akt2 recruitment in HCC1937 human breast cancer cells , which do not express PTEN , the primary PIP3 phosphatase ( Kabuyama et al . , 1996; Lee et al . , 2018; Maehama and Dixon , 1998; Myers et al . , 1998 ) . Akt recruitment increased in response to IGF1 and increased further with GDC0068 ( Figure 4E ) . This was consistent with the engagement of Akt-mediated feedback in the absence of PTEN . Both our modelling and experimental data suggested that Akt regulates the ability of PI3K to synthesise PIP3 . As PI3K is activated at the PM , we determined whether Akt controls PI3K localisation . Subcellular fractionation revealed an increase in the abundance of PI3K p85 and p110 at the PM upon insulin stimulation , which was potentiated in the presence of Akt inhibitors GDC0068 and MK2206 ( Figure 4F ) . These data were corroborated by assessment of PI3K p85 localisation using immunofluorescence/TIRFM . This technique was less sensitive than subcellular fractionation , as it did not reveal a change in PM localised PI3K with 1 nM insulin; however , an increase was detectable in the presence of the Akt inhibitors GDC0068 and MK2206 ( Figure 4G ) . Together , these data are consistent with the Akt to IRS/PI3K negative feedback model . Specifically , upon acute growth factor stimulation , Akt translocates to the PM where it is phosphorylated and activated . Akt then engages a negative feedback mechanism that limits PM-associated PI3K and consequently lowers PIP3 abundance . We next investigated how Akt controls PI3K localisation . The speed of the feedback suggested that Akt-mediated phosphorylation of PI3K itself could drive changes in its localisation . However , PI3K does not contain an Akt substrate motif ( R-X-R-X-X-S/T , where R represents arginine , X represents any amino acid , and S/T represents serine/threonine ) ( Alessi et al . , 1996b ) . Furthermore , translocation of PI3K to the PM is thought to be primarily regulated by interactions with other proteins , rather than post-translational modification of PI3K itself ( Rordorf-Nikolic et al . , 1995 ) . IRS1 and IRS2 are the primary adaptor proteins responsible for recruiting PI3K to the PM following insulin stimulation ( White , 2003 ) . Consequently , we explored whether Akt regulates IRS1/2 behaviour . First , we determined whether Akt controls the abundance of IRS1-eGFP and IRS2-eGFP at the PM using TIRFM . Following insulin stimulation , the amount of IRS1/2 at the PM decreased ( Figure 5A , B ) . However , in the presence of Akt inhibitors ( GDC0068 and MK2206 ) , IRS1/2 were retained at the PM ( Figure 5A , B ) . Importantly , rapamycin was unable to inhibit the insulin-stimulated decrease in PM-associated IRS1/2 ( Figure 5C , D ) , demonstrating that this process is mTORC1/S6K-independent . Consistent with these observations , subcellular fractionation also demonstrated an Akt-dependent decrease in the abundance of endogenous IRS1 at the PM and extended these findings to reveal that IRS1 moves from the PM to the cytosol upon insulin stimulation , rather than to internal membranes ( high density microsome [HDM]/low density microsome [LDM] fractions ) ( Figure 5E ) . We attribute the PM localisation of IRS1/2 prior to insulin stimulation to its PH domain , as deletion of the IRS1 PH domain ( DelPH IRS1-eGFP ) decreased its abundance at the PM in unstimulated cells ( Figure 5F ) . This is likely due to the affinity of the IRS1 PH domain for PI ( 4 , 5 ) P2 ( Dhe-Paganon et al . , 1999 ) . Together , these data indicate that IRS1/2 is PM-localised in unstimulated cells via a PH domain-dependent interaction , and upon insulin stimulation Akt induces the translocation of IRS1/2 from the PM to the cytosol . We next dissected whether these changes in IRS1/2 localisation were a product of Akt modulating the localisation or activation of the IR . Consistent with direct regulation of IRS by Akt , treatment with Akt inhibitors did not alter IR localisation ( Figure 5G ) or activation as measured by tyrosine phosphorylation ( Figure 5H ) . Furthermore , deletion of the PTB domain of IRS1 ( DelPTB IRS1-eGFP ) , which is responsible for its interaction with the IR ( Eck et al . , 1996 ) , did not impact IRS1 removal from the PM upon insulin stimulation ( Figure 5I ) . These data suggest that Akt removes IRS from the PM via disruption of IR-independent interactions . Despite occurring independent of the IR , we hypothesised that these changes in IRS localisation would ultimately serve to shrink the pool of IRS available to interact with activated IR . Following immunoprecipitation of endogenous IRS1 or IRS2 followed by mass spectrometry , the IR was only detected in the presence of insulin , consistent with a strong insulin-dependent interaction ( Figure 5J ) . However , the amount of IR detected was augmented in the presence of Akt inhibitors GDC0068 and MK2206 ( approximately 17- and 7-fold for IRS1; 19- and 12-fold for IRS2 , respectively; Figure 5J ) , suggesting that Akt controls the level of interaction between the IR and IRS . These data suggest that Akt limits PI3K-mediated PIP3 production by depleting PM-associated IRS , to reduce the pool of IRS available to interact with activated IR . We next explored how Akt controls IRS localisation . Phosphorylation is frequently responsible for changes in protein localisation ( Cohen , 2002 ) and we have previously identified an abundance of insulin-regulated phosphorylation sites on IRS1/2 ( Humphrey et al . , 2013 ) . Thus , we hypothesised that Akt phosphorylates IRS to release it from the PM and engage negative feedback . Collectively , IRS1 and IRS2 contained 11 phosphorylation sites within an Akt substrate motif ( R-X-R-X-X-S/T ) ( Alessi et al . , 1996b ) . To investigate whether these phosphorylation sites were responsible for the depletion of PM-associated IRS1/2 upon insulin stimulation , we concurrently mutated these Ser/Thr residues on human IRS1-eGFP ( herein 6P IRS1 ) and human IRS2-eGFP ( herein 5P IRS2 ) to alanine , to prevent phosphorylation ( Figure 6A , B ) . We then co-expressed 6P IRS1 or 5P IRS2 with TagRFP-T-Akt2 in adipocytes and utilised TIRFM to measure changes in their localisation . Following insulin stimulation , WT IRS1/2 dissociated from the PM; however , increased PM abundance was observed for both 6P IRS1 and 5P IRS2 ( Figure 6C , D ) . Furthermore , expression of 6P IRS1 or 5P IRS2 in adipocytes resulted in hyper-recruitment of Akt to the PM compared to cells expressing WT IRS ( Figure 6C , D ) . This response mimicked GDC0068 treatment ( Figure 2D ) , causing athreefold increase in PM Akt upon insulin stimulation . These data suggest that phosphorylation of IRS1/2 facilitates its removal from the PM and drives negative feedback . To investigate which of these 11 phosphorylation sites were responsible for the dissociation of IRS from the PM , we individually mutated each of these Ser/Thr residues on IRS1-eGFP and IRS2-eGFP to alanine to prevent phosphorylation . We then co-expressed this mutant with TagRFP-T-Akt2 and monitored their localisation using TIRFM . No single mutation phenocopied 6P IRS1 or 5P IRS2 , with each mutant having only a subtle effect on Akt and IRS localisation ( Figure 6—figure supplement 1A–F , Figure 6—figure supplement 2A–E ) . This indicated that at least two insulin-stimulated phosphorylation events cooperate to release IRS1/2 from the PM and limit downstream signal propagation . We next investigated whether any of the 11 IRS phosphorylation sites making up the 6P IRS1 and 5P IRS2 were directly phosphorylated by Akt . Our criteria for an Akt substrate were that the IRS residue ( 1 ) can be phosphorylated by Akt in vitro , ( 2 ) cannot be phosphorylated in the presence of GDC0068/MK2206 ( Akt inhibitors ) in cells , and ( 3 ) can be phosphorylated in the presence of rapamycin/rapalink ( mTORC1 inhibitors ) in cells , to exclude S6K as the upstream kinase – S6K is activated downstream of Akt and also recognises the R-X-R-X-X-S/T substrate motif ( Alessi et al . , 1996b ) . In this context IRS residues phosphorylated by S6K were not of interest as this feedback mechanism is mTORC1/S6K-independent ( Figures 1 and 5C–D ) . To address these criteria , we performed three experiments which relied on quantifying IRS1/2 phosphorylation using mass spectrometry . Quantifying all 11 IRS1/2 phosphorylation sites across all experiments was technically challenging , due to limited sequence coverage . However , four of the five IRS2 phosphorylation sites were identified in all experiments , and so we focused on these phosphorylation sites hereafter . IRS2 S365 ( S362 in mouse ) was phosphorylated by Akt in vitro and sensitive to GDC0068/MK2206 in cells ( Figure 6—figure supplement 3A ) . However , IRS2 S365 phosphorylation was also rapalink/rapamycin sensitive ( Figure 6—figure supplement 3A ) , implicating it as an S6K substrate . IRS2 S1149 ( S1138 in mouse ) phosphorylation was not increased following insulin stimulation and was unable to be phosphorylated by Akt in vitro ( Figure 6—figure supplement 3B ) , suggesting it is not an Akt substrate . However , IRS2 S306 ( S303 in mouse ) and S577 ( S573 in mouse ) were phosphorylated by Akt in vitro , and in cells were sensitive to GDC0068/MK2206 , but not rapamycin/rapalink ( Figure 6E , F ) . These data were sufficient to classify IRS2 S306 and S577 as novel Akt substrates . We hypothesised that Akt-mediated phosphorylation of IRS2 S306 and S577 synergistically depletes PM-associated IRS2 to drive negative feedback . When IRS2 S306 or S577 were individually mutated to alanine ( S306A IRS2 or S577A IRS2 ) , IRS2 still dissociated from the PM upon insulin stimulation , but with a higher endpoint compared to WT IRS2 , and both mutants subtly increased Akt translocation to the PM ( Figure 6—figure supplement 2A , D ) . However , concurrent mutation of both phosphorylation sites ( S306/577A IRS2 ) prevented PM dissociation of IRS2 ( Figure 6G ) , mimicking Akt inhibition with GDC0068 and MK2206 ( Figure 5A , B ) . Furthermore , expression of S306/577A IRS2 resulted in hyper-recruitment of Akt and PDPK1 to the PM compared to cells expressing WT IRS2 ( Figure 6G , H ) , consistent with increased PIP3 levels . Intriguingly , mutation of the corresponding phosphorylation sites in IRS1 ( S270/527A ) did not prevent its dissociation from the PM upon insulin stimulation and only had a subtle effect on Akt recruitment ( Figure 6—figure supplement 4 ) , demonstrating alternate regulation between the IRS isoforms . These data suggest that Akt-mediated phosphorylation of IRS2 S306 and S577 synergistically dissociates PM-associated IRS2 to limit PIP3 production and Akt activation . We propose a model where in unstimulated cells , a pool of IRS is localised to the PM via a PH domain-dependent interaction . Upon insulin binding its receptor , Akt is rapidly activated ( by the canonical IRS/PI3K pathway ) and directly phosphorylates IRS proteins at key regulatory residues such as IRS2 S306 and S577 . This results in the translocation of IRS from the PM to the cytosol and depletes the pool available to interact with the IR . Ultimately , these events limit PM-associated PI3K and PIP3 synthesis ( Figure 7 ) . The PI3K/Akt signalling network is critical to the survival of all eukaryotic cells , and as such the consequences of its dysregulation are severe ( Hers et al . , 2011; Manning and Toker , 2017 ) . Aberrant activation of PI3K/Akt signalling underlies a variety of complex diseases , such as cancer and type 2 diabetes ( Hers et al . , 2011; Manning and Toker , 2017 ) . Consequently , tight regulation of the PI3K/Akt pathway is critical . Here , we describe an acutely engaged , powerful negative feedback signal that emanates from Akt to IRS1/2 and limits signal flow in a broad range of cell types . Loss of this feedback results in a profound increase in PIP3 production by PI3K . As well as Akt , this feedback has substantial impacts on other PIP3-dependent proteins such as PDPK1 and GAB2 . Beyond increasing our basic understanding of this crucial signalling node , this discovery has important implications , particularly in cancer therapy , where Akt has long been a major drug target . Here , we implicate six phosphorylation sites in IRS1 and five phosphorylation sites in IRS2 , which are within an Akt substrate motif ( R-X-R-X-X-S/T ) , as major contributors of Akt-mediated negative feedback . In particular , we identify IRS2 as a novel substrate of Akt , and show Akt-mediated phosphorylation of IRS2 at S306 ( S303 in mouse ) and S577 ( S573 in mouse ) are primary drivers of negative feedback . We have shown that the phosphorylation of IRS removes it from the PM by disrupting its receptor-independent interactions ( Figure 5I ) , and there are several possibilities as to how this occurs . The negative charge of these modifications may repel IRS from the negative electrostatic surface charge of the PM . This has been shown to be the case for other proteins such as MARCKS ( Goldenberg and Steinberg , 2010 ) . Alternatively , phosphorylation of IRS may promote 14-3-3 binding and sequester IRS away from an interacting protein/lipid at the PM . 14-3-3 has been shown to accompany the movement of the IRS/PI3K complex from membranes to the cytosol ( Xiang et al . , 2002 ) . Intriguingly , IRS2 phosphorylated at S577 has been shown to bind 14-3-3 ( Neukamm et al . , 2012 ) . We suspect that by one of these mechanisms IRS phosphorylation depletes the PM localised pool of IRS which is available to interact with the IR . In addition to negative feedback , it is possible that this movement of IRS to the cytosol might facilitate translocation of the IRS signalling complex to another site , which could be important for signal propagation . Here , we have shown that the phosphorylation of two IRS2 serine residues by Akt act in synergy to induce IRS2 translocation to the cytosol ( Figure 6G ) . We have also shown that there is further synergy between the IRS1/2 phosphorylation sites within an R-X-R-X-X-S/T motif ( Figure 6C–D ) . It is possible that this cooperation extends further to IRS1/2 serine/threonine residues phosphorylated by other kinases . A variety of other kinases , including insulin-independent kinases , have been shown to phosphorylate IRS such as JNK , ERK1/2 , PKCs , S6K , and mTORC1 ( Copps and White , 2012 ) . Phosphosite Plus ( Hornbeck et al . , 2015 ) reports 110 serine/threonine phosphorylation sites on human IRS1 and 132 serine/threonine phosphorylation sites on human IRS2 , almost all of which are located within the unstructured tail of IRS1/2 . Consequently , IRS phosphorylation on distinct S/T residues could serve as a site of crosstalk between distinct pathways/various kinases , integrating their activation status and modifying the strength of insulin signalling accordingly . Intriguingly , metabolic stress has previously been shown to increase IRS1 serine/threonine phosphorylation and modulate IRS1 tyrosine phosphorylation ( Hançer et al . , 2014 ) . Thus , it is possible that phosphorylation of these other sites could also be a means to induce IRS translocation to the cytosol and impair insulin signalling . It is important to note , however , that modulation of IRS serine/threonine phosphorylation is likely not the mechanism of insulin resistance ( Fazakerley et al . , 2019; Hoehn et al . , 2008 ) . We envisage that the described feedback mechanism may have co-evolved with IRS proteins as an obligate intermediate between the IR and PI3K to enhance the signalling capacity of the IR in several ways . First , it may provide the basis for more precise temporal control of insulin signalling . This may be pertinent in that insulin’s principal role as a metabolic regulator occurs over a timescale of minutes , while the control of other biological processes such as proliferation or differentiation may not require such fine control as they have different temporal demands . Second , it provides means to specifically regulate insulin signalling whilst keeping PI3K activation by other signalling pathways intact . Finally , as IRS1/2 can be phosphorylated by a range of other kinases ( Copps and White , 2012 ) , it is possible that the described mechanism transcends to other kinases and enables crosstalk between different growth factor pathways . Previous studies have identified mTORC1/S6K-dependent feedback signals that regulate Akt activity ( Carlson et al . , 2004; Copps and White , 2012; Harrington et al . , 2004; Shah et al . , 2004; Shah and Hunter , 2006; Tremblay and Marette , 2001 ) . However , the feedback signal reported here does not need mTORC1 activation; rapamycin had no effect on acute Akt recruitment , phosphorylation ( Figure 1G , H ) , or IRS1/2 localisation ( Figure 5C , D ) . Rather , our data suggest an Akt-derived feedback signal that is activated acutely ( ~1 min ) following growth factor exposure . We suspect that described negative feedbacks emanating from mTORC1/S6K are much slower than the negative feedback described in this study , particularly those that rely on the degradation of IRS . Recent findings by our lab and others have shown that in the absence of S474 phosphorylation , Akt is active and sustains T309 phosphorylation ( Beg et al . , 2017; Jacinto et al . , 2006; Kearney et al . , 2019 ) . Indeed , here we show that negative feedback is engaged in the absence of S474 phosphorylation ( Figure 2I ) . Correspondingly , we have previously indicated a role for Akt phosphorylated at T309 alone in driving a positive feedback loop to activate mTORC2 and enhance S474 phosphorylation ( Yang et al . , 2015 ) . Collectively these findings demonstrate that Akt phosphorylated at T309 , but not S474 is capable of substrate phosphorylation and sufficient for eliciting feedback mechanisms . Losing negative feedback from Akt to IRS resulted in a profound increase in PIP3 levels ( Figure 3A ) . The catastrophic consequences of such an increase in signal flow is exemplified in cancer . Hyperactivation of Akt due to upstream genetic lesions ( e . g . , PI3K , PTEN ) or mutation in Akt itself ( e . g . , E17K ) can result in uncontrolled regulation of processes such as cell proliferation and survival ( Altomare and Testa , 2005 ) . As Akt is frequently hyperactivated in human cancers , it has been the target of numerous cancer therapeutics . However , despite the perceived potential , no Akt inhibitors have been approved for use ( Nitulescu et al . , 2018 ) . Akt inhibitors such as MK2206 and GDC0068 have been tested in clinical trials; however , severe side effects were experienced and substantial tumour shrinkage was generally not observed ( Saura et al . , 2017; Xing et al . , 2019; Yap et al . , 2011 ) . The data presented herein demonstrate that the therapeutic targeting of kinases such as Akt can have unexpected effects on other signalling networks , resulting from loss of feedback and crosstalk . For example , inhibition of Akt increased insulin-stimulated PIP3 levels by more than fivefold ( Figure 3A ) and enhanced PM recruitment of PH domain containing proteins PDPK1 and GAB2 ( Figure 3B–D ) . PDPK1 is the master regulator of other kinases such as PKC , S6K , and SGK , and so PDPK1 mislocalisation at the PM would likely influence these networks . Thus , the central importance of PIP3 beyond Akt signalling ( Czech , 2000 ) may have contributed to the lack of efficacy Akt inhibitors have had in the clinic . These data highlight the need for a more detailed understanding of feedback and crosstalk across signalling networks in order to generate effective therapeutics . We constructed nine mechanistic models to investigate different possible network structures of the PI3K/Akt signalling pathway . All models included components representing proximal insulin signalling proteins ( IR , IRS , PI3K ) . The activation of this pathway is initiated by insulin binding the IR . Then , IRS bind the IR and recruit PI3K to the PM . In the model , the IRS/PI3K node represents proximal insulin signalling . Activated PI3K phosphorylates PIP2 , converting it to PIP3 . This is negatively regulated by PTEN . In the model , Akt is activated by PDPK1 and mTORC2 through phosphorylation at T309 and S474 , respectively . mTORC2 has two independent activation mechanisms: ( 1 ) binding of SIN1 to PIP3 that releases its inhibition on mTOR kinase activity ( Liu et al . , 2015 ) and ( 2 ) phosphorylation of SIN1 T86 residue by Akt ( Humphrey et al . , 2013; Yang et al . , 2015 ) . Activated Akt phosphorylates PRAS40 , which results in mTORC1 activation . In the model , Akt singly phosphorylated at T309 , and doubly phosphorylated at T309 and S474 has kinase activity . However , Akt singly phosphorylated at S474 is not active . All models were trained with the data in Figure 1B and C , D ( pAkt T309 , pAkt S474 , and pPRAS40 T246 ) . Additionally , model 1 included negative feedback from mTORC1 to IRS/PI3K . Model 2 included negative feedback from mTORC1 to IRS/PI3K and was also trained with the data in Figure 1G . Model 3 incorporated no negative feedback . Model 4 included negative feedback from PDPK1 to IRS/PI3K . Model 5 included negative feedback from PDPK1 to PTEN . Model 6 included negative feedback from mTORC2 to IRS/PI3K . Model 7 included negative feedback from mTORC2 to PTEN . Model 8 included negative feedback from Akt to PTEN . Model 9 included negative feedback from Akt to IRS/PI3K . Detailed schematic diagrams of these models are illustrated in Figures 1A and 2A . These models were formulated using ODEs . The rate equations , ODEs , and the best-fitted parameter sets for each network model are given in Supplementary file 1 . The code for the modelling has been deposited to Github ( Ghomlaghi et al . , 2021 ) . The model construction and calibration processes were implemented in MATLAB ( The MathWorks Inc 2019a ) and the IQM toolbox ( http://www . intiquan . com/intiquan-tools/ ) was used to compile the IQM file for a MEX file which makes the simulation much faster . The quality of a mathematical model is generally justified by its ability to recapitulate known experimental data . Model calibration ( or model training ) is the process of estimation of the model’s parameters . This process produces a ‘best-fitted’ model that best recapitulates biological observations used for model calibration . Model calibration was done by estimating the model parameter values to minimise an objective function J that quantifies the difference between model simulation results and corresponding experimental measurements:J ( p ) =∑j=1Mwj∑i=1Nyj , iD-yj ( ti , p ) σj , i2 Here , M denotes number of available experimental data sets for fitting and N is the number of time points in each experimental data set . yj ( ti , p ) is simulation result of the model for the component j in the network at the time point ti while parameter set p is used for the simulation . Finally , yj , iD is the mean value of the experimental data of component j at time point ti with the error variance σj , i . wj is the weight of the component j . A genetic algorithm ( GA ) was used to optimise the objective function ( Man et al . , 1996; Reali et al . , 2017; Shin et al . , 2014 ) . This was done by using the Global Optimization Toolbox and the function ga in MATLAB . Selection rules in GA select the individual solutions with the best fitness values ( called ‘elite solutions’ ) from the current population . The elite count was set to 5% of the population size . Crossover rules combine two parents to generate offspring for the next generation . The crossover faction was set at 0 . 8 . Mutation rules apply random changes to individual parents to generate the population of the next generation . For the mutation rule , we generated a random number from a Gaussian distribution with mean 0 and standard deviation σk , which was applied to the individuals of the current generation . The standard deviation function ( σk ) is given by the recursive formula as follows:σk=σk-1 ( 1-kG ) where k is the kth generation , G is the number of generation , and σ0=1 . To derive the best-fitted parameter sets , we carried out repeated GA runs with population size of 200 and the generation number set to 800 . Identifiability analysis for all tested models ( models 4–9; Figure 2A ) was performed using a well-established method based on profile likelihood ( Maiwald et al . , 2016; Rateitschak et al . , 2012; Raue et al . , 2009 ) . This method is able to detect both structural and practical non-identifiable parameters through calculating confidence intervals defined by a threshold in the profile likelihoods , as detailed previously ( Rateitschak et al . , 2012 ) . A parameter is considered to be identifiable if the confidence interval of its estimate is finite ( Rateitschak et al . , 2012 ) . Another important advantage of this approach is that it is computationally efficient and thus suitable for medium-to-large ( non-minimal ) models such as those considered in this paper ( Rateitschak et al . , 2012 ) . For normally distributed observational noise , this function corresponds to the maximum likelihood estimate of θ . The profile likelihood of a parameter θ is given by Maiwald et al . , 2016; Rateitschak et al . , 2012; Raue et al . , 2009:χPL2 ( θi ) =minθj≠i⁡χ2 ( θ ) which represents a function in θi of least increase in the residual sum of squares χ ( θ ) . Profile likelihood-based confidence interval ( CI ) can be derived via:CI ( θ ) ={θ\|χPL2 ( θ ) −χPL2 ( θ^ ) <Δα}where Δα=χ2 ( α , df ) is the threshold , α is a confidence level ( the α quantile of the χ2 distribution ) , df is the degree of freedom ( df=1 for pointwise confidence interval and df=#ofparameters for simultaneous confidence intervals , respectively ) . θ^ denotes the best-fitted parameter set . Supplementary file 2 contains the identifiability analysis results for models 4–9 . In each plot , the black dashed lines depict pointwise confidence levels of α=95% . The results show for each parameter whether it is structurally ( indicated by a flat curve in both directions ) or practically ( indicated by a flat curve only in one direction ) non-identifiable ( Rateitschak et al . , 2012 ) . Overall , these results demonstrate that all the models are non-identifiable , meaning each model has at least one non-identifiable parameter . As most large ODE-based models in systems biology are unidentifiable to some extent , the results here were expected given the detailed scope of our models , which were designed to capture the important biological mechanisms within the insulin signalling network , including multiple feedback/feedforward loops . Because generally there is a trade-off between model identifiability and level of biological details , although the models in this study could be made more identifiable through model abstraction , such a process is inevitably at the cost of sacrificing specific biological details and may weaken the models’ explanatory and predictive power . TagRFP-T-Akt2 consists of human Akt2 tagged with TagRFP-T at its N-terminus as described previously ( Norris et al . , 2017 ) . TagRFP-T-Akt2 W80A , W80A-T309A , W80A-S474A , W80A-T309A-S474A , and W80A-K181A were generated using site-directed mutagenesis ( Sanchis et al . , 2008 ) . PDPK1-eGFP consists of human PDPK1 tagged with eGFP at its C-terminus . To generate PDPK1-eGFP , R777-E159 Hs . PDPK1 was kindly gifted from Dominic Esposito ( Addgene plasmid #70443 ) and was used as a template to amplify human PDPK1 . Human PDPK1 was placed in the pEGFP-C1 vector using Gibson assembly cloning ( Gibson , 2011 ) . PH-Gab2-GFP was a gift from Sergio Grinstein ( Addgene plasmid #35147 ) . Constitutively active p110 ( p110* ) was kindly gifted by Morris Birnbaum . pMIG FLAG-W80A and FLAG-W80A-T309A Akt2 were generated as previously described ( Kearney et al . , 2019 ) . IRS1-eGFP consists of human IRS1 tagged with eGFP at its C-terminus . To generate IRS1-eGFP , R777-E109 Hs . IRS1 was kindly gifted from Dominic Esposito ( Addgene plasmid #70393 ) and was used as a template to amplify human IRS1 . Human IRS1 was placed in the pEGFP-C1 vector using Gibson assembly cloning ( Gibson , 2011 ) . IRS2-eGFP consists of human IRS2 tagged with eGFP at its C-terminus . To generate IRS2-eGFP , R777-E111 Hs . IRS2 was kindly gifted from Dominic Esposito ( Addgene plasmid #70395 ) and was used as a template to amplify human IRS2 . Human IRS2 was placed in the pEGFP-C1 vector using Gibson assembly cloning ( Gibson , 2011 ) . Mutations in IRS1-eGFP and IRS2-eGFP were generated using site-directed mutagenesis ( Sanchis et al . , 2008 ) . DelPH IRS1-eGFP consists of IRS1 without its PH domain ( residues 2–115 removed , based on Dhe-Paganon et al . , 1999 ) and was generated using Gibson assembly cloning ( Gibson , 2011 ) . DelPTB IRS1-eGFP consists of IRS1 without its PTB domain ( residues 161–265 removed , based on Eck et al . , 1996 ) and was generated using Gibson assembly cloning ( Gibson , 2011 ) . To generate IRS2-FLAG , IRS2-eGFP was used as a template , and eGFP replaced with a FLAG tag ( DYKDDDDK ) using Gibson assembly cloning ( Gibson , 2011 ) . To generate PDPK1-TagRFP-T , TagRFP-T-Akt2 was used as a template to amplify TagRFP-T , and this was inserted into PDPK1-eGFP to replace eGFP , using Gibson assembly cloning ( Gibson , 2011 ) . Plasmid and primer DNA sequences are provided in Supplementary file 3 . 3T3-L1 fibroblasts obtained from the Howard Green Laboratory ( Harvard Medical School ) were cultured in high glucose Dulbecco’s modified eagle medium ( DMEM ) ( Gibco by Life Technologies ) supplemented with 10% ( v/v ) fetal bovine serum ( FBS ) ( Gibco by Life Technologies ) , and 1× GlutaMAX ( Gibco by Life Technologies ) at 37°C and 10% CO2 . Cells were differentiated into adipocytes as described previously ( Fazakerley et al . , 2015; Norris et al . , 2018 ) and used for experiments 7–12 days after initiation of differentiation . 3T3-L1 adipocytes stably expressing FLAG-W80A or FLAG-W80A-T309A Akt2 were generated using retrovirus as previously described and characterised previously ( Kearney et al . , 2019 ) . HEK293E , HeLa , and HCC1937 cell lines were obtained from the American Type Culture Collection and grown in the medium described above to culture 3T3-L1 cells . MCF7 cells were a gift from Associate Prof . Jeff Holst ( Centenary Institute ) and were validated by STR . Cells were maintained in Minimum Essential Media ( Gibco by Life Technologies Cat#10370–021 ) , with the addition of 10% ( v/v ) FBS , 1× GlutaMAX , and 1 mM sodium pyruvate ( Gibco by Life Technologies ) at 37°C and 5% CO2 . Cells were routinely tested for mycoplasma contamination and found to be contamination-free . 3T3-L1 adipocytes were serum-starved with DMEM containing 1× GlutaMAX and 0 . 2% BSA ( w/v ) for 2 hr . Cells were then exposed to drugs/insulin . Cells were then placed on ice , washed with cold PBS , lysed with 1% ( w/v ) SDS in PBS containing protease inhibitors ( Roche Applied Science ) and phosphatase inhibitors ( 2 mM Na3VO4 , 1 mM Na4O7P2 , and 10 mM NaF ) , and tip probe-sonicated . Lysates were centrifuged at 13 , 000× g for 15 min at 4°C . The lipid layer was removed , and protein content was quantified using the Pierce BCA Protein Assay Kit ( Thermo Scientific ) ; 10 μg of lysate was then resolved by SDS-PAGE and transferred to PVDF membranes . Membranes were blocked and immunoblotted as described previously ( Fazakerley et al . , 2015 ) . Densitometry analysis was performed using ImageStudioLite version 5 . 2 . 5 ( LI-COR ) . Band intensities were normalised to the loading control . Statistical tests were performed using GraphPad Prism version 7 . 0 . For the quantification of the blots in Figure 1C , D , the time courses were normalised to the mean intensity of all samples ( within a blot ) . Next , biological replicates were normalised to the maximum of the mean of all responses ( across blots ) within a dose . As some of the 1 and 100 nM time courses were acquired separately , the difference in magnitude between the doses was determined by the three biological 1 and 100 nM replicates that were acquired concurrently and run on the same gels . The representative blot is an example of a paired experiment . 3T3-L1 adipocytes were electroporated 6–8 days post-differentiation with 6–10 μg of plasmid and placed onto the Matrigel-coated µ-Dish 35 mm , high Glass Bottom coverslips ( Ibidi ) as described previously ( Norris et al . , 2017 ) . For other cell types , cells were transfected using Lipofectamine 2000 ( Thermo Scientific ) . Twenty-four hours later , cells were serum-starved for 2 hr and then incubated at 37°C with Krebs-Ringer-phosphate-HEPES buffer ( 0 . 6 mM Na2HPO4 , 0 . 4 mM NaH2PO4 , 120 mM NaCl , 6 mM KCl , 1 mM CaCl2 , 1 . 2 mM MgSO4 , and 12 . 5 mM HEPES [pH 7 . 4] ) supplemented with 10 mM glucose , 1× minimum essential medium amino acids ( Gibco by Life Technologies ) , 1× GlutaMAX , and 0 . 2% ( w/v ) BSA . While imaging , temperature and humidity were then maintained using an Okolab cage incubator and temperature control . The cells were treated using a custom-made perfusion system . Images were acquired with a CFI Apochromat TIRF 60× oil , NA 1 . 49 objective , using the Nikon Ti-LAPP H-TIRF module angled to image ∼90 nm into cells . Images were acquired approximately every 15 s . To quantify changes in the PM recruitment of each protein of interest , we measured the average pixel intensity ( and subtracted background intensity ) for each cell over the time course using Fiji ( Schindelin et al . , 2012 ) . Each cellular response to stimuli was normalised to its average intensity over the basal period . The cell-to-cell heterogeneity in Akt recruitment responses ( described previously; Norris et al . , 2021 ) can make the comparison of several large population TIRF responses difficult to interpret if presented as mean ± SD . These data are presented as mean ± SEM to aid interpretation . Rate constants ( Figure 4B–D ) were calculated using Graphpad Prism version 7 . 0 , by fitting an exponential curve to the data ( plateau followed by one-phase decay ) . To assess the relative cell surface level of IRS1 ( Figure 5F ) , TIRF and epifluorescence images were acquired and for each cell its median TIRF intensity ( corrected for background ) was normalised to its median epifluorescence intensity ( corrected for background ) using Fiji ( Schindelin et al . , 2012 ) . 3T3-L1 adipocytes were serum-starved with DMEM containing 1× GlutaMAX and 0 . 2% ( w/v ) BSA for 2 hr . Cells were then preincubated with drugs or vehicle controls and stimulated with 1 nM insulin . Lipids were extracted from cells and measured using an ELISA kit ( Echelon Biosciences ) . 1 × 10 cm dish of 3T3-L1 adipocytes was used for each sample . For each sample , PIP3 mass was normalised to PIP ( 4 , 5 ) P2 mass . PIP ( 4 , 5 ) P2 is highly abundant in cells ( Guillou et al . , 2007 ) and thus is only marginally affected by changes in PI3K activity ( Condliffe et al . , 2005 ) . As has been done previously ( Clark et al . , 2011; Costa et al . , 2015; Guillou et al . , 2007 ) , we normalised to PI ( 4 , 5 ) P2 mass to account for differences in extraction efficiency between samples , and control for total cellular phosphoinositides . 3T3-L1 adipocytes were serum-starved for 2 hr , and then exposed to DMSO , 10 μM MK2206 , or 10 μM GDC0068 for 5 min , followed by 1 nM insulin for 10 min . Cells were placed on ice , washed with cold PBS , and harvested in cold HES buffer ( 20 mM HEPES , 1 mM EDTA , 250 mM sucrose , pH 7 . 4 ) containing phosphatase ( 2 mM Na3VO4 , 1 mM Na4O7P2 , 10 mM NaF ) and protease ( Roche Applied Science ) inhibitors . All subsequent steps were carried out at 4°C . Cells were homogenised by passing through a 22-gauge needle 10 times and a 27-gauge needle six times prior to centrifugation at 500× g for 10 min . The supernatant was centrifuged at 13 , 550× g for 12 min to pellet the PM and mitochondria/nuclei , while the supernatant contained the cytosol , LDM fraction , and HDM fraction . The supernatant was then centrifuged at 21 , 170× g for 17 min to pellet the HDM fraction . That supernatant was then centrifuged at 235 , 200× g for 75 min to obtain the cytosol fraction ( supernatant ) and the LDM fraction ( pellet ) . The PM and mitochondria/nuclei pellet was resuspended in HES buffer and again centrifuged at 13 , 550× g for 12 min . The pellet was then resuspended in HES buffer , layered over high sucrose HES buffer ( 20 mM HEPES , 1 mM EDTA , 1 . 12 M sucrose , pH 7 . 4 ) , and centrifuged at 111 , 160× g for 60 min in a swing-out rotor . The PM fraction was collected at the interface between the sucrose layers , and pelleted by centrifugation at 235 , 200× g for 75 min . All pellets were resuspended in HES buffer containing phosphatase and protease inhibitors . Protein concentrations for each fraction were determined using the Pierce BCA Protein Assay Kit ( Thermo Scientific ) . 3T3-L1 adipocytes were seeded onto Matrigel-coated eight-well microslides ( Ibidi ) . Forty-eight hours later , cells were serum-starved for 2 hr and then exposed to DMSO , 10 μM MK2206 , or 10 μM GDC0068 for 5 min , followed by 1 nM insulin for 10 min . The coverslips were then briefly immersed in ice-cold PBS and fixed with 4% paraformaldehyde in PBS at room temperature for 15 min . Cells were then washed twice with room temperature PBS and quenched with 200 mM glycine for 10 min . Cells were then blocked and permeabilised with 2% BSA/0 . 1% saponin in PBS for 30 min . Cells were incubated with the anti-PI3K p85 ( Cell Signaling Technology CST4257S ) primary antibody ( 1:100 in 2% BSA/0 . 1% saponin in PBS ) for 1 hr at room temperature . Cells were then washed with 2% BSA/0 . 1% saponin in PBS three times , and then incubated with anti-rabbit-IgG conjugated to Alexa Fluor 555 ( 1:200 in 2% BSA/0 . 1% saponin in PBS ) at room temperature for 1 hr in the dark . Cells were then washed five more with PBS and then stored and imaged in 5% glycerol/2 . 5% 1 , 4-diazabicyclo[2 . 2 . 2]octane in PBS . Images were acquired using the Nikon Ti-LAPP H-TIRF module angled to image ∼90 nm into cells . To quantify relative changes in PM PI3K p85 , for each cell we measured the median pixel intensity ( corrected for background ) using Fiji ( Schindelin et al . , 2012 ) . 3T3-L1 adipocytes were serum-starved for 2 hr and then exposed to DMSO , 10 μM MK2206 , or 10 μM GDC0068 for 5 min , followed by 1 nM insulin for 10 min . Cells were washed three times with ice-cold PBS and lysed in cold lysis buffer ( 1% ( v/v ) NP40 , 10% ( v/v ) glycerol , 137 mM NaCl , 25 mM Tris pH 7 . 4 ) containing phosphatase ( 2 mM Na3VO4 , 1 mM Na4O7P2 , 10 mM NaF ) and protease ( Roche Applied Science ) inhibitors . All subsequent steps were performed at 4°C . Lysates were passed through a 22-gauge needle 10 times , followed by a 27-gauge needle six times . Lysates were then solubilised for 15 min on ice prior to centrifugation at 18 , 000× g for 20 min to remove lipid and cell debris; 850 μg of each supernatant was then incubated with 2 μL of antibody ( IRS1; Cell Signaling Technology CST3407S or IRS2; Cell Signaling Technology CST3089S ) or the same amount of Rabbit IgG control ( Santa Cruz ) for 2 hr with rotation . 50 μL of Dynabeads Protein G ( Invitrogen ) were added into each antibody-lysate mixture and incubated for 2 hr with rotation . Beads were washed once with lysis buffer and then four times with PBS . Beads were incubated in 25 μL elution buffer 1 ( 2 M urea , 5 mM TCEP , 20 mM 2-chloroacetamide , 5 μg/mL trypsin , 50 mM Tris-HCl , pH 7 . 5 ) for 30 min at room temperature , and then 100 μL elution buffer 2 ( 2 M urea , 50 mM Tris-HCl , pH 7 . 5 ) was added . Eluate was collected into a LowBind Eppendorf tube and digested for 16 hr at room temperature . Peptides were then acidified by adding TFA to a final concentration of 1% ( v/v ) and stored at 4°C prior to LC-MS/MS . IRS2-FLAG was transfected into HEK293E cells using Lipofectamine 2000 ( Thermo Scientific ) . Twenty-four hours later , cells were serum-starved for 2 hr and treated with 10 μM MK2206 for 30 min . Cells were placed on ice , washed with cold PBS , and harvested in cold lysis buffer ( 1% ( v/v ) NP40 , 10% ( v/v ) glycerol , 137 mM NaCl , 25 mM Tris pH 7 . 4 ) containing phosphatase ( 2 mM Na3VO4 , 1 mM Na4O7P2 , 10 mM NaF ) and protease ( Roche Applied Science ) inhibitors . All subsequent steps were carried out at 4°C . Cells were homogenised by passing through a 22-gauge needle 10 times and a 27-gauge needle six times prior to solublisation on ice for 15 min . Then samples were centrifuged at 18 , 000× g for 20 min; 1 mg of the supernatant was incubated with 5 μg of anti-FLAG antibody ( Sigma-Aldrich ) , on a rotator for 2 hr . Then , this was mixed with protein G agarose beads ( GE Healthcare ) on a rotator for a further 2 hr . Beads were washed four times with lysis buffer , once with kinase buffer ( 25 mM Tris-HCl [pH 7 . 5] , 10 mM MgCl2 , 5 mM beta-glycerophosphate , 0 . 1 mM Na3VO4 , 1 mM DTT ) , and then dried . To elute the protein from the beads , 0 . 4 μg/μL of 3× FLAG peptide ( Sigma-Aldrich ) in kinase buffer was added to each sample and incubated for 1 hr with rapid agitation ( 1500 rpm using an Eppendorf ThermoMixer C ) . The eluate was removed . To determine the concentration of IRS2-FLAG obtained , an aliquot of eluate and Albumin standards ( Thermo Scientific ) were resolved by SDS-PAGE and stained using SYPRO Ruby Protein Gel Stain ( Bio-Rad ) . Remaining eluate was stored at −20°C for further analysis . 100 ng of immunoprecipitated IRS2-FLAG protein , 30 ng of recombinant active Akt2 ( SignalChem ) , and 100 μM ATP were mixed in kinase buffer ( 25 mM Tris-HCl [pH 7 . 5] , 10 mM MgCl2 , 5 mM beta-glycerophosphate , 0 . 1 mM Na3VO4 , 1 mM DTT ) . Samples were incubated at 30°C for 1 hr with rapid agitation ( 500 rpm ) . Samples were placed at 70°C for 10 min , then cooled on ice to room temperature . Proteins were reduced and alkylated by the addition of 10 mM TCEP ( Thermo Scientific , Bond-Breaker TCEP solution , Neutral pH ) and 40 mM 2-chloroacetamide ( Sigma-Aldrich ) and incubated at 45°C for 5 min . Samples were cooled on ice to room temperature , and 1% ( w/v ) SDC ( in 25 mM Tris pH 7 . 5 ) added to the samples; 10 ng of trypsin and 10 ng of LysC were added and samples shaken at 37°C with rapid agitation ( 2000 rpm ) for 18 hr . Samples were then mixed with equal volume of 1% TFA in ethyl acetate ( 45 µL ) and vortexed to dissolve precipitated SDC . Peptides were desalted using StageTips ( Rappsilber et al . , 2003 ) using SDB-RPS solid-phase extraction discs ( Empore ) . Briefly , 200 µL tips were packed with two layers of SDB-RPS material and placed into a 3D-printed 96-well StageTips adapter ( Harney et al . , 2019 ) for centrifugation . StageTips were equilibrated with sequential 50 µL washes of 100% acetonitrile , 30% MeOH with 1% TFA , and 0 . 2% TFA in water by centrifugation at 1000× g for 2 min . Peptides were then loaded onto the StageTips by centrifugation at 1000× g for 2 min . StageTips were washed sequentially with 1% TFA in ethyl acetate , 1% TFA in isopropanol and 0 . 2% TFA in 5% acetonitrile , and eluted into PCR strip tubes with 5% ammonium hydroxide in 60% ACN . Peptides were concentrated to dryness in a vacuum concentrator at 45°C for 30 min . Peptides were resuspended in MS loading buffer ( 10 µL 3% ACN/0 . 1% TFA ) prior to LC-MS/MS analysis . For GDC0068/MK2206 experiments , 3T3-L1 adipocytes were serum-starved for 2 hr and then exposed to vehicle ( DMSO ) , 10 μM MK2206 or 10 μM GDC0068 for 5 min , followed by 1 nM insulin for 10 min . Cells were harvested in ice-cold SDC lysis buffer ( 4% sodium deoxycholate/100 mM Tris pH 8 . 5 ) , boiled at 95°C for 5 min , centrifuged at 18 , 000× g for 15 min , and the layer of fat removed prior to determining protein concentration by BCA assay . For rapamycin/rapalink experiments , HEK-293E cells were maintained in DMEM , with 4 . 5 g glucose/L , 2 mM L-GlutaMAX , and 10% FBS . Cells were passaged for six doublings in stable isotope labelling by amino acids in cell culture ( SILAC ) DMEM containing three different isotopic versions of lysine and arginine supplemented with 10% dialysed FBS , generating ‘triple‐labelled’ SILAC cells ( Ong et al . , 2002 ) . Cells were serum‐starved for 4 hr together with either 100 nM rapamycin , 3 nM RapaLink1 , or vehicle ( DMSO ) , and then treated with 100 nM insulin for 10 min . Experiments were performed with four biological replicates and label switching . Cells were harvested in ice-cold GdmCl lysis buffer ( 6 M GdmCl , 100 mM Tris pH 8 . 8 , 10 nM TCEP , 40 mM CAA ) . Protein concentration was estimated by BCA assay , and SILAC samples mixed accordingly in equal ratios . Samples were processed using the EasyPhos method ( Humphrey et al . , 2015a ) and phosphopeptides were resuspended in MS loading buffer ( 0 . 3% v/v TFA/2% ( v/v ) ACN ) prior to LC-MS/MS . For endogenous IRS1/2 immunoprecipitation and IRS2-FLAG in vitro kinase assay samples , peptides were analysed using a Dionex HPLC coupled to a Q-Exactive HF-X benchtop Orbitrap mass spectrometer ( Thermo Fisher Scientific ) . Peptides were injected onto an in-house packed 75 µm ID × 40 cm column packed with 1 . 9 μm C18 ( ReproSil Pur C18-AQ , Dr Maisch ) and separated by a binary gradient of buffer A ( 0 . 1% formic acid ) and buffer B ( 0 . 1% formic acid/80% ACN ) . Peptides were separated by a gradient of 5–30% ( IP ) or 5–40% ( in vitro kinase assay ) buffer B at a flow rate of 300 or 400 nL/min . Eluting peptides were directly analysed with one full scan ( 350–1400 m/z , R = 60 , 000 ) . The top 5 ( in vitro kinase assay ) or 15 ( IP ) most intense precursors were fragmented with a collision energy of 27% and MS2 spectra collected at a resolution of 15 , 000 . For phosphoproteomics , phosphopeptides were loaded onto in-house fabricated 40 cm columns with a 75 µM inner diameter , packed with 1 . 9 µM C18 ReproSil Pur AQ particles ( Dr Maisch GmbH ) using EASY-nLC 1000 HPLC . Column temperature was maintained at 60°C using a column oven ( Sonation , GmbH ) . Peptides were separated using a binary buffer system comprising 0 . 1% formic acid ( buffer A ) and 80% ACN plus 0 . 1% formic ( buffer B ) , at a flow rate of 350 nL/min , with a gradient of 3–19% or 3–20% buffer B over 60 min or 85 min , respectively ( for the GDC0068/MK2206 and rapamycin/rapalink experiments ) , followed by 19–41% or 20–45% buffer B over 30 or 45 min , resulting in gradients of approximately 1 . 5 or 2 hr . Peptides were analysed on a Q Exactive HF or HF-X benchtop Orbitrap mass spectrometer ( Thermo Fisher Scientific ) , with one full scan ( 300–1600 m/z , R = 60 , 000 ) at a target of 3e6 ions , followed by up to 5 ( HF ) or 10 ( HF-X ) data-dependent MS/MS scans with higher-energy collisional dissociation ( HCD ) fragmentation . MS2 scan settings: target 1e5 ions , max ion fill time 120 ms ( HF ) or 50 ms ( HF-X ) , isolation window 1 . 6 m/z , normalised collision energy ( NCE ) 25% ( HF ) or 27% ( HF-X ) , intensity threshold 3 . 3e5 ions ( HF ) or 4e5 ions ( HF-X ) , with fragments detected in the Orbitrap ( R = 15 , 000 ) . Dynamic exclusion ( 40 s , HF; 30 s HF-X ) and apex trigger ( 4–7 s , HF; 2–4 s HF-X ) were enabled . RAW MS data was analysed using using MaxQuant ( Cox and Mann , 2008 ) with searches performed against the UniProt database ( June 2020 [in vitro kinase assay] , June 2019 [IRS1/2 IP] , November 2016 [rapamycin/rapalink phosphoproteome] , and March 2018 [GDC0068/MK2206 phosphoproteome] releases ) with a false discovery rate of <0 . 01 at the protein , peptide , site , and PSM levels . Default settings in MaxQuant were used , with the addition of ‘Phospho ( STY ) ’ as a variable modification and SILAC labels ( Arg 0/Lys 0 , Arg 6/Lys 4 , Arg 10 , Lys 8 ) for the rapamycin/rapalink phosphoproteome samples . ‘Match between runs’ was enabled with a 0 . 7 min match time window ( default ) . Data was filtered , normalised , and analysed using R , Tableau Prep , and Graphpad Prism . For the phosphoproteomics with GDC0068/MK2206 , LFQ Intensities were log2-transformed and median normalised . For endogenous IRS1/2 immunoprecipitation , intensities were log2-transformed and then normalised to the intensity of IRS in each sample . Then , values were normalised to the mean of the insulin treated sample . RAW and MaxQuant processed data have been deposited in the PRIDE proteomeXchange repository and can be accessed at https://www . ebi . ac . uk/pride/archive/ , using the accession PXD023441 . The reactions , rate equations , differential equations , and parameter sets required to reproduce the models can be found in Supplementary file 1 . The code for the modelling has been deposited to Github ( Nguyen Lab , 2021a , copy archived at swh:1:rev:09b5d4f838bf60e790c10843fec901516845d7e2 , Nguyen Lab , 2021b ) . Plasmids generated in this study will be made available upon request . Any further information and requests for resources should be directed to james . burchfield@sydney . edu . au or david . james@sydney . edu . au .	For the body to work properly , cells must constantly ‘talk’ to each other using signalling molecules . Receiving a chemical signal triggers a series of molecular events in a cell , a so-called ‘signal transduction pathway’ that connects a signal with a precise outcome . Disturbing cell signalling can trigger disease , and strict control mechanisms are therefore in place to ensure that communication does not break down or become erratic . For instance , just as a thermostat turns off the heater once the right temperature is reached , negative feedback mechanisms in cells switch off signal transduction pathways when the desired outcome has been achieved . The hormone insulin is a signal for growth that increases in the body following a meal to promote the storage of excess blood glucose ( sugar ) in muscle and fat cells . The hormone binds to insulin receptors at the cell surface and switches on a signal transduction pathway that makes the cell take up glucose from the bloodstream . If the signal is not engaged diseases such as diabetes develop . Conversely , if the signal cannot be adequately switched of cancer can develop . Determining exactly how insulin works would help to understand these diseases better and to develop new treatments . Kearney et al . therefore set out to examine the biochemical ‘fail-safes’ that control insulin signalling . Experiments using computer simulations of the insulin signalling pathway revealed a potential new mechanism for negative feedback , which centred on a molecule known as Akt . The models predicted that if the negative feedback were removed , then Akt would become hyperactive and accumulate at the cell’s surface after stimulation with insulin . Further manipulation of the ‘virtual’ insulin signalling pathway and studies of live cells in culture confirmed that this was indeed the case . The cell biology experiments also showed how Akt , once at the cell surface , was able to engage the negative feedback and shut down further insulin signalling . Akt did this by inactivating a protein required to pass the signal from the insulin receptor to the rest of the cell . Overall , this work helps to understand cell communication by revealing a previously unknown , and critical component of the insulin signalling pathway .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "computational", "and", "systems", "biology" ]	2021	Akt phosphorylates insulin receptor substrate to limit PI3K-mediated PIP3 synthesis
Secretory and endolysosomal fusion events are driven by SNAREs and cofactors , including Sec17/α-SNAP , Sec18/NSF , and Sec1/Munc18 ( SM ) proteins . SMs are essential for fusion in vivo , but the basis of this requirement is enigmatic . We now report that , in addition to their established roles as fusion accelerators , SM proteins Sly1 and Vps33 directly shield SNARE complexes from Sec17- and Sec18-mediated disassembly . In vivo , wild-type Sly1 and Vps33 function are required to withstand overproduction of Sec17 . In vitro , Sly1 and Vps33 impede SNARE complex disassembly by Sec18 and ATP . Unexpectedly , Sec17 directly promotes selective loading of Sly1 and Vps33 onto cognate SNARE complexes . A large thermodynamic barrier limits SM binding , implying that significant conformational rearrangements are involved . In a working model , Sec17 and SMs accelerate fusion mediated by cognate SNARE complexes and protect them from NSF-mediated disassembly , while mis-assembled or non-cognate SNARE complexes are eliminated through kinetic proofreading by Sec18 . Membrane fusion , the final stage of intracellular vesicular traffic , is tightly regulated so that cargos are delivered to destination compartments in an accurate and timely manner ( Bonifacino and Glick , 2004 ) . The core proteins required for fusion are conserved from the yeast vacuole to the synaptic active zone ( Table 1 ) . These include compartment-specific SNARE and SM ( Sec1/Munc18 ) proteins , the SNARE disassembly ATPase Sec18/NSF ( N-ethylmaleimide-sensitive factor ) , and its essential recruitment adapter Sec17/α-SNAP ( soluble NSF attachment protein; Jackson and Chapman , 2008; Jahn and Fasshauer , 2012; Südhof and Rothman , 2009; Ungar and Hughson , 2003; Wickner and Schekman , 2008 ) . 10 . 7554/eLife . 02272 . 003Table 1 . Nomenclature of general and compartment-specific SNAREs and SNARE cofactors employed in this study , and their equivalents in mammalian synaptic exocytosisDOI: http://dx . doi . org/10 . 7554/eLife . 02272 . 003YeastMammalGeneralGeneralAAA-family ATPaseSec18NSFSec18 adapterSec17α-SNAPGolgiVacuoleSynaptic exocytosisSM proteinSly1Vps33Munc18-1Qa-SNARESed5Vam3SyntaxinQb-SNAREBos1Vti1SNAP-25 ( N-domain ) Qc-SNAREBet1Vam7SNAP-25 ( C-domain ) R-SNARESec22Nyv1Synaptobrevin ( VAMP2 ) The Q/R taxonomy of SNARE domain subfamilies is derived from Fasshauer et al . ( 1998 ) . During docking , SNAREs on apposed vesicle and target membranes oligomerize in trans , ‘zippering’ into an ultrastable coiled-coil bundle . SNARE zippering pulls the membranes into tight apposition , locally deforming and dehydrating the bilayers to initiate fusion and compartmental mixing ( Hanson et al . , 1997; Nichols et al . , 1997; Sutton et al . , 1998; Fasshauer et al . , 2002 ) . Following fusion , individual SNAREs are entrapped within stable , fusion-inactive cis-complexes . To separate the SNAREs and energize them for additional instances of trans-complex assembly and membrane fusion , Sec17 binds the cis-SNARE complex , in turn recruiting Sec18 . Sec18 , a hexameric AAA-family ATPase , disassembles the cis-SNARE complex and ejects Sec17 ( Sollner et al . , 1993; Mayer et al . , 1996; Hanson et al . , 1997; Littleton et al . , 1998; Grote et al . , 2000; Wimmer et al . , 2001; Marz et al . , 2003; Cipriano et al . , 2013 ) . SNAREs alone can fuse membranes in vitro ( Weber et al . , 1998 ) , but fusion in vivo requires additional cofactors including regulatory small G proteins , compartment-specific tethers , and proteins of the SM family ( Jackson and Chapman , 2006; Wickner and Schekman , 2008; Südhof and Rothman , 2009; Yu and Hughson , 2010; Jahn and Fasshauer , 2012 ) . SM proteins are SNARE-interacting ∼600 residue proteins with a highly conserved tertiary fold ( Carr and Rizo , 2010; Rizo and Südhof , 2012 ) . Four SM subfamilies are essential for fusion within specific subcellular domains: ER and Golgi ( Sly1; Cao and Barlowe , 2000; Dascher et al . , 1991 ) ; plasma membrane ( Sec1/Munc18; Grote et al . , 2000; Harrison et al . , 1994; Hosono et al . , 1992; Novick et al . , 1981; Verhage et al . , 2000; Weimer et al . , 2003 ) ; endosomes ( Vps45; Cowles et al . , 1994; Piper et al . , 1994 ) ; and late endolysosomal organelles ( Vps33; Banta et al . , 1990; Wada et al . , 1990 ) . The in vivo SM requirement is so general and so stringent that SMs are now considered , along with SNAREs , to be components of the core fusion machinery ( Südhof and Rothman , 2009 ) . However , the biochemical mechanisms underlying the in vivo SM requirement are opaque . Various hypotheses have been proposed to explain the function of SMs in fusion . A major reason for the proliferation of models is that different SMs have divergent SNARE binding modalities ( Carr and Rizo , 2010; Rizo and Südhof , 2012 ) . However , accruing evidence suggests that SMs share a core ability to bind cognate ternary or quaternary SNARE bundles ( Carr et al . , 1999; Scott et al . , 2004; Carpp et al . , 2006; Togneri et al . , 2006; Dulubova et al . , 2007; Kramer and Ungermann , 2011; Lobingier and Merz , 2012 ) . These observations prompted the conjecture that the central , evolutionarily conserved function of SM proteins involves their direct association with assembling pre-fusion trans-SNARE complexes ( Carr and Rizo , 2010; Rizo and Südhof , 2012 ) . Indeed , Sec1 and Munc18-1 accelerate SNARE-mediated liposome fusion by several-fold ( Scott et al . , 2004; Shen et al . , 2007; Rathore et al . , 2010 ) . This acceleration is contingent on initial reaction conditions , and recent experiments show that Munc18-1 stimulates liposome fusion more efficiently in concert with the specialist exocytosis cofactors Munc13 and synaptotagmin ( Ma et al . , 2013 ) . Similarly , Vps33 , Vps45 , and Sly1 accelerate liposome fusion , but with nearly absolute requirements for additional factors including Rab proteins and tethering factors ( Hickey et al . , 2009; Ohya et al . , 2009; Furukawa and Mima , 2014 ) . It remains unclear how SM proteins accelerate SNARE-mediated fusion , and it is unknown whether the kinetic stimulation observed in vitro is sufficient to explain absolute requirements for SMs in vivo . In a different and not mutually exclusive role , SM proteins might interact with SNARE recycling factors . In vitro , Sec18 and Sec17 can disassemble pre-fusion trans-SNARE complexes and can prevent the fusion of intact yeast lysosomal vacuoles or liposomes ( Ungermann et al . , 1998; Rohde et al . , 2003; Mima et al . , 2008; Stroupe et al . , 2009 ) . Premature SNARE disassembly was impeded by the Vps-C tethering complex HOPS ( Xu et al . , 2010 ) . This protective activity of HOPS was hypothesized to reside within its SM subunit Vps33 , which is necessary and sufficient for HOPS binding to the vacuole SNARE complex ( Lobingier and Merz , 2012 ) . Similarly , Munc18-1 , acting in concert with Munc13 and synaptotagmin , facilitated SNARE-mediated liposome fusion in the presence of otherwise inhibitory concentrations of NSF and α-SNAP ( Sec18 and Sec17; Ma et al . , 2013 ) . In the absence of NSF and α-SNAP , Munc18 and Munc13 had little or no effect on the extent of SNARE-mediated liposome fusion when compared to fusion driven solely by SNAREs and synaptotagmin . These findings led to proposals that SMs protect pre-fusion SNARE complexes from premature disassembly while exposing post-fusion complexes and mis-assembled or non-cognate pre-fusion complexes to disassembly by Sec18 ( Mima et al . , 2008; Starai et al . , 2008; Rizo and Südhof , 2012 ) . However , central predictions of these models are still untested . Direct protection of a SNARE complex by an SM has not been experimentally demonstrated , and it is unknown whether SM proteins functionally interact ( or compete ) with the disassembly machinery in living cells . Using Saccharomyces cerevisiae as an experimental platform , we tested the hypothesis that SMs functionally interact not only with SNAREs , but also with Sec17 and Sec18 . Through a combination of genetic manipulations in vivo , and in vitro assays of SNARE complex assembly and disassembly , we establish that SM proteins directly impair Sec18-mediated SNARE disassembly . In the course of these studies we discovered that Sec17 directly promotes selective loading of at least two different SM proteins onto cognate SNARE complexes . Moreover , an extraordinarily steep temperature dependence limits SM loading onto SNARE complexes , implying that SM-SNARE complex formation entails significant conformational transitions . The thermal dependence of SM loading may partially explain why SNARE–Sec17–SM complexes eluded detection in previous studies . In vitro reconstitution experiments led to models in which SM proteins , in conjunction with additional SNARE cofactors , functionally oppose Sec17 and Sec18 activity ( Mima et al . , 2008; Stroupe et al . , 2009; Xu et al . , 2010; Ma et al . , 2013 ) . To probe for antagonism between SMs and SNARE disassembly factors in vivo , we turned to the late endolysosomal SM Vps33 . We recently characterized a hypomorphic VPS33 allele , vps33car . Vps33car ( G297V ) is an analog of the Drosophila Vps33a ( G249V ) mutant , encoded by carnation1 , probably the first SM allele ever isolated ( Patterson , 1932; Sevrioukov et al . , 1999 ) . Null mutant vps33Δ cells , or vps33R281A functional nulls ( Lobingier and Merz , 2012 ) , have severe trafficking defects , lack identifiable vacuolar lysosomes ( vps class C morphology; Raymond et al . , 1992 ) , and are inviable at 37°C . In contrast , vps33car mutants retain partial function , with milder defects in vacuolar cargo sorting , moderate ( Class B ) vacuole fragmentation , and slow growth at 37°C ( Lobingier and Merz , 2012 ) . When Sec17 and Sec18 were overproduced in wild-type VPS33 cells , growth was normal at either standard temperature ( 30°C ) or at 37°C ( Figure 1A ) . In marked contrast , Sec17 and Sec18 overproduction in mutant vps33car cells caused severe growth defects at 37°C ( Figure 1A ) . As an additional control we overproduced Sec17 and Sec18 in another Class B vps mutant , vps41Δ . Vps41 is , with Vps33 , a subunit of the HOPS tethering complex . There was little or no growth defect when Sec17 and Sec18 were overproduced in vps41Δ cells ( Figure 1A ) . 10 . 7554/eLife . 02272 . 004Figure 1 . Partial Vps33 deficiency sensitizes cells to overproduction of SNARE disassembly proteins . ( A ) Limiting dilution growth assay on synthetic media agar plates incubated at 37°C . ( B ) Growth curves in selective , synthetic liquid media ( YNB lacking uracil and containing 0 . 05% casamino acids and 2% dextrose , with or without 1 mM ZnCl2 ) . Data points represent the means of n = 4 samples . ( C ) LUCID analysis of luminal sorting efficiency of the vacuole cargo Sna3-fLuc . Note that vps33car is a hypomorphic allele with partial loss-of-function , while vps33 R281A is a functional null with total loss of function . Box plots summarize n = 5 biological replicates , except for vps33 R281A ( n = 4 ) . **p<0 . 01 ( one-way ANOVA ) . fLuc , firefly luciferase . 2µ , high copy plasmid vector . DOI: http://dx . doi . org/10 . 7554/eLife . 02272 . 004 The late endosome and vacuolar lysosome are required for metal tolerance in S . cerevisiae . For this reason , growth in the presence of added Zn2+ is a classical indicator of intact endolysosomal function . Wild-type VPS33 cells grew at equal rates with or without 1 mM Zn2+ . Overproduction of Sec17 , Sec18 , or both together had almost no effect on growth of VPS33 cells , with or without added Zn2+ ( Figure 1B , top panel ) . vps33car mutants grew almost as well as VPS33 cells , but when Sec17 , or Sec17 and Sec18 were overproduced together , the vps33car mutants grew slowly ( Figure 1B , middle panel ) , a defect markedly enhanced by 1 mM Zn2+ ( Figure 1B , bottom panel ) . Thus , Sec17 or Sec17 and Sec18 overproduction strongly exacerbate defects in endolysosomal function , even when Vps33 function is only partially impaired . To test for synthetic trafficking defects we employed LUCID , a quantitative assay of traffic from the Golgi to the late endosome . LUCID uses a chimeric reporter , the cargo protein Sna3 fused to firefly luciferase ( fLuc ) . Sna3-fLuc accumulates when endolysosomal traffic or cargo sorting into multivesicular bodies is impaired ( Nickerson et al . , 2012; Paulsel et al . , 2013 ) . Renilla luciferase is co-expressed to control for expression and nonspecific protein turnover . Sec17 overproduction in vps33car cells , but not in wild-type cells , significantly impaired Sna3-fLuc sorting ( Figure 1C ) . Together , the data show that full Vps33 function is required to withstand either Sec17 or Sec17 and Sec18 overproduction . To test for functional interactions between Sec17 and Sec18 and another SM , we studied a conditional mutant of Sly1 , the Golgi SM . sly1ts mutant cells grow almost as well as wild-type cells at permissive temperature ( 26°C ) but cannot grow at elevated temperatures ( Cao and Barlowe , 2000 ) . Wild-type SLY1 cells grew normally when Sec17 and Sec18 were overproduced , alone or together ( Figure 2 ) . In contrast , overproduction of Sec17 , or Sec17 and Sec18 together , profoundly impaired the growth of sly1ts mutants , even at permissive temperatures ( 24–26°C; Figure 2 ) . In independent work , sly1ts sec18-1 double mutants were inviable ( Kosodo et al . , 2003 ) . We conclude that full , wild-type Sly1 and Vps33 function is required to buffer cells against perturbations of the SNARE disassembly machinery . 10 . 7554/eLife . 02272 . 005Figure 2 . Partial Sly1 deficiency sensitizes cells to overproduction of SNARE disassembly proteins . ( A ) Limiting dilution growth assay on plasmid-selective , synthetic media agar plates at 24° , 30° and 37°C . ( B ) Growth curves of yeast in selective , synthetic liquid media at 26°C . Data points each represent the mean of nine replicate samples . 2µ , high copy plasmid vector . DOI: http://dx . doi . org/10 . 7554/eLife . 02272 . 005 To test the hypothesis that SM proteins directly regulate the activities of Sec17 and Sec18 , we established an in vitro assay of SNARE complex disassembly ( Figure 3A ) . Vacuole and Golgi SM proteins , cognate SNAREs , and Sec17 and Sec18 were individually purified ( Table 1; Figure 3—figure supplement 1 ) . Golgi or vacuole SNARE complexes ( Table 1 ) were then assembled on immobilized Qa-SNAREs . We emphasize that the SNARE constructs , assembled on affinity supports to probe protein–protein interactions , encoded only cytoplasmic domains , not transmembrane segments . Consequently , the complexes formed from these proteins cannot be described using the membrane-dependent topological terms cis- and trans- . Using conditions optimized for Vps33–SNARE complex binding ( Lobingier and Merz , 2012 ) , SNARE complexes were incubated with Sec17 , with or without the cognate SM ( Vps33 or Sly1 ) . Sec18 was then added to initiate disassembly . In the absence of SMs , Sec18 rapidly and completely disassembled the SNARE complexes ( Figure 3B , C , lanes 1–5 ) . Disassembly required ATP and was blocked when Mg2+ was sequestered by EDTA ( Figure 3B , C , compare lanes 5 and 6 ) . Pre-incubation with SMs ( Sly1 or Vps33 ) delayed , but did not prevent , SNARE disassembly ( Figure 3B , C , lanes 7–11 ) . The effect of Sly1 was quantified: pre-incubation with the Sly1 decreased the rate of disassembly by 63 ± 9% ( Figure 3—figure supplement 2 ) . The Sed5 ( Qa-SNARE ) used for the experiments shown in Figure 3B , D contained only the SNARE domain , and not the Habc or N-peptide domains . However , Sly1 also protected Golgi SNARE complexes assembled on Sed5 full cytoplasmic domain rather than Sed5 SNARE domain ( Figure 3—figure supplement 3 ) . In control reactions the ability of the Sly1 and Vps33 preparations to impede SNARE disassembly was heat labile ( Figure 3D , E , compare lanes 3 and 5 ) . Omitting the pre-incubation step reduced both SM binding and SNARE complex protection ( Figure 3—figure supplement 4 ) . 10 . 7554/eLife . 02272 . 006Figure 3 . SM proteins oppose Sec18-mediated SNARE disassembly . ( A ) Schematic of SNARE disassembly assay . SNARE complexes assembled onto immobilized Qa-SNAREs ( Vam3 cytoplasmic domain or Sed5 SNARE domain ) were pre-incubated in the presence of Sec17 , with or without added SM ( Vps33 or Sly1 ) . Sec18 was added to initiate disassembly . The remaining resin-bound material was washed , collected , and analyzed by SDS-PAGE at indicated times . ( A and B ) Sec17 ( 20 μM ) and Sly1 ( 10 μM ) or Vps33 ( 2 . 5 μM ) were pre-incubated with SNARE complexes ( 500 nM ) at 30°C for 60 min in Disassembly Buffer . Sec18 ( 300 nM ) was then added . Under these conditions each Sec18 hexamer catalyzed disassembly of >10 SNARE complexes . In negative controls ( lanes 6 and 12 ) , Mg2+ was chelated with EDTA prior to Sec18 addition . ( D and E ) SNARE complex disassembly was assayed as in B and C , but with variable SM protein concentrations as indicated . In ΔSly1 or ΔVps33 lanes , the SM solutions were heated in a boiling water bath for 10 min , plunged into ice-water , and then clarified at 20 k × g to rule out effects of potential heat-stable contaminants in the preparations . DOI: http://dx . doi . org/10 . 7554/eLife . 02272 . 00610 . 7554/eLife . 02272 . 007Figure 3—figure supplement 1 . Purified Proteins . Purified components used in SNARE complex disassembly and binding assays were separated by SDS-PAGE and stained with Coomassie blue . DOI: http://dx . doi . org/10 . 7554/eLife . 02272 . 00710 . 7554/eLife . 02272 . 008Figure 3—figure supplement 2 . Quantification of SNARE complex protection by Sly1 . Immobilized SNARE complexes ( 125 pmol in a 250 µl reaction ) were pre-incubated for 1 hr at 30°C with saturating Sec17 , and with or without Sly1 . As in Figure 3 , disassembly was initiated by adding Sec18 ( 12 . 5 pmol of hexamer ) . Resin-bound material was washed , collected , and bound proteins were separated by SDS-PAGE and visualized with SYPRO-Ruby . Standard curves of purified proteins were used to determine the amount complex remaining at the indicated times . In the presence of Sly1 the disassembly rate was 0 . 83 ± 0 . 04 complexes per Sec18 hexamer per min . In the absence of Sly1 the rate was 1 . 35 ± 0 . 01 complexes per Sec18 hexamer per min . Both calculations assume a Sec18 specific activity of 50% and ∼20 turnovers per Sec18 hexamer . Comparable rates of neuronal SNARE disassembly by NSF and α-SNAP ( ∼1 complex per Sec18 hexamer per min , in 100 mM KCl ) were recently reported by Cipriano et al . ( 2013 ) ( Figure 2G ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02272 . 00810 . 7554/eLife . 02272 . 009Figure 3—figure supplement 3 . Sly1 protection of SNARE complexes assembled on Sed5 containing Habc domain and N-peptide . Golgi SNARE complex ( 500 nM ) was assembled on immobilized Sed5 SNARE domain ( GST-Sed5FL ) containing a Habc domain and N-peptide . SNARE complexes were incubated with Sec17 ( 20 μM ) , Sly1 ( 10 μM ) , or both for 60 min at 30°C in SM Assay Buffer supplemented with 1 mM ATP and 2 mM MgCl2 . 300 nM Sec18 was then added for the indicated period of time , unbound material was washed out , and the bound proteins were analyzed by SDS-PAGE and Coomassie blue staining . DOI: http://dx . doi . org/10 . 7554/eLife . 02272 . 00910 . 7554/eLife . 02272 . 010Figure 3—figure supplement 4 . Pre-incubation of Sly1 with SNARE complexes increases the fraction of SNARE complex resistant to Sec18-mediated disassembly . SNARE complex disassembly was performed as described in Figure 3 , with the exception that Sly1 was in the reaction for lane 4 was added simultaneously with Sec18 . In lane 5 , Sly1 was pre-incubated for the standard 1 hr prior to addition of Sec18 . Dashed vertical lines indicate samples from a single experiment , run on parallel gels . Note that GST-Sed5 serves as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 02272 . 010 Substantial amounts of Vps33 and Sly1 remained bound to immobilized Qa-SNAREs even after complete SNARE complex disassembly ( Figure 3B , C , lanes 11 ) . This raised the possibility that the SMs could capture Qa-SNAREs in an assembly-active state . Thus , it was necessary to test whether bound SM proteins reduce the rate of SNARE complex disassembly or , alternatively , accelerate SNARE complex re-assembly . To evaluate these alternatives , SNARE complexes were completely disassembled using Sec17 , Sec18 , and ATP . Sec18 activity was then quenched with EDTA and the reactions were incubated for an additional 30 min ( Figure 4A ) . At the low concentrations of free SNAREs liberated by disassembly , no re-assembly of SNARE complexes was detected within 30 min in either the absence or presence of SM ( Figure 4B , C , compare lanes 1 and 3 , and lanes 2 and 4 ) . We next tested the competence of the Qa-SNARE for de novo assembly following Sec18-mediated disassembly . After complete disassembly , additional Qb , Qc , and R-SNAREs were added along with the EDTA quench ( Figure 4B , C , compare lanes 5 and 6 ) . Under these conditions re-assembly occurred , but the rates of re-assembly were not increased by either Vps33 or Sly1 , in accord with previous reports that these SMs do not accelerate SNARE assembly in solution ( Kosodo et al . , 2002; Peng and Gallwitz , 2002; Hickey and Wickner , 2010 ) . 10 . 7554/eLife . 02272 . 011Figure 4 . Vps33 and Sly1 do not accelerate SNARE complex re-assembly in solution . ( A ) Cartoon schematic of the re-assembly assay . ( B and C ) SNARE complexes ( 500 nM ) were assembled as in Figure 3 . Following pre-incubation of SNARE complexes with 20 μM Sec17 and SM ( 10 μM Sly1 or 2 . 5 μM Vps33 , as indicated ) , SNARE complexes were completely disassembled for 30 min by Sec18 . Disassembly was terminated with EDTA , and SNARE complex re-assembly was assayed after a further 30 min at 30°C . As indicated , some re-assembly reactions ( lanes 5 and 6 ) were supplemented with soluble SNAREs ( Qb , Qc , and R; ∼3 μM each ) , which were added along with the EDTA quench . DOI: http://dx . doi . org/10 . 7554/eLife . 02272 . 011 We conclude that Vps33 and Sly1 kinetically impair , but do not prevent , Sec18-mediated SNARE disassembly . Because Vps33 is both necessary and sufficient for HOPS binding to SNARE core bundles ( Lobingier and Merz , 2012 ) , Vps33 likely accounts for the ability of HOPS to shield trans-SNARE complexes from premature disassembly ( Xu et al . , 2010 ) . Because Vps33 and Sly1 exhibit similar activities , protection of trans-SNARE complexes from Sec18/NSF may be a more general feature of SM biochemistry . Liposome fusion experiments with SMs from the other two SM subfamilies , Vps45 and Munc18-1 , are also consistent with this interpretation ( Ohya et al . , 2009; Ma et al . , 2013 ) . Up to three copies of Sec17/α-SNAP bind per SNARE complex bundle ( Hanson et al . , 1997; Fleming et al . , 1998; Marz et al . , 2003; Vivona et al . , 2013 ) , and α-SNAP can competitively displace the presynaptic Ca2+ sensor synaptotagmin from neuronal SNARE complexes ( Sollner et al . , 1993 ) . Thus , it was surprising to observe both Sec17 and Vps33 bound to SNARE complexes in our disassembly assays ( e . g . , Figure 3C , lanes 7 and 12 ) . Similarly , both Sec17 and Sly1 associated with Golgi SNARE complex , even though the Sed5 ( Qa-SNARE ) construct used lacked the N-peptide , a high-affinity recruitment site for the Sly1 ( e . g . , Figure 3B , lanes 7 and 12 ) . To test whether Sec17 and SMs compete for binding , immobilized vacuolar SNARE complexes were assayed for binding of Vps33 , Sec17 , or both . As we previously reported ( Lobingier and Merz , 2012 ) , Vps33 binds the vacuole SNARE complex with low µM affinity ( Figure 5A , lane 2 ) . But rather than competing , Sec17 addition strongly stimulated Vps33 binding to the SNARE complex ( Figure 5A , lane 4 ) . Importantly , the stoichiometry of Sec17 binding was unaltered when Vps33 was also bound ( Figure 5A , compare lanes 3 and 4 ) . 10 . 7554/eLife . 02272 . 012Figure 5 . Sec17 promotes Vps33 binding to vacuole SNARE complex . ( A ) SNARE complex ( 500 nM ) was assembled on Vam3-GST . Sec17 ( 20 μM ) , Vps33 ( 2 . 5 μM ) , or both were incubated with the SNARE complex for 1 hr at 30°C in SM Assay Buffer . Unbound material was washed out , then bound material was separated by SDS-PAGE and visualized with Coomassie blue . ( B ) The dose–response for Sec17 stimulation of Vps33 binding to SNARE complexes was assayed as in A , but Sec17 concentration was varied ( 0 . 5–20 μM ) while Vps33 was held constant ( 2 . 5 μM ) . ( C ) Vps33 binding to SNARE complex with or without Sec17 was assayed as in A and B , except that Vps33 concentration was varied and protein bands were stained and quantified using SYPRO Ruby . The fractional saturation of total Vps33–SNARE complex binding was plotted vs free ( total minus bound ) Vps33 . Fits of a one-site binding model yielded Kdobs = 300 ± 10 nM for Vps33 binding to the SNARE complex in the presence of Sec17 , and Kdobs = 1 . 6 ± 0 . 10 µM without Sec17 . Two-site or cooperative binding models did not substantially improve the fits . ( D ) To estimate the stoichiometry of SNARE–Sec17–Vps33 binding , complexes were assembled under saturation binding conditions , separated by SDS-PAGE , and analyzed using SYPRO Ruby stain . The band intensities were quantified using standard curves generated with individual purified proteins . ( E ) Vps33 binding is stimulated by SNARE-associated Sec17 . Complexes were assayed in lanes 1 and 2 as in A . In lane 3 ( Sec17 pre-bound ) , Sec17 was bound to SNARE complexes for 60 min at 30°C . Unbound Sec17 was washed out and 2 . 5 μM Vps33 was then added for an additional 60 min at 30°C . In lane 4 , 20 µM BSA was substituted for Sec17 . ( F ) Cooperativity of assembly . GST-Vps33 ( 500 nM ) was immobilized and incubated with 20 μM Sec17 , soluble SNARE complex , or both . Bound material was separated by SDS-PAGE and stained with Coomassie blue , or analyzed by immunoblot . DOI: http://dx . doi . org/10 . 7554/eLife . 02272 . 012 Sec17 stimulation of Vps33 binding to the SNARE complex depended on the Sec17 concentration and tracked with Sec17 occupancy on the complex ( Figure 5B ) . Vps33-SNARE complex binding was saturable ( Figure 5C ) , and Sec17 increased the apparent affinity of Vps33 for SNARE complex by more than 5-fold ( from KD ( obs ) = 1 . 60 ± 0 . 10 μM to 0 . 30 ± 0 . 01 μM ) . As these are non-equilibrium measurements , we caution that they may systematically underestimate absolute SNARE-SM affinities . Under saturation binding conditions , Sec17–Vps33–SNARE complexes assembled in apparent 1:3:1 stoichiometry ( Figure 5D ) . The above results argue that SNARE-bound Sec17 stimulates Vps33 binding . To rule out an alternative possibility , that free Sec17 in solution enhances Vps33 binding activity , Sec17 was bound to SNARE complexes , and unbound Sec17 was washed out before Vps33 was added ( Figure 5E ) . Vps33 bound to SNARE complex equally well in the presence of free-plus-bound Sec17 ( Figure 5E , lane 2 ) or to SNARE–Sec17 complex from which excess unbound Sec17 had been removed ( Figure 5E , lane 3 ) . Thus , Vps33 binding is stimulated by Sec17 on the SNARE complex , not by Sec17 in solution . In a further control , addition of bovine serum albumin ( BSA ) in place of Sec17 had no effect on Vps33 binding to SNARE complex ( Figure 5E , lane 4 ) . To test whether Vps33 binds Sec17 directly , GST-Vps33 was immobilized and assayed for binding of soluble SNARE complex , Sec17 , or both ( Figure 5F ) . SNARE complex and Sec17 bound efficiently to immobilized Vps33 only when both were present ( Figure 5E , compare lane 4 to lanes 2 and 3 ) . SNARE complexes and Sec17 therefore bind Vps33 through a cooperative mechanism involving all three components . A co-complex between SNAREs , Sec17 , and an SM is unprecedented . Thus , it was essential to test whether similar results might be obtained with a divergent SM protein and its cognate SNAREs . We again turned to Sly1 , the Golgi SM . Sly1 was previously shown to avidly bind the terminal N-peptide of the Qa-SNARE Sed5 . However , N-peptide binding is dispensable for Sly1 function in vivo ( Bracher and Weissenhorn , 2002; Peng and Gallwitz , 2002 , 2004 ) . To examine N-peptide-independent Sly1 binding , Golgi SNARE complex was assembled on an immobilized , truncated Sed5 SNARE domain that lacks both the N-peptide and Habc domains ( GST-Sed5SNARE domain ) . As in previous work ( Peng and Gallwitz , 2004 ) little or no Sly1 bound to the Golgi SNARE bundle . In the presence of Sec17 , however , Sly1 binding was dramatically stimulated ( Figure 6A , compare lanes 2 and 4 ) . 10 . 7554/eLife . 02272 . 013Figure 6 . Sec17 promotes Sly1 binding to Golgi SNARE complex . ( A ) Golgi SNARE complex ( 500 nM ) was assembled on immobilized Sed5 SNARE domain ( GST-Sed5SNARE domain ) lacking the N-peptide and Habc segments . SNARE complexes were incubated with Sec17 ( 20 μM ) , Sly1 ( 10 μM ) , or both for 60 min at 30°C . Unbound proteins were washed out , and bound proteins were separated by SDS-PAGE and stained with Coomassie blue . ( B ) The dose–response for Sec17 stimulation of Sly1 binding to SNARE complexes was assayed as in A , but Sec17 concentration was varied ( 2 . 5–20 μM ) while Sly1 was held constant ( 10 μM ) . In lanes 7 and 8 , the full cytoplasmic domain of Sed5 ( Sed5FL , including the Habc and N-peptide segments; 500 nM ) was immobilized to test for direct binding of Sec17 to Sed5 and Sly1 in the absence of assembled SNARE complex . ( C ) Sly1 binding to SNARE complex with or without Sec17 was assayed as in A and B , except that Sly1 concentration was varied and protein bands were stained and quantified using SYPRO Ruby . The fractional saturation of total Sly1-SNARE complex binding was plotted vs free ( total minus bound ) Sly1 . Fits of a one-site binding model yielded with an apparent Kdobs = 1 . 0 ± 0 . 1 µM for Sly1 binding to the SNARE complex in the presence of Sec17 . It was not possible to fit the no-Sec17 condition . Two-site or cooperative binding models did not substantially improve the fits . ( D ) Stoichiometry of SNARE-Sec17-Sly1 complexes assembled under saturation conditions was estimated using standard curves of purified proteins of known concentrations . ( E ) Sly1 binding is stimulated by SNARE-associated Sec17 . Binding was assayed in lanes 1 and 2 as in A . In lane 3 , Sec17 was pre-bound to SNARE complexes for 60 min at 30°C . Unbound Sec17 was then washed out and Sly1 ( 10 μM ) was added for an additional 60 min at 30°C . In lane , 4 BSA ( 20 µM ) was substituted for Sec17 . ( F ) Cooperativity of assembly . GST-Sly1 ( 500 nM ) was immobilized and incubated with 20 μM Sec17 , SNARE complex , or both . Bound material was separated by SDS-PAGE and stained with Coomassie blue or analyzed by immunoblot . DOI: http://dx . doi . org/10 . 7554/eLife . 02272 . 013 As with Vps33 , Sly1 binding to SNARE complex depended on the Sec17 concentration and tracked with Sec17 occupancy on the complex ( Figure 6B , lanes 1-6 ) . Sec17 did not associate with , or impede formation of , binary complexes between Sly1 and the N-peptide of full-length Sed5 ( Figure 6B , lanes 7 and 8 ) . Thus , Sec17 stimulates Sly1–SNARE complex association , but Sec17 does not efficiently bind Sly1 or Sed5 in the absence of an assembled SNARE complex . Sly1 bound to the Sec17–SNARE complex saturably and with moderate affinity ( Figure 6C; KD ( obs ) = 1 . 0 ± 0 . 1 µM ) . We emphasize that this binding was measured under conditions where Sly1–Sed5 N-peptide binding cannot occur , as the Sed5 construct lacks the N-peptide and Habc domains . In the absence of Sec17 , little or no Sly1 bound the SNARE bundle at concentrations up to 15 µM . Under saturation conditions the SNARE complex , Sec17 , and Sly1 assembled in apparent 1:3:1 stoichiometry , consistent with the results for Vps33 ( Figures 5D and 6D ) . As with Vps33 , Sly1 binding was promoted by SNARE-bound Sec17 rather than Sec17 in solution ( Figure 6E ) . The Sec17 concentration required to saturate Golgi SNARE complex was about four-fold greater than for vacuole SNARE complex ( compare the Sec17 curves in Figures 5B and 6B ) . Consistent with this observation , Golgi SNARE complex pre-bound to Sec17 and then washed retained less Sec17 , and commensurately less Sly1 , versus pulldowns in which Sec17 was in excess ( Figure 6E , compare lanes 2 and 3 ) . No increase in Sly1 binding to Golgi SNARE complexes was observed when BSA was added instead of Sec17 ( Figure 6E , lane 4 ) . Sly1 , like Vps33 , binds SNARE–Sec17 complexes through a cooperative mechanism . GST-tagged Sly1 was immobilized and assayed for binding of soluble SNARE complex , Sec17 , or both ( Figure 6F ) . SNARE complex was retained by GST-Sly1 only in the presence of Sec17 , and Sec17 was retained most efficiently on GST-Sly1 when SNARE complexes were present . There may be some affinity between GST-Sly1 and Sec17 in the absence of SNARE complex under these conditions . However , Sec17 did not detectably interact with Sly1 when Sly1 was tethered to the GST-Sed5 N-peptide ( Figure 6B , lanes 7 and 8 ) . We conclude that Sec17 promotes Sly1 and Vps33 loading onto SNARE complexes through a cooperative mechanism . The resulting assemblies contain three Sec17 molecules and one SM per quaternary SNARE bundle . Sec17 and its mammalian homolog α-SNAP are capable of engaging all SNARE complexes . In contrast , the SM proteins operate at specific organelles and preferentially recognize cognate , pathway-specific SNAREs and SNARE complexes ( Carr and Rizo , 2010; Rizo and Südhof , 2012 ) . Sec17 can bind SNARE complexes as a trimer , raising the possibility that Vps33 and Sly1 recognize composite features of the Sec17 multimer surface but do not touch the underlying SNAREs . If this model is correct , Sec17-stimulated SM binding to SNARE complexes should exhibit no selectivity for the underlying SNARE bundle . In an alternative model , SM proteins touch and recognize cognate SNARE complexes during Sec17-stimulated binding . This model predicts that Sec17-stimulated SM binding should be SNARE-selective and compartment-specific . To evaluate these models , Golgi and vacuole SNARE complexes were assembled and assayed for Vps33 and Sly1 binding in the absence or presence of Sec17 ( Figure 7 ) . In the presence of Sec17 , Vps33 preferentially bound to vacuole SNARE complexes ( Figure 7A , lanes 3 and 3' ) . In the reciprocal experiment , Sec17 promoted selective Sly1 binding to cognate Golgi SNARE complexes ( Figure 7B , lanes 3 and 3' ) . Sec17 therefore promotes selective recognition of cognate SNARE complexes by both Vps33 and Sly1 . These findings further underscore the cooperativity of Sec17 , SM , and SNARE complex co-assembly ( Figures 5 and 6 ) and indicate that even when a Sec17 trimer is bound , Vps33 and Sly1 recognize organelle-specific determinants on the underlying SNARE complex . 10 . 7554/eLife . 02272 . 014Figure 7 . SM proteins touch and recognize cognate SNARE-Sec17 complexes . Golgi and vacuole SNARE complexes ( 500 nM ) were assembled on affinity supports and assayed for binding in the absence or presence of Sec17 ( 20 μM ) . ( A ) Assay of Vps33 ( 2 . 5 μM ) binding . ( B ) Assay of Sly1 ( 10 µM ) binding . Golgi complexes were assembled on Sed5 SNARE domain lacking the N-peptide and Habc segments . Binding reactions were incubated 2 hr at 30°C , unbound material was washed out , and the bound proteins were analyzed by SDS-PAGE and Coomassie blue staining . DOI: http://dx . doi . org/10 . 7554/eLife . 02272 . 014 Unexpectedly , Sly1 binding to the SNARE-Sec17 complex increased markedly from 4° to 30°C , the physiological growth temperature ( Figure 8A , compare lanes 4 , 7 , and 10 ) . In assays performed across a range of temperatures ( Figure 8B ) , Sly1 loading onto SNARE–Sec17 complex decreased by an order of magnitude as temperature dropped from 26 . 5°C to 20°C . In control experiments there was no comparable temperature dependence ( Figure 8B: binding of Sly1 to Sed5 N-peptide , and binding of Sec17 to SNARE complex ) . Below , we show that thermal stimulation of Sly1–SNARE complex binding is dominated by changes in association rather than dissociation kinetics . Most biochemical processes have a thermal coefficient Q10 of 2–4 ( change in activity over 10°C temperature gradient; see Hille , 1991 ) . Sly1 binding to the Sec17-SNARE complex has an extrapolated Q10 of ∼30 , indicating that the Sly1 binding mechanism entails traversal of a large thermodynamic barrier . 10 . 7554/eLife . 02272 . 015Figure 8 . Thermal dependence of SM-SNARE complex association . ( A ) Golgi SNARE complexes were assembled on immobilized Sed5 SNARE domain ( lacking the Sed5 N-peptide ) , then assayed for Sly1 ( 12 µM ) binding in the absence and presence of Sec17 ( 12 µM ) . The binding reactions were incubated for 60 min at 4° , 24° or 30°C . ( B ) Sly1 binding to Golgi SNARE complexes ( S . C . ) was evaluated across a range of temperatures . Note that reciprocal temperature is plotted in units of 1/K , with warmer temperatures on the left side of the plot . Immobilized SNARE complexes ( 500 nM ) were assayed for binding of sub-saturating amounts of Sly1 ( 8 µM , in the presence of 20 µM Sec17 ) , or for binding of Sec17 alone ( 7 µM ) . In an additional control , Sly1 ( 6 µM ) was assayed for binding to the N-peptide of Sed5 ( FL; full-length cytoplasmic domain; 500 nM ) . Bound material was separated by SDS-PAGE , visualized with SYPRO Ruby , and quantified using standard curves of purified proteins of known concentrations . ( C ) The thermal dependence of Vps33–SNARE association was assayed as in A , except that Vps33 was present at 1 µM and Sec17 was present at 2 µM . ( D ) Vps33 binding to SNARE complexes was evaluated across a range of temperatures , similar to B . As indicated , Vps33 was present at 1 . 5 µM and Sec17 was present at 20 µM . ( E ) The kinetics of Sly1 binding at 30°C to Sec17–SNARE complex were analyzed and quantified using conditions and protein concentrations as in panel B . ( F ) Sly1 association kinetics are controlled by temperature . Immobilized Golgi SNARE complex ( 500 nM ) was incubated with Sly1 ( 10 µM ) and Sec17 ( 20 µM ) at the indicated temperatures for the indicated times . Bound proteins were separated by SDS-PAGE and stained with Coomassie blue . DOI: http://dx . doi . org/10 . 7554/eLife . 02272 . 01510 . 7554/eLife . 02272 . 016Figure 8—figure supplement 1 . Stability of Sly1-Sec17-SNARE complexes at 4° and 30°C . Golgi SNARE complex ( 500 nM ) was assembled on immobilized Sed5 SNARE domain ( GST-Sed5 ) lacking the N-peptide and Habc segments . SNARE complexes were incubated with Sec17 ( 20 μM ) , Sly1 ( 10 μM ) , or both , and incubated for 60 min at 30°C . Unbound protein was washed out and resins were incubated in SM Assay Buffer at either 4° or 30°C for a further 60 min . Resins were washed again at the post-wash incubation temperature ( either 4° or 30°C ) , and the bound proteins were analyzed by SDS-PAGE and Coomassie blue staining . DOI: http://dx . doi . org/10 . 7554/eLife . 02272 . 01610 . 7554/eLife . 02272 . 017Figure 8—figure supplement 2 . Pre-incubation at 30°C does not trigger conversion of Sly1 into a persistently activated state . ( A ) Golgi SNARE complex ( 500 nM ) was assembled on immobilized Sed5 SNARE domain ( GST-Sed5 ) lacking the N-peptide and Habc segments . SNARE complexes were then pre-incubated with Sec17 ( 20 μM ) at room temperature for 60 min prior to addition of Sly1 ( 10 μM ) . Sly1 was pre-incubated for 60 min either at room temperature ( conditions 1 and 2 ) or 30 C ( conditions 3 and 4 ) . In condition 3 , Sly1 pre-incubation for 60 min at 30 C is followed by binding at room temperature ( RT ) . Condition 1 shows baseline binding when Sly1 has only been incubated at RT , while condition 4 shows binding of Sly1 at 30 C ( standard experimental condition ) . To control for the addition of 30°C in condition 4 , condition 2 shows binding of Sly1 when the SM has been pre-incubated at RT and an equivalent volume of buffer , added at timepoint zero , was pre-incubated at 30°C . DOI: http://dx . doi . org/10 . 7554/eLife . 02272 . 017 As with Sly1 , Vps33 binding increased with temperature ( Figure 8C , compare lanes 4 , 7 , and 10 , and Figure 8D , blue triangles ) . In contrast , Sec17 efficiently bound the vacuole SNARE complex at 4°C . Because Vps33 has moderate affinity for vacuole SNARE complex ( Figure 5C; Lobingier and Merz , 2012 ) , it was also possible to assess the temperature sensitivity of SNARE complex–Vps33 binding without Sec17 . Although Sec17 stimulated Vps33 binding ( Figure 8D , blue triangles ) , Vps33 bound the SNARE complex most efficiently at elevated ( physiological ) temperatures even when no Sec17 was present ( Figure 8D , red circles ) . Thus , elevated temperature directly enhances Vps33 association with its SNARE complex . To evaluate whether SM binding efficiency is limited by on- or off-rates , SNARE–Sec17–Sly1 co-complexes were formed at 30°C for 60 min , washed in either 30°C buffer or 4°C buffer , and incubated an additional 60 min at the same temperature as the prior wash . There was little Sly1 dissociation at 4°C and slightly more at 30°C ( Figure 8—figure supplement 1 ) , indicating that elevated temperature promotes productive SM–SNARE association rather than stabilizing extant SM–SNARE complexes . At 90% saturating levels , Sly1 bound the Sec17–SNARE complex with a half-time of ∼20 min . Under similar conditions , Sly1 binding to the Sed5 N-peptide and Sec17 binding to Golgi SNARE complexes were both complete within 1 min ( Figure 8E ) . When Sly1 was loaded onto SNARE–Sec17 complexes at 24° instead of 30°C , the time required to reach steady-state binding increased from 1 to 4 hr ( Figure 8F ) . These observations indicate that the temperature dependence of SM binding is due to rate-limiting steps in SM–SNARE association rather than dissociation . A possible interpretation is that Vps33 and Sly1 slowly interconvert between ground states unable to bind SNARE complexes and activated , binding-competent , states . SNARE-bound Sec17—and perhaps other docking factors—would then elicit or stabilize activated SM conformations to accelerate SM loading onto cognate SNARE complexes . Because pre-incubation of Sly1 at 30°C did not accelerate subsequent Sly1 binding to Sec17-SNARE complexes ( Figure 8—figure supplement 2 ) , we suggest that elevated temperature increases the frequency of interconversion between the putative ground and binding-active conformations , rather than by stabilizing an activated conformation . Pioneering biochemical and genetic studies led to the idea that parallel N-to-C zippering of SNARE complexes in trans pulls membranes together to initiate fusion ( Hanson et al . , 1997; Nichols et al . , 1997; Poirier et al . , 1998; Sutton et al . , 1998 ) . SNAREs are sufficient to drive basal fusion of liposomes ( Figure 9A , reaction i ) and impose a layer of compartmental specificity ( Weber et al . , 1998; McNew et al . , 2000 ) . In these minimal systems , purified SM proteins accelerate fusion ( reaction ii; Scott et al . , 2004; Shen et al . , 2007; Furukawa and Mima , 2014 ) . However , the significant basal activities of SNAREs and the relatively modest rate enhancements conferred by SMs have been difficult to reconcile with absolute requirements for SM function in vivo . 10 . 7554/eLife . 02272 . 018Figure 9 . Working model . ( A ) Subreactions of SNARE-driven fusion . ( i ) Basal fusion , as with SNARE proteoliposomes . ( ii ) SM stimulation of the forward basal fusion reaction . Note that the SM may stimulate trans-complex assembly , the fusogenic activity of extant complexes , or both . ( iii ) Disassembly of nascent pre-fusion complexes by Sec17 and Sec18 impairs fusion . ( B ) SM stimulation of fusion in vivo . ( iv ) Sec17 accelerates SM loading onto cognate SNARE complexes , resulting in more efficient fusion and shielding of the complex from premature disassembly by Sec18 . The location of the SM on the SNARE–Sec17 complex , and the dissociation of the SM from the post-fusion complex , are speculative . ( v ) The SM does not efficiently bind an improperly assembled , damaged , or non-cognate SNARE complex , exposing the complex to kinetic proofreading by Sec17 and Sec18 . DOI: http://dx . doi . org/10 . 7554/eLife . 02272 . 018 Sec17 , Sec18 , and ATP completely suppress basal SNARE-mediated liposome fusion ( Figure 9A , reaction iii ) , likely by binding and prematurely disassembling trans-SNARE complexes or their precursors ( Rohde et al . , 2003; Starai et al . , 2008; Ungermann et al . , 1998; Xu et al . , 2010; but see also; Weber et al . , 2000 ) . In these systems the SM , in combination with other factors including HOPS , Munc13 , and synaptotagmin , facilitates efficient fusion in the presence of Sec17 and Sec18 , suggesting that SMs have a minimum of two fusion-promoting functions . First , SMs promote the assembly of trans-SNARE complexes or make them more fusogenic once assembled . Second , SMs protect trans complexes from premature disassembly by Sec18 ( Figure 9B , reaction iv ) . Because multiple cofactors were present in previous studies , it was unclear whether SNARE complex protection from disassembly was attributable directly to SMs , or a property emergent from multiple proteins ( Starai et al . , 2008; Ohya et al . , 2009; Xu et al . , 2010; Ma et al . , 2013 ) . Our experiments now show that SMs functionally interact with the SNARE disassembly machinery in vivo , and affirm that SMs from at least two of the four subfamilies are sufficient to decrease rates of Sec18-mediated SNARE disassembly in vitro . Contrary to expectations , however , Vps33 and Sly1 did not compete with Sec17 for SNARE complex binding . Instead , SNARE-bound Sec17 accelerated SM loading , resulting in the efficient formation of SM–Sec17–SNARE co-complexes with 1:3:1 stoichiometry . Sec17 is therefore a multifunctional SNARE complex adapter , capable of recruiting not only the disassembly ATPase Sec18 but also compartment-specific SMs that oppose Sec18-mediated SNARE disassembly . Our results and other converging lines of experimentation suggest a working model in which Sec17 binding to pre-fusion complexes leads to alternative fates ( Figure 9B ) . When an SM recognizes and binds a Sec17–SNARE complex ( reaction iv ) , it promotes fusion in at least three ways . First , the bound SM directly impedes SNARE disassembly by Sec18 . Second , the SM augments the SNARE complex's fusogenic activity . Third , by accelerating fusion , the SM shortens the pre-fusion complex's lifetime , thereby shortening its temporal exposure to disassembly by Sec18 ( kinetic partitioning; Hardy and Randall , 1991 ) . In this view of fusion , SMs function as true enzymes: they bind and stabilize on-pathway intermediates , protect these intermediates from side reactions , and accelerate the conversion of intermediates to end products . Crucially , this model predicts the annihilation of pathway-specific fusion when SMs are deleted from living cells—a result that holds across many different organisms and fusion pathways . When SM function is compromised , pre-fusion SNARE complexes should be less fusogenic , persist for longer times , and be more exposed to premature disassembly by Sec17 and Sec18 ( Figure 9B , reaction v ) . A further implication of our working model is that SNAREs , Sec17 , SMs , and Sec18 have precisely the features required to implement a kinetic proofreading system ( Hopfield , 1974 ) . Kinetic proofreading entails a sequence of independent , driven discrimination events . For SNARE-mediated fusion in vivo , there would be a minimum of two discrimination events . First , the nucleation of the trans-SNARE complex ( an intermediate analogous to a Michaelis complex ) has significant but not absolute intrinsic selectivity , derived mainly from packing interactions at the core of the SNARE bundle . Second , on-pathway SNARE bundles are positively selected by SM proteins . Sec17 detects the general shape and charge distribution of assembled SNARE bundles ( Marz et al . , 2003 ) , while SMs identify pathway-specific SNARE configurations , protecting them from Sec18 and augmenting their forward fusogenic activity . Error products—off-pathway , incorrectly assembled , or damaged SNARE complexes—would not be efficiently bound by SMs ( Figure 9B , reaction v ) . These complexes would fuse more slowly or not at all , and would be fully exposed to Sec17 and Sec18-mediated disassembly . Although SNAREs selectively drive fusion when complexed with cognate partners ( McNew et al . , 2000 ) , individual SNAREs readily enter into non-cognate , off-pathway , antiparallel or mis-registered complexes with substantial thermal stability and commensurately low off-rates ( Brunger et al . , 2009; Furukawa and Mima , 2014 ) . Moreover , certain R-SNAREs ( Nyv1 , Snc2 , Sec22 ) drive fusion with non-cognate Q-SNAREs ( McNew et al . , 2000; Izawa et al . , 2012 ) . Compartment-specific tethers confer additional selectivity by accelerating the forward rate of cognate SNARE pairing , but this may not be sufficient . Normal cellular transactions are replete with opportunities to assemble erroneous SNARE complexes that could drive inappropriate fusion or trigger unregulated and irreversible organelle aggregation . For example , mammalian endosomes progressively associate with endoplasmic reticulum ( ER ) until 98% of late endosomes and lysosomes are adjacent to ER , generally within 30 nm or less ( Friedman et al . , 2013 ) . Proofreading would emplace a last line of defense against SNARE assembly errors and consequent defects in cell architecture and function . We emphasize that our working schema is necessarily simplified and that pathway-specific specializations are likely to occur . At neuronal synapses , α-SNAP ( Sec17 ) and the Ca2+ sensor synaptotagmin compete for SNARE complex binding ( Sollner et al . , 1993 ) . Within this specialized context ( Fasshauer et al . , 1998; Jackson and Chapman , 2008; Südhof and Rothman , 2009 ) , synaptotagmin may interact with the SM ( Munc18-1 ) to protect pre-fusion complexes in a manner analogous to Sec17 and Vps33 or Sly1 . The complete sequence of SNARE assembly events during priming , docking , and trans complex assembly has not been definitively established for any in vivo pathway . Off-pathway cis-SNARE complexes ( e . g . , Qa2-Qb-Qc; Fasshauer and Margittai , 2004; Fasshauer et al . , 1997; Margittai et al . , 2001; Pobbati et al . , 2006 ) , undergo futile cycles of assembly and Sec18/NSF-mediated disassembly , and it has been suggested that SMs might positively select activated pre-docking intermediates ( Carr et al . , 1999; Carr and Rizo , 2010; Kramer and Ungermann , 2011; Ma et al . , 2013; Furukawa and Mima , 2014 ) . In this context it is notable that high levels of Vps33 and Sly1 remain associated with Qa-SNARE domains following complex disassembly by Sec17 and Sec18 ( Figure 3 ) , raising the possibility that SNARE disassembly is coupled to the formation of SM-SNARE subcomplexes prior to docking . The biochemical experiments in this study were done with complexes in solution , free of membranes . However , the geometry of SNARE juxtamembrane domains in cis or trans , and especially the geometries of the membrane surfaces before and after fusion , may control whether Sec17 and SMs synergize or compete for binding . Consistent with this idea , Sec17 competes with the HOPS complex for binding to post-fusion cis-SNARE complexes on the yeast vacuole , as shown in meticulous co-isolation experiments ( Collins et al . , 2005 ) . On the other hand , Sec17 interacts cooperatively with poised , partially-zipped trans-SNARE complexes on docked vacuoles , triggering fusion ( Schwartz and Merz , 2009 ) . This Sec17-dependent fusion requires Rab signaling and the HOPS effector complex—including the SM Vps33—but is totally independent of Sec18 and ATP ( Schwartz and Merz , 2009 ) . Vps33 and Sly1 binding to core SNARE bundles , with or without Sec17 , is slow and rate-limited by temperature ( Figure 8 ) , implying that significant conformational transitions are required for assembly of SM–SNARE complexes . The requirement for physiological ( warm ) temperatures in our in vitro assays may explain in part why SM–Sec17–SNARE interactions were not previously detected . In vivo , the relevant transitions may occur in a concerted manner; alternatively , they may involve ratchet-like sequential SM association with a series of SNARE assembly intermediates , with each sub-step traversing a smaller energy barrier . The central cavities of SMs are proposed binding sites for SNARE helical bundles and vary in the sizes of their openings , implying conformational flexibility ( Bracher and Weissenhorn , 2001; Bar-On et al . , 2011; Baker et al . , 2013 ) . Our observations provide new empirical support for models in which SM proteins and perhaps SNAREs must undergo substantial conformational transitions to productively associate . The nature of these transitions , and the specific biochemical events that allow SM binding and on-pathway fusion to occur over physiologically relevant time scales , remain to be elucidated . SNAREs and SM open reading frames were cloned into bacterial expresssion vectors using T4 DNA ligase ( New England Biolabs , Beverly , MA ) as previously described ( Lobingier and Merz , 2012 ) . In all cases , constructs lacked the SNARE transmembrane domains . In brief , the cytoplasmic domain of Vam3 ( aa1-264 ) was cloned into NcoI/SacI-digested pRSF-1b with no N-terminal tag and C-terminal GST separated from the SNARE by a TEV cleavage site . N-terminal GST tags for the full cytoplasmic domain of Sed5 ( aa1-319 ) , or a Sed5SNARE domain ( aa170-319 ) construct lacking the N-peptide and Habc domain , were cloned into BamHI/XhoI-cut pGST-Parallel1 . The soluble domains of SNAREs were cloned as N-terminal His6-tagged fusions into pHIS-Parallel1: Bos1 ( aa1-222 ) , Sec22 ( aa1-188 ) were ligated into BamHI/XhoI-cut vector , while Nyv1 ( aa1-231 ) was cloned into NcoI/SacI-digested vector . The SNARE domain of Vam7 ( aa190-316 ) was inserted in-frame into a BamHI/PstI-cut His6-GFP-TEV sequence . The soluble domains of Bet1 ( aa1-123 ) and Vti1 ( aa1-194 ) were cloned into BamHI/XhoI-cut or NcoI/PstI-cut ( respectively ) pRSF-1b carrying an N-terminal His7-MBP tag . Full-length Sly1 ( aa1-667 ) was cloned for expression in NcoI/SacI-digested pHIS-Parallel1 . Vps33 was cloned for expression in the baculovirus system as described ( Brett et al . , 2008; Lobingier and Merz , 2012 ) . Plasmids overexpressing SNARE disassembly machinery in yeast were made by gap repair recombination of PCR products containing SEC17 and/or SEC18 , each with 500 bases of promoter sequence and 300 bases of terminator sequence , into high copy ( 2µ ) vectors pDN526 or pDN524 at unique SacI and HindIII restriction sites , respectively ( see Table 2 ) . We verified overexpression constructs by DNA sequencing and Western blotting of yeast lysates . 10 . 7554/eLife . 02272 . 019Table 2 . Yeast lines and plasmids employed in this studyDOI: http://dx . doi . org/10 . 7554/eLife . 02272 . 019NameGenotypeReference or sourceS . cerevisiae SEY6210MATα leu2-3112 ura3-52 his3-200 trp1-901 lys2-801 suc2-9Robinson et al . , 1988 WSY41SEY6210; vps41Δ1::LEU2Cowles et al . , 1997 BY4742MATα his3Δ1 leu2Δ0 lys2Δ0 ura3Δ0ATCC BLY3BY4742; pep4Δ::KAN VPS33-ttx-GFP::NATLobingier and Merz , 2012 BLY5BY4742; pep4Δ::KAN vps33 R281A-ttx-GFP::NATLobingier and Merz , 2012 BLY6BY4742; pep4Δ::KAN vps33car [G297V]-ttx-GFP::NATLobingier and Merz , 2012 CBY267S288C; MATα ade2-1 ura3-1 trp1-1 leu2-3112 can1-100Cao et al . , 1998 RSY268 ( CBY268 ) S288C; MATα ade2-1 ura3-1 trp1-1 leu2-3112 can1-100 sly1tsCao et al . , 1998Plasmids pDN526ApR 2µ URA3Nickerson et al . , 2012 pDN313SEC18 ( pDN526 ) This study pDN314SEC17 ( pDN526 ) This study pDN315SEC17 SEC18 ( pDN526 ) This study pDN524ApR 2µ TRP1This study pDN316SEC18 ( pDN524 ) This study pDN317SEC17 ( pDN524 ) This study pDN318SEC17 SEC18 ( pDN524 ) This study pGO735ApR CEN LEU2 PGK1pr::RLuc SNA3-FLuc ( pRS415 ) G Odorizzi ( CU-Boulder ) pRP1pRSF KmR His7-MBP- ( tev ) -Lobingier and Merz , 2012 pBL14pBL12 KmR VAM3 ( 1-264 ) - ( tev ) -GSTLobingier and Merz , 2012 pBL19pRP1 KmR His7-MBP- ( tev ) -VTI1 ( 1-194 ) Lobingier and Merz , 2012 pBL20pHIS Parallel1 ApR His6- ( tev ) -NYV1 ( 1-231 ) Lobingier and Merz , 2012 pBL22pBL12 KmR His6-GFPA207K- ( tev ) -Vam7 ( 190-316 ) Lobingier and Merz , 2012 pBL25pGST Parallel1 ApR GST- ( tev ) -SED5SNARE ( 170-319 ) This study pBL26pHIS Parallel1 ApR His6- ( tev ) -BOS1 ( 1-222 ) This study pBL27pHIS Parallel1 ApR His6- ( tev ) -SE22 ( 1-188 ) This study pBL49pRP1 KmR His7-MBP- ( tev ) -Bet1 ( 1-123 ) This study pBL50pGST Parallel1 ApR GST- ( tev ) -SED5SNARE ( 1-319 ) This study pBL51pHIS Paralle1 ApR His6- ( tev ) -Sly1This study pSec17pTYB12 ApR CBD- ( intein ) -Sec17Schwartz and Merz , 2009 Vps33 was expressed and purified from insect cells using the Baculovirus system as described ( Lobingier and Merz , 2012 ) . All other proteins were expressed in E . coli that harbored a pRIL codon-bias correction plasmid with the exception of Bos1 , which was expressed in Rosetta2 pLys cells . Cells were inoculated at 0 . 05 OD600 , grown to 1 . 0–1 . 2 OD600 in Terrific Broth , and expression was induced with 100 µM IPTG for overnight expression at 21°C ( His6-Sly1 , His6-Nyv1 and His7-MBP-Vti1 , GST-Sed5SNARE domain , His6-Bos1 , His7-MBP-Bet1 , His6-Sec22 ) , 500 µM IPTG for 4–5 hr at 30°C ( His6-GFP-Vam7SNARE ) , or 1 mM IPTG for 3 hr at 37°C ( Vam3-GST and GST-Sed5 ) . Cells expressing His-tagged proteins were lysed by sonication in Buffer A ( 50 mM HEPES , 200 mM NaCl , 10% [m/v] glycerol , 5 mM 2-meracptoethanol , 25 mM imidazole , 0 . 5% TritonX-100 , pH 7 . 4 ) supplemented with protease inhibitors . Cell lysates were clarified by centrifugation for 25 min at 18 , 500×g at 4°C . The supernatant was incubated with Ni-NTA HP resin ( GE Heathcare , Piscataway , NJ ) for 10 min at 4°C . The resins were washed extensively in Buffer A followed by washes in Buffer B ( 20 mM HEPES , 200 mM NaCl , 10% [m/v] glycerol , 2 mM 2-mercaptoethanol , 35 mM imidazole , pH 7 . 4 ) . His-tagged proteins were eluted from the resin with Buffer B supplemented with 400 mM imidazole , and then exchanged into Storage Buffer ( 20 mM HEPES , 200 mM NaCl , 10% [m/v] glycerol , 2 mM 2-mercaptoethanol , pH 7 . 4 ) and snap-frozen in liquid nitrogen . Cells containing the GST-tagged SNAREs were lysed in Storage Buffer supplemented with protease inhibitors and 5 mM EDTA , and the clarified lysate was frozen in liquid nitrogen . SNARE complexes were formed by binding 125 pmol of GST-SNARE ( 5 . 8 µg of GST-Sed5SNARE domain or 7 . 1 µg of Vam3-GST ) to glutathione sepharose 4B resin ( GE Healthcare ) for 2 hr at 4°C . Resins were washed twice with SM Assay Buffer: 20 mM HEPES , 150 mM NaCl , 2 mM 2-mercaptoethanol , 0 . 05% ( m/v ) Anapoe-X-100 ( also called Triton-X-100; Affymetrix , Santa Clara , CA ) , pH 7 . 4 . A ≥fivefold molar excess of Qb- , Qc- , and R-SNAREs was incubated overnight at 4°C with the GST-SNARE . For vacuole SNARE complexes , these were soluble domains of Vti1 and Nyv1 and the SNARE domain of Vam7: His7-MBP-Vti1 , His6-GFP-Vam7SNARE , and His6-Nyv1 . For Golgi SNARE complexes , these were soluble domains of Bos1 , Bet1 , and Sec22: His6-Bos1 , His7-MBP-Bet1 , His6-Sec22 . Unbound SNAREs were removed from SNARE complexes by washing the resins with SM assay buffer twice at 4°C and twice at room temperature . Sec17 , the SM protein , or both were added at the indicated concentration to binding reactions containing immobilized SNARE complex ( 500 nM final , unless otherwise specified ) . Pulldowns were performed at 30°C for 1 hr , the resins were washed three times , and eluted with SM Assay Buffer supplemented with 20 mM reduced glutathione , pH 7 . 4 . Samples were boiled in SDS-loading buffer , and separated using 12% SDS-PAGE for experiments using Sly1 or 10% SDS-PAGE for experiments using Vps33 . Unless otherwise indicated , all gels shown were stained with Coomassie brilliant blue and imaged on an Epson 4490 transmission scanner . All experiments were repeated three times or more; representative gels are shown . Quantification of protein binding was performed using SYPRO-Ruby stain ( Invitrogen , Carlsbad , CA ) and a standard curve of each relevant protein , and gels were imaged using a Gel Doc XR+ ( Bio-Rad , Hercules , CA ) . Where indicated , experiments with Vps33 were done twice rather than 3 or more times due to limitations in the amount of available protein . Data from three experiments ( Sly1 ) or two experiments ( Vps33 ) were plotted as fractional saturation of SM protein binding to immobilized SNARE complex , relative the total concentration of free SM protein in solution . Kdobs values and Hill coefficients were estimated by nonlinear fitting ( GraphPad Prism v . 5 ) of a single-site binding model with Hill coefficient to the data . Two-site models did not substantially improve the quality of the fits . Thermal coefficients ( Q10 ) of SM association with SNARE or Sec17-SNARE complexes were calculated asQ10= ( X2/X1 ) 10/ ( T2−T1 ) where X1 and X2 are the binding efficiencies at lower and higher temperatures , T1 and T2 ( Hille , 2001 ) . The Q10 values were extrapolated from the slope of the steepest part of the temperature-binding curve . Sec18 activity was assayed in SNARE Disassembly Buffer , which was SM Assay Buffer with 1 mM ATP and 2 mM MgCl2 , pH 7 . 4 . Unless otherwise noted , the SM protein and Sec17 were allowed to bind to SNARE complexes for 1 hr at 30°C before addition of Sec18 . Disassembly reactions were then incubated for the indicated times and quenched by washing the samples in ice-cold SM Assay Buffer with 10 mM EDTA final , pH 7 . 4 . Remaining resin-bound proteins were eluted with SM Assay Buffer containing 20 mM reduced glutathione , pH 7 . 4 at room temperature . To assay re-assembly of SNARE complexes , SNARE complexes were disassembled for 30 min at 30°C . Sec18 activity was then quenched with a final concentration of 10 mM EDTA , reactions were supplemented with additional soluble Qb , Qc , and R-SNAREs as indicated , and incubated at 30°C for 30 min . Limiting plate dilutions were performed by growing yeast carrying plasmid vectors overnight at permissive temperature in synthetic media lacking Ura and with 2% ( m/v ) glucose and 0 . 05% ( m/v ) casamino acids . OD600 was measured for each culture to permit equalization of cell mass . Cells were serially diluted onto synthetic media dropout plates . Growth curves in liquid were obtained as described ( Paulsel et al . , 2013 ) but used synthetic media lacking Ura and with 2% glucose and 0 . 05% casamino acids . To prevent cell clumping in the Bioscreen-C machine ( Growth Curves USA ) , 0 . 2% Nonidet-P40 was added to synthetic media ( McIntosh et al . , 2011 ) . For studies of Zn2+ sensitivity , synthetic medium was prepared with asparagine at molar equivalence to , and in place of , the usual ammonium and adjusted to pH 6 . 5 . This allowed 1 mM ZnCl2 to remain soluble . YODA software ( Olsen et al . , 2010 ) was used to analyze growth curve data . Cargo protein sorting was assayed using LUCID ( Nickerson et al . , 2012 ) .	Eukaryotic organisms , from single-celled yeast to humans , divide their cells into membrane-bound compartments ( organelles ) of distinct function . To move from one compartment to another , or to enter or exit a cell , large molecules like proteins are packaged into small membrane sacs called vesicles . To release its cargo , the membrane of a vesicle must fuse with the membrane of the correct destination compartment . The SNARE family of proteins plays a key role in this fusion process . As the membranes of a vesicle and target compartment come close , SNARE proteins located on each membrane form a SNARE complex that tethers the vesicle in place and causes the two membranes fuse . SNARE proteins do not act alone in this process: the SM family of proteins also plays an essential role in SNARE-mediated membrane fusion . However , it is still not clear exactly why the SM proteins are needed . Lobingier et al . used the yeast model organism and biochemical studies with purified proteins to show that SM proteins help SNARE complexes form at the right time by regulating the delicate balance between SNARE complex formation and disassembly . This is achieved through the interplay of SM proteins and two other proteins ( Sec17 and Sec18 ) . Sec17 is known to load Sec18 onto SNARE complexes to break them apart . Lobingier et al . showed that Sec17 can also load SM proteins on SNARE complexes . This hinders Sec18 action , and so helps to keep the SNARE complexes intact . Because each SM protein tested only binds to the SNARE complex that should function at the membrane where the SM protein resides , these findings suggest SM proteins perform quality control at potential sites of membrane fusion .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "cell", "biology" ]	2014	SM proteins Sly1 and Vps33 co-assemble with Sec17 and SNARE complexes to oppose SNARE disassembly by Sec18
Rhythmic behaviors vary across individuals . We investigated the sources of this output variability across a motor system , from the central pattern generator ( CPG ) to the motor plant . In the bilaterally symmetric leech heartbeat system , the CPG orchestrates two coordinations in the bilateral hearts with different intersegmental phase relations ( Δϕ ) and periodic side-to-side switches . Population variability is large . We show that the system is precise within a coordination , that differences in repetitions of a coordination contribute little to population output variability , but that differences between bilaterally homologous cells may contribute to some of this variability . Nevertheless , much output variability is likely associated with genetic and life history differences among individuals . Variability of Δϕ were coordination-specific: similar at all levels in one , but significantly lower for the motor pattern than the CPG pattern in the other . Mechanisms that transform CPG output to motor neurons may limit output variability in the motor pattern . Variability across individuals and across cell types in underlying intrinsic and synaptic properties is now viewed as a hallmark of neuronal networks , even those that produce stereotyped output . Indeed , the hunt is on to find the mechanisms and rules that permit constant network output through coordinated regulation , both developmentally and homeostatically , of highly variable membrane and synaptic conductances ( Davis , 2013; Marder et al . , 2015; O'Leary et al . , 2013 , 2014; Ransdell et al . , 2013; Schulz et al . , 2006 ) . But how constant is network output across individuals ? Not very seems to be the answer when looking at the literature more closely ( e . g . , locomotion in mice [Bellardita and Kiehn , 2015] and zebrafish [Masino and Fetcho , 2005; Wiggin et al . , 2014]; food processing in crabs [Hamood et al . , 2015; Hamood and Marder , 2015; Yarger and Stein , 2015] , crawling in fly larvae [Pulver et al . , 2015] ) . The central pattern generating networks of invertebrates have provided some of the best evidence supporting the notion of constant output with underlying variability of conductances ( Goaillard et al . , 2009; Marder et al . , 2015; Prinz et al . , 2004; Ransdell et al . , 2013 ) . Here phase of firing of component neurons is considered a critical aspect of a functional motor pattern , and phase is by no means constant . Even though it is not correlated with period , phase varies considerably across animals as shown in the stomatogastric nervous system ( STNS ) and in our work on the leech heartbeat system ( Bucher et al . , 2005; Norris et al . , 2006; Norris et al . , 2007b; Wenning et al . , 2004a , 2004b ) . Indeed , we were forced to the conclusion that each animal arrives at a unique solution to producing a functional heartbeat motor pattern; based on phase differences in the premotor pattern and synaptic strength patterns from the central pattern generator ( CPG ) to motor neurons ( Norris et al . , 2011; Wright and Calabrese , 2011b ) . Thus , not only are underlying conductances variable but activity itself is variable and any attempt to elucidate mechanisms of regulation must consider what the target limits for regulation are ( ‘What is good enough ? ’; Marder et al . , 2006 ) , what are the sources of variability in network output , and how variability at one level in a network influences the variability on another . Here we focus mainly on the latter two questions , having previously established the range of functional output ( Norris et al . , 2006; Norris et al . , 2007a; Norris et al . , 2007b; Wenning et al . , 2004a; Wenning et al . , 2014 ) . Leech heartbeat presents an amenable system for answering these questions because all relevant neurons of the CPG and motor neurons are identified and easily recorded , and the motor plant ( here the hearts ) also can be directly monitored ( recent review: Calabrese et al . , 2016 ) . Moreover , the system is strictly feedforward – CPG to motor neurons , to heart muscle – , its elements are bilaterally symmetrical , it operates without phasic sensory feedback ( Calabrese , 1977; Calabrese , 1979 ) , and it has already been demonstrated to be highly variable in output across individuals at each level . A unique aspect to this system is that the coordination of the CPG , motor neurons , and hearts differs at any given time on the two sides – rear-to-front peristaltic versus synchronous – with periodic switches in coordination between sides ( Norris et al . , 2006; Norris et al . , 2007b; Wenning et al . , 2004a; Wenning et al . , 2014; Figure 1 , Figure 1—figure supplement 1 and Figure 1—video 1 ) . Thus , while rhythmic circuit output is continuous , it presents episodic coordination states on each body side . The sources of variability that we considered were ( 1 ) inherent variability owing to the stochastic nature of biological processes ( analyzing the cycle-to-cycle variabilities within a coordination state episode ) , ( 2 ) repetition variability as the same function is performed multiple times by the same elements ( comparing across coordination state episodes ) , ( 3 ) variability within an individual due to differences between genetically identical bilaterally homologous neurons and muscles , and ( 4 ) population variability including , but not limited to , genetic variability and variability in individual experience ( comparing across animals ) . We then compare these sources across levels and coordination states . We show that cycle-to-cycle variabilities in phase were low at all levels , in both coordination states , and on both sides . Thus , activity within an individual is precise . We confirm and quantify the large variability in phase across individuals at each level and show that high variability at one level is not necessarily fed forward to the next . In seeking to elucidate the sources of this population variability , we show that repetitions of a coordination state have low variability and thus contribute little . On the other hand , we show that when the same motor act is performed by bilaterally homologous neurons and muscles , variability can be as large as in the population itself depending on the level and coordination . Medicinal leeches have two bilateral heart tubes which run the entire length of the animal ( Maranto and Calabrese , 1984a , 1984b; Thompson and Stent , 1976a ) . Segmental heart ( HE ) motor neurons innervate the hearts along their length , timing and coordinating their constrictions . The HE motor neurons are controlled by a heartbeat CPG that produces a bilaterally asymmetric activity pattern ( Calabrese , 1977 ) . On one side , CPG premotor interneurons fire bursts in a peristaltic rear-to-front progression and in near synchrony on the other . Motor neurons fire correspondingly leading to peristaltic and synchronous motor patterns , in turn leading to an asymmetric beat pattern of the hearts ( Wenning et al . , 2004a , 2004b ) . The beat period in leeches is about 4 . 5 to 11 s , which translates into several thousand heartbeats per day . Embedded in this ongoing activity is the periodic alternation between the two coordinations about every 70–350 s creating episodes of coordination that span 15 to 60 beat cycles . We define a switch cycle of a given side when it has completed both coordinations , one after the other . The core heartbeat CPG consists of 7 bilaterally paired segmental heart interneurons ( HN ) linked by inhibitory synapses and electrical coupling ( Figure 1; review: Calabrese , 2010 ) . Beat timing is determined by the mutually inhibitory bilateral pairs ( Right/Left ) of interneurons in ganglion 3 ( HN ( R/L , 3 ) ) and ganglion 4 ( HN ( R/L , 4 ) ) linked by coordinating interneurons HN ( R/L , 1 ) and HN ( R/L , 2 ) , which form a beat timing network and ensure bilaterally symmetrical timing ( Calabrese , 1977; Calabrese , 1979; Masino and Calabrese , 2002a , 2002b , 2002c ) . Output to the segmental heart motor neurons , which occur in segments 3 to 18 , is provided by inhibitory input from premotor interneurons . Each HE motor neuron makes an excitatory connection to the ipsilateral heart section in its home segment . The HN ( R/L , 5 ) interneurons switch the network . The switch interneuron on the synchronous side bursts with beat timing while the one on the peristaltic side is silent ( Calabrese , 1977; Gramoll et al . , 1994 ) . Switches in coordination occur when the silent switch interneuron starts to burst and the bursting switch interneuron simultaneously becomes silent . Because the switch interneurons make bilateral connections to the middle premotor interneurons HN ( 6 ) and HN ( 7 ) , phasing in these CPG premotor interneurons is dominated by the single active switch interneuron ( Figure 1 ) . In summary , a single CPG consisting of bilateral homologous pairs of HN interneurons produces an asymmetric premotor pattern , motor pattern , and beat pattern that episodically and periodically switches between left synchronous/right peristaltic and left peristaltic/right synchronous states of coordination ( Figure 2 A2 , B2 , C2 ) . For this study , we focused on segments 8 to 14 , because here each HE motor neuron receives input from all four ipsilateral front and middle premotor HN interneurons ( HN ( 3 ) , HN ( 4 ) , ( HN ( 6 ) , HN ( 7 ) ( Figure 1; Norris et al . , 2007a , 2007b; Thompson and Stent , 1976a ) . In segments 3 to 6 and 15 to 18 the HE motor neurons receive input from additional HNs ( Norris et al . , 2007a; Wenning et al . , 2011 ) . We collected data from three levels ( CPG , motor neurons , and motor plant ) and from the two coordinations on the two body sides resulting in 12 scenarios ( 3 × 2 × 2 = 12 ) . The Project Database , illustrated in Figure 2—figure supplement 1A–D , was compiled since 2008 and partially reported ( Norris et al . , 2011; Wenning et al . , 2014; Wright and Calabrese , 2011a ) , but all Bilateral Recordings and all analysis are novel . For the sake of clarity , we report and discuss the data , and present the figures , on all three levels for one coordination ( peristaltic ) and for one body side ( left; except when discussing bilateral variability ) . All data ( left and right side , both coordinations ) accompany the relevant Figures as source data ( in table format ) . There were no differences in the main conclusions for the right body side . Differences in the data for synchronous vs . peristaltic coordination are pointed out and discussed . All N’s reported represent the number of different animals recorded . We focused on phase , which is a critical output characteristic of any coordinated motor program and its underlying neuronal circuitry . Phase and period do not correlate in the CPG pattern or in the motor pattern across animals in our Project Database ( N = 153; data not shown ) or in the beat patterns of both adult and juvenile leeches ( Wenning et al . , 2004a , 2004b ) . The phase difference between the activity phases of two segments ( Δϕ ) is a good metric for characterizing the two coordination states ( Norris et al . , 2006; Norris et al . , 2007a; Norris et al . , 2007b; Norris et al . , 2011; Wenning et al . , 2004a; Wenning et al . , 2004b; Wright and Calabrese , 2011b ) . For the CPG pattern , we recorded from two pairs of heart interneurons – HN ( L/R , 4 ) and HN ( L/R , 7 ) – ( Figure 2A1 ) , for the motor pattern we recorded from two pairs of motor neurons – HE ( L/R , 8 ) and the HE ( L/R , 12 ) – ( Figure 2B1 ) , and for the beat pattern , we extracted the digitized optical signals for heart ( L/R , 8 ) and heart ( L/R , 12 ) ( Wenning et al . , 2014 ) ; Figure 2C1 ) . All sample recordings ( Figure 2A2 , B2 , C2 ) start with the left side in peristaltic coordination ( rear-to-front delay ) and the right in synchronous coordination . In the CPG , the HN ( L , 7 ) interneuron bursts lead those of the HN ( L , 4 ) interneuron , while the HN ( R , 4 ) interneuron slightly leads the HN ( R , 7 ) interneuron . Similarly , the HE ( L , 12 ) motor neuron bursts lead those of the HE ( L , 8 ) motor neuron while the HE ( R , 8 ) motor neuron bursts slightly lead those of the HE ( R , 12 ) motor neuron . Finally , heart ( L , 12 ) starts to constrict before heart ( L , 8 ) while the right side is in synchronous coordination with the hearts ( R , 12 ) and ( R , 8 ) constricting almost synchronously . Figure 1—figure supplement 1 and Figure 1—video 1 show the beat pattern for this preparation . In all recordings , the two sides switch coordinations simultaneously to left synchronous/right peristaltic ( Figure 2A2 , B2 , C2 ) . The triangles of Figure 2A3 , B3and C3 show the average Δϕ between the left front and rear segment of one switch cycle , i . e . between the HN ( L , 4 ) and HN ( L , 7 ) interneurons ( Figure 2A3 ) , between the HE ( L , 8 ) and HE ( L , 12 ) motor neurons ( Figure 2B3 ) , and between heart ( L , 8 ) and heart ( L , 12 ) ( Figure 2C3 ) . The circular phase plots of Figure 2A4 , B4 , C4 illustrate the cycle-to-cycle variability for all cycles in peristaltic coordination . Each bout of behavior , either peristaltic or synchronous , has between 15 and 60 neuronal bursts ( CPG pattern , motor pattern ) and rhythmic constrictions ( beat pattern ) . How variable are the patterns across the bursts or beats within one bout of behavior ? Figure 2 illustrates the individual Δϕ between two segments of these individual bursts and constrictions , referred to as cycle-to-cycle variability , and their variance , for a single preparation for the CPG pattern ( A4 ) , the motor pattern ( B4 ) , and for the beat pattern ( C4 ) . Figure 3A shows the variances for all preparations used in this study for two subsequent switch cycles . Cycle-to-cycle variances of the output patterns were equally low on the two sides , in both coordinations , and on all levels ( Figure 3—source data 1 ) . But what does ‘low’ mean ? We reasoned that the timing network had the lowest variabilities in the heartbeat system . In the same 26 animals where we recorded the CPG pattern ( Figure 2—figure supplement 1D ) we calculated the cycle-to-cycle variances of the phase difference between the two HN ( 4 ) interneurons , which are part of the timing network and which form a half-center oscillator with strong mutual , inhibitory connections ( Figure 1; Calabrese , 1977; Hill et al . , 2001; Sorensen et al . , 2004 ) . The average cycle-to-cycle variance of the Δϕ between the two HN ( 4 ) interneurons was similar to the average cycle-to-cycle variances of the Δϕ between two premotor interneurons , two motor neurons , and two heart segments ( Figure 3A ) . In these same recordings , we assessed period variability and found the coefficient of variation to be on average of less than 5% ( Figure 3B ) . The low cycle-to-cycle variability indicates a highly coordinated and precise motor system and allowed us to use the average Δϕ of a given coordination in an individual to assess the population , repetition , and bilateral variability . We plotted the average Δϕ between two segments as detailed above in circular phase plots using the animals of this study ( Figure 2—figure supplement 1D and Figure 4 ) , for the Project Database ( Figure 2—figure supplement 1A and Figure 5A ) , and for the Simultaneous Recordings from the HN and HE neurons ( Figure 2—figure supplement 1B and Figure 5B ) . We had shown previously that the average Δϕ differences between the front and middle premotor HN interneurons ( HN ( 4 ) and HN ( 7 ) ) were larger than those between the HE ( 8 ) and the HE ( 12 ) motor neurons ( Wright and Calabrese , 2011a ) . We obtained the same results in three data sets: ( 1 ) Bilateral Recordings , Figure 2—figure supplement 1C and Figure 4B: unpaired t-test , p<0 . 001; ( 2 ) Project Database , Figure 2—figure supplement 1A and Figure 5A: unpaired t-test , p<0 . 001; ( 3 ) Simultaneous Recordings of the CPG pattern and the motor pattern , Figure 2—figure supplement 1B and Figure 5C: paired t-test , p<0 . 001 . Finally , the intersegmental Δϕ of the CPG pattern also exceeded that of the motor pattern ( paired t-test , p<0 . 001 ) in the nine simultaneous Bilateral Recordings of the CPG pattern and the motor pattern ( Figure 2—figure supplement 1C; both coordinations , both sides; data not shown ) . Next , we determined the angular variances of these intersegmental phase differences for the animals in which we made bilateral recordings ( Figure 2—figure supplement 1C ) , calculated the population variance for each level , and determined the confidence interval for each level with bootstrapping ( 10 , 000 times with replacement ) ( significance level: 0 . 05; Figure 4C ) . The population variances of these intersegmental Δϕ were substantial ( among the largest variances determined in this study ) indicating considerable variability in the population . The smaller population of bilateral recordings reported here ( Figure 2—figure supplement 1C ) nevertheless represents the larger populations ( Figure 2—figure supplement 1A , B ) reasonably well , because their confidence intervals overlap extensively ( compare Figure 4 and Figure 5 ) . We had previously shown that at none of the network levels was intersegmental Δϕ correlated with the cycle period ( Norris et al . , 2006; Wenning et al . , 2004a; Wenning et al . , 2004b ) . We corroborated these results for the CPG pattern and the motor pattern using our Project Database ( Figure 2—figure supplement 1A ) and found no correlation over a cycle period range of 4 to 13 s for the 129 HN interneurons and of 5 to 11 s for the 83 heart motor neurons ( data not shown ) . The variance of the motor pattern in the bilateral recordings of Figure 4C appears to be lower than that of the CPG or motor plant although the confidence intervals overlap . To clarify whether the motor pattern does indeed have a lower variance than the CPG pattern , we used our larger databases . We calculated the angular variances using the entire Project Database ( Figure 2—figure supplement 1A ) and found that variances were higher in the CPG pattern than in the motor pattern ( 0 . 0038 , CPG pattern; 0 . 0020 , motor pattern ) with no overlap in the bootstrapped 95% confidence intervals ( Figure 5B ) . To eliminate the possibility that this difference results from CPG and motor pattern recordings being made in different preparations , we calculated the angular variances from simultaneous ( mostly unilateral ) recordings of the CPG pattern and the motor pattern ( Figure 2—figure supplement 1B ) . We found that variances were also higher in the CPG pattern than in the motor pattern ( 0 . 0043 , CPG pattern , 0 . 0018 motor pattern ) , again with no overlap in the bootstrapped 95% confidence intervals ( Figure 5D ) . Two aspects of the CPG output determine the motor pattern for the HE motor neurons of segments 8 to 14: the Δϕ of the premotor interneurons of the CPG and their synaptic strength ( Wright and Calabrese , 2011a , 2011b ) . For this study , we quantified Δϕ and synaptic strength for two of the four pairs of the premotor interneurons on both sides simultaneously ( Figure 2A1 ) . These factors combine with the intrinsic properties of the HE motor neurons and their electrical coupling between bilateral homologs to determine when an individual HE motor neuron fires in a heartbeat cycle ( Shafer and Calabrese , 1981; Wright and Calabrese , 2011b ) . While the phase difference of the premotor HN interneurons is the same in all the segments considered here , the synaptic strength of their connections to motor neurons progressively changes across segments although there is considerable individual variability ( Norris et al . , 2006 , 2007a; Norris et al . , 2011; Wright and Calabrese , 2011a , 2011b ) . In peristaltic coordination , the phase progression of the premotor bursting pattern determines the maximal phase range , and segment-specific synaptic strength pattern , intrinsic properties and coupling determines the phase realized between two ipsilateral motor neurons ( Wright and Calabrese , 2011b ) . The Δϕ that the motor neurons achieve is a portion of the Δϕ of the premotor interneurons of the CPG depending on the number of segments considered ( Figure 4 and Figure 5 ) . Therefore , because the Δϕ of the motor pattern ( HE ( 8 ) to HE ( 12 ) ) is significantly smaller than that of the CPG pattern ( Figure 4—source data 1 ) , it expresses less of the CPG’s variability . Under this hypothesis , as more or less of the CPG’s Δϕ is expressed then more or less of its variability is expressed . We found that this was the case . In some of our bilateral HE recordings we recorded the HE ( 14 ) ( N = 15 ) or the HE ( 10 ) ( N = 9 ) along with the HE ( 8 ) . Indeed , variances increased as the motor pattern’s Δϕ approached that of the CPG as more segments intervened between recorded motor neurons ( Figure 4—figure supplement 1 ) . While ongoing , leech heartbeat is episodic; the CPG , motor neurons , the hearts alternate between two coordination states at regular intervals ( Figure 2; Calabrese , 2010 ) . How similar is the same coordination when repeated on the same side a few minutes later by the exact same neurons ? We assessed this repetition variability at all levels using the Bilateral Recordings and the Intact Animal Database where we had imaged the beat pattern on both sides ( Figure 2—figure supplement 1C , D ) . We subtracted the average intersegmental Δϕ of one switch cycle from that of another , subsequent switch cycle ( ΔΔϕ = Δϕ1 - Δϕ2; Figure 6A ) and calculated the variance of that distribution . This difference between the two intersegmental phase differences ( ΔΔϕ ) is 0 when the two switch cycles have identical intersegmental phase differences ( Δϕs ) . The circular phase plots of Figure 6B show the ΔΔϕ between two consecutive switch cycles across animals for all levels , and show that , indeed , the average phase difference is near 0 . We found that consecutive bouts of the same pattern differed in half of the preparations ( CPG pattern: 10 of 24; motor pattern: 17 of 32; beat pattern: 5 of 9; unpaired t-tests; Figure 6—source data 1 ) . To evaluate the difference between repetitions , we compared these ΔΔϕ to the average intersegmental phase difference Δϕ , and found the ΔΔϕ to be an order of magnitude smaller ( ΔΔϕ vs Df : 0 . 028 vs 0 . 23 , CPG pattern; 0 . 021 vs 0 . 13 , motor pattern; 0 . 023 vs 0 . 21 , beat pattern ) . To determine how the repetition variance compared with that across animals , we calculated the variance of scrambled pairs of switch cycles 1 and 2 using all preparations in the dataset and repeated this procedure 10 , 000 times . At all levels , the repetition variance in the original population was much smaller than in the scrambled populations ( Figure 6C; p<0 . 005 for all levels ) . The differences between repetitions that we did find can be attributed not to their large size but to the overall low cycle-to-cycle variances ( Figure 3 ) . Repetition variances were equally low on the two sides and in both coordinations . The one exception which had a high repetition variance ( right heart , synchronous ) is most likely due to an outlier in this small sample . On all levels , across coordinations and sides , repetition variances were significantly lower than those in the population ( Figure 6 and Figure 6—source data 1 ) . Our results show that the repetition variances were small and suggest that they do not contribute substantially to the population variance . The leech heartbeat system is composed of bilaterally homologous elements ( interneurons , motor neurons , and hearts; Figure 1 and Figure 2 ) which allowed us to assess how similar the same coordination is when executed by the genetically identical contralateral homologs of the same neurons and muscles . We assessed this bilateral variability at all levels using the Bilateral Recordings and the animals where we had imaged the beat pattern on both sides ( Figure 2—figure supplement 1C , D ) . We subtracted the average intersegmental Δϕ of one side from that of the other side ( within coordination and level ) ( ΔΔϕ = ΔϕR - ΔϕL; Figure 7A ) and calculated the variance of that distribution . This difference between the two intersegmental phase differences ( ΔΔϕ ) is 0 when the two sides have identical intersegmental phase differences ( Δϕs ) . The circular phase plots of Figure 7B show the ΔΔϕ between the two sides across animals for each level . The mean ΔΔϕ at each level was near zero and evenly distributed , eliminating the possibility of a dominant form of handedness in the heartbeat system . Bilateral variances at each level were substantial ( Figure 7C ) . We found that the mean of the absolute values of the ΔΔϕs ( \|ΔΔϕ\| ) between sides were about 2–3 times larger than the \|ΔΔϕ\|s between repetitions ( CPG pattern: 0 . 06 vs 0 . 03; motor pattern: 0 . 04 vs 0 . 02; beat pattern: 0 . 06 vs 0 . 02; compare Figure 6—source data 1 and Figure 7—source data 1 ) . Hence bilateral variances at each level were 3-4fold higher than the repetition variances ( in 10−3 phase squared: CPG pattern , 5 . 1 vs 1 . 5; motor pattern , 2 . 1 vs 0 . 8; beat pattern , 4 . 5 vs 0 . 5; compare Figure 6—source data 1 and Figure 7—source data 1 ) . Moreover , we found that the bouts of the same coordination on the two sides differed in most cases ( CPG pattern: 20 of 26; motor pattern: 29 of 33; beat pattern: 8 of 11; unpaired t-tests; Figure 7—source data 1 ) , yet , the average phase difference is similar on the two sides ( Figure 7—figure supplement 1 ) . To determine how the bilateral variance compared with the population variance , we calculated the variances of scrambled pairs ( one from the left , one from the right side ) using all preparations in the dataset and repeated this procedure 10 , 000 times . In the CPG pattern and in the beat pattern , the bilateral variance in our data set was not significantly different from the scrambled populations ( Figure 7C; p=0 . 18 and p=0 . 15 , respectively ) . In the motor pattern , however , the bilateral variance was significantly smaller than in the scrambled populations ( p<0 . 001 ) . We obtained the same result when using the HE ( 8 ) to HE ( 14 ) motor pattern where population variance was higher ( Figure 4—figure supplement 1; plot not shown ) . In synchronous coordination , on all levels of the network , bilateral variances were lower in the original population than those of the scrambled data ( Figure 7C ) . These results suggest that at least for the CPG and the motor plant when in peristaltic coordination , differences between homologous elements , as reflected in the bilateral variances , may contribute significantly to the population variance . Across levels , intersegmental phase differences on one side do not correlate with those on the other side ( data not shown ) . The CPG network distributes its output over an ensemble of motor neurons in a stereotyped pattern of synaptic connections ( Calabrese , 1977; Thompson and Stent , 1976a ) . The HN ( 4 ) and HN ( 7 ) interneurons we recorded for this study make connections to all motor neuron pairs of segments 8 to 18 ( Calabrese , 1977; Shafer and Calabrese , 1981; Thompson and Stent , 1976b ) ( Figure 1 ) . The synaptic strengths of the individual premotor HN interneurons have distinct average segmental profiles but vary across animals ( Norris et al . , 2007a , 2007b , 2011 ) . For example , on average , synaptic strength of the HN ( 4 ) interneuron is highest in the HE ( 8 ) motor neuron and weakens towards more posterior heart motor neurons while the synaptic strength of the HN ( 7 ) interneuron is highest between the HE ( 10 ) and HE ( 14 ) motor neurons . Synaptic strengths do not change with changes in coordination states ( Norris et al . , 2007a ) . Figure 8 shows the connection strength of the premotor heart interneurons to several ipsilateral motor neurons on both sides using a subgroup of the bilateral recordings from the HN interneurons ( ‘Synaptics’ , Figure 2—figure supplement 1C ) . Comparing the two sides let us assess variabilities emerging during development in bilaterally homologous neurons . Specifically , we voltage-clamped three pairs of heart motor neurons ( HE ( R/L , 8 ) , HE ( R/L , 10 ) , and HE ( R/L , 12 ) ) , one after the other , while simultaneously recording extracellularly from two pairs of premotor interneurons ( HN ( R/L , 4 ) and HN ( R/L , 7 ) ) . Using spike-triggered averaging ( described in Norris et al . , 2007a ) , we determined the synaptic strength from the HN interneurons to their ipsilateral motor neurons in at least 10 bursts per interneuron/motor neuron pair . Figure 8A shows the synaptic strength from the HN ( 4 ) and HN ( 7 ) interneurons to each of the three ipsilateral heart motor neurons . As detailed above , synaptic strength of the HN ( 4 ) interneuron tends to be higher in the HE ( 8 ) than in the HE ( 12 ) motor neuron while synaptic strength of the HN ( 7 ) interneuron tends to be higher in the HE ( 12 ) than in the HE ( 8 ) motor neuron , and synaptic strengths tend to be about equal in the HE ( 10 ) . Synaptic strengths on the two sides vary and are not identical – but how similar are they ? To compensate for differences in the quality of voltage-clamp recording of individual neurons we computed the proportion of the synaptic strength due to the HN ( 4 ) premotor interneuron on each side for each motor neuron ( Figure 8B ) . Right and left body side did not differ ( paired t-test; p=0 . 23 HE ( 8 ) , p=0 . 91 HE ( 10 ) , p=0 . 99 HE ( 12 ) ) , indicating no systematic bilateral bias . The mean p values after randomly combining the two sides 10 000 times were HE ( 8 ) : 0 . 25 , HE ( 10 ) : 0 . 89 , and HE ( 12 ) : 0 . 98 ( Figure 8B ) . The results show that despite a lot of variation in individual strength ( nS; Figure 8A ) , the two sides do not differ in proportional synaptic strength ( Figure 8B ) indicating that bilateral variability does not seem to contribute to the population variability in synaptic strength . In the motoneuronal network of Drosophila larvae motor neurons of the same cell type contacting a common interneuron can have quite different numbers of synapses but whether the number of synapses correlate with synaptic strength is unknown ( Couton et al . , 2015 ) . Cycle-to-cycle variability within individuals reflects imprecision in the underlying neuronal networks and muscles . This variability is thought to arise principally from the stochastic nature of all biochemical processes , particularly ion channels , synaptic release , and neuromuscular systems ( Marder and Calabrese , 1996; O'Leary et al . , 2014; Roffman et al . , 2012 ) . In leech heartbeat , the cycle-to-cycle variabilities of intersegmental phase differences ( Δϕs ) were small ( Figure 3 and Figure 3—source data 1 ) . Across levels and in both coordinations they were only slightly higher than that for the bilateral pair of HN ( 4 ) interneurons , which are part of the strongly , monosynaptically , and reciprocally interconnected timing network ( Figure 1 , Figure 3A ) . Variances are reported from the gastric pattern in the isolated crab STNS within animals but this is convolved across episodes over many days ( Hamood and Marder , 2015 ) . In this episodic motor pattern , cycle-to-cycle variabilities were substantially larger than those in the ongoing leech heartbeat CPG pattern and motor pattern . Cycle-to-cycle variance of Δϕ in the swim episodes of larval zebrafish seems also rather large judging from the representative example shown ( Wiggin et al . , 2014 ) . Many motor programs are shaped , and often stabilized , on a cycle-to-cycle basis by sensory feedback . Prominent examples are locomotion e . g . , insects ( Büschges , 2005 ) , mastication in decapod crustaceans ( Marder et al . , 2014 ) , and feeding in Aplysia ( Cullins et al . , 2015a , Cullins et al . , 2015b ) . Recording from two reporter motor neurons in the Aplysia feeding circuit , Cullins et al . , 2015a found that sensory feedback increases the variability within animals , i . e . , decreasing stereotypy , but decreases variability across animals so that a common solution space for functional output emerges ( Cullins et al . , 2015a ) . Intriguingly , the variability of individual motor components is negatively correlated with their importance in behavioral performance ( Lu et al . , 2015 ) . In leech heartbeat , the premotor HN interneurons on one side dictate a common relative timing ( phase ) of the premotor inputs to all ipsilateral heart motor neurons ( Figure 1; Maranto and Calabrese , 1984a , 1984b; Norris et al . , 2007a , 2007b ) . Any local sensory input to the CPG representing a segmental perturbation , if present , would thus affect all motor neurons . Our recordings of the CPG and motor neurons are in the absence of any potential feedback . The period of leech heartbeat is modulated by a variety of inputs . Period decreases with higher metabolic demand ( e . g . , locomotion ) and higher temperatures , and increases in higher ambient oxygen ( Arbas and Calabrese , 1984 , 1990; Davis , 1986 ) . At the same time , as in other rhythmic behaviors ( e . g . , lobster and crab STNS , Bucher et al . ( 2005 ) ; Hamood et al . , 2015; crawling in larval Drosophila , Pulver et al . ( 2015 ) ; zebrafish swimming , Masino and Fetcho , 2005 ) intersegmental phase differences do not correlate with cycle period . We emphasize that this ‘phase constancy’ as embodied in a lack of correlation between phase and period across animals should not be construed as suggesting that phase does not vary within individuals or indeed the lack of correlation of period and phase . In crabs , the phase relations across many gastric episodes across many days in the isolated STNS are significantly correlated with gastric frequency as seen from two individual examples ( Hamood and Marder , 2015 ) . In larval zebrafish , swimming occurs in episodes with declining cycle period ( Masino and Fetcho , 2005 ) making this preparation ideal to determine whether intersegmental phase delay scales with the cycle period in an individual episode . The population variability in the leech , at all levels , was about 2–3 times higher than the cycle-to-cycle variability ( Figure 3 and Figure 4 ) . Similarly , in the pyloric pattern of the isolated STNS , variances were much lower within than across animals ( Hamood et al . , 2015 ) . In the episodic gastric pattern , however , variances within animals , but again convolved across episodes across many days , were comparable to those across animals ( Hamood et al . , 2015 ) . Variability increases when modulatory input is compromised ( decentralization ) , in both the pyloric and the gastric pattern in the crab STNS ( Hamood and Marder , 2015 ) . In contrast , variability in interlimb phase decreases sharply in galloping mice after ablating all V0 commissural interneurons resulting in a bounding gait ( Bellardita and Kiehn , 2015 ) . Two potential sources for the large variability in phase across animals at each level of the leech heartbeat system are variability in the episodic repeats of the same motor program within an individual , and variability between bilaterally homologous elements , also within an individual . Episodic variability is seen as the two alternating patterns are repeated periodically by the same neurons and muscles ( Figure 2 ) . Repetition variances were significantly lower than expected from the population variance across both coordinations and levels , and therefore seem not to contribute to the population variability ( Figure 6 and Figure 6—source data 1 ) . Because in much of the literature episodes within an animal and across animals are convolved it is not easy to determine whether repetition variability ( i . e . , inter-episode variability ) contributes to population variability . We suspect it does . For example , interlimb phase variance in galloping mice was assessed convolving 51 episodes across three animals so the large variance shown probably reflects large repetition variance ( Bellardita and Kiehn , 2015 ) . Similarly , in an elegant study on the motor and constriction patterns in fly larvae , covering forward and backward crawling , episodes and animals are convolved in the analysis presented , and the considerable inter-episode variability was not quantified ( Pulver et al . , 2015 ) though it certainly exists ( personal communication , Stefan Pulver ) . In zebrafish literature again episodes and animals are convolved making it difficult to parse repetition variance and population variance ( Masino and Fetcho , 2005; Wiggin et al . , 2014 ) . The heroic enterprise to record from several motor neurons in the STNS in vivo over several days ( Yarger and Stein , 2015 ) does not tease apart the phase variability across episodes and across animals for the episodic gastric motor pattern . Variability between homologous cells within an individual can also contribute to variability seen in a population . For example , in the crab heartbeat system , the five ( presumably ) homologous motor neurons – which are part of the CPG in this system – can express different levels of a common set of membrane conductances to achieve synchrony ( Lane et al . , 2016; Ransdell et al . , 2013 ) . This system seems more tightly regulated in phase than the leech heartbeat neuronal network , reflecting the synchronized rather than segment-specific nature of its motor output . We took advantage of the bilateral layout in the leech heartbeat system to test whether variability in homologous elements can contribute to population variability . In the leech , neurons arise from bilaterally paired columns of blast cells derived from bilateral pairs of stem cells ( teloblasts ) each of which arises from a symmetric division . Such symmetry in the origin of bilateral blast cells gives rises to the concept of bilaterally homologous structures and cells . Blast cells differentiate into neurons , epithelia and muscles in each segment ( Stent and Weisblat , 1985; Weisblat and Shankland , 1985 ) . Yet , there is evidence for stochastic events during leech development , for example the bilateral OP neuroblasts undergo symmetric divisions but for each the fate of the daughter cells is determined by position , and also in the formation of unpaired neurons , where either the left or the right neuron dies ( Blair et al . , 1990; Stent and Weisblat , 1985; Weisblat and Shankland , 1985 ) . At all levels , bilaterally homologous elements ( neurons and muscles ) participate in heartbeat with a single CPG orchestrating two different coordinations of motor neurons and heart muscles . Bilateral variances differed depending on level ( CPG , motor neurons , motor plant ) and coordination ( peristaltic , synchronous ) . In synchronous coordination , on all levels , the variance of differences in intersegmental Δϕ between sides ( ΔΔϕ ) was significantly less than expected from the population ( Figure 7—source data 1 ) , and therefore unlikely to contribute to the population variability . In peristaltic coordination , however , the variance between sides ( ΔΔϕ ) in the CPG pattern and in the beat pattern did not differ from the population but differed significantly from the population in the motor pattern ( Figure 7 and Figure 7—source data 1 ) . Therefore , such variability between homologous elements may contribute to the population variability at least for the CPG and the beat patterns when peristaltic . This bilateral variability might arise from stochastic processes during development as envisioned e . g . , in the models of O’Leary and Marder ( O'Leary et al . , 2013; O'Leary et al . , 2014 ) . Nevertheless , an important source of population variability is likely genetic and life history differences inherent in our population . The sources of this high bilateral variability in the CPG and the hearts are likely to be different . For example , the bilateral homologs of the premotor interneurons of the CPG vary in the synaptic strengths of their connectivity pattern ( Roffman et al . , 2012 ) and likely in their intrinsic conductances . Each heart segment receives excitatory input from its ipsilateral motor neuron , these synapses may also vary in strength , and heart muscles may vary in their intrinsic conductances . Moreover , the exact timing of that heart segment’s constriction appears to depend on load ( Wenning et al . , 2014 ) , which is unlikely to be identical on both sides in a soft-bodied animal like the leech ( Wenning and Meyer , 2007 ) . Despite the feedforward nature of the leech heartbeat system – CPG , motor neurons , heart muscle – we found that the high phase variance in the CPG in peristaltic coordination did not translate into an equally high variance in the motor pattern ( Figure 4 and Figure 5 ) . This puzzling result begs the question how the variance of the motor pattern is reduced . The peristaltic phase difference that the HE ( 8 ) to HE ( 12 ) motor neurons achieve is significantly smaller than that of the premotor interneurons of the CPG ( Figure 4 and Figure 5 ) . The phase progression of the premotor bursting pattern determines the maximal phase range , but the segment-specific synaptic strength pattern , intrinsic properties , and coupling determines the phase realized between two motor neurons ( Wright and Calabrese , 2011b ) . The premotor phase differences and the synaptic strength patterns interact so that the phase difference of the motor neurons progresses smoothly across the segments ( Wright and Calabrese , 2011b ) , and this smoothing and segment-to-segment attenuation of the CPG phase difference may limit the variance of intersegmental Δϕs , especially for nearby segments . In support of this conclusion , we observed that when a larger or smaller number of segments intervene between motor neurons are assessed , then the variance in the motor pattern reflects more or less the corresponding variance of the CPG ( Figure 4—figure supplement 1 ) . Thus , in segmentally distributed motor patterns ( such as the swim networks of lampreys , fish , leeches , crayfish swimmerets and locomotor patterns in insects ) it is important to define or , better even , to compare different sets of segments over which variability is assessed ( Büschges , 2005; Grillner and El Manira , 2015; Ingebretson and Masino , 2013; Kristan et al . , 2005; Mullins et al . , 2011; Pulver et al . , 2015; Smarandache-Wellmann et al . , 2014; Wiggin et al . , 2012 ) . We interrogated a feed-forward motor control system to determine the output variability of phase among individuals at each level - CPG , motor neurons , muscles - and found that it varied at each level , which nevertheless did not obscure recognition of distinct coordinations . We attempted to identify some of the sources of this variability in output activity . It is unlikely due to variability in performing the same function multiple times since the repetition variances are low everywhere in the system . Some of this population variability may be due to differences between homologous cells in an individual . We observed that when the same motor act is performed by bilaterally homologous CPG neurons and heart muscles , variability in peristaltic phase on the two sides is as large as in the population itself . In other cases ( peristaltic motor pattern and synchronous patterns at all levels ) , it appears likely that output variability is mainly associated with genetic and life history difference among individuals in the population . Across levels , phase variability was coordination-specific: similar at all levels in the synchronous but significantly lower for the motor pattern than for the CPG pattern in peristaltic coordination . Mechanisms involved in the transform from CPG to the motor neurons may limit the range of output variability in the motor pattern . We provide a roadmap for others that may wish to analyze variability in motor system and argue that existing data sets on the locomotor and other motor patterns of invertebrates and vertebrates can be teased apart to determine the sources of output variability . Adult leeches ( Hirudo sp . ) were obtained from commercial suppliers ( Leeches USA , Westbury , NY , or Niagara Medical Leeches ( www . leeches . biz/contact ) and kept in artificial pond water at 16°C . Prior to all procedures , leeches were cold-anesthetized in crushed ice for about 10 min . Dissections were done in ice-cold leech saline ( composition in mM: 115 NaCl , 4 KCl , 1 . 8 CaCl2 , 10 glucose , and 10 HEPES buffer , adjusted to pH 7 . 4 with NaOH ) . Animals were superfused with leech saline during the electrophysiological experiments . For video-imaging , intact , adult leeches were flattened and covered with artificial pond water ( details in Wenning et al . , 2014 ) . Experiments were done at room temperature ( 21–22°C ) . In what follows ganglion and segment numbers refer to midbody segments . Body side is indicated by R and L , i . e . , HE ( R , 8 ) is the heart motor neuron in segment eight on the right side , and heart ( L , 8 ) is the heart in segment eight on the left side . Referring to both sides is indicated by L/R , i . e . , HN ( L/R , 3 ) refers to the bilateral pair of heart interneurons in segment 3 . We define a switch cycle as the time ( or the number of neuronal bursts and heart constrictions , respectively ) a given side needs to complete both coordination states , from peristaltic to synchronous to peristaltic or vice versa , one after the other . Thus , five switches are needed to record two consecutive complete switch cycles . The Project Database contains recordings of 153 preparations: HN interneurons in 129 animals and HE motor neurons in 83 animals ( Figure 2—figure supplement 1A ) . The Project Database contains a common set ( N = 59 ) , in which both HN interneurons and HE motor neurons were recorded simultaneously ( Figure 2—figure supplement 1B; 4-point recordings ) . In a subset of the HN recordings , bilateral HN interneurons were recorded ( N = 17; 4-point recordings ) , and in a subset of the HE recordings bilateral HE motor neurons were recorded ( N = 24; 4-point-recordings ) ( Figure 2—figure supplement 1A ) . In the common set of 59 HN/HE recordings , a subset of 9 were bilateral simultaneous recordings ( Figure 2—figure supplement 1B , C; 8-point recordings ) . In this study , we report new bilateral recording of HN interneurons from a total of 26 animals and of HE motor neurons of 33 animals ( Figure 2—figure supplement 1C ) . These include data on bilateral simultaneous recordings from the HE ( 8 ) and HE ( 14 ) motor neurons ( N = 15 ) and from the HE ( 8 ) and HE ( 10 ) motor neurons ( N = 9 ) , which were recorded simultaneously with the HN ( 12 ) neurons ( 6-point recordings ) . We present new data on bilateral measurements of synaptic currents from a subset of the 26 animals in which bilateral HN recordings were made ( N = 16; ‘Synaptics’ , Figure 2—figure supplement 1C; 5-point recordings ) . We present the bilateral beat pattern for the first time from previously imaged intact adult leeches ( Wenning et al . , 2014 ) ; N = 12; Figure 1—figure supplement 1 and Figure 1—video 1; ‘Intact Animal Database’ , Figure 2—figure supplement 1D ) . Electrodes were pulled on a Flaming/Brown micropipette puller ( P-97 , Sutter Instruments; http://www . sutter . com ) from borosilicate glass ( 1 mm OD , 0 . 75 mm ID; A-M Systems; http://www . a-msystems . com ) . For extracellular recordings , suction electrodes were filled with leech saline and placed in a suction electrode holder ( E series , Warner Instruments; http://www . warneronline . com ) . To ensure a tight fit between cell and electrode , electrode tips were drawn to approximately the diameter of the cell body of the HE motor neuron ( 30 µm ) or the HN interneuron ( 15 to 20 µm ) , respectively . The electrode tip was brought in contact with the cell body and light suction was applied until the cell body was inside the electrode . Extracellular signals were monitored with a differential AC amplifier ( model 1700 , A-M Systems ) at a gain of 1000 with the low- and high-frequency cutoffs set at 100 and 1 , 000 Hz , respectively . Noise was reduced with a 60 Hz notch filter . A second amplifier ( model 410 , Brownlee Precision; http://www . brownleeprecision . com ) amplified the signal appropriately for digitization . Intracellular recording techniques and voltage clamp protocols were conventional and are described in detail in ( Norris et al . , 2007a , 2011 ) . At the end of each voltage-clamp experiment , the electrode was withdrawn from the motor neuron . The experiment was accepted if the electrode potential was within ±5 mV of ground . Thus , holding potentials were accurate within ±5 mV . Data were digitized ( >5 kHz sampling rate ) , using a digitizing board ( Digi-Data 1200 or 1550 Series Interface ( http://www . moleculardevices . com ) , and acquired using pCLAMP software ( http://www . moleculardevices . com ) on a personal computer . Electrophysiological recordings were done in isolated chains of ganglia . Those ganglia in which we recorded HN interneurons or HE motor neurons extracellularly were desheathed . We recorded from the HN ( L/R , 4 ) and HN ( L/R , 7 ) interneurons in 26 animals . In two animals , only one switch cycle was recorded so we assessed the repetition variability in 24 recordings . In 9 of the 26 animals we recorded simultaneously the HE ( L/R , 8 ) and HE ( L/R , 12 ) motor neurons ( Figure 2—figure supplement 1C ) . We recorded from the HE ( L/R , 8 ) and HE ( L/R , 12 ) motor neurons in 33 animals . In one animal , only one switch cycle was recorded so we assessed the repetition variability in 32 recordings . In 9 of the 33 animals we recorded simultaneously the HN ( L/R , 4 ) and HN ( L/R , 7 ) interneurons ( see above ) ( Figure 2—figure supplement 1C ) . In 20 animals , we attempted to voltage-clamp 6 motor neurons , one after another: the HE ( L/R , 8 ) , HE ( L/R , 10 ) , and HE ( L/R , 12 ) motor neurons to determine the synaptic strength from the HN ( L/R , 4 ) and HN ( L/R , 7 ) heart interneurons recorded simultaneously . Using spike-triggered averaging ( for a detailed description of the methods see Norris et al . , 2006 , 2007a ) , we determined the synaptic strength in the left and right HE motor neurons ( HE ( 8 ) : N = 9 , HE ( 10 ) : N = 8 , and HE ( 12 ) : N = 10 ) in a total of 16 animals , i . e . in some animals , we successfully voltage-clamped several pairs of HE motor neurons , one after the other ( Figure 2—figure supplement 1C ) . Video-imaging of intact , adult leeches provided the optical signals to determine the bilateral beat patterns , in the motor plant ( i . e . , the two hearts ) in midbody segments 7 to 14 ( N = 12; Figure 1—Video 1 ) ( Figure 2—figure supplement 1D ) . These data were compiled previously , and data acquisition and analysis were described in detail in ( Wenning et al . , 2011 , 2014 ) . As for the motor pattern , we used segments 8 and 12 . Imaging was limited to 10 min , which did not yield two complete switch cycles in all 12 animals . We assessed the repetition variability in 9 ( left side ) and 8 animals ( right side ) , respectively . One animal was identified as an outlier ( outside the 1 . 5*interquartile range on a Whisker barrel plot ) because of the irregular sequence of constrictions in segments 8 and 9 on both sides in synchronous coordination . To further examine the intersegmental phase differences and the variability across individuals , we used our current Project Database which includes unilateral recordings from the HN ( 4 ) and HN ( 7 ) interneurons and from the HE ( 8 ) and HE ( 12 ) motor neurons since bilateral recordings are not necessary for this analysis ( Figure 2—figure supplement 1A; N = 129 , CPG pattern; N = 83 , motor pattern ) . Some of these data were published ( Norris et al . , 2007a , 2011; Wenning et al . , 2014 ) and some were presented at the Society for Neuroscience meeting in 2016 ( Norris et al . , 2016 ) . We used specific points in time to calculate the intersegmental phase difference Δϕ between bursts of the two pairs of HN interneurons , the two pairs of HE motor neurons , and the constrictions of the two pairs of heart segments . The detailed description of the methods and the custom-made MATLAB codes have been published ( Cymbalyuk et al . , 2002; Masino and Calabrese , 2002c; Norris et al . , 2007b; Wenning et al . , 2004a , 2004b ) , so we summarize briefly here . To characterize the bursting patterns of the HN interneurons and HE motor neurons , spikes were detected based on threshold and then grouped into bursts ( interburst interval ≥1 s ) . Stray spikes were eliminated . The middle spike ( based on count ) served as the phase marker for an individual burst . We then calculated burst period ( T ) and phase ( ϕ ) . The burst period was defined as the interval in seconds from middle spike to middle spike of consecutive bursts ( e . g . Figure 2 , A2 for the HN premotor interneurons ) . Phases were referenced to an absolute phase reference ( ϕ = 0 ) , the HN ( 4 ) interneuron on the right side . The phases of the HN ( 4 ) premotor interneuron on the left side , those of the two HN ( 7 ) premotor interneurons , and , if applicable , those of the four HE motor neurons were determined on a cycle-to-cycle basis . Phase differences ( HN ( R , 4 ) - HN ( R , 7 ) , HN ( L 4 ) - HN ( L , 7 ) , HE ( R , 8 ) – HE ( R , 12 ) , and HE ( L , 8 ) - HE ( L , 12 ) , respectively , were defined as the difference between the time of their middle spikes ( ti ) and the time of the middle spike of the reference segment ( tr ) in the same cycle divided by the reference cycle period ( Tr ) expressed as ( ϕr-i = [ ( ti –tr ) /Tr] ) . For the beat pattern of the hearts , the phase reference was the maximum rate of rise ( MRR ) of the digitized optical signals of an individual heartbeat cycle ( Figure 1—figure supplement 1B , inset ) . The MRR corresponds to emptying ( systole ) and maximal heart constriction ( Wenning et al . , 2014 ) . We determined the intersegmental Δϕ between heart segments 8 and 12 on the left and on the right side , using the ipsilateral heart segment 8 as the phase reference . Inhibitory postsynaptic currents ( IPSCs ) from the ipsilateral HN ( 4 ) and HN ( 7 ) heart interneurons were recorded in three pairs of heart motor neurons ( HE ( R/L , 8 ) , HE ( R/L , 10 ) , and HE ( R/L , 12 ) ) . To normalize across different holding potentials , IPSCs were converted to , and reported as , conductances ( reversal potential: −62 mV; Angstadt and Calabrese , 1991 ) . Angular variances were calculated by using the Cartesian average of polar phase vectors . Circular ( or polar ) plots were drawn , and the statistics were calculated using the Pandora Toolbox ( Günay et al . , 2009 ) . The code , custom scripts , and the data can be found in: Sources of variability in a motor system , Calabrese , 2018 ) . Using vector summation , we found the mean vector phase and length r ( as a measure of concentration ) with 1-r as a measure of dispersion . From the vector length , we calculated the angular variance s2 = 2 ( 1-r ) ( in radians squared ) and from this the angular standard deviation s ( in radians ) ( Zar , 1974 ) . We report these values in phase units squared or phase units by dividing by 4π2 and 2π , respectively . To assess the cycle-to-cycle variability , we used the intersegmental phase differences between two segments ( Δϕ ) burst-by-burst and beat-by-beat , respectively , within body side and coordination . Data on consecutive switch cycles were kept separate . This analysis yielded 4 data sets per body side for each level ( peristaltic 1 , synchronous 1 , peristaltic 2 , synchronous 2 ) ; 24 sets total . To assess the population variability , we used the average intersegmental Δϕ of a single switch cycle on each level , for each side and coordination . To assess whether these population variances differed between levels we resampled the data ( with replacement; 10 , 000 times; ‘bootstrapping’ ) , calculated the 95% confidence intervals , and compared variances between two levels . This analysis yielded 4 data sets for each level , 12 sets total . To assess the repetition variability , we first determined whether the cycle-to-cycle intersegmental Δϕ on one side in a given coordination differed between two consecutive switch cycles ( unpaired t-test ) . To compare the repetition variability across animals , we calculated the difference of the average intersegmental Δϕ between two consecutive switch cycles ( the Δ of Δϕ ) , within coordination and side . Note that repetitions of the same coordination are separated by the time it takes to execute the other coordination , about 2–4 min . For example , for the CPG we calculatedΔΔϕ=Δϕcycle1 ( HN ( L , 4 ) −HN ( L , 7 ) ) −Δϕcycle2 ( HN ( L , 4 ) −HN ( L , 7 ) ) We plotted the individual ΔΔϕ in a circular phase plot and calculated their variance . This analysis yielded 4 data sets for each level , 12 sets total . We repeated this procedure 10 , 000 times after randomly combining ( scrambling ) two switch cycles from different animals , and calculated the p values from the z score of the normal distribution of the scrambled populations . The motor pattern and the beat pattern were treated the same way . To assess the bilateral variability , we used one switch cycle . We first determined whether the cycle-to-cycle intersegmental Δϕ differed between the two sides ( unpaired t-test ) . To compare the bilateral variability across animals , we calculated the difference of the average intersegmental Δϕ within coordination between the right and the left side ( the Δof Δϕ ) . Note that the two sides do not execute the same coordination at the same time . For the CPG , we calculatedΔΔϕ=Δϕright ( HN ( L , 4 ) −HN ( L , 7 ) ) −Δϕleft ( HN ( L , 4 ) −HN ( L , 7 ) ) We plotted the individual ΔΔϕ in a circular phase plot , and calculated their variance . This analysis yielded two data sets ( one for each coordination ) per level , 6 sets total . We repeated this procedure 10 , 000 times after scrambling the data from the right and left body sides from different animals , and calculated the p values from the z score of the normal distribution of the scrambled populations . The motor pattern and the beat pattern were treated the same way . We assessed the variability of the synaptic input from premotor heart interneurons HN ( R , 4 ) and HN ( R , 7 ) to heart motor neurons HE ( R , 8 ) , HE ( R , 10 ) and HE ( R , 12 ) , and from heart interneurons HN ( L , 4 ) and HN ( L , 7 ) to heart motor neurons HE ( L , 8 ) , HE ( L , 10 ) and HE ( L , 12 ) . This analysis yielded two data sets per motor neuron pair , 6 sets total ( Figure 8A ) . To evaluate the synaptic input further and to eliminate any differences due to the quality of the voltage-clamp recordings , we calculated the proportion of the synaptic strength due to the HN ( 4 ) premotor interneuron on each side using its ratio to the sum of the synaptic inputs of both premotor interneurons HN ( 4 ) /[HN ( 4 ) + HN ( 7 ) ] , separately for each side ( Figure 8B ) . We used a paired t-test to assess whether the two sides were different . Next , we scrambled the data from the right and left body sides from different animals ( 10 , 000 times ) , and calculated their p values using a paired t-test . We used means ±SD for descriptive statistics .	Many of our everyday behaviors are rhythmic actions , such as walking , breathing and chewing . Networks of neurons called Central Pattern Generators , or CPGs , are in charge of rhythmic behaviors . CPGs send instructions to cells called motor neurons , which in turn tell muscles to contract in a particular sequence to produce rhythmic behaviors . Rhythmic behaviors follow stereotyped patterns: we recognize walking when we see it . But they also vary between individuals: we can recognize the specific gait or ‘walk’ of a friend . Wenning et al . set out to discover where this variability in rhythmic behaviors comes from , using the leech heartbeat system as a model . Leeches have two hearts , or more precisely two heart tubes that run along the entire length of the body , one on either side . The two heart tubes beat with different patterns , but under the direction of the CPGs and motor neurons , they swap patterns with each other every few minutes . The CPG neurons that generate these rhythms , the motor neurons that respond , and the heart muscles themselves , i . e . each level of the system , can all be tracked in leeches . Wenning et al . showed that within each leech , the activity of the CPG neurons , motor neurons and muscles associated with a heart tube varies little . Even when the activity of one of these levels varies less than another , for example between CPG and motor neurons , it is not necessarily reflected in the next level of the system . In some cases , however , variability is seen between opposite sides . Moreover , the rhythmic activity of CPG neurons , motor neurons , and muscle cells in one leech differs greatly from that of another . This likely reflects differences in the genes and life history of the animals . Wenning et al . provide a roadmap for others to use in identifying sources of variability in rhythmic movements . Applying this approach to existing data sets could help tease apart variability in diverse rhythmic behaviors in a variety of animals .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2018	Output variability across animals and levels in a motor system
The ability to computationally predict whether a compound treats a disease would improve the economy and success rate of drug approval . This study describes Project Rephetio to systematically model drug efficacy based on 755 existing treatments . First , we constructed Hetionet ( neo4j . het . io ) , an integrative network encoding knowledge from millions of biomedical studies . Hetionet v1 . 0 consists of 47 , 031 nodes of 11 types and 2 , 250 , 197 relationships of 24 types . Data were integrated from 29 public resources to connect compounds , diseases , genes , anatomies , pathways , biological processes , molecular functions , cellular components , pharmacologic classes , side effects , and symptoms . Next , we identified network patterns that distinguish treatments from non-treatments . Then , we predicted the probability of treatment for 209 , 168 compound–disease pairs ( het . io/repurpose ) . Our predictions validated on two external sets of treatment and provided pharmacological insights on epilepsy , suggesting they will help prioritize drug repurposing candidates . This study was entirely open and received realtime feedback from 40 community members . The cost of developing a new therapeutic drug has been estimated at 1 . 4 billion dollars ( DiMasi et al . , 2016 ) , the process typically takes 15 years from lead compound to market ( Reichert , 2003 ) , and the likelihood of success is stunningly low ( Hay et al . , 2014 ) . Strikingly , the costs have been doubling every 9 years since 1970 , a sort of inverse Moore’s law , which is far from an optimal strategy from both a business and public health perspective ( Scannell et al . , 2012 ) . Drug repurposing — identifying novel uses for existing therapeutics — can drastically reduce the duration , failure rates , and costs of approval ( Ashburn and Thor , 2004 ) . These benefits stem from the rich preexisting information on approved drugs , including extensive toxicology profiling performed during development , preclinical models , clinical trials , and postmarketing surveillance . Drug repurposing is poised to become more efficient as mining of electronic health records ( EHRs ) to retrospectively assess the effect of drugs gains feasibility ( Wang et al . , 2015; Xu et al . , 2015; Brilliant et al . , 2016; Tatonetti et al . , 2012 ) . However , systematic approaches to repurpose drugs based on mining EHRs alone will likely lack power due to multiple testing . Similar to the approach followed to increase the power of genome-wide association studies ( GWAS ) ( Stephens and Balding , 2009; Sawcer , 2008 ) , integration of biological knowledge to prioritize drug repurposing will help overcome limited EHR sample size and data quality . In addition to repurposing , several other paradigm shifts in drug development have been proposed to improve efficiency . Since small molecules tend to bind to many targets , polypharmacology aims to find synergy in the multiple effects of a drug ( Roth et al . , 2004 ) . Network pharmacology assumes diseases consist of a multitude of molecular alterations resulting in a robust disease state . Network pharmacology seeks to uncover multiple points of intervention into a specific pathophysiological state that together rehabilitate an otherwise resilient disease process ( Hopkins , 2008; Hopkins , 2007 ) . Although target-centric drug discovery has dominated the field for decades , phenotypic screens have more recently resulted in a comparatively higher number of first-in-class small molecules ( Swinney and Anthony , 2011 ) . Recent technological advances have enabled a new paradigm in which mid- to high-throughput assessment of intermediate phenotypes , such as the molecular response to drugs , is replacing the classic target discovery approach ( Iskar et al . , 2012; Lamb , 2007; Qu and Rajpal , 2012 ) . Furthermore , integration of multiple channels of evidence , particularly diverse types of data , can overcome the limitations and weak performance inherent to data of a single domain ( Hodos et al . , 2016 ) . Modern computational approaches offer a convenient platform to tie these developments together as the reduced cost and increased velocity of in silico experimentation massively lowers the barriers to entry and price of failure ( Hurle et al . , 2013; Liu et al . , 2013 ) . Hetnets ( short for heterogeneous networks ) are networks with multiple types of nodes and relationships . They offer an intuitive , versatile , and powerful structure for data integration by aggregating graphs for each relationship type onto common nodes . In this study , we developed a hetnet ( Hetionet v1 . 0 ) by integrating knowledge and experimental findings from decades of biomedical research spanning millions of publications . We adapted an algorithm originally developed for social network analysis and applied it to Hetionet v1 . 0 to identify patterns of efficacy and predict new uses for drugs . The algorithm performs edge prediction through a machine learning framework that accommodates the breadth and depth of information contained in Hetionet v1 . 0 ( Himmelstein and Baranzini , 2015a; Sun et al . , 2011 ) . Our approach represents an in silico implementation of network pharmacology that natively incorporates polypharmacology and high-throughput phenotypic screening . One fundamental characteristic of our method is that it learns and evaluates itself on existing medical indications ( i . e . a 'gold standard’ ) . Next , we introduce previous approaches that also performed comprehensive evaluation on existing treatments . A 2011 study , named PREDICT , compiled 1933 treatments between 593 drugs and 313 diseases ( Gottlieb et al . , 2011 ) . Starting from the premise that similar drugs treat similar diseases , PREDICT trained a classifier that incorporates five types of drug-drug and two types of disease-disease similarity . A 2014 study compiled 890 treatments between 152 drugs and 145 diseases with transcriptional signatures ( Cheng et al . , 2014 ) . The authors found that compounds triggering an opposing transcriptional response to the disease were more likely to be treatments , although this effect was weak and limited to cancers . A 2016 study compiled 402 treatments between 238 drugs and 78 diseases and used a single proximity score — the average shortest path distance between a drug’s targets and disease’s associated proteins on the interactome — as a classifier ( Guney et al . , 2016 ) . We build on these successes by creating a framework for incorporating the effects of any biological relationship into the prediction of whether a drug treats a disease . By doing this , we were able to capture a multitude of effects that have been suggested as influential for drug repurposing including drug-drug similarity ( Gottlieb et al . , 2011; Li and Lu , 2012 ) , disease-disease similarity ( Gottlieb et al . , 2011; Chiang and Butte , 2009 ) , transcriptional signatures ( Lamb , 2007; Qu and Rajpal , 2012; Cheng et al . , 2014; Lamb et al . , 2006; Iorio et al . , 2013 ) , protein interactions ( Guney et al . , 2016 ) , genetic association ( Nelson et al . , 2015; Sanseau et al . , 2012 ) , drug side effects ( Campillos et al . , 2008; Nugent et al . , 2016 ) , disease symptoms ( Zhou et al . , 2014 ) , and molecular pathways ( Pratanwanich and Lió , 2014 ) . Our ability to create such an integrative model of drug efficacy relies on the hetnet data structure to unite diverse information . On Hetionet v1 . 0 , our algorithm learns which types of compound–disease paths discriminate treatments from non-treatments in order to predict the probability that a compound treats a disease . We refer to this study as Project Rephetio ( pronounced as rep-het-ee-oh ) . Both Rephetio and Hetionet are portmanteaus combining the words repurpose , heterogeneous , and network with the URL het . io . We obtained and integrated data from 29 publicly available resources to create Hetionet v1 . 0 ( Figure 1 ) . The hetnet contains 47 , 031 nodes of 11 types ( Table 1 ) and 2 , 250 , 197 relationships of 24 types ( Table 2 ) . The nodes consist of 1552 small molecule compounds and 137 complex diseases , as well as genes , anatomies , pathways , biological processes , molecular functions , cellular components , perturbations , pharmacologic classes , drug side effects , and disease symptoms . The edges represent relationships between these nodes and encompass the collective knowledge produced by millions of studies over the last half century . For example , Compound–binds–Gene edges represent when a compound binds to a protein encoded by a gene . This information has been extracted from the literature by human curators and compiled into databases such as DrugBank , ChEMBL , DrugCentral , and BindingDB . We combined these databases to create 11 , 571 binding edges between 1389 compounds and 1689 genes . These edges were compiled from 10 , 646 distinct publications , which Hetionet binding edges reference as an attribute . Binding edges represent a comprehensive catalog constructed from low-throughput experimentation . However , we also integrated findings from high-throughput technologies — many of which have only recently become available . For example , we generated consensus transcriptional signatures for compounds in LINCS L1000 and diseases in STARGEO . While Hetionet v1 . 0 is ideally suited for drug repurposing , the network has broader biological applicability . For example , we have prototyped queries for ( a ) identifying drugs that target a specific pathway , ( b ) identifying biological processes involved in a specific disease , ( c ) identifying the drug targets responsible for causing a specific side effect , and ( d ) identifying anatomies with transcriptional relevance for a specific disease ( Himmelstein , 2016j ) . Each of these queries was simple to write and took less than a second to run on our publicly available Hetionet Browser . Although it is possible that existing services provide much of the aforementioned functionality , they offer less versatility . Hetionet differentiates itself in its ability to flexibly query across multiple domains of information . As a proof of concept , we enhanced the biological process query ( b ) , which identified processes that were enriched for disease-associated genes , using multiple sclerosis ( MS ) as an example disease . The verbose Cypher code for this query is shown below:MATCH path = //Specify the type of path to match ( n0:Disease ) -[e1:ASSOCIATES_DaG]- ( n1:Gene ) -[:INTERACTS_GiG]- ( n2:Gene ) -[:PARTICIPATES_GpBP]- ( n3:BiologicalProcess ) WHERE //Specify the source and target nodes n0 . name = 'multiple sclerosis' AND n3 . name = 'retina layer formation' //Require GWAS support for the Disease-associates-Gene relationship AND 'GWAS Catalog' in e1 . sources //Require the interacting gene to be upregulated in a relevant tissue AND exists ( ( n0 ) -[:LOCALIZES_DlA]- ( :Anatomy ) -[:UPREGULATES_AuG]- ( n2 ) ) RETURN path The query above identifies genes that interact with MS GWAS-genes . However , interacting genes are discarded unless they are upregulated in an MS-related anatomy ( i . e . anatomical structure , e . g . organ or tissue ) . Then relevant biological processes are identified . Thus , this single query spans four node and five relationship types . The integrative potential of Hetionet v1 . 0 is reflected by its connectivity . Among the 11 metanodes , there are 66 possible source–target pairs . However , only 11 of them have at least one direct connection . In contrast , for paths of length 2 , 50 pairs have connectivity ( paths types that start on the source node type and end on the target node type , see Figure 1C ) . At length 3 , all 66 pairs are connected . At length 4 , the source–target pair with the fewest types of connectivity ( Side Effect to Symptom ) has 13 metapaths , while the pair with the most connectivity types ( Gene to Gene ) has 3542 pairs . This high level of connectivity across a diversity of biomedical entities forms the foundation for automated translation of knowledge into biomedical insight . Hetionet v1 . 0 is accessible via a Neo4j Browser at https://neo4j . het . io . This public Neo4j instance provides users an installation-free method to query and visualize the network . The Browser contains a tutorial guide as well as guides with the details of each Project Rephetio prediction . Hetionet v1 . 0 is also available for download in JSON , Neo4j , and TSV formats ( Himmelstein , 2017a ) . The JSON and Neo4j database formats include node and edge properties — such as URLs , source and license information , and confidence scores — and are thus recommended . One aim of Project Rephetio was to systematically evaluate how drugs exert their therapeutic potential . To address this question , we compiled a gold standard of 755 disease-modifying indications , which form the Compound–treats–Disease edges in Hetionet v1 . 0 . Next , we identified types of paths ( metapaths ) that occurred more frequently between treatments than non-treatments ( any compound–disease pair that is not a treatment ) . The advantage of this approach is that metapaths naturally correspond to mechanisms of pharmacological efficacy . For example , the Compound–binds–Gene–associates–Disease ( CbGaD ) metapath identifies when a drug binds to a protein corresponding to a gene involved in the disease . We evaluated all 1206 metapaths that traverse from compound to disease and have length of 2–4 ( Figure 2A ) . To control for the different degrees of nodes , we used the degree-weighted path count ( DWPC , see Materials and methods ) — which downweights paths going through highly connected nodes ( Himmelstein and Baranzini , 2015a ) — to assess path prevalence . In addition , we compared the performance of each metapath to a baseline computed from permuted networks . Hetnet permutation preserves node degree while eliminating edge specificity , allowing us to isolate the portion of unpermuted metapath performance resulting from actual network paths . We refer to the permutation-adjusted performance measure as Δ AUROC . A positive Δ AUROC indicates that paths of the given type tended to occur more frequently between treatments than non-treatments , after accounting for different levels of connectivity ( node degrees ) in the hetnet . In general terms , Δ AUROC assesses whether paths of a given type were informative of drug efficacy . Overall , 709 of the 1206 metapaths exhibited a statistically significant Δ AUROC at a false discovery rate cutoff of 5% . These 709 metapaths included all 24 metaedges , suggesting that each type of relationship we integrated provided at least some therapeutic utility . However , not all metaedges were equally present in significant metapaths: 259 significant metapaths included a Compound–binds–Gene metaedge , whereas only four included a Gene–participates–Cellular Component metaedge . Table 3 lists the predictiveness of several metapaths of interest . Refer to the Discussion for our interpretation of these findings . We implemented a machine learning approach to translate the network connectivity between a compound and a disease into a probability of treatment ( Himmelstein , 2016k; Himmelstein , 2017b ) . The approach relies on the 755 known treatments as positives and 29 , 044 non-treatments as negatives to train a logistic regression model . Note that 179 , 369 non-treatments were omitted as negative training observations because they had a prior probability of treatment equal to zero ( see Materials and methods ) . The features consisted of a prior probability of treatment , node degrees for 14 metaedges , and DWPCs for 123 metapaths that were well suited for modeling . A cross-validated elastic net was used to minimize overfitting , yielding a model with 31 features ( Figure 2B ) . The DWPC features with negative coefficients appear to be included as node-degree-capturing covariates , i . e . they reflect the general connectivity of the compound and disease rather than specific paths between them . However , the 11 DWPC features with non-negligible positive coefficients represent the most salient types of connectivity for systematically modeling drug efficacy . See the metapaths with positive coefficients in Table 3 for unabbreviated names . As an example , the CcSEcCtD feature assesses whether the compound causes the same side effects as compounds that treat the disease . Alternatively , the CbGeAlD feature assesses whether the compound binds to genes that are expressed in the anatomies affected by the disease . We applied this model to predict the probability of treatment between each of 1538 connected compounds and each of 136 connected diseases , resulting in predictions for 209 , 168 compound–disease pairs ( Himmelstein et al . , 2016a ) , available at http://het . io/repurpose/ . The 755 known disease-modifying indications were highly ranked ( AUROC = 97 . 4% , Figure 3 ) . The predictions also successfully prioritized two external validation sets: novel indications from DrugCentral ( AUROC = 85 . 5% ) and novel indications in clinical trial ( AUROC = 70 . 0% ) . Together , these findings indicate that Project Rephetio has the ability to recognize efficacious compound–disease pairs . Predictions were scaled to the overall prevalence of treatments ( 0 . 36% ) . Hence a compound–disease pair that received a prediction of 1% represents a twofold enrichment over the null probability . Of the 3980 predictions with a probability exceeding 1% , 586 corresponded to known disease-modifying indications , leaving 3394 repurposing candidates . For a given compound or disease , we provide the percentile rank of each prediction . Therefore , users can assess whether a given prediction is a top prediction for the compound or disease . In addition , our table-based prediction browser links to a custom guide for each prediction , which displays in the Neo4j Hetionet Browser . Each guide includes a query to display the top paths supporting the prediction and lists clinical trials investigating the indication . There are currently two FDA-approved medications for smoking cessation ( varenicline and bupropion ) that are not nicotine replacement therapies . PharmacotherapyDB v1 . 0 lists varenicline as a disease-modifying indication and nicotine itself as a symptomatic indication for nicotine dependence , but is missing bupropion . Bupropion was first approved for depression in 1985 . Owing to the serendipitous observation that it decreased smoking in depressed patients taking this drug , Bupropion was approved for smoking cessation in 1997 ( Harmey et al . , 2012 ) . Therefore , we looked whether Project Rephetio could have predicted this repurposing . Bupropion was the ninth best prediction for nicotine dependence ( 99 . 5th percentile ) with a probability 2 . 50-fold greater than the null . Figure 4 shows the top paths supporting the repurposing of bupropion . Atop the nicotine dependence predictions were nicotine ( 10 . 97-fold over null ) , cytisine ( 10 . 58-fold ) , and galantamine ( 9 . 50-fold ) . Cytisine is widely used in Eastern Europe for smoking cessation due to its availability at a fraction of the cost of other pharmaceutical options ( Cahill et al . , 2016 ) . In the last half decade , large-scale clinical trials have confirmed cytisine’s efficacy ( West et al . , 2011; Walker et al . , 2014 ) . Galantamine , an approved Alzheimer’s treatment , is currently in Phase 2 trial for smoking cessation and is showing promising results ( Ashare et al . , 2016 ) . In summary , nicotine dependence illustrates Project Rephetio’s ability to predict efficacious treatments and prioritize historic and contemporary repurposing opportunities . Several factors make epilepsy an interesting disease for evaluating repurposing predictions ( Khankhanian and Himmelstein , 2016 ) . Antiepileptic drugs work by increasing the seizure threshold — the amount of electric stimulation that is required to induce seizure . The effect of a drug on the seizure threshold can be cheaply and reliably tested in rodent models . As a result , the viability of most approved drugs in treating epilepsy is known . We focused our evaluation on the top 100 scoring compounds — referred to as the epilepsy predictions in this section — after discarding a single combination drug . We classified each compound as anti-ictogenic ( seizure suppressing ) , unknown ( no established effect on the seizure threshold ) , or ictogenic ( seizure generating ) according to medical literature ( Khankhanian and Himmelstein , 2016 ) . Of the top 100 epilepsy predictions , 77 were anti-ictogenic , eight were unknown , and 15 were ictogenic ( Figure 5A ) . Notably , the predictions contained 23 of the 25 disease-modifying antiepileptics in PharamcotherapyDB v1 . 0 . Many of the 77 anti-ictogenic compounds were not first-line antiepileptic drugs . Instead , they were used as ancillary drugs in the treatment of status epilepticus . For example , we predicted four halogenated ethers , two of which ( isoflurane and desflurane ) are used clinically to treat life-threatening seizures that persist despite treatment ( Mirsattari et al . , 2004 ) . As inhaled anesthetics , these compounds are not appropriate as daily epilepsy medications , but are feasible for refractory status epilepticus where patients are intubated . Given this high precision ( 77% ) , the eight compounds of unknown effect are promising repurposing candidates . For example , acamprosate — whose top prediction was epilepsy — is a taurine analog that promotes alcohol abstinence . Support for this repurposing arose from acamprosate’s inhibition of the glutamate receptor and positive modulation of the GABAA receptor ( Figure 5C ) . If effective against epilepsy , acamprosate could serve a dual benefit for recovering alcoholics who experience seizures from alcohol withdrawal . While certain classes of compounds were highly represented in our epilepsy predictions , such benzodiazepines and barbiturates , there was also considerable diversity ( Khankhanian and Himmelstein , 2016 ) . The 100 predicted compounds encompassed 26 third-level ATC codes ( Knaus , 2016 ) , such as antiarrhythmics ( quinidine , classified as anti-ictogenic ) and urologicals ( phenazopyridine , classified as unknown ) . Furthermore , 25 of the compounds were chemically distinct , i . e . they did not resemble any of the other epilepsy predictions ( Figure 5B ) . Next , we investigated which components of Hetionet contributed to the epilepsy predictions ( Khankhanian and Himmelstein , 2016 ) . In total , 392 , 956 paths of 12 types supported the predictions . Using several different methods for grouping paths , we were able to quantify the aggregate biological evidence . Our algorithm primarily drew on two aspects of epilepsy: its known treatments ( 76% of the total support ) and its genetic associations ( 22% of support ) . In contrast , our algorithm drew heavily on several aspects of the predicted compounds: their targeted genes ( 44% ) , their chemically similar compounds ( 30% ) , their pharmacologic classes , their palliative indications ( 5% ) , and their side effects ( 4% ) . Specifically , 266 , 192 supporting paths originated with a Compound–binds–Gene relationship . Aggregating support by these genes shows the extent that 121 different drug targets contributed to the predictions ( Khankhanian and Himmelstein , 2016 ) . In order of importance , the predictions targeted GABAA receptors ( 15 . 3% of total support ) , cytochrome P450 enzymes ( 5 . 6% ) , the sodium channel ( 4 . 6% ) , glutamate receptors ( 3 . 8% ) , the calcium channel ( 2 . 7% ) , carbonic anhydrases ( 2 . 5% ) , cholinergic receptors ( 2 . 1% ) , and the potassium channel ( 1 . 4% ) . Besides cytochrome P450 , which primarily influences pharmacokinetics ( Johannessen and Landmark , 2010 ) , our method detected and leveraged bonafide anti-ictogenic mechanisms ( Rogawski and Löscher , 2004 ) . Figure 5C shows drug target contributions per compound and illustrates the considerable mechanistic diversity among the predictions . Also notable are the 15 ictogenic compounds in our top 100 predictions . Nine of the ictogenic compounds share a tricyclic structure ( Figure 5B ) , five of which are tricyclic antidepressants . While the ictogenic mechanisms of these antidepressants are still unclear ( Johannessen Landmark et al . , 2016 ) , Figure 5C suggests their anticholinergic effects may be responsible ( Himmelstein , 2017d ) , in accordance with previous theories ( Dailey and Naritoku , 1996 ) . We also ranked the contribution of the 1137 side effects that supported the epilepsy predictions through 117 , 720 CcSEcCtD paths . The top five side effects — ataxia ( 0 . 069% of total support ) , nystagmus ( 0 . 049% ) , diplopia ( 0 . 045% ) , somnolence ( 0 . 044% ) , and vomiting ( 0 . 043% ) — reflect established adverse effects of antiepileptic drugs ( Zadikoff et al . , 2007; Wu and Thijs , 2015; ROFF HILTONHilton et al . , 2004; Placidi et al . , 2000; Jahromi et al . , 2011 ) . In summary , our method simultaneously identified the hallmark side effects of antiepileptic drugs while incorporating this knowledge to prioritize 1538 compounds for anti-ictogenic activity . We created Hetionet v1 . 0 by integrating 29 resources into a single data structure — the hetnet . Consisting of 11 types of nodes and 24 types of relationships , Hetionet v1 . 0 brings more types of information together than previous leading-studies in biological data integration ( Gligorijević and Pržulj , 2015 ) . Moreover , we strove to create a reusable , extensible , and property-rich network . While all the resources we include are publicly available , their integration was a time-intensive undertaking and required careful consideration of legal barriers to data reuse . Hetionet allows researchers to begin answering integrative questions without having to first spend months processing data . Our public Neo4j instance allows users to immediately interact with Hetionet . Through the Cypher language , users can perform highly specialized graph queries with only a few lines of code . Queries can be executed in the web browser or programmatically from a language with a Neo4j driver . For users that are unfamiliar with Cypher , we include several example queries in a Browser guide . In contrast to traditional REST APIs , our public Neo4j instance provides users with maximal flexibility to construct custom queries by exposing the underlying database . As data has grown more plentiful and diverse , so has the applicability of hetnets . Unfortunately , network science has been naturally fragmented by discipline resulting in relatively slow progress in integrating heterogeneous data . A 2014 analysis identified 78 studies using multilayer networks — a superset of hetnets ( heterogeneous information networks ) with the potential for additional dimensions , such as time . However , the studies relied on 26 different terms , 9 of which had multiple definitions ( Kivela et al . , 2014; Himmelstein et al . , 2015b ) . Nonetheless , core infrastructure and algorithms for hetnets are emerging . Compared to the existing mathematical frameworks for multilayer networks that must deal with layers other than type ( such as the aspect of time ) ( Kivela et al . , 2014 ) , the primary obligation of hetnet algorithms is to be type aware . One goal of our project has been to unite hetnet research across disciplines . We approached this goal by making Project Rephetio entirely available online and inviting community feedback throughout the process ( Himmelstein et al . , 2015c ) . Integrating every resource into a single interconnected data structure allowed us to assess systematic mechanisms of drug efficacy . Using the max performing metapath to assess the pharmacological utility of a metaedge ( Figure 2A ) , we can divide our relationships into tiers of informativeness . The top tier consists of the types of information traditionally considered by pharmacology: Compound–treats–Disease , Pharmacologic Class–includes–Compound , Compound–resembles–Compound , Disease–resembles–Disease , and Compound–binds–Gene . The upper-middle tier consists of types of information that have been the focus of substantial medical study , but have only recently started to play a bigger role in drug development , namely the metaedges Disease–associates–Gene , Compound–causes–Side Effect , Disease–presents–Symptom , Disease–localizes–Anatomy , and Gene–interacts–Gene . The lower-middle tier contains the transcriptomics metaedges such as Compound–downregulates–Gene , Anatomy–expresses–Gene , Gene→regulates→Gene , and Disease–downregulates–Gene . Much excitement surrounds these resources due to their high-throughput and genome-wide scope , which offers a route to drug discovery that is less biased by existing knowledge . However , our findings suggest that these resources are only moderately informative of drug efficacy . Other lower-middle tier metaedges were the product of time-intensive biological experimentation , such as Gene–participates–Pathway , Gene–participates–Molecular Function , and Gene–participates–Biological Process . Unlike the top tier resources , this knowledge has historically been pursued for basic science rather than primarily medical applications . The weak yet appreciable performance of the Gene–covaries–Gene suggests the synergy between the fields of evolutionary genomics and disease biology . The lower tier included the Gene–participates–Cellular Component metaedge , which may reflect that the relevance of cellular location to pharmacology is highly case dependent and not amenable to systematic profiling . The performance of specific metapaths ( Table 3 ) provides further insight . For example , significant emphasis has been put on the use of transcriptional data for drug repurposing ( Iorio et al . , 2013 ) . One common approach has been to identify compounds with opposing transcriptional signatures to a disease ( Qu and Rajpal , 2012; Sirota et al . , 2011 ) . However , several systematic studies report underwhelming performance of this approach ( Gottlieb et al . , 2011; Cheng et al . , 2014; Guney et al . , 2016 ) — a finding supported by the low performance of the CuGdD and CdGuD metapaths in Project Rephetio . Nonetheless , other transcription-based methods showed some promise . Compounds with similar transcriptional signatures were prone to treating the same disease ( CuGuCtD and CdGdCtD metapaths ) , while compounds with opposing transcriptional signatures were slightly averse to treating the same disease ( CuGdCtD and CdGuCtD metapaths ) . In contrast , diseases with similar transcriptional profiles were not prone to treatment by the same compound ( CtDdGuD and CtDuGdD ) . By comparably assessing the informativeness of different metaedges and metapaths , Project Rephetio aims to guide future research towards promising data types and analyses . One caveat is that omics-scale experimental data will likely play a larger role in developing the next generation of pharmacotherapies . Hence , were performance reevaluated on treatments discovered in the forthcoming decades , the predictive ability of these data types may rise . Encouragingly , most data types were at least weakly informative and hence suitable for further study . Ideally , different data types would provide orthogonal information . However , our model for whether a compound treats a disease focused on 11 metapaths — a small portion of the hundreds of metapaths available . While parsimony aids interpretation , our model did not draw on the weakly-predictive high-throughput data types — which are intriguing for their novelty , scalability , and cost-effectiveness — as much as we had hypothesized . Instead our model selected types of information traditionally considered in pharmacology . However , unlike a pharmacologist whose area of expertise may be limited to a few drug classes , our model was able to predict probabilities of treatment for all 209 , 168 compound–disease pairs . Furthermore , our model systematically learned the importance of each type of network connectivity . For any compound–disease pair , we now can immediately provide the top network paths supporting its therapeutic efficacy . A traditional pharmacologist may be able to produce a similar explanation , but likely not until spending substantial time researching the compound’s pharmacology , the disease’s pathophysiology , and the molecular relationships in between . Accordingly , we hope certain predictions will spur further research , such as trials to investigate the off-label use of acamprosate for epilepsy , which is supported by one animal model ( Farook et al . , 2008 ) . As demonstrated by the 15 ictogenic compounds in our top 100 epilepsy predictions , Project Rephetio’s predictions can include contraindications in addition to indications . Since many of Hetionet v1 . 0’s relationship types are general ( e . g . the Compound–binds–Gene relationship type conflates antagonist with agonist effects ) , we expect some high scoring predictions to exacerbate rather than treat the disease . However , the predictions made by Hetionet v1 . 0 represent such substantial relative enrichment over the null that uncovering the correct directionality is a logical next step and worth undertaking . Going forward , advances in automated mining of the scientific literature could enable extraction of precise relationship types at omics scale ( Ehrenberg et al . , 2016; Himmelstein et al . , 2016b ) . Future research should focus on gleaning orthogonal information from data types that are so expansive that computational methods are the only option . Our CuGuCtD feature — measuring whether a compound upregulates the same genes as compounds which treat the disease — is a good example . This metapath was informative by itself ( Δ AUROC = 4 . 4% ) but was not selected by the model , despite its orthogonal origin ( gene expression ) to selected metapaths . Using a more extensive catalog of treatments as the gold standard would be one possible approach to increase the power of feature selection . Integrating more types of information into Hetionet should also be a future priority . The ‘network effect’ phenomenon suggests that the addition of each new piece of information will enhance the value of Hetionet’s existing information . We envision a future where all biological knowledge is encoded into a single hetnet . Hetionet v1 . 0 was an early attempt , and we hope the strong performance of Project Rephetio in repurposing drugs foreshadows the many applications that will thrive from encoding biology in hetnets . Nodes encode entities . We extracted nodes from standard terminologies , which provide curated vocabularies to enable data integration and prevent concept duplication . The ease of mapping external vocabularies , adoption , and comprehensiveness were primary selection criteria . Hetionet v1 . 0 includes nodes from five ontologies — which provide hierarchy of entities for a specific domain — selected for their conformity to current best practices ( Malone et al . , 2016 ) . We selected 137 terms from the Disease Ontology ( Schriml et al . , 2012; Kibbe et al . , 2015 ) ( which we refer to as DO Slim ( Himmelstein and Li , 2015d; Himmelstein , 2016g ) ) as our disease set . Our goal was to identify complex diseases that are distinct and specific enough to be clinically relevant yet general enough to be well annotated . To this end , we included diseases that have been studied by GWAS and cancer types from TopNodes_DOcancerslim ( Wu et al . , 2015 ) . We ensured that no DO Slim disease was a subtype of another DO Slim disease . Symptoms were extracted from MeSH by taking the 438 descendants of Signs and Symptoms ( Himmelstein and Pankov , 2015a; Himmelstein , 2016h ) . Approved small molecule compounds with documented chemical structures were extracted from DrugBank version 4 . 2 ( Law et al . , 2014; Himmelstein , 2015b; Himmelstein , 2016i ) . Unapproved compounds were excluded because our focus was repurposing . In addition , unapproved compounds tend to be less studied than approved compounds making them less attractive for our approach where robust network connectivity is critical . Finally , restricting to small molecules with known documented structures enabled us to map between compound vocabularies ( see Mappings ) . Side effects were extracted from SIDER version 4 . 1 ( Kuhn et al . , 2016; Himmelstein , 2015c; Himmelstein , 2016j ) . SIDER codes side effects using UMLS identifiers ( Bodenreider , 2004 ) , which we also adopted . Pharmacologic Classes were extracted from the DrugCentral data repository ( Ursu et al . , 2017; Himmelstein et al . , 2016d ) . Only pharmacologic classes corresponding to the ‘Chemical/Ingredient’ , ‘Mechanism of Action’ , and ‘Physiologic Effect’ FDA class types were included to avoid pharmacologic classes that were synonymous with indications ( Himmelstein et al . , 2016d ) . Protein-coding human genes were extracted from Entrez Gene ( Maglott et al . , 2011; Himmelstein et al . , 2015h; Himmelstein , 2016l ) . Anatomical structures , which we refer to as anatomies , were extracted from Uberon ( Mungall et al . , 2012 ) . We selected a subset of 402 Uberon terms by excluding terms known not to exist in humans and terms that were overly broad or arcane ( Malladi et al . , 2015; Himmelstein , 2016m ) . Pathways were extracted by combining human pathways from WikiPathways ( Kutmon et al . , 2016; Pico et al . , 2008 ) , Reactome ( Fabregat et al . , 2016 ) , and the Pathway Interaction Database ( Schaefer et al . , 2009 ) . The latter two resources were retrieved from Pathway Commons ( RRID:SCR_002103 ) ( Cerami et al . , 2011 ) , which compiles pathways from several providers . Duplicate pathways and pathways without multiple participating genes were removed ( Pico and Himmelstein , 2015; Himmelstein and Pico , 2016a ) . Biological processes , cellular components , and molecular functions were extracted from the Gene Ontology ( Ashburner et al . , 2000 ) . Only terms with 2–1000 annotated genes were included . Before adding relationships , all identifiers needed to be converted into the vocabularies matching that of our nodes . Oftentimes , our node vocabularies included external mappings . For example , the Disease Ontology includes mappings to MeSH , UMLS , and the ICD , several of which we submitted during the course of this study ( Himmelstein , 2015e ) . In a few cases , the only option was to map using gene symbols , a disfavored method given that it can lead to ambiguities . When mapping external disease concepts onto DO Slim , we used transitive closure . For example , the UMLS concept for primary progressive multiple sclerosis ( C0751964 ) was mapped to the DO Slim term for multiple sclerosis ( DOID:2377 ) . Chemical vocabularies presented the greatest mapping challenge ( Himmelstein , 2015b ) , since these are poorly standardized ( Hersey et al . , 2015 ) . UniChem’s ( Chambers et al . , 2013 ) Connectivity Search ( Chambers et al . , 2014 ) was used to map compounds , which maps by atomic connectivity ( based on First InChIKey Hash Blocks ( Heller et al . , 2013 ) ) and ignores small molecular differences . Anatomy–downregulates–Gene and Anatomy–upregulates–Gene edges ( Himmelstein et al . , 2016f; Himmelstein and Bastian , 2015e; Himmelstein and Bastian , 2015f ) were extracted from Bgee ( Bastian et al . , 2008 ) , which computes differentially expressed genes by anatomy in post-juvenile adult humans . Anatomy–expresses–Gene edges were extracted from Bgee and TISSUES ( Santos et al . , 2015; Himmelstein and Jensen , 2015g; Himmelstein and Jensen , 2015h ) . Compound–binds–Gene edges were aggregated from BindingDB ( Chen et al . , 2001; Gilson et al . , 2016 ) , DrugBank ( Law et al . , 2014; Wishart et al . , 2006 ) , and DrugCentral ( Ursu et al . , 2017 ) . Only binding relationships to single proteins with affinities of at least 1 μM ( as determined by Kd , Ki , or IC₅₀ ) were selected from the October 2015 release of BindingDB ( Himmelstein and Gilson , 2015i; Himmelstein et al . , 2015d ) . Target , carrier , transporter , and enzyme interactions with single proteins ( i . e . excluding protein groups ) were extracted from DrugBank 4 . 2 ( Himmelstein , 2016i; Himmelstein and Protein , 2015j ) . In addition , all mapping DrugCentral target relationships were included ( Himmelstein et al . , 2016d ) . Compound–treats–Disease ( disease-modifying indications ) and Compound–palliates–Disease ( symptomatic indications ) edges are from PharmacotherapyDB as described in Intermediate resources . Compound–causes–Side Effect edges were obtained from SIDER 4 . 1 ( Kuhn et al . , 2016; Himmelstein , 2015c; Himmelstein , 2016j ) , which uses natural language processing to identify side effects in drug labels . Compound–resembles–Compound relationships ( Himmelstein , 2016i; Himmelstein and Chen , 2015k; Himmelstein et al . , 2015q ) represent chemical similarity and correspond to a Dice coefficient ≥0 . 5 ( Dice , 1945 ) between extended connectivity fingerprints ( Rogers and Hahn , 2010; Morgan , 1965 ) . Pharmacologic Class–includes–Compound edges were extracted from DrugCentral for three FDA class types ( Ursu et al . , 2017; Himmelstein et al . , 2016d ) . Compound–downregulates–Gene and Compound–upregulates–Gene relationships were computed from LINCS L1000 as described in Intermediate resources . Disease–associates–Gene edges were extracted from the GWAS Catalog ( Himmelstein and Baranzini , 2016b ) , DISEASES ( Himmelstein and Jensen , 2015l; Himmelstein and Jensen , 2016c ) , DisGeNET ( Himmelstein , 2015f; Himmelstein and Piñero , 2016d ) , and DOAF ( Himmelstein , 2015g; Himmelstein , 2016s ) . The GWAS Catalog compiles disease–SNP associations from published GWAS ( MacArthur et al . , 2017 ) . We aggregated overlapping loci associated with each disease and identified the mode reported gene for each high confidence locus ( Himmelstein , 2015h; Himmelstein et al . , 2015v ) . DISEASES integrates evidence of association from text mining , curated catalogs , and experimental data ( Pletscher-Frankild et al . , 2015 ) . Associations from DISEASES with integrated scores ≥ 2 were included after removing the contribution of DistiLD . DisGeNET integrates evidence from over 10 sources and reports a single score for each association ( Piñero et al . , 2015; Piñero et al . , 2017 ) . Associations with scores ≥ 0 . 06 were included . DOAF mines Entrez Gene GeneRIFs ( textual annotations of gene function ) for disease mentions ( Xu et al . , 2012 ) . Associations with three or more supporting GeneRIFs were included . Disease–downregulates–Gene and Disease–upregulates–Gene relationships ( Himmelstein et al . , 2015a; Himmelstein et al . , 2016j ) were computed using STARGEO as described in Intermediate resources . Disease–localizes–Anatomy , Disease–presents–Symptom , and Disease–resembles–Disease edges were calculated from MEDLINE co-occurrence ( Himmelstein and Pankov , 2015a; Himmelstein , 2016u ) . MEDLINE is a subset of 21 million PubMed articles for which designated human curators have assigned topics . When retrieving articles for a given topic ( MeSH term ) , we activated two non-default search options as specified below: majr for selecting only articles where the topic is major and noexp for suppressing explosion ( returning articles linked to MeSH subterms ) . We identified 4 , 161 , 769 articles with two or more disease topics; 696 , 252 articles with both a disease topic ( majr ) and an anatomy topic ( noexp ) ( Himmelstein , 2015i ) ; and 363 , 928 articles with both a disease topic ( majr ) and a symptom topic ( noexp ) . We used a Fisher’s exact test ( Fisher , 1922 ) to identify pairs of terms that occurred together more than would be expected by chance in their respective corpus . We included co-occurring terms with p<0 . 005 in Hetionet v1 . 0 . Gene→regulates→Gene directed edges were generated from the LINCS L1000 genetic interference screens ( see Intermediate resources ) and indicate that knockdown or overexpression of the source gene significantly dysregulated the target gene ( Himmelstein and Chung , 2015q; Himmelstein et al . , 2016k ) . Gene–covaries–Gene edges represent evolutionary rate covariation ≥0 . 75 ( Priedigkeit et al . , 2015; Himmelstein and Partha , 2015r; Himmelstein , 2016w ) . Gene–interacts–Gene edges ( Himmelstein et al . , 2015z; Himmelstein and Baranzini , 2016e ) represent when two genes produce physically interacting proteins . We compiled these interactions from the Human Interactome Database ( Rual et al . , 2005; Venkatesan et al . , 2009; Yu et al . , 2011; Rolland et al . , 2014 ) , the Incomplete Interactome ( Menche et al . , 2015 ) , and our previous study ( Himmelstein and Baranzini , 2015a ) . Gene–participates–Biological Process , Gene–participates–Cellular Component , and Gene–participates–Molecular Function edges are from Gene Ontology annotations ( Huntley et al . , 2015 ) . As described in Intermediate resources , annotations were propagated ( Himmelstein et al . , 2015g; Himmelstein et al . , 2015f ) . Gene–participates–Pathway edges were included from the human pathway resources described in the Nodes section ( Pico and Himmelstein , 2015; Himmelstein and Pico , 2016a ) . Whether a certain type of relationship has directionality is defined at the metaedge level . Directed metaedges are only necessary when they connect a metanode to itself and correspond to an asymmetric relationship . In the case of Hetionet v1 . 0 , the sole directed metaedge was Gene→regulates→Gene . To demonstrate the implications of directionality , Hetionet v1 . 0 contains two relationships between the genes HADH and STAT1: HADH–interacts–STAT1 and HADH→regulates→STAT1 . Both edges can be represented in the inverse orientation: STAT1–interacts–HADH and STAT1←regulates←HADH . However due to directed nature of the regulates relationship , STAT1→regulates→HADH is a distinct edge , which does not exist in the network . Similarly , HADH–associates–obesity and obesity–associates–HADH are inverse orientations of the same underlying undirected relationship . Accordingly , the following path exists in the network: obesity–associates–HADH→regulates→STAT1 , which can also be inverted to STAT1←regulates←HADH–associates–obesity . In the process of creating Hetionet , we produced several datasets with broad applicability that extended beyond Project Rephetio . These resources are referred to as intermediate resources and described below . STARGEO is a nascent platform for annotating and meta-analyzing differential gene expression experiments ( Hadley et al . , 2017 ) . The STAR acronym stands for Search-Tag-Analyze Resources , while GEO refers to the Gene Expression Omnibus ( Edgar et al . , 2002; Barrett et al . , 20122013 ) . STARGEO is a layer on top of GEO that crowdsources sample annotation and automates meta-analysis . Using STARGEO , we computed differentially expressed genes between healthy and diseased samples for 49 diseases ( Himmelstein et al . , 2015a; Himmelstein et al . , 2016j ) . First , we and others created case/control tags for 66 diseases . After combing through GEO series and tagging samples , 49 diseases had sufficient data for case-control meta-analysis: multiple series with at least three cases and three controls . For each disease , we performed a random effects meta-analysis on each gene to combine log₂ fold-change across series . These analyses incorporated 27 , 019 unique samples from 460 series on 107 platforms . Differentially expressed genes ( false discovery rate ≤0 . 05 ) were identified for each disease . The median number of upregulated genes per disease was 351 and the median number of downregulated genes was 340 . Endogenous depression was the only of the 49 diseases without any significantly dysregulated genes . LINCS L1000 profiled the transcriptional response to small molecule and genetic interference perturbations . To increase throughput , expression was only measured for 978 genes , which were selected for their ability to impute expression of the remaining genes . A single perturbation was often assayed under a variety of conditions including cell types , dosages , timepoints , and concentrations . Each condition generates a single signature of dysregulation z-scores . We further processed these signatures to fit into our approach ( Himmelstein et al . , 2016m; Himmelstein et al . , 2016n ) . First , we computed consensus signatures — which meta-analyze multiple signatures to condense them into one — for DrugBank small molecules , Entrez genes , and all L1000 perturbations ( Himmelstein and Chung , 2015q; Himmelstein et al . , 2016k ) . First , we discarded non-gold ( non-replicating or indistinct ) signatures . Then , we meta-analyzed z-scores using Stouffer’s method . Each signature was weighted by its average Spearman’s correlation to other signatures , with a 0 . 05 minimum , to de-emphasize discordant signatures . Our signatures include the 978 measured genes and the 6489 imputed genes from the ‘best inferred gene subset’ . To identify significantly dysregulated genes , we selected genes using a Bonferroni cutoff of p=0 . 05 and limited the number of imputed genes to 1000 . The consensus signatures for genetic perturbations allowed us to assess various characteristics of the L1000 dataset . First , we looked at whether genetic interference dysregulated its target gene in the expected direction ( Himmelstein , 2016c ) . Looking at measured z-scores for target genes , we found that the knockdown perturbations were highly reliable , while the overexpression perturbations were only moderately reliable with 36% of overexpression perturbations downregulating their target . However , imputed z-scores for target genes barely exceeded chance at responding in the expected direction to interference . Hence , we concluded that the imputation quality of LINCS L1000 is poor . However , when restricting to significantly dyseregulated targets , 22 out of 29 imputed genes responded in the expected direction . This provides some evidence that the directional fidelity of imputation is higher for significantly dysregulated genes . Finally , we found that the transcriptional signatures of knocking down and overexpressing the same gene were positively correlated 65% of the time , suggesting the presence of a general stress response ( Himmelstein et al . , 2016o ) . Based on these findings , we performed additional filtering of signifcantly dysregulated genes when building Hetionet v1 . 0 . Compound–down/up-regulates–Gene relationships were restricted to the 125 most significant per compound-direction-status combination ( status refers to measured versus imputed ) . For genetic interference perturbations , we restricted to the 50 most significant genes per gene-direction-status combination and merged the remaining edges into a single Gene→regulates→Gene relationship type containing both knockdown and overexpression perturbations . We created PharmacotherapyDB , an open catalog of drug therapies for disease ( Himmelstein , 2016a; Himmelstein et al . , 2016p; Himmelstein et al . , 2016q ) . Version 1 . 0 contains 755 disease-modifying therapies and 390 symptomatic therapies between 97 diseases and 601 compounds . This resource was motivated by the need for a gold standard of medical indications to train and evaluate our approach . Initially , we identified four existing indication catalogs ( Himmelstein et al . , 2015e ) : MEDI-HPS which mined indications from RxNorm , SIDER 2 , MedlinePlus , and Wikipedia ( Wei et al . , 2013 ) ; LabeledIn which extracted indications from drug labels via human curation ( Khare et al . , 2014; Khare et al . , 2015; Himmelstein and Khare , 2015s ) ; EHRLink which identified medication–problem pairs that clinicians linked together in electronic health records ( McCoy et al . , 2012; Himmelstein , 2015j ) ; and indications from PREDICT , which were compiled from UMLS relationships , drugs . com , and drug labels ( Gottlieb et al . , 2011 ) . After mapping to DO Slim and DrugBank Slim , the four resources contained 1388 distinct indications . However , we noticed that many indications were palliative and hence problematic as a gold standard of pharmacotherapy for our in silico approach . Therefore , we recruited two practicing physicians to curate the 1388 preliminary indications ( Himmelstein et al . , 2015j ) . After a pilot on 50 indications , we defined three classifications: disease modifying meaning a drug that therapeutically changes the underlying or downstream biology of the disease; symptomatic meaning a drug that treats a significant symptom of the disease; and non-indication meaning a drug that neither therapeutically changes the underlying or downstream biology nor treats a significant symptom of the disease . Both curators independently classified all 1388 indications . The two curators disagreed on 444 calls ( Cohen’s κ = 49 . 9% ) . We then recruited a third practicing physician , who reviewed all 1388 calls and created a detailed explanation of his methodology ( Himmelstein et al . , 2015j ) . We proceeded with the third curator’s calls as the consensus curation . The first two curators did have reservations with classifying steroids as disease modifying for autoimmune diseases . We ultimately considered that these indications met our definition of disease modifying , which is based on a pathophysiological rather than clinical standard . Accordingly , therapies we consider disease modifying may not be used to alter long-term disease course in the modern clinic due to a poor risk–benefit ratio . We created a browser ( http://git . dhimmel . com/gene-ontology/ ) to provide straightforward access to Gene Ontology annotations ( Himmelstein et al . , 2015g; Himmelstein et al . , 2015f ) . Our service provides annotations between Gene Ontology terms and Entrez Genes . The user chooses propagated/direct annotation and all/experimental evidence . Annotations are currently available for 37 species and downloadable as user-friendly TSV files . We committed to openly releasing our data and analyses from the origin of the project ( Spaulding et al . , 2015 ) . Our goals were to contribute to the advancement of science ( Hrynaszkiewicz , 2011; Molloy , 2011 ) , maximize our impact ( McKiernan et al . , 2016; Piwowar and Vision , 2013 ) , and enable reproducibility ( Stodden et al . , 2016; Stodden and Miguez , 2014; Baggerly , 2010 ) . These objectives required publicly distributing and openly licensing Hetionet and Project Rephetio data and analyses ( Hrynaszkiewicz and Cockerill , 2012; Hagedorn et al . , 2011 ) . Since we integrated only public resources , which were overwhelmingly funded by academic grants , we had assumed that our project and open sharing of our network would not be an issue . However , upon releasing a preliminary version of Hetionet ( Himmelstein and Jensen , 2015u ) , a community reviewer informed us of legal barriers to integrating public data . In essence , both copyright ( rights of exclusivity automatically granted to original works ) and terms of use ( rules that users must agree to in order to use a resource ) place legally binding restrictions on data reuse . In short , public data is not by default open data . Hetionet v1 . 0 integrates 29 resources ( Table 4 ) , but two resources were removed prior to the v1 . 0 release . Of the total 31 resources ( Himmelstein et al . , 2015i ) , 5 were United States government works not subject to copyright , and 12 had licenses that met the Open Definition of knowledge version 2 . 1 . Four resources allowed only non-commercial reuse . Most problematic were the remaining nine resources that had no license — which equates to all rights reserved by default and forbids reuse ( Oxenham , 2016 ) — and one resource that explicitly forbid redistribution . Additional difficulty resulted from license incompatibles across resources , which was caused primarily by non-commercial and share-alike stipulations . Furthermore , it was often unclear who owned the data ( Elliott , 2005 ) . Therefore , we sought input from legal experts and chronicled our progress ( Himmelstein et al . , 2015i; Himmelstein , 2015k; Himmelstein et al . , 2016r; Himmelstein , 2015a; Himmelstein , 2015d ) . Ultimately , we did not find an ideal solution . We had to choose between absolute compliance and Hetionet: strictly adhering to copyright and licensing arrangements would have decimated the network . On the other hand , in the United States , mere facts are not subject to copyright , and fair use doctrine helps protect reuse that is transformative and educational . Hence , we choose a path forward which balanced legal , normative , ethical , and scientific considerations . If a resource was in the public domain , we licensed any derivatives as CC0 1 . 0 . For resources licensed to allow reuse , redistribution , and modification , we transmitted their licenses as properties on the specific nodes and relationships in Hetionet v1 . 0 . For all other resources — for example , resources without licenses or with licenses that forbid redistribution — we sent permission requests to their creators . The median time till first response to our permission requests was 16 days , with only two resources affirmatively granting us permission . We did not receive any responses asking us to remove a resource . However , we did voluntarily remove MSigDB ( Liberzon et al . , 2011 ) , since its license was highly problematic ( Himmelstein , 2015d ) . As a result of our experience , we recommend that publicly funded data should be explicitly dedicated to the public domain whenever possible . From Hetionet , we derived five permuted hetnets ( Himmelstein , 2016b ) . The permutations preserve node degree but eliminate edge specificity by employing an algorithm called XSwap to randomly swap edges ( Hanhijärvi et al . , 2009 ) . To extend XSwap to hetnets ( Himmelstein and Baranzini , 2015a ) , we permuted each metaedge separately , so that edges were only swapped with other edges of the same type . We adopted a Markov chain approach , whereby the first permuted hetnet was generated from Hetionet v1 . 0 , the second permuted hetnet was generated from the first , and so on . For each metaedge , we assessed the percent of edges unchanged as the algorithm progressed to ensure that a sufficient number of swaps had been performed to randomize the network ( Himmelstein , 2016b ) . Permuted hetnets are useful for computing the baseline performance of meaningless edges while preserving node degree ( Himmelstein , 2015l ) . Since , our use of permutation focused on assessing Δ AUROC , a small number of permuted hetnets was sufficient , as the variability in a metapath’s AUROC across the permuted hetnets was low . Traditional relational databases — such as SQLite , MySQL , and PostgreSQL — excel at storing highly structured data in tables . Connectivity between tables is accomplished using foreign-key references between columns . However , for many biomedical applications the connectivity between entities is of foremost importance . Furthermore , enforcing a rigid structure of what attributes an entity may possess is less important and often unnecessarily prohibitive . Graph databases focus instead on capturing connectivity ( relationships ) between entities ( nodes ) . Accordingly , graph databases such as Neo4j offer greater ease when modeling biomedical relationships and superior performance when traversing many levels of connectivity ( Yoon et al . , 2017; Jaiswal , 2013 ) . Until recently , graph database adoption in bioinformatics was limited ( Have and Jensen , 2013 ) . However lately , the demand to model and capture biological connectivity at scale has led to increasing adoption ( Lysenko et al . , 2016; Balaur et al . , 2016; Summer et al . , 2016; Mungall et al . , 2017 ) . We used the Neo4j graph database for storing and operating on Hetionet and noticed major benefits from tapping into this large open source ecosystem ( Himmelstein , 2015m ) . Persistent storage with immediate access and the Cypher query language — a sort of SQL for hetnets — were two of the biggest benefits . To facilitate our migration to Neo4j , we updated hetio — our existing Python package for hetnets ( Himmelstein , 2016g ) — to export networks into Neo4j and DWPC queries to Cypher . In addition , we created an interactive GraphGist for Project Rephetio , which introduces our approach and showcases its Cypher queries . Finally , we created a public Neo4j instance ( Himmelstein , 2016i ) , which leverages several modern technologies such Neo4j Browser guides , cloud hosting with HTTPS , and Docker deployment ( Belmann et al . , 2015; Beaulieu-Jones and Greene , 2017 ) . Project Rephetio relied on the previously published DWPC metric to generate features for compound–disease pairs . The DWPC measures the prevalence of a given metapath between a given source and target node ( Himmelstein and Baranzini , 2015a ) . It is calculated by first extracting all paths from the source to target node that follow the specified metapath . Next , each path is weighted by taking the product of the node degrees along the path raised to a negative exponent . This damping exponent — the sole parameter — thereby determines the extent that paths through high-degree nodes are downweighted: we chose w = 0 . 4 based on our past optimizations ( Himmelstein and Baranzini , 2015a ) . The DWPC equals the sum of the path weights ( referred to as path-degree products ) . Traversing the hetnet to extract all paths between a source and target node , which we performed in Neo4j , is the most computationally intensive step in computing DWPCs ( Himmelstein and Lizee , 2016t ) . For future work , we are exploring matrix multiplication approaches , which could improve runtime several orders of magnitude . Project Rephetio made several refinements to metapath-based hetnet edge prediction compared to previous studies ( Himmelstein and Baranzini , 2015a; Sun et al . , 2011 ) . First , we transformed DWPCs by mean scaling and then taking the inverse hyperbolic sine ( Burbidge et al . , 1988 ) to make them more amenable to modeling ( Himmelstein et al . , 2016s ) . Second , we bifurcated the workflow into an all-features stage and an all-observations stage ( Himmelstein , 2016k ) . The all-features stage assesses feature performance and does not require computing features for all negatives . Here , we selected a random subset of 3020 ( 4 × 755 ) negatives . Little error was introduced by this optimization , since the predominant limitation to performance assessment was the small number of positives ( 755 ) rather than negatives . Based on the all-features performance assessment ( Himmelstein , 2015n ) , we selected 142 DWPCs to compute on all observations ( all 209 , 168 compound–disease pairs ) . The feature selection was designed to remove uninformative features ( according to permutation ) and guard against edge-dropout contamination ( Himmelstein , 2016h ) . Third , we included 14 degree features , which assess the degree of a specific metaedge for either the source compound or target disease . To improve the interpretability of the predictions , we developed a method for decomposing a prediction into its network support ( Himmelstein , 2016e ) . This information is deployed to our Neo4j Browser guides , allowing users to assess the biomedical evidence contributing to a given prediction . First , we used logistic regression terms to quantify the contribution of metapaths that positively support a prediction . Second , we decomposed a metapath’s contribution , according to its DWPC , into specific paths contributions . Finally , we aggregated paths based on their source ( first ) or target ( last ) edge to quantify the contribution of specific edges of the source compound or target disease ( Himmelstein , 2016f ) . Using the acamprosate–epilepsy prediction as an example , we first quantified metapath contributions: 40% of the prediction was supported by CbGbCtD paths , 36% by CbGaD paths , 11% by CcSEcCtD paths , 8% by CbGpPWpGaD paths , and 5% by CbGeAlD paths . Second , we calculated path contributions: Acamprosate–binds–GRM5–associates–epilepsy syndrome was the most supportive path , contributing 11% of the prediction . Finally , we aggregated path contributions to calculate that the source edge of Acamprosate—binds—GRM5 contributed 23% of the prediction , while the target edge of epilepsy syndrome–treats–Felbamate contributed 12% . The 755 treatments in Hetionet v1 . 0 are not evenly distributed between all compounds and diseases . For example , methotrexate treats 19 diseases and hypertension is treated by 68 compounds . We estimated a prior probability of treatment — based only on the treatment degree of the source compound and target disease — on 744 , 975 permutations of the bipartite treatment network ( Lizee and Himmelstein , 2016a ) . Methotrexate received a 79 . 6% prior probability of treating hypertension , whereas a compound and disease that both had only one treatment received a prior of 0 . 12% . Across the 209 , 168 compound–disease pairs , the prior predicted the known treatments with AUROC = 97 . 9% . The strength of this association threatened to dominate our predictions . However , not modeling the prior can lead to omitted-variable bias and confounded proxy variables . To address the issue , we included the logit-transformed prior , without any regularization , as a term in the model . This restricted model fitting to the 29 , 799 observations with a nonzero prior — corresponding to the 387 compounds and 77 diseases with at least one treatment . To enable predictions for all 209 , 168 observations , we set the prior for each compound–disease pair to the overall prevalence of positives ( 0 . 36% ) . This method succeeded at accommodating the treatment degrees . The prior probabilities performed poorly on the validation sets with AUROC = 54 . 1% on DrugCentral indications and AUROC = 62 . 5% on clinical trials . This performance dropoff compared to training shows the danger of encoding treatment degree into predictions . The benefits of our solution are highlighted by the superior validation performance of our predictions compared to the prior ( Figure 3 ) . We evaluated our predictions on four sets of indications as shown in Figure 3 . Only the Clinical Trial and DrugCentral indication sets were used for external validation , since the Disease Modifying and Symptomatic indications were included in the hetnet . As an aside , several additional indication catalogs have recently been published , which future studies may want to also consider ( Himmelstein et al . , 2015e; Brown and Patel , 2017; Shameer et al . , 2017; Sharp , 2017 ) . We conducted our study using Thinklab — a platform for real-time open collaborative science — on which this study was the first project ( Himmelstein et al . , 2015c ) . We began the study by publicly proposing the idea and inviting discussion ( Himmelstein et al . , 2015k ) . We continued by chronicling our progress via discussions . We used Thinklab as the frontend to coordinate and report our analyses and GitHub as the backend to host our code , data , and notebooks . On top of our Thinklab team consisting of core contributors , we welcomed community contribution and review . In areas where our expertise was lacking or advice would be helpful , we sought input from domain experts and encouraged them to respond on Thinklab where their comments would be CC BY licensed and their contribution rated and rewarded . In total , 40 non-team members commented across 86 discussions , which generated 622 comments and 191 notes ( Figure 6 ) . Thinklab content for this project totaled 145 , 771 words or 918 , 837 characters ( Himmelstein and Lizee , 2016v ) . Using an estimated 7000 words per academic publication as a benchmark , Project Rephetio generated written content comparable in volume to 20 . 8 publications prior to its completion . We noticed several other benefits from using Thinklab including forging a community of contributors ( Patil and Siegel , 2009 ) ; receiving feedback during the early stages when feedback was most actionable ( Mietchen et al . , 2015 ) ; disseminating our research without delay ( Powell , 2016; Vale , 2015 ) ; opening avenues for external input ( Allison et al . , 2016 ) ; facilitating problem-oriented teaching ( Himmelstein et al . , 2016t; Waldrop , 2015 ) ; and improving our documentation by maintaining a publication-grade digital lab notebook ( Giles , 2012 ) . Thinklab began winding down operations in July 2017 and has switched to a static state . While users will no longer be able to add comments , the corpus of content remains browsable at https://think-lab . github . io and available in machine-readable formats at dhimmel/thinklytics . The preprint for this study is available at doi . org/bs4f ( Himmelstein et al . , 2016u ) . The manuscript was written in markdown , originally on Thinklab at doi . org/bszr ( Himmelstein et al . , 2016v ) . In August 2017 , we switched to using the Manubot system to generate the manuscript . With Manubot , a GitHub repository ( dhimmel/rephetio-manuscript ) tracks the manuscript’s source code , while continuous integration automatically rebuilds the manuscript upon changes . As a result , the latest version of the manuscript is always available at dhimmel . github . io/rephetio-manuscript . Additionally , readers can leave feedback or questions for the Project Rephetio team via GitHub Issues . All software and datasets from Project Rephetio are publicly available on GitHub , Zenodo , or Figshare ( Himmelstein et al . , 2017b ) . Additional documentation for these materials is available in the corresponding Thinklab discussions . For reader convenience , software , datasets , and Thinklab discussions have been cited throughout the manuscript as relevant . Copies of the most relevant Github repositories are archived at: https://github . com/elifesciences-publications/hetionet; https://github . com/elifesciences-publications/integrate; https://github . com/elifesciences-publications/learn; https://github . com/elifesciences-publications/hetio and https://github . com/elifesciences-publications/rephetio-manuscript .	Of all the data in the world today , 90% was created in the last two years . However , taking advantage of this data in order to advance our knowledge is restricted by how quickly we can access it and analyze it in a proper context . In biomedical research , data is largely fragmented and stored in databases that typically do not “talk” to each other , thus hampering progress . One particular problem in medicine today is that the process of making a new therapeutic drug from scratch is incredibly expensive and inefficient , making it a risky business . Given the low success rate in drug discovery , there is an economic incentive in trying to repurpose an existing drug that has already been shown to be safe and effective towards a new disease or condition . Himmelstein et al . used a computational approach to analyze 50 , 000 data points – including drugs , diseases , genes and symptoms – from 19 different public databases . This approach made it possible to create more than two million relationships among the data points , which could be used to develop models that predict which drugs currently in use by doctors might be best suited to treat any of 136 common diseases . For example , Himmelstein et al . identified specific drugs currently used to treat depression and alcoholism that could be repurposed to treat smoking addition and epilepsy . These findings provide a new and powerful way to study drug repurposing . While this work was exclusively performed with public data , an expanded and potentially stronger set of predictions could be obtained if data owned by pharmaceutical companies were incorporated . Additional studies will be needed to test the predictions made by the models .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology" ]	2017	Systematic integration of biomedical knowledge prioritizes drugs for repurposing
While small molecule inhibitors of the bacterial ribosome have been instrumental in understanding protein translation , no such probes exist to study ribosome biogenesis . We screened a diverse chemical collection that included previously approved drugs for compounds that induced cold sensitive growth inhibition in the model bacterium Escherichia coli . Among the most cold sensitive was lamotrigine , an anticonvulsant drug . Lamotrigine treatment resulted in the rapid accumulation of immature 30S and 50S ribosomal subunits at 15°C . Importantly , this was not the result of translation inhibition , as lamotrigine was incapable of perturbing protein synthesis in vivo or in vitro . Spontaneous suppressor mutations blocking lamotrigine activity mapped solely to the poorly characterized domain II of translation initiation factor IF2 and prevented the binding of lamotrigine to IF2 in vitro . This work establishes lamotrigine as a widely available chemical probe of bacterial ribosome biogenesis and suggests a role for E . coli IF2 in ribosome assembly . The bacterial ribosome is a 2 . 6-MDa ribonucleoprotein complex responsible for protein translation , which sediments as a 70S particle composed of a small ( 30S ) and a large ( 50S ) subunit . While there is a relatively thorough understanding of the structure and function of the ribosome during translation ( Moore , 2012 ) , the molecular events underlying its assembly remain largely enigmatic . Ribosome biogenesis , which consumes up to 40% of the cell's energy in rapidly growing Escherichia coli ( Maguire , 2009 ) , involves the coordinated transcription , modification , and folding of rRNA transcripts; translation , modification , and folding of r-proteins; binding of r-proteins to the appropriate rRNA scaffolds; and binding and release of ribosome biogenesis factors . In vivo , these events occur in parallel and represent a highly dynamic system of interrelated processes that occur cooperatively to narrow the assembly landscape of the ribosome ( Holmes and Culver , 2005; Williamson , 2005; Kim et al . , 2014 ) . Ribosome biogenesis factors are proteins that transiently bind to assembling ribosomal particles to increase the efficiency of subunit maturation ( Bunner et al . , 2010 ) and prevent the entry of immature subunits into the translation cycle ( Strunk et al . , 2011; Boehringer et al . , 2012; Lebaron et al . , 2012; Strunk et al . , 2012 ) . E . coli has approximately 60 of such factors . Genetic perturbation has been the conventional route to probe the function of these proteins but has drawbacks . Genetic inactivation is typically permanent , often ‘all or none’ in scope , and for essential genes is fraught with the difficulty of creating conditional alleles . Further , due to the coordination of 30S and 50S subunit biogenesis , and regulatory feedback from the translational capacity of the cell ( Yamagishi and Nomura , 1988; Gaal et al . , 1997 ) , genetic probes of ribosome assembly are prone to wide-ranging impacts and pleiotropic phenotypes ( Lerner and Inouye , 1991 ) . Small molecules are finding increasing use in a research paradigm that emphasizes the value of these as probes of biology . Such chemicals can exert their effects on a time scale of seconds and be added or removed from cell systems at will . Further , small molecules can be dosed to achieve varying levels of target inhibition and as such can be elegant probes of protein function . While existing antibiotics provide a surfeit of probes for on-going efforts to understand the mechanistic details of protein translation , no chemical probes exist for the study of ribosome biogenesis . Small molecule inhibitors of ribosome biogenesis could provide important new tools for the study of this complex process , particularly those events controlled by uncharacterized protein assembly factors . Additionally , chemical inhibitors of bacterial ribosome biogenesis might serve as leads for an entirely new mechanistic class of antibiotics ( Comartin and Brown , 2006 ) . In this study , we report the discovery and characterization of a chemical inhibitor of bacterial ribosome biogenesis . Using a diverse chemical library that included previously approved drugs and compounds of known bioactivity , we enriched for molecules that induced cold sensitive growth inhibition in the model bacterium E . coli . Indeed , numerous studies have revealed that genetic defects in ribosome assembly result in cold sensitive growth phenotypes ( Bryant and Sypherd , 1974; Dammel and Noller , 1995; Jones et al . , 1996; Bubunenko et al . , 2006; Connolly et al . , 2008; Clatterbuck Soper et al . , 2013 ) . We too performed validating efforts , reported herein , of the cold sensitivity of strains from the Keio collection , a comprehensive compendium of E . coli deletion strains . A subsequent chemical screen determined that the anticonvulsant drug lamotrigine induced a strongly cold sensitive growth phenotype . Treatment with this molecule resulted in the accumulation of immature ribosomal subunits in a time-dependent manner without inhibiting protein translation . Spontaneous suppressors of lamotrigine activity mapped exclusively to translation initiation factor IF2 , encoded by infB . These mutations , found in the poorly characterized and evolutionarily divergent domain II of IF2 , obviated the binding of lamotrigine to IF2 in vitro . This work establishes lamotrigine as a widely available chemical probe of bacterial ribosome biogenesis and suggests a role for E . coli IF2 in this process . Where cold sensitive growth has previously been identified as a dominant phenotype for defects in ribosome biogenesis , we set out to first validate such an enrichment strategy with a screen of the E . coli Keio collection ( Baba et al . , 2006 ) , a comprehensive set of non-essential gene deletion strains ( Figure 1—source data 1 ) . We looked for strains that were sensitized to growth at 15°C compared to 37°C ( Figure 1—figure supplement 1A , B ) . A cold sensitivity factor was subsequently generated for each clone , defined as the ratio of growth at 37°C to growth at 15°C , normalized to the mean growth ratio measured for the entire collection ( Figure 1A ) . Strains that displayed a cold sensitivity factor in the top 3 . 5% ( 155 clones ) were analyzed using clusters of orthologous groups ( Tatusov et al . , 1997 , 2003 ) to categorize the cellular function of each deleted gene ( Figure 1—figure supplement 1C , Supplementary file 1A ) . To highlight the relative proportion of genes in each functional class , the number of cold sensitive genes in each was divided by the total number of non-essential genes in that same category ( Figure 1B ) . This normalization procedure highlighted ribosome-related genes as exceptionally sensitive to low temperatures , as >20% of genes in this functional class were found to be cold sensitive . Importantly , this screen was also successful in identifying the vast majority of previously reported cold sensitive ribosome biogenesis genes ( Supplementary file 1B ) , providing support that screening compounds for cold sensitivity would enrich for those related to ribosome function and biogenesis . 10 . 7554/eLife . 03574 . 003Figure 1 . The ribosome is a primary target of cold stress . ( A ) Screen of the E . coli Keio collection for cold sensitivity . Each strain's cold sensitivity factor is defined as the ratio of growth at 37°C to growth at 15°C . Cold sensitivity factors for each strain were normalized to 1 , based on the mean of all cold sensitivity factors calculated for the entire collection . Growth at each temperature was calculated based on the average of two replicates . A gray box highlights strains exhibiting cold sensitivity in the top 3 . 5% ( 155 strains ) . ( B ) The 155 cold sensitive genes from ( A ) were grouped based on clusters of orthologous groups classifications . The percentage of cold sensitive genes in each functional class was defined as the number of cold sensitive genes in that class divided by the total number of non-essential E . coli genes in that same functional class . By permuting the classification assignments , we determined that the proportion of cold sensitive genes in the translation class ( 21% ) was significant with a bootstrapped p-value < 1e−6 . DOI: http://dx . doi . org/10 . 7554/eLife . 03574 . 00310 . 7554/eLife . 03574 . 004Figure 1—source data 1 . Screen of the E . coli Keio collection . Cold sensitivity factors for each strain were normalized to 1 , based on the mean of all cold sensitivity factors calculated . DOI: http://dx . doi . org/10 . 7554/eLife . 03574 . 00410 . 7554/eLife . 03574 . 005Figure 1—figure supplement 1 . Primary data from the screen of the E . coli Keio collection . ( A ) Replicate plot of Keio strains grown in duplicate at 37°C for 24 hr . ( B ) Replicate plot of Keio strains grown in duplicate at 15°C for 48 hr . Cells were grown in LB media supplemented with 50 μg/ml kanamycin for the aforementioned durations and subsequently read at 600 nm using a Perkin Elmer EnVision 96-well plate reader . Cells were grown in a final volume of 100 μl per well . ( C ) Distribution of functional classes amongst the top 3 . 5% of strains identified as cold sensitive . Classes are grouped according to the following: information storage and transfer ( light gray ) ; cellular processes ( mid gray ) ; metabolism ( dark gray ) ; genes of unknown function ( very dark gray ) according to clusters of orthologous groups . DOI: http://dx . doi . org/10 . 7554/eLife . 03574 . 005 Having validated our cold sensitivity enrichment strategy , we proceeded to screen a diverse chemical collection to identify molecules that exhibited a cold sensitive growth inhibition phenotype . This collection , assembled from a variety of vendors , included some 30 , 000 compounds . These were largely diverse synthetic molecules with a subset of 3500 previously approved drugs and chemicals with known biological activity ( Figure 2—source data 1 ) . E . coli was grown in LB media at 15°C and 37°C in the presence of 10 μM of each compound ( Figure 2—figure supplement 1 ) . To select compounds for follow-up , we identified those that strongly inhibited growth at 15°C ( >3σ below the mean OD600 at 15°C ) , yet displayed little growth inhibition at 37°C ( <2σ below the mean OD600 at 37°C ) . These criteria resulted in 49 active compounds ( Figure 2A ) . We removed all antibiotics with known mechanisms of action and filtered the active molecules for diversity in chemical structure . This led to a short-list of 38 active molecules , which were analyzed in dose at 37°C and 15°C . The anticonvulsant drug lamotrigine displayed the largest change in minimum inhibitory concentration ( MIC ) upon temperature downshift , increasing in potency more than 50-fold from >512 μM at 37°C to 7 . 8 μM at 15°C ( Figure 2B , Figure 2—figure supplement 2 ) . 10 . 7554/eLife . 03574 . 006Figure 2 . Lamotrigine induces profound cold sensitivity in E . coli . ( A ) Screen of ∼30 , 000 small molecules at 10 μM against E . coli for cold sensitivity . Compounds found within the gray region were selected for secondary screening . Hit inclusion boundaries are defined as molecules residing >3σ below the mean OD600 at 15°C and <2σ below the mean OD600 at 37°C . Growth at each temperature was calculated based on the average of two replicates . ( B ) Dose-response analysis of lamotrigine at 37°C ( black dots ) and 15°C ( white dots ) . Error bars represent the error of two biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 03574 . 00610 . 7554/eLife . 03574 . 007Figure 2—source data 1 . Small molecule screen for cold senstivity . OD values were normalized across screening plates to account for plate-to-plate variations . DOI: http://dx . doi . org/10 . 7554/eLife . 03574 . 00710 . 7554/eLife . 03574 . 008Figure 2—figure supplement 1 . Primary data from the small molecule screen . ( A ) Replicate plot of E . coli BW25113 grown in the presence of 10 μM of each molecule from a collection of ∼30 , 000 at 37°C for 24 hr , in duplicate . ( B ) Replicate plot of E . coli BW25113 grown in the presence of 10 μM of each molecule from a collection of ∼30 , 000 at 15°C for 48 hr , in duplicate . Cells were grown in LB media and subsequently read at 600 nm using a Perkin Elmer EnVision 96-well plate reader . Cells were grown in a final volume of 100 μl per well . DOI: http://dx . doi . org/10 . 7554/eLife . 03574 . 00810 . 7554/eLife . 03574 . 009Figure 2—figure supplement 2 . Temperature dependence of lamotrigine activity in E . coli . ( A ) E . coli BW25113 was grown in M9 ( left column ) and LB ( middle column ) until early stationary phase in the presence of varying concentrations of lamotrigine at 42°C . Cells were also grown at 37°C ( B ) , 30°C ( C ) , 25°C ( D ) , 20°C ( E ) , and 15°C ( F ) . Blue lines represent no-drug control cultures . Black lines represent cultures treated with MIC quantities of lamotrigine . Dose-response curves of cells grown in LB ( black dots ) and M9 ( white dots ) in the presence of lamotrigine are also shown for each temperature . Cells were grown in a final volume of 100 μl with continuous shaking and read at 600 nm every 10 min using a Tecan Sunrise 96-well plate reader . DOI: http://dx . doi . org/10 . 7554/eLife . 03574 . 009 To determine whether lamotrigine resulted in cold sensitivity through perturbation of the ribosome , we harvested ribosomal particles from early-log cultures of E . coli treated with 2× MIC of lamotrigine for 1 hr and 6 hr at 15°C in LB media and resolved them using sucrose density centrifugation . We note that the doubling time of wild-type E . coli at 15°C in LB media was 6 hr . Cultures were pulse labeled with [14C]-uridine immediately upon drug treatment to visualize the accumulation of newly synthesized particles . Since previous reports have shown that inhibitors of protein translation can cause accumulation of immature ribosomal particles ( Siibak et al . , 2009 , 2011; Sykes et al . , 2010 ) , we also tested a panel of antibiotics ( Figure 3—figure supplement 1A–D ) with known mechanism of action for comparison . Mock treatment of cells with DMSO and simultaneous pulsing with [14C]-uridine allowed for the visualization of 30S , 50S , and 70S particle accumulation after 1 hr and 6 hr of growth post-treatment ( Figure 3A ) . After 1 hr of treatment , small quantities of newly synthesized particles were present , and after 6 hr , cells had accumulated labeled ribosomal particles to near steady-state levels . Cultures treated with chloramphenicol ( Figure 3B ) , erythromycin ( Figure 3C ) , and tetracycline ( Figure 3D ) displayed a substantial accumulation of non-native ribosomal particles after just 1 hr of treatment , illustrating that inhibition of translation can indirectly inhibit ribosomal subunit assembly by limiting the availability of r-proteins . Interestingly , we found that the addition of 2× MIC of vancomycin to E . coli resulted in a detectable perturbation of the ribosome profile ( Figure 3E ) . However , the presence of a ∼40S particle after 6 hr of treatment is likely the result of cell lysis induced by the inhibition of peptidoglycan synthesis ( Stokes and Brown , unpublished data ) . Treatment with lamotrigine resulted in the accumulation of non-native ribosomal particles after 1 hr of incubation and did so in a time-dependent manner ( Figure 3F ) . Further investigations revealed that treatment of E . coli with 2× MIC of lamotrigine for only 5 min ( ∼1% of the doubling time ) caused a significant accumulation of these non-native particles ( Figure 3G , H ) . Consistent with the cold sensitive phenotype induced by lamotrigine , these pre-30S and pre-50S particles that accumulated at 15°C were not present after treatment at 37°C ( Figure 3—figure supplement 1E , F ) . 10 . 7554/eLife . 03574 . 010Figure 3 . Lamotrigine treatment results in the accumulation of non-native ribosomal particles . ( A ) Cells were treated with DMSO ( vehicle ) and immediately pulse labeled with [14C]-uridine . Cells were harvested after 1 hr ( left ) and 6 hr ( right ) of treatment , and ribosomal particle accumulation was monitored using UV absorbance at 260 nm ( black trace ) and scintillation counting ( gray trace ) . Also shown are treatments with 2× MIC chloramphenicol ( B ) ; 2× MIC erythromycin ( C ) ; 2× MIC tetracycline ( D ) ; 2× MIC vancomycin ( E ) ; and 2× MIC lamotrigine ( F ) . ( G ) Early-log cultures of E . coli were treated with DMSO ( solid line ) or 2× MIC lamotrigine ( hashed line ) , pulse labeled with [14C]-uridine , and incubated for 5 min . Ribosomal particles were separated on a sucrose gradient and monitored using UV absorbance . ( H ) These gradients were also analyzed via scintillation counting . DOI: http://dx . doi . org/10 . 7554/eLife . 03574 . 01010 . 7554/eLife . 03574 . 011Figure 3—figure supplement 1 . Temperature-dependent antibiotic activity in E . coli . ( A ) E . coli BW25113 was grown in LB media at 37°C for 24 hr ( black dots ) and 15°C for 48 hr ( white dots ) in duplicate in the presence of varying concentrations of chloramphenicol , ( B ) erythromycin , ( C ) tetracycline , and ( D ) vancomycin . Cells were grown in a final volume of 100 μl . Minimum inhibitory concentration is defined as the lowest concentration of antibiotic required to prevent growth by >95% , as analyzed by OD600 . Error bars represent the error of two biological replicates . ( E ) Sucrose gradients of ribosomal particles from early-log cultures of E . coli treated with DMSO at 37°C . Cells were treated with DMSO and immediately pulse labeled with [14C]-uridine . Cells were harvested after 1 hr ( ∼3 doublings ) of treatment and ribosomal particle accumulation was monitored using UV absorbance at 260 nm ( black trace ) and scintillation counting ( gray trace ) . ( F ) Cells were also treated with 2× MIC ( 15 . 6 μM ) of lamotrigine and ribosomes analyzed in the same manner . ( G ) Sucrose gradients of ribosomal particles from early-log cultures of E . coli treated with DMSO and immediately labeled with [3H]-lamotrigine to a final concentration of 0 . 2 μCi/ml . Cells were grown at 15°C in 25 ml of LB and harvested after 6 hr of treatment , after which they were lysed and the ribosomal particles separated through a sucrose gradient . The gradient was passed through a UV cell measuring absorbance at 260 nm ( black trace ) and 500 μl fractions were subsequently collected and scintillation counted to localize [3H]-lamotrigine ( gray dots ) . ( H ) Same as ( G ) , except cells were treated with 1× MIC unlabeled lamotrigine in place of DMSO . DOI: http://dx . doi . org/10 . 7554/eLife . 03574 . 011 To determine if lamotrigine directly associated with ribosomal particles , early-log cultures of E . coli were treated with [3H]-lamotrigine in the absence ( Figure 3—figure supplement 1G ) or presence ( Figure 3—figure supplement 1H ) of 1× MIC of unlabeled lamotrigine . Ribosomal particles were then separated on a sucrose gradient and individual fractions counted to localize [3H]-lamotrigine . Radiolabeled compound was found exclusively in the soluble fractions eluting early in the gradient , suggesting that lamotrigine does not interact directly with mature or non-native ribosomal particles . We reasoned that ribosomal particles accumulating during treatment could be immature subunits or degradation products of weakly assembled ribosomes . Thus , we analyzed rRNA processing and r-protein content of all particles that accumulated upon lamotrigine treatment . Because previous investigations have shown that the cleavage of 5′ and 3′ termini of rRNA is among the final events in ribosomal subunit assembly ( Lindahl , 1973; Mangiarotti et al . , 1974; Srivastava and Schlessinger , 1988 ) , we first performed 5′ primer extension reactions using rRNA purified from sucrose gradients of lamotrigine-treated cells . Early-log cultures of E . coli were grown in the presence of 2× MIC of lamotrigine at 15°C in LB media for 5 min , 1 hr , and 6 hr , at which time the ribosomal particles were resolved on sucrose gradients , and the rRNA corresponding to each discrete particle was purified and reverse transcribed using 5′ carboxyfluorescein-tagged primers . A 16S rRNA-specific primer was used to analyze the 30S subunit rRNA in pre-30S , 30S , and 70S fractions , whereas a 23S rRNA-specific primer was used to analyze the 50S subunit rRNA in pre-30S , pre-50S , 50S , and 70S fractions . Figure 4A displays the 5′ cleavage events during the processing of 16S and 23S rRNAs ( Shajani et al . , 2011 ) . We note here that our experiments were unable to detect the first 16S cleavage event of 49 nucleotides by Rnase E , and that all immature 16S rRNA species described contain a full-length 5′ terminus of 115 nucleotides . 10 . 7554/eLife . 03574 . 012Figure 4 . Non-native ribosomal particles are immature 30S and 50S subunits . ( A ) 5′ cleavage sites of 16S and 23S rRNA . ( B ) 5′ primer extension analysis of ribosomal particles harvested from DMSO- and lamotrigine-treated E . coli . Early-log cells were treated with DMSO for 6 hr or 2× MIC lamotrigine for 5 min , ribosomal particles were separated on a sucrose gradient , and rRNA was fractionated according to increasing sedimentation rates as indicated ( pre-30S , 30S , pre-50S , 50S , and 70S ) . Particle detection by reverse transcription used a 16S rRNA specific primer ( light gray ) or a 23S rRNA specific primer ( gray and dark gray ) . Proportion of immature rRNA was calculated as ( immature 16S rRNA species/total 16S rRNA species ) and ( immature 23S rRNA species/total 23S rRNA species ) . +7 and +3 represent immature 23S rRNA containing an additional 7 nucleotides and 3 nucleotides at the 5′ terminus , respectively . Error bars represent the error of two biological replicates . ( C ) Quantitative cDNA production of rRNA species within pre-30S regions from DMSO- and lamotrigine-treated E . coli . Early-log cells were treated with DMSO for 6 hr or 2× MIC lamotrigine for 5 min , 1 hr , and 6 hr . Ribosomal particles were separated on a sucrose gradient , and rRNA purified from a single pre-30S fraction from each treatment was reverse transcribed in parallel using 16S- and 23S-specific primers . p16S represents immature 16S rRNA . Error bars represent the error of two biological replicates . ( D ) Quantitation of ribosomal protein occupancy within individual fractions collected from sucrose gradients . Fractions are colored from blue ( lowest density portion of the gradient ) to red ( highest density portion of the gradient ) . Each open circle represents a unique peptide measurement; closed circles denote median values . Occupancy profiles for early ( S15 , L24 ) and late binding ( S3 , L28 ) proteins are compared between sucrose gradients analyzed using DMSO ( top ) or lamotrigine-treated ( bottom ) cells . ( E ) R-protein occupancy of ribosomal particles harvested from sucrose density gradient fractions of DMSO- ( red ) and lamotrigine-treated ( green ) E . coli . Data are plotted as a heat map using the median occupancy values ( see results ) corrected for the amount of sample analyzed in each fraction and normalized to scale from 0 ( white ) to 1 . 0 ( darkest shade ) . Small subunit ( left ) fractions span the pre-30S to the pre-50S regions of the sucrose gradient . Large subunit fractions ( right ) span the late-30S to the late-50S regions of the sucrose gradient . A representative 70S fraction is included in each data set . Absorbance measured at 260 nm is plotted for the region analyzed above each heat map . ( F ) Mass spectrometric localization of RbfA and DeaD in sucrose density gradient fractions of DMSO- ( red ) and lamotrigine-treated ( green ) E . coli . Relative protein abundance was calculated as described in ‘Materials and methods’ . DOI: http://dx . doi . org/10 . 7554/eLife . 03574 . 01210 . 7554/eLife . 03574 . 013Figure 4—source data 1 . R-protein occupancy across sucrose gradients , normalized to the maximum value observed . DOI: http://dx . doi . org/10 . 7554/eLife . 03574 . 01310 . 7554/eLife . 03574 . 014Figure 4—figure supplement 1 . 5′ primer extension of lamotrigine-treated E . coli . ( A ) 5′ primer extension analysis of ribosomal particles harvested from E . coli treated with 2× MIC lamotrigine for 1 hr . Early-log cells were treated with lamotrigine , ribosomal particles were separated on a sucrose gradient , and rRNA was fractionated according to increasing sedimentation rates as indicated ( pre-30S , 30S , pre-50S , 50S , and 70S ) . Particle detection by reverse transcription used a 16S rRNA specific primer ( light gray ) or a 23S rRNA specific primer ( gray and dark gray ) . Proportion of immature rRNA is calculated as ( immature 16S rRNA species/total 16S rRNA species ) and ( immature 23S rRNA species/total 23S rRNA species ) . +7 and +3 represent immature 23S rRNA containing an additional 7 nucleotides and 3 nucleotides at the 5′ terminus , respectively . Error bars represent the error of two biological replicates . ( B ) Same as ( A ) , except cells were treated with 2× MIC lamotrigine for 6 hr . ( C ) Quantitative cDNA production of rRNA species within pre-30S regions from DMSO- and lamotrigine-treated E . coli . Early-log cells were treated with DMSO for 6 hr or 2× MIC lamotrigine for 5 min , 1 hr , and 6 hr . Ribosomal particles were separated on a sucrose gradient , and rRNA purified from a single pre-30S fraction from each treatment was reverse transcribed in parallel using 16S- and 23S-specific primers . p16S represents immature 16S rRNA . Error bars represent the error of two biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 03574 . 01410 . 7554/eLife . 03574 . 015Figure 4—figure supplement 2 . R-protein mass spectrometry of ribosomal particles from lamotrigine-treated E . coli . ( A ) R-protein occupancy of small subunit proteins across sucrose gradients from DMSO ( top ) and lamotrigine-treated ( bottom ) cells . Each open circle represents a unique peptide measurement in a given fraction . Closed circles highlight the median value for each peptide in a given fraction . Individual fractions are colored from blue ( lowest density portion of the gradient ) to red ( highest density portion of the gradient ) . ( B ) Same as ( A ) , except monitoring abundance of large subunit proteins . ( C ) Over-represented ( green ) and under-represented ( red ) small subunit r-proteins of the pre-30S particle that accumulates during lamotrigine treatment are highlighted on the Nomura assembly map ( Held et al . , 1974; Chen et al . , 2012 ) . Assembly groups are colored according to Chen and Williamson ( 2013 ) . Low occupancy proteins highlighted on the 30S subunit ( PDB 2AVY ) cluster around the neck of the 30S subunit . ( D ) Under-represented ( red ) large subunit r-proteins from the pre-50S particle that accumulates during lamotrigine treatment are highlighted on the Nierhaus assembly map ( Herold and Nierhaus , 1987; Chen et al . , 2012 ) . Depleted proteins highlighted on the 50S subunit ( PDB 2Y11 ) generally cluster in a ring around the L1 arm , central protuberance , and L11 arm . DOI: http://dx . doi . org/10 . 7554/eLife . 03574 . 015 Primer extension analysis of cells treated with DMSO for 6 hr is depicted in Figure 4B ( top panel ) . Analysis of the pre-30S , 30S , and 70S regions of the gradient using the 16S rRNA-specific primer revealed that increasing sedimentation rate correlated with a decreased proportion of immature 16S rRNA relative to total 16S rRNA . The presence of immature 16S rRNA sedimenting in the pre-30S region suggested a heterogeneous composition of 30S particles at various stages of maturation . At the maximum of the 30S peak approximately 40% of 16S rRNA was unprocessed . The 16S rRNA found in fractions corresponding to the 70S subunit was >95% processed , as expected . Overall , a similar trend was seen when analyzing the processing of 23S rRNA in the pre-30S , pre-50S , 50S , and 70S regions of the gradient; increasing sedimentation rate correlated with a decreased proportion of immature 23S rRNA relative to total 23S rRNA . The pre-30S and pre-50S regions of the gradient were devoid of quantifiable 23S rRNA , suggesting very little 50S precursor accumulation in unperturbed cells . Compared to cells treated with DMSO , those treated with 2× MIC of lamotrigine for 5 min ( Figure 4B , bottom panel ) , 1 hr ( Figure 4—figure supplement 1A ) , and 6 hr ( Figure 4—figure supplement 1B ) contained similar proportions of immature to total 16S and 23S rRNA in the 30S , 50S , and 70S regions of the gradient . Furthermore , lamotrigine-treated cells displayed almost identical proportions of immature to total 16S rRNA in the pre-30S region , relative to DMSO-treated cells . Unlike DMSO-treatment , however , lamotrigine treatment resulted in the accumulation of immature 23S rRNA in the pre-30S and pre-50S regions . While the presence of unprocessed 23S rRNA in the pre-50S region strongly suggested an immature 50S subunit sedimenting at ∼40S , its presence in the pre-30S region raised questions of whether the dominant species in the pre-30S peak ( Figure 3F ) was derived from 16S or 23S rRNA . While calculating proportions of immature to total rRNA of the same species ( [immature 16S/total 16S] and [immature 23S/total 23S] ) provides detail of rRNA processing efficiency in each region of the gradient , it does not inform on the absolute quantity of one species ( 16S rRNA ) relative to the other ( 23S rRNA ) . To answer this question , we quantified absolute cDNA fluorescence from 5′ primer extension reactions ( Figure 4C , Figure 4—figure supplement 1C ) . Samples of rRNA purified from single fractions of sucrose gradients were reverse transcribed in parallel reactions using either the 16S- or 23S-specific primer . In this study , cDNA production is proportional to the amount of rRNA transcript in the sample and therefore reflects absolute quantities of each rRNA species ( 16S and 23S ) present . The quantity of immature 16S rRNA from DMSO-treated cells was minor relative to cells treated with lamotrigine . This is consistent with previous reports , which have shown that immature ribosomal particles in unperturbed cells account for only a small proportion of total ribosomal material ( Mulder et al . , 2010; Chen et al . , 2012 ) . Furthermore , while immature 23S rRNA slowly accumulated in this region as a function of lamotrigine treatment length , immature 16S rRNA did so at a significantly greater rate . These results indicate that , by far , the major species of rRNA residing within the pre-30S region in lamotrigine-treated cells was unprocessed 16S rRNA . Thus , 5′ primer extension results strongly suggested that lamotrigine treatment results in the accumulation of an immature 30S subunit that sediments at ∼25S and an immature 50S subunit that sediments at ∼40S . To further test this hypothesis , we used quantitative mass spectrometry to determine the relative stoichiometry of r-proteins across sucrose gradients of DMSO- and lamotrigine-treated cells . Early-log cultures of E . coli grown at 15°C in 14N-labeled LB media were treated with DMSO or 2× MIC of lamotrigine for 6 hr , lysed , and the ribosomal particles were separated through sucrose gradients . Fractions spanning the pre-30S to the 70S regions were spiked with a fixed concentration of 70S ribosomes purified from cells grown in 15N-labeled media . These spiked samples were then digested with trypsin and prepared for mass spectrometry . This approach resulted in multiple independent peptide measurements for each r-protein in every fraction ( Figure 4D , Figure 4—figure supplement 2A , B ) . Protein occupancy was calculated as 14N/[14N + 15N] . Direct inspection of the protein occupancy profiles revealed distinct patterns for early- ( e . g . , S15 , L24 ) and late- ( e . g . , S3 , L28 ) binding proteins . In the DMSO samples , all r-proteins within a given subunit displayed highly correlated occupancy patterns with maximal occupancy corresponding to ‘peak’ fractions as determined by rRNA absorbance ( Figure 4D ) . In contrast , treatment with lamotrigine resulted in significant occupancy of the early-binding proteins in pre-30S and pre-50S fractions , whereas the late-binding r-proteins exhibited relatively unperturbed profiles . Indeed , protein S15 is found at significantly greater occupancy in the pre-30S fractions upon lamotrigine treatment ( dark blue ) . This effect on early binding proteins is particularly pronounced in the L24 profile with peak occupancy shifted six fractions earlier in the gradient ( from orange to green ) . To facilitate further analysis , this large data set ( ∼20 , 000 measurements ) was compressed to a 53-protein × 28-fraction heat map using the median protein occupancy value for each protein in each fraction ( Figure 4E , Figure 4—source data 1 ) . As expected , the 70S peak from both DMSO- and lamotrigine-treated cells exhibited stoichiometric occupancy of each r-protein . In both DMSO- and lamotrigine-treated samples , we observed sub-stoichiometric occupancy of the late-binding r-proteins S21 , S2 , and to a lesser extent S3 , within the 30S peak . Notably , this effect was enhanced in the lamotrigine-treated samples . The depletion of these proteins is consistent with prior in vivo analysis of small subunit biogenesis at 37°C , which found S2 , S3 , and S21 to be the latest-binding small subunit proteins ( Figure 4—figure supplement 2C; Chen and Williamson , 2013 ) . Consistent with sucrose gradient traces monitoring UV absorbance and [14C]-uridine incorporation , we observed a subtle broadening of the 30S protein occupancy peak upon treatment with lamotrigine . Analysis of the leading edge of this peak revealed an enrichment of relatively early-binding proteins S13 , S7 , S16 , S8 , S15 , S6 , and S18 , consistent with the presence of 30S subunit assembly intermediates in lamotrigine-treated cells . Inspection of large subunit protein occupancy revealed drastic changes as a result of lamotrigine treatment , resulting in the accumulation of an immature particle depleted of the late-binding r-proteins L35 , L36 , L16 , L30 , and L28 as well as earlier-binding proteins , L34 , L6 , and L21 ( Figure 4—figure supplement 2D ) . Particles with this heterogeneous protein composition could not be found in the DMSO-treated samples , indicating that if they do form they rapidly convert to mature particles that migrate later in the gradient . Formally , these r-protein occupancy patterns could have resulted either from immature assembly intermediates or from the degradation of mature particles initiated by the removal of late-binding proteins . To distinguish between these possibilities , we used mass spectrometry to determine the occupancy pattern for the ribosome biogenesis factors RbfA and DeaD ( Figure 4F ) . These proteins were completely absent from mature 70S particles , and thus we reasoned that their presence could be used as markers of immature particles . The 30S-specific maturation factor RbfA co-sedimented with the 30S particles in both DMSO- and lamotrigine-treated samples . Further , the 50S-biogenesis factor DeaD co-migrated with the pre-50S peak in the lamotrigine-treated samples and the 50S peak in the DMSO-treated samples . These data further suggest that the pre-30S and pre-50S particles are immature subunits and not the result of degradation of mature ribosomal particles . To identify lamotrigine's target in vivo , we generated suppressor mutants and sequenced the resulting genomes to identify the mutation ( s ) that were responsible for resistance . Briefly , E . coli BW25113 was grown at 15°C in the presence of 5× MIC ( 39 μM ) of lamotrigine in LB media to saturation . Putative suppressors were then serially passaged in the presence and absence of lamotrigine to purify and to ensure mutation stability . After 20 independent strains had been isolated , three were selected at random , sequenced using an Illumina MiSeq platform , and analyzed against the E . coli MG1655 genome using BreSeq . At this time , the chromosome of BW25113 had yet to be sequenced , thus this strain was sequenced in parallel to be used as a reference genome . The sequencing data revealed mutations solely in domain II of initiation factor IF2 ( Figure 5A , Figure 5—figure supplement 1A ) . Subsequent Sanger sequencing of the infB genes from each of the remaining 17 suppressor strains revealed that all mutations mapped to domain II of IF2 and fell into only four categories . Three classes of mutant contained in-frame chromosomal deletions in this region and one mutant class contained a short duplication . 10 . 7554/eLife . 03574 . 016Figure 5 . Lamotrigine binds to wild type but not mutant IF2 in a G-nucleotide-dependent manner . ( A ) General domain organization of Enterobacteriaceae IF2-α with lamotrigine suppressor mutations mapped against the parental E . coli BW25113 sequence . ( B ) Experimental design of [3H]-lamotrigine association assay . ( C ) Relative association of [3H]-lamotrigine to wild-type E . coli IF2 and lamotrigine suppressor IF2 ( mutant #3 ) under varying conditions . CPMs of the experimental samples were normalized to the baseline flow-through of [3H]-lamotrigine in buffer . Error bars represent the error of three biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 03574 . 01610 . 7554/eLife . 03574 . 017Figure 5—figure supplement 1 . Genetic determinants of lamotrigine activity . ( A ) Structures of E . coli IF2N ( PDB 1ND9 ) and Methanothermobacter thermautotrophicus IF2/eIF5B ( PDB 1G7R ) depicting the location of lamotrigine suppressor mutations in E . coli IF2 . All mutations are localized within domain II . ( B ) Example , growth curves of E . coli BW25113 ( black trace ) and lamotrigine suppressor mutant #3 ( gray trace ) in 150 μl LB media at 15°C shaking at 200 rpm . Cells were read at OD600 every 10 min throughout the duration of the experiment . ( C ) Example , potency analysis of lamotrigine against E . coli BW25113 ( white dots ) and lamotrigine suppressor mutant #3 ( black dots ) at 15°C in LB media . Cells were grown for 48 hr prior to reading OD600 . Error bars represent the error of two biological replicates . ( D ) Example , sucrose gradient of ribosomal particles from early-log cultures of lamotrigine suppressor mutant #3 treated with 2× MIC of lamotrigine at 15°C . Cells were treated with lamotrigine and immediately pulse labeled with [14C]-uridine . Cells were harvested after 6 hr of treatment , and ribosomal particle accumulation was monitored using UV absorbance at 260 nm ( black trace ) and scintillation counting ( gray trace ) . ( E ) IF2 homologs from various bacterial species were aligned using the MuscleWS multiple sequence alignment plugin through Jalview version 2 . 8 . Residues 181 to 206 ( E . coli numbering ) are shown , depicting the conservation of this region exclusively in the Enterobacteriaceae . ( F ) Using Jalview , a distance tree relating the various IF2 homologs was calculated based on average distance using the percent identity . IF2 homology between the Enterobacteriaceae predicts lamotrigine potency at 15°C . See Supplementary file 2 for potency analyses . DOI: http://dx . doi . org/10 . 7554/eLife . 03574 . 017 To understand the phenotypic characteristics of these four unique lamotrigine suppressor strains , cells from each class were first analyzed for growth rate and resistance to lamotrigine . All displayed wild-type growth at 15°C in the absence of lamotrigine and little sensitivity to lamotrigine treatment up to 512 μM in LB media ( Figure 5—figure supplement 1B , C ) . We subsequently analyzed these strains to determine the composition of ribosomal particles upon lamotrigine treatment . Suppressor strains were grown to early-log phase , treated with 2× MIC of lamotrigine , and grown for 6 hr at 15°C . Immediately after the addition of lamotrigine to the cultures , cells were pulse labeled with [14C]-uridine to monitor accumulation of non-native ribosomal particles . Treatment of suppressor strains with lamotrigine did not result in the accumulation of non-native ribosomal particles ( Figure 5—figure supplement 1D ) . To test the hypothesis that IF2 was the target of lamotrigine , we conducted in vitro binding studies using recombinant E . coli IF2 and [3H]-lamotrigine ( Figure 5B ) . Wild type and mutant forms of E . coli IF2 were purified and incubated with [3H]-lamotrigine in the presence of GDP or GTP . After incubation for 3 hr at 15°C , the reaction mixtures were passed through a pre-cooled Sephadex G-25 column , and the flow-through was collected and scintillation counted to detect the presence of lamotrigine-IF2 complexes . Lamotrigine was found to associate with wild-type E . coli IF2 in a G-nucleotide-dependent manner , with lamotrigine-IF2 complex formation favored in the presence of GDP over GTP ( Figure 5C ) . We note here that previous studies have not reported measurable GTP turnover by IF2 in the absence of ribosomal subunits ( Severini et al . , 1991 ) . Analyses of lamotrigine binding with mutant IF2 in the presence of GTP and GDP failed to reveal association ( Figure 5C ) . Consistent with these results , lamotrigine was found to have activity solely against members of the Enterobacteriaceae ( Supplementary file 2 ) , which is the only bacterial family that contains the domain II sequence outlined in Figure 5A . Interestingly , the potency of lamotrigine at 15°C against the Enterobacteriaceae was directly correlated with IF2-α sequence homology , further suggesting that the sequence of domain II defines essential structural features for lamotrigine association ( Figure 5—figure supplement 1E , F ) . Given the known role of IF2 in protein translation , we tested whether lamotrigine was indirectly perturbing ribosome biogenesis by inhibiting IF2-dependent translation . We first monitored [35S]-methionine incorporation into bulk cellular protein . Early-log cultures of E . coli grown in M9 minimal media were treated with lamotrigine and a collection of known antibiotics for 2 . 6 hr at 15°C ( doubling time = 16 hr ) . Immediately after the addition of drug , cells were pulsed with [35S]-methionine to monitor the production of newly synthesized proteins . Cells were then pelleted , washed , lysed , and treated with TCA . The precipitated proteins were captured on glass filters and counted . These investigations revealed that lamotrigine had no impact on [35S]-methionine incorporation , even when cells were treated with 8× MIC of lamotrigine ( Figure 6A , C , black dots ) . Similarly , cells treated with these same concentrations of lamotrigine at 37°C for three doublings did not display any inhibition of translation ( Figure 6—figure supplement 1A ) . We found that when cells were treated with 8× MIC of tetracycline , chloramphenicol , and erythromycin , there was a marked decrease in protein labeling . As expected given its known mechanism of action , cells treated with 8× MIC of vancomycin did not display inhibition of protein biosynthesis after 2 . 6 hr of treatment . 10 . 7554/eLife . 03574 . 018Figure 6 . Accumulation of immature ribosomal subunits is not the result of translation inhibition . ( A ) [35S]-methionine incorporation into early-log cells grown for 2 . 6 hr in M9 media at 15°C . Immediately prior to the radioactivity pulse , cultures were treated with 8× MIC of each antibiotic . [35S]-methionine incorporation was quantified by liquid scintillation counting . Error bars represent the error of two biological replicates . ( B ) Cell-free coupled transcription/translation reactions in the presence of 8× MIC of each antibiotic . Samples were incubated at 15°C for 4 hr , at which time reactions were halted on ice , excess luciferin was added , and luminescence was monitored . Error bars represent the error of two biological replicates . ( C ) [35S]-methionine incorporation ( black dots ) and cell-free luminescence ( white dots ) as a function of lamotrigine concentration . Samples were prepared as described in ( A ) and ( B ) . ( D ) Cell-free coupled transcription/translation reactions in the presence of increasing concentrations of evernimicin at 37°C ( black dots ) and 15°C ( white dots ) . Reactions were assembled and analyzed as in ( B ) . ( E ) Analysis of growth rate and r-protein synthesis rate as a function of lamotrigine and chloramphenicol concentrations . Synthesis rates were determined for each ribosomal protein using quantitative mass spectrometry . For each condition , r-protein synthesis rates ( 50 measurements per treatment ) are presented as a notched box and whisker plot centered at the growth rate ( 1/hr ) observed for that treatment ( ‘Materials and methods’ ) . Black dots represent synthesis rates of individual proteins in excess of ( 1 . 5 × inner quartile range ) of that data set . Light to dark shades of red represent 2× , 3× , 4× , 5× , and 6× MIC of chloramphenicol . Light to dark shades of green represent 2× , 3× , 4× , 5× , and 6× MIC of lamotrigine . Values were normalized to the DMSO control and log transformed . DOI: http://dx . doi . org/10 . 7554/eLife . 03574 . 01810 . 7554/eLife . 03574 . 019Figure 6—source data 1 . In vivo r-protein synthesis rates measured using qMS . Synthesis rates are log2 transformed after normalization to the measured synthesis rate of E . coli in the presence of DMSO . Growth rates are reported as 1/doubling time ( hr ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03574 . 01910 . 7554/eLife . 03574 . 020Figure 6—figure supplement 1 . Effects of lamotrigine on translation in E . coli . ( A ) [35S]-methionine incorporation into early-log cells grown for 2 hr ( ∼3 doublings ) in M9 media at 37°C . Immediately prior to the radioactivity pulse , cultures were treated with increasing concentrations of lamotrigine . [35S]-methionine incorporation was quantified by liquid scintillation counting . Error bars represent the error of two biological replicates . ( B ) Cell-free coupled transcription/translation reactions in the presence of increasing concentrations of lamotrigine . Samples were incubated at 37°C for 1 hr , at which time reactions were halted on ice , excess luciferin was added , and luminescence was monitored . Error bars represent the error of two biological replicates . ( C ) Kinetics of cell-free transcription/translation system at 15°C . 10 μl reactions containing 1% DMSO were read every 30 min for 5 hr to establish a linear range of luciferase production . Error bars represent the error of two biological replicates . ( D ) Example , pulse-labeling data showing the incorporation of 50% 15N into an r-protein peptide as a function of time . Similar data were gathered for all r-protein peptides from cells treated with DMSO; 2× , 3× , 4× , 5× , and 6× MIC lamotrigine; and 2× , 3× , 4× , 5× , and 6× MIC chloramphenicol . ( E ) R-protein degradation as a function of time in DMSO-treated cultures . Protein degradation is defined as ( 14N intensity/15N intensity ) for each peptide . Light shades to dark shades represent 0 , 4 , 8 , and 16 hr of 50% 15N pulse . Small circles represent individual peptide measurements . Large circles denote the median measurement for that sample . ( F ) R-protein synthesis as a function of time in DMSO-treated cultures . Protein synthesis is defined as ( 50% 15N intensity/15N intensity ) for each peptide . Light shades to dark shades represent 0 , 4 , 8 , and 16 hr of 50% 15N pulse . ( G and I ) are the same as ( E ) , except for 6× MIC lamotrigine and 6× MIC chloramphenicol , respectively . ( H and J ) are the same as ( F ) , except for 6× MIC lamotrigine and 6× MIC chloramphenicol , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 03574 . 020 To determine if lamotrigine had a direct effect on protein biosynthesis in vitro , we employed a commercially available E . coli K-12 cell-free transcription/translation system producing luciferase . Reactions in the presence of 8× MIC of lamotrigine and the aforementioned antibiotics were incubated at 15°C for 4 hr ( see Figure 6—figure supplement 1C for in vitro translation kinetics at 15°C ) , at which time luciferin was added to quantify the luciferase produced . As expected , all translation inhibitors blocked the production of luciferase while vancomycin did not ( Figure 6B ) . Lamotrigine failed to block the production of luciferase at either 15°C or 37°C ( Figure 6B , C , white dots , Figure 6—figure supplement 1B ) . To ensure that the in vitro translation assay required IF2 activity , we tested the effect of evernimicin , an oligosaccharide antibiotic known to inhibit IF2-dependent 70S initiation complex formation ( McNicholas et al . , 2000; Belova et al . , 2001 ) . Evernimicin prevented luciferase synthesis in vitro at both 15°C and 37°C ( Figure 6D ) . Having ruled out a direct effect on bulk protein biosynthesis , we wondered if lamotrigine might have a specific effect on r-protein synthesis that could lead to the accumulation of immature ribosome subunits . To test this , we measured the synthesis rate of each r-protein in vivo using a mass spectrometry-based pulse labeling technique . At 15°C , cells were grown in 14N-labeled M9 minimal media to mid-log phase at which point they were diluted twofold into 15N-labeled M9 media and concurrently treated with DMSO , lamotrigine , or chloramphenicol . Cells were harvested after 1 hr , 2 . 6 hr , 4 hr , 8 hr , and 16 hr and spiked with equal quantities of 15N-labeled 70S ribosomes as an internal reference standard . After cell lysis , these spiked samples were digested with trypsin for analysis by mass spectrometry . Using a Fourier transform deconvolution algorithm ( Sperling et al . , 2008; Chen et al . , 2012 ) , we independently quantified the r-proteins produced before the pulse ( 14N ) and those synthesized post-pulse ( 50% 15N ) from the cellular lysate ( Figure 6—figure supplement 1D ) . Inspection of 14N abundance as a function of time revealed that most ribosomal proteins were stable over this time course ( Figure 6—figure supplement 1E , G , I ) , consistent with our prior work ( Chen et al . , 2012 ) . We then carefully inspected the rate of 50% 15N incorporation into each ribosomal protein in each treatment condition . Using a linear approximation of the synthesis rate based on the 4- , 8- , and 16-hr time points , we found that each protein was synthesized at a similar rate in the DMSO- and lamotrigine-treated cells up to 6× MIC . However , we found significant inhibition of r-protein synthesis with increasing concentrations of chloramphenicol , our positive control compound ( Figure 6E , Figure 6—figure supplement 1F , H , J , Figure 6—source data 1 ) . To establish if the pre-30S and pre-50S particles that accumulate upon lamotrigine treatment represented immature subunits on pathway to maturity , we endeavored to monitor the impact of relieving inhibition by lamotrigine . We hypothesized that cells relieved of lamotrigine stress would assemble immature 30S and 50S particles into mature 30S and 50S subunits . E . coli was grown to early-log phase in LB media at 15°C and treated with either 2× MIC of lamotrigine or DMSO as a mock treatment . After 5 min , [14C]-uridine was added and the cells were grown an additional 3 hr , at which point cells were pelleted , washed , and resuspended in fresh LB media supplemented with a 1000-fold excess of non-labeled uridine . Cells were harvested immediately preceding the chase , and after 30 min , 1 hr , 2 hr , and 3 hr of this chase period ( Figure 7A ) . 10 . 7554/eLife . 03574 . 021Figure 7 . Immature ribosomal particles sediment as mature subunits upon removal of lamotrigine stress . Particles were analyzed by sedimentation over sucrose gradients and analyzed with radioactivity detection . ( A ) Experimental design of pulse-chase analysis of E . coli treated with 2× MIC lamotrigine . ( B ) Cells were treated with DMSO and concurrently pulsed with [14C]-uridine for 3 hr in LB media at 15°C prior to media exchange and unlabeled uridine chase . This time course reveals that no additional radiolabel was incorporated into ribosomal subunits during the chase period . ( C ) Cells were treated as in ( B ) , except with 2× MIC lamotrigine in place of DMSO , revealing that radiolabeled pre-30S and pre-50S particles matured to 30S and 50S particles over the duration of chase period . DOI: http://dx . doi . org/10 . 7554/eLife . 03574 . 021 In each DMSO-treated sample ( Figure 7B ) , we found significant quantities of [14C]-uridine-labeled 30S and 50S subunits . Because the quantity of labeled subunits did not change as a function of the length of the chase , these particles likely represent fully mature subunits that have simply dissociated . Interestingly , DMSO-treated cells harvested immediately after the 3-hr pulse and before the addition of the chase show a slight decrease in the levels of complete 70S ribosomes relative to any of the samples harvested post-chase ( Figure 7B ) . This small but significant change likely results from the presence of an intracellular pool of [14C]-uridine , which is incorporated into 70S particles during the initial 30-min chase . This pool may consist of free nucleotides that are not washed away during the chase or , as described previously , may exist as transcribed rRNA that has not completed the assembly process ( Chen et al . , 2012; Chen and Williamson , 2013 ) . Cells harvested at subsequent times during the chase period showed no change in the quantities of 30S , 50S , and 70S particles , indicating that all newly synthesized rRNA is incorporating exclusively non-labeled uridine . This result allowed us to analyze the maturation of the lamotrigine-induced pre-30S and pre-50S particles , confident that they were generated during the initial pulse and not synthesized de novo between 30 min and 3 hr post-pulse . We next analyzed the ability of cells treated with lamotrigine to process pre-30S and pre-50S particles ( Figure 7C ) . Cells harvested immediately after the 3-hr pulse period displayed a significant accumulation of pre-30S and pre-50S material . As shown earlier ( Figure 3F ) , we also noticed a large decrease in the relative accumulation of 70S ribosomes during drug treatment . Some 30 min after removal of lamotrigine and non-labeled uridine chase , the relative proportions of ribosomal particles began to adjust . Specifically , the levels of pre-30S and pre-50S particles decreased with a corresponding increase in 70S ribosomes . This trend continued throughout the 3-hr chase period , after which there were no apparent differences between DMSO-treated and lamotrigine-treated cells . Interestingly , after 1 hr of non-labeled uridine chase , a cluster of three particles that sedimented at approximately 40S ( discussed above as the pre-50S ) , 45S , and 50S appeared . With each successive time point , the levels of 50S increased at the expense of the other particles , suggesting our time course had captured cells actively assembling the pre-50S particles into 50S subunits . Understanding bacterial ribosome assembly has proven to be a challenging undertaking . Involving nearly 60 protein factors , the process is rapid , highly efficient , and studies to date suggest that assembly intermediates are elusive and do not accumulate in significant amounts ( Mulder et al . , 2010; Chen et al . , 2012 ) . The genetic inactivation of ribosome biogenesis factors has provided an opportunity to perturb the process in order to better understand the action of chemical modification and chaperone functions in the assembly process ( Shajani et al . , 2011 ) . Nevertheless , many of these factors are essential and resist genetic manipulation . Further , genetic inactivation has poor temporal resolution and is not ideally suited to probe the coordinated action of these factors in time and space . Indeed , our understanding of ribosome function has benefited enormously from a great number and variety of small molecule probes of chemical and conformational steps of protein translation ( Tenson and Mankin , 2006; Wilson , 2009 ) . Chemical inhibitors of the assembly process would similarly provide important new probes of ribosome biogenesis . Herein , we report the discovery and characterization of a small molecule inhibitor of ribosome assembly in E . coli under cold temperature growth conditions . The inhibitor , lamotrigine , is a widely available anticonvulsant drug whose target in E . coli is domain II of the initiation factor IF2 . In all , this work provides the first small molecule probe of ribosome assembly and points to a novel role for IF2 in E . coli ribosome biogenesis . To find small molecule inhibitors of ribosome assembly , we developed a cell-based platform to first enrich for inhibitors of ribosome assembly and function by screening for compounds that led to a cold sensitive growth phenotype . The screen was inspired by numerous previous reports of cold sensitive mutants in ribosome-related genes and was validated with a screen of the E . coli Keio collection . In our screen of the Keio collection , ribosome genes were overwhelmingly enriched and , of the known cold sensitive ribosome biogenesis genes , we were successful in identifying the vast majority of these . We next screened a diverse collection of ∼30 , 000 small molecules , including many known drugs and bioactive compounds , to identify growth inhibitory compounds with increased potency at 15°C relative to 37°C . Of 38 structurally diverse active compounds from this screen , lamotrigine induced the most profound cold sensitive growth inhibition . Sedimentation analysis revealed that lamotrigine induced the rapid accumulation of non-native ribosomal particles with apparent sedimentation rates of ∼25S ( pre-30S ) and ∼40S ( pre-50S ) , prompting an in-depth analysis of their composition . Using 5′ primer extension of rRNA and r-protein mass spectrometry , these particles were found to be immature 30S and 50S subunits , respectively . These immature subunits lacked r-proteins associated with the neck of the 30S subunit and the body of the 50S subunit around the L1 arm , central protuberance , and L11 arm ( Figure 4—figure supplement 2C , D ) . With these regions encompassing the functional centers of each subunit , it is tempting to speculate that lamotrigine may perturb late steps in 30S and 50S subunit assembly . It has recently been suggested that these sites are among the last to mature ( Jomaa et al . , 2011; Guo et al . , 2013; Li et al . , 2013; Jomaa et al . , 2014 ) consistent with this hypothesis . Whole genome sequencing of spontaneous suppressor mutants capable of robust growth in the presence of lamotrigine revealed mutations in domain II near the N-terminus of initiation factor IF2 . Further , in vitro binding studies indicated that lamotrigine binds to IF2 in a nucleotide-dependent fashion and that suppressor mutations abrogated binding . Interestingly , domain II is conserved solely among IF2 proteins from the Enterobacteriaceae and has yet to be assigned a definitive function . Indeed , only one study suggests a role for this region in binding strongly to 30S , 50S , and 70S ribosomal particles relative to the other Enterobacteriaceae IF2 domains ( Moreno et al . , 1999 ) . While lamotrigine treatment resulted in the rapid accumulation of immature 30S and 50S subunits , the target IF2 led us to wonder if lamotrigine might be an inhibitor of protein translation . We speculated that the observed ribosome biogenesis phenotype might be an indirect effect of blocking r-protein production , as described previously for antibiotics known to inhibit translation ( Siibak et al . , 2009 , 2011; Sykes et al . , 2010 ) . Using multiple orthogonal approaches , both in vitro and in vivo , we were unable to detect any translational inhibition by lamotrigine even when using concentrations far above the MIC . These results were in contrast to assays with known translational inhibitors , including evernimicin , a known inhibitor of IF2-dependent 70S initiation complex formation . The finding that IF2 is the target of lamotrigine is intriguing in light of emerging information on the role of its eukaryotic counterpart , eIF5B , in 40S subunit assembly in yeast ( Lebaron et al . , 2012; Strunk et al . , 2012 ) . In Saccharomyces cerevisiae , eIF5B associates with immature 40S subunits in a translation-like checkpoint , wherein immature 40S particles bind to mature 60S subunits prior to final maturation . Given that pre-30S and pre-50S particles accumulate during lamotrigine stress , a bacterial model may include association of two immature particles prior to maturation of the functional centers within the 30S and 50S subunits . Alternatively , it is possible that IF2 is involved in the maturation of 30S and 50S subunits independently . Regardless of the precise events mediated by IF2 , our data strongly support a central role for the enigmatic and divergent domain II of E . coli IF2 in the assembly of both subunits . Interestingly , our observations help rationalize the previously unexplained finding that overexpression of IF2 in a ΔyjeQ background of E . coli partially suppresses the mutant slow growth phenotype and restores ribosome profiles to wild type ( Campbell and Brown , 2008 ) . Similarly to lamotrigine-treated cells , the 30S particles of cells lacking YjeQ display significantly depleted occupancy of S21 , S1 , S2 , and S3 ( Jomaa et al . , 2011 ) , suggesting that IF2 may perform an overlapping role in late 30S maturation during cold stress . Our results also parallel work dating back almost two decades , which showed that truncation of the N-terminus of E . coli IF2 , containing domain II led to cold sensitive growth ( Laalami et al . , 1991a , 1991b ) . With much of the machinery involved in ribosome biogenesis ( Bharat et al . , 2006; Schaefer et al . , 2006 ) and translation ( Anger et al . , 2013 ) conserved among bacteria and eukaryotes , the maintenance of IF2/eIF5B function in ribosome biogenesis through evolution is surely plausible . Given that domain II of IF2 is highly divergent , this work raises questions of how diverse bacterial species carry out the temperature-dependent functions of IF2 described herein . Domain II may have a purpose that is uniquely important to the Enterobacteriaceae under cold stress or , alternatively , species-specific proteins that mimic the N-terminus of IF2 from Enterobacteriaceae may perform this activity . Taken together , this work establishes lamotrigine as a first-in-class small molecule inhibitor of bacterial ribosome biogenesis . Moreover , we have identified domain II of IF2 as the molecular target of lamotrigine , suggesting an as-yet-uncharacterized ribosome assembly function for this canonical translation initiation factor . We posit that lamotrigine will serve as an important tool in expanding our understanding of the molecular details of IF2 in ribosome biogenesis and functions as a proof-of-concept molecule in the development of novel antibiotics . Overnight cultures of E . coli BW25113 ( including Keio strains ) grown in LB media at 37°C were diluted 1/1000 in fresh LB , and incubated at 15°C ( 48 hr ) and 37°C ( 24 hr ) in duplicate without shaking in a final volume of 100 μl . Cells were grown in Corning ( Corning , NY ) Costar 96-well clear-bottom plates . For the small molecule screen , compounds were added to E . coli BW25113 to a final concentration of 10 μM . All screens were performed in duplicate . Molecules , dissolved in DMSO , were sourced from ChemBridge ( San Diego , CA ) , Maybridge ( Waltham , MA ) , MicroSource Discovery Systems ( Gaylordsville , CT ) , Prestwick Chemicals ( Washington , DC ) , and Biomol-Enzo Life Sciences ( Farmingdale , NY ) . Liquid handling was performed using a Beckman Coulter ( Brea , CA ) FXP Laboratory Automated Workstation . After incubation , plates were read using a Perkin Elmer ( Waltham , MA ) EnVision plate reader at 600 nm . 25 ml cultures of early-log E . coli BW25113 ( OD = 0 . 2 ) grown in LB media at 15°C were treated with the appropriate concentration of each antibiotic ( purchased from Sigma , St . Louis , MO ) and , when applicable , pulse labeled with [14C]-uridine ( purchased from American Radiolabeled Chemicals , St . Louis , MO ) to a final concentration of 0 . 2 μCi/ml ( specific activity 55 mCi/mmol ) . Cells were incubated as necessary , harvested by centrifugation , and lysed using a Constant Systems ( Daventry , England ) cell disruptor at 13 kpsi in 3 ml ice-cold ribosome buffer ( 20 mM Tris–HCl , pH 7 . 0 , 10 . 5 mM MgOAc , 100 mM NH4Cl , 3 mM β-mercaptoethanol ) . Cell lysates were clarified using a Beckman Coulter MLA-80 rotor at 24 , 000 rpm for 45 min , at which time they were loaded onto 35 ml 10–40% sucrose gradients and centrifuged for 18 hr at 18 , 700 rpm in a Thermo ( Waltham , MA ) SureSpin rotor . The volume of lysate added to each gradient was adjusted based on OD600 of the DMSO-treated control culture to ensure reproducibility across experiments . Gradients were either fractionated using an AKTA Prime FPLC ( GE Healthcare , Little Chalfont , England ) outfitted with a continuous flow UV cell at 260 nm or analyzed via continuous flow UV and scintillation counting using an AKTA Prime FPLC in series with a Perkin Elmer 150TR flow scintillation analyzer . Sucrose density gradients , loaded with clarified cell lysates normalized for OD600 , were ran as described above . Total rRNA from 500 μl sucrose gradient fractions was purified using phenol chloroform extraction followed by sodium acetate precipitation and dissolved in 5 μl of water . 1 μl of rRNA from each sample was added to 9 μl of water and 1 μl ( 2 . 4 pmol ) of the necessary primer was added . Each 11 μl reaction was incubated at 80°C for 10 min and allowed to cool to room temperature in order to denature the rRNA . rRNA was subsequently reverse transcribed at 45°C for 24 hr using RevertAid H Minus Reverse Transcriptase from Thermo Scientific in a reaction volume of 20 μl according to the manufacturers instructions . cDNA products from each reaction were precipitated using sodium acetate and 90% ethanol and washed once in 70% ethanol . Purified cDNA samples were analyzed via capillary electrophoresis using a GeneScan 350 TAMRA size standard ( Thermo Scientific ) . 16S rRNA and 23S rRNA primers containing a 5′ 6-oxyfluorescein marker were purchased from Sigma . 16S rRNA primer sequence: 5′-CTGTTACCGTTCGACTTG-3′ . 23S rRNA primer sequence: 5′-CTTATCGCAGATTAGCACG-3′ . Reference standard ribosomal particles were prepared by growing E . coli strain NCM3722 in supplemented M9 ( 48 mM Na2HPO4 , 22 mM KH2PO4 , 8 . 5 mM NaCl , 10 mM MgCl2 , 10 mM MgSO4 , 5 . 6 mM glucose , 50 µM Na3·EDTA , 25 mM CaCl2 , 50 µM FeCl3 , 0 . 5 µM ZnSO4 , 0 . 5 µM CuSO4 , 0 . 5 µM MnSO4 , 0 . 5 µM CoCl2 , 0 . 04 µM d-biotin , 0 . 02 µM folic acid , 0 . 08 µM vitamin B1 , 0 . 11 µM calcium pantothenate , 0 . 4 nM vitamin B12 , 0 . 2 µM nicotinamide , and 0 . 07 µM riboflavin ) bearing 7 . 6 mM of either 14N or 15N-labeled ( NH4 ) 2SO4 . Cells were harvested at OD = 0 . 5 and lysed in buffer A ( 20 mM Tris–HCl , 100 mM NH4Cl , 10 mM MgCl2 , 0 . 5 mM EDTA , 6 mM β-mercaptoethanol; pH 7 . 5 ) using a mini bead beater . Clarified lysates ( 5 ml ) were layered above a 5 ml sucrose cushion ( 20 mM Tris–HCl , 500 mM NH4Cl , 10 mM MgCl2 , 0 . 5 mM EDTA , 6 mM β-mercaptoethanol , 37% sucrose; pH 7 . 5 ) and were spun for 22 hr at 37 . 2k rpm in a Ti 70 . 1 rotor . Pellets bearing 70S ribosomes were solubilized in buffer A at 4°C and saved at −80°C . Lamotrigine- and DMSO-treated ribosomal particles were separated on a sucrose density gradient and fractions were collected as described above . A mixed reference standard bearing 10 pmol of 14N-labeled and 30 pmol of 15N-labeled 70S ribosomal particles was added to 20 pmol of each experimental fraction . The use of this mixed reference ensured that every 15N-labeled peptide bore a 14N-labeled peptide pair irrespective of the abundance of that peptide in the experimental sample . Additionally , one sample bearing only the reference standard was mixed with an equal volume of buffer A . These samples were then prepared for LC/MS via precipitation , reduction , alkylation , and tryptic digestion as described previously ( Jomaa et al . , 2014 ) . Peptides were eluted from a C18 column using a concave acetonitrile gradient and detected using first an Agilent ( Santa Clara , CA ) G1969A ESI-TOF and second , to improve proteomic coverage and to identify non-ribosomal proteins , using an AB/Sciex ( Framingham , MA ) 5600 Triple-TOF run in MS2 mode . In each case , the entire isotope distribution of each extracted MS1 spectrum was fit using a Least Squares Fourier Transform Convolution algorithm ( Sperling et al . , 2008 ) providing accurate quantitation of the 14N and 15N species' abundance . To account for the reference standard's contribution to the measured 14N peptide abundance , each spectrum was normalized using the paired 15N abundance . Having measured the reference standard alone in triplicate , we then subtracted these normalized spectra , resulting in the corrected peptide abundance for each peptide in each experimental sample . Data sets from the ESI-TOF and Triple-TOF were merged and filtered for interference from co-eluting peptides . As a proof of principle , a series of standards bearing various quantities of 14N-labeled 70S particles were also analyzed to assess the linearity of our detection technique ( Gulati et al . , 2014 ) . Non-ribosomal proteins were quantified across the gradient using the aforementioned MS2 Triple-TOF data sets . These data sets were acquired as IDA experiments with 200 ms MS1 scans followed by 50 MS2 scans , each with 50 ms of ion accumulation . Precursor ions were excluded from MS2 analysis 12 s after one occurrence . In each fraction , spectral counts for each non-ribosomal protein were normalized to the total number of spectral counts in that fraction . These values were then normalized to the maximal spectral counts in any gradient fraction , and the occupancy profile was smoothed using a 3-fraction sliding Gaussian window . 30 ml mid-log cultures of E . coli BW25113 ( OD = 0 . 4 ) grown in M9 media were pulsed with 50% 15N by adding 30 ml of M9 containing 15N ammonium chloride as the sole nitrogen source . Cells were introduced to the necessary concentrations of chloramphenicol or lamotrigine during the pulse by supplementing the 15N M9 with antibiotic . Cells were pulsed for 0 hr , 1 hr , 2 . 6 hr , 4 hr , 8 hr , and 16 hr , at which times 10 ml of each culture was removed and the cells harvested via centrifugation . Cell pellets were frozen at −80°C prior to processing for mass spectrometry . Pulse-labeled cells were spiked with 20 pmol of 15N-labeled 70S ribosomal particles and prepared for analysis on the ESI-TOF as described above . Extracted MS1 spectra were fit using Least Squares Fourier Transform Convolution algorithm with three species: 0% 15N ( pre-pulse ) , 50% 15N ( post-pulse ) , and 100% 15N ( Sperling et al . , 2008 ) . Each species was normalized to the reference resulting in the following: pre-pulse material [0%/100%] , post-pulse synthesis [50%/100%] , and total material ( [0% + 50%]/100% ) . Synthesis rates were calculated for each r-protein independently by fitting the median post-pulse synthesis measurement for the 0 , 4 , 8 , 16-hr time points to a line . The synthesis rate of each of r-protein in the lamotrigine or chloramphenicol treatment was normalized to that of the DMSO treatment and log-transformed . The resultant values were presented as notched box and whisker plots , centered at the growth rate for each treatment ( Figure 6E ) . Whiskers extend to the most extreme data point within 1 . 5 times the inner quartile range ( IQR ) whereas notches extend from the median 1 . 57 × IQR/ ( number of points ) 1/2 . All aforementioned data analysis was performed using a series of Python scripts available at https://github . com/joeydavis/StokesDavis_eLife_2014 . Dense overnight cultures of E . coli BW25113 were diluted 1/1000 in 10 ml of LB media supplemented with 39 μM lamotrigine and grown at 15°C until cultures became dense . This occurred after 7 days of incubation . Potential suppressor clones from these cultures were subsequently passaged three times on LB agar . Single colonies from LB agar plates were then re-streaked onto LB agar supplemented with 39 μM lamotrigine to assess mutation stability and purify individual suppressor clones . Individual colonies were isolated based on colony diameter and analyzed via UV absorbance at 600 nm in liquid LB media to determine growth kinetics and lamotrigine MICs at 15°C . Kinetic growth assays were conducted in a temperature-controlled Tecan ( Mannedorf , Switzerland ) Sunrise plate reader . The ribosomal particles of these clones were analyzed using sucrose density gradient centrifugation as described above . Genomic DNA from wild-type E . coli BW25113 and lamotrigine suppressor mutants were purified using a Qiagen ( Venlo , Netherlands ) Gentra Puregene kit and sequenced using an Illumina ( San Diego , CA ) MiSeq platform . Paired-end 250 bp read data for wild type and mutant samples were aligned to the E . coli MG1655 chromosome ( NC_000913 ) using BowTie2 , and mutations were visualized and annotated using BreSeq and Tablet . Wild type and mutant #3 E . coli infB genes were cloned into the pDEST17 plasmid containing an N-terminal His tag using the Invitrogen ( Carlsbad , CA ) Gateway cloning system . Protein expression was conducted in E . coli BL21-AI cells grown in LB at 15°C . Expression was induced at OD ∼0 . 6 using 0 . 2% arabinose , and cells were harvested after 16 hr of induction . Cells were lysed using a Constant Systems cell disruptor at 20 kpsi in IF2 lysis buffer ( 50 mM HEPES–KOH , pH 7 . 4 , 1 M NH4Cl , 10 mM MgCl2 , 0 . 1% Triton X-100 , 7 mM β-mercaptoethanol ) containing EDTA-free protease inhibitor tablets from Roche ( Basel , Switzerland ) . Cell lysates were clarified via centrifugation at 20 , 000 rpm for 45 min in a Beckman Coulter JA-25 . 50 rotor . Clarified lysates were loaded onto a 1 ml GE Healthcare HisTrap FF column and eluted with IF2 elution buffer ( 50 mM HEPES–KOH , pH 7 . 4 , 1 M NH4Cl , 10 mM MgCl2 , 7 mM β-mercaptoethanol , 400 mM imidazole ) . Purified wild type and mutant IF2 was buffer exchanged into ice-cold ribosome buffer using an Amicon ( Millipore , Billerica , MA ) Ultracel filtration unit with a 50 kDa cutoff filter . 50 µl reactions containing 2 mg/ml BSA , 20 µM IF2 , 200 nM [3H]-lamotrigine ( specific activity 5 Ci/mmol; purchased from American Radiolabeled Chemicals ) , and 30 mM G-nucleotide ( purchased from Sigma ) were incubated at 15°C in ribosome buffer for 3 hr . Reactions were loaded onto 200 µl pre-wet Sephadex G-25 resin beds ( resin purchased from GE Healthcare ) and centrifuged at 400×g for 3 min . Flow-through samples were scintillation counted using Perkin Elmer Ultima Gold scintillation fluid . 1 ml early-log cultures of E . coli BW25113 ( OD = 0 . 2 ) grown in M9 media were treated with 8× MIC of various antibiotics and were concurrently pulse labeled with [35S]-methionine ( purchased from Perkin Elmer ) to a final concentration of 5 μCi/ml ( specific activity 1175 Ci/mmol ) . Cells were incubated for 2 . 6 hr at 15°C , at which time they were harvested via centrifugation and washed twice in 1 ml 0 . 85% saline . Cells were lysed using 100 μl Millipore BugBuster Master Mix reagent and proteins were precipitated using 25 μl ice-cold 25% TCA . Protein pellets were then washed twice in 25 μl ice-cold 10% TCA and passed through Whatman ( GE Healthcare ) GF/C filters using a Millipore vacuum manifold . Filters were washed three times in ice-cold 10% TCA , dried overnight at room temperature , and scintillation counted using Perkin Elmer Ultima Gold scintillation fluid . Cell-free translation was conducted using the E . coli S30 transcription–translation system for circular DNA from Promega ( Finchburg , WI ) according the manufacturer's instructions . 10 μl reactions containing the necessary antibiotics and plasmid DNA encoding the firefly luciferase gene were incubated either at 15°C for 4 hr or 37°C for 1 hr . Reactions were halted on ice for 5 min prior to addition of 25 μl of room-temperature luciferin ( purchased from Promega ) . Immediately after the addition of luciferin , samples were analyzed for luminescence output in a Nunc ( Roskilde , Denmark ) 384-well clear bottom plate using a Tecan Ultra Evolution luminometer . The GenBank accession numbers for the IF2 variants described in this paper are KJ752767 ( wild-type E . coli BW25113 IF2 ) ; KJ52768 ( mutant #1 IF2 ) ; KJ752769 ( mutant #2 IF2 ) ; KJ752770 ( mutant #3 IF2 ) ; and KJ752771 ( mutant #4 IF2 ) .	Inside cells , molecular machines called ribosomes make proteins from instructions that are provided by genes . The ribosomes themselves are made up of about 50 proteins and three RNA molecules that need to be assembled like a 3-D jigsaw . In bacteria , a group of proteins called ribosome biogenesis factors help to assemble these pieces correctly . To study how a biological process works , scientists often look at what happens when a component is missing or not working properly . However , this approach cannot be used to study how ribosomes are made because stopping protein production entirely will kill the cell . Another approach is to use chemicals to temporarily stop or slow down a biological process , but researchers are yet to find a chemical that can do this for ribosome assembly . To address this problem , Stokes et al . ‘screened’ 30 , 000 chemicals in an effort to find one or more that could affect ribosome assembly in bacteria . The screen revealed that a drug called lamotrigine—which is used to treat epilepsy and other conditions in humans—could stop the assembly of ribosomes , but did not affect the production of proteins by completed ribosomes . The experiments also suggest that initiation factor 2 , a protein that is involved in the production of other proteins , may also have a role in ribosome assembly . Another recent study found that the equivalent of initiation factor 2 in yeast acts as a quality control checkpoint during ribosome assembly , so the bacterial version may also perform a similar role . It is also be possible that lamotrigine might be used to help develop a novel mechanistic class of antibiotics .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "microbiology", "and", "infectious", "disease" ]	2014	Discovery of a small molecule that inhibits bacterial ribosome biogenesis
Cellular senescence is a crucial tumor suppressor mechanism . We discovered a CAPERα/TBX3 repressor complex required to prevent senescence in primary cells and mouse embryos . Critical , previously unknown roles for CAPERα in controlling cell proliferation are manifest in an obligatory interaction with TBX3 to regulate chromatin structure and repress transcription of CDKN2A-p16INK and the RB pathway . The IncRNA UCA1 is a direct target of CAPERα/TBX3 repression whose overexpression is sufficient to induce senescence . In proliferating cells , we found that hnRNPA1 binds and destabilizes CDKN2A-p16INK mRNA whereas during senescence , UCA1 sequesters hnRNPA1 and thus stabilizes CDKN2A-p16INK . Thus CAPERα/TBX3 and UCA1 constitute a coordinated , reinforcing mechanism to regulate both CDKN2A-p16INK transcription and mRNA stability . Dissociation of the CAPERα/TBX3 co-repressor during oncogenic stress activates UCA1 , revealing a novel mechanism for oncogene-induced senescence . Our elucidation of CAPERα and UCA1 functions in vivo provides new insights into senescence induction , and the oncogenic and developmental properties of TBX3 . Senescence is defined as irreversible arrest of cell growth and loss of replicative capacity ( Hayflick , 1965 ) . Senescent cells have a large , flattened morphology and a characteristic secretory phenotype . They may be multinucleate , exhibit nuclear distortion , and contain senescence-associated heterochromatin foci ( SAHFs ) ( Kosar et al . , 2011 ) . Senescence can be induced by various stimuli such as DNA damage , metabolic or oxidative stress , or expression of oncoproteins ( Larsson , 2005; Kuilman et al . , 2010; Coppé et al . , 2011 ) . The p16/retinoblastoma protein ( RB ) and p53 tumor suppressor pathways are key regulators of senescence induction and maintenance in many cell types ( Narita et al . , 2003 ) . p14ARF-p53 activates p21 , whereas the p16INK4a-RB pathway culminates in E2F transcriptional target repression and senescence ( DeGregori , 2004 ) . Expression of CDKN2A-p14ARF and CDKN1A-p21CIP is repressed by the related transcription factors TBX2 and TBX3; this is the postulated mechanism for senescence bypass of Bmi1−/− and SV40 transformed mouse embryonic fibroblasts by overexpressed TBX2 and TBX3 , respectively ( Jacobs et al . , 2000; Brummelkamp et al . , 2002; Prince et al . , 2004 ) . Mutations in human TBX3 cause a constellation of severe birth defects called ulnar-mammary syndrome ( Bamshad et al . , 1997 ) . Efforts to understand the molecular biogenesis of this developmental disorder uncovered additional functions for TBX3 beyond transcriptional repression ( Fan et al . , 2009; Frank et al . , 2013; Kumar et al . , 2014 ) as well as critical roles in adult tissue homeostasis ( Frank et al . , 2012 ) . The pleiotropic effects of TBX3 gain and loss of function suggest its molecular activities are context and cofactor dependent . Despite the biologic importance of TBX3 , few interacting proteins or target genes have been discovered , and the mechanisms underlying its regulation of cell fate , cell cycle , and carcinogenesis are obscure . We found that TBX3 associates with CAPERα ( Coactivator of AP1 and Estrogen Receptor ) , a protein identified in a liver cirrhosis patient who developed hepatocellular carcinoma ( Imai et al . , 1993 ) . CAPERα regulates hormone responsive expression and alternative splicing of minigene reporters in vitro ( Jung et al . , 2002; Dowhan et al . , 2005 ) but its in vivo functions are unknown . We show that a CAPERα/TBX3 repressor complex is required to prevent premature senescence of primary cells and regulates the activity of core senescence pathways in mouse embryos . We discovered co-regulated targets of this complex in vivo and during oncogene-induced senescence ( OIS ) , including a novel tumor suppressor , the lncRNA UCA1 . UCA1 is sufficient to induce senescence and does so in part by sequestering hnRNP A1 to specifically stabilize CDKN2A-p16INK mRNA . Our finding that CAPERα/TBX3 regulates p16 levels by dual , reinforcing mechanisms position CAPERα/TBX3 and UCA1 upstream of multiple members of the p16/RB pathway in the regulatory hierarchy that controls cell proliferation , fate and senescence . We recently discovered that TBX3 ( human ) and Tbx3 ( mouse ) interact with RNA-binding and splicing factors ( Kumar et al . , 2014 ) . Among these , mass spectrometry of anti-TBX3 immunoprecipitated ( IP'd ) proteins identified CAPERα ( Figure 1A ) . Since TBX3 functions in mammary development and may contribute to the pathogenesis of breast and other hormone responsive cancers ( Douglas and Papaioannou , 2013 ) , its interaction with an ERα co-activator drove further investigation . 10 . 7554/eLife . 02805 . 003Figure 1 . CAPERα and TBX3 directly interact via the TBX3 repressor domain . ( A ) Representative spectrum for CAPERα identified in anti-TBX3 co-IP of HEK293 cell lysates . Mass spec analysis identified six specific CAPERα peptides , providing 8 . 5% sequence coverage of the protein . This spectrum shows fragmentation of one of these peptides , CPSIAAAIAAVNALHGR , with diagnostic b- and y-series ions shown in red and blue , respectively . indicates carbamidomethylation . ( B ) Anti-CAPERα immunoblot ( IB ) analysis of anti-CAPERα immunoprecipitated ( IP'd , lane 2 ) e10 . 5 mouse embryo lysates . Black arrowheads indicate IgG heavy chain and red indicate protein of interest ( CAPERα or TBX3 ) . ( C ) Anti-Tbx3 IB of anti-Tbx3 ( lane 4 ) and anti-Caperα ( lane 5 ) IP'd mouse embryo lysates . Rabbit ( r ) -IgG ( lanes1 , 6 ) and mouse ( m ) -IgG ( lane 7 ) are negative controls . ( D ) In vitro MBP pull down assay: MBP and MBP-Tbx3 bound amylose affinity columns were incubated with GST or GST-CAPERα . Bound proteins were eluted , subjected to SDS-PAGE followed by IB with anti-CAPERα antibody . ( E–G ) Colocalization of Tbx3 and Caperα in vivo shown by immunohistochemical analysis of sectioned e10 . 5 mouse embryo: embryonic dorsal root ganglion ( DRG , E ) , proximal ( F ) , and distal ( G ) limb bud with anti-Tbx3 ( red ) and anti-Caperα ( green ) antibodies and DAPI ( blue ) . White arrowheads in G label representative ectodermal and mesenchymal cells with cytoplasmic Tbx3 and nuclear Caperα . ( H ) Schematic representation of mouse Tbx3 overexpression constructs . Tbx3 DNA binding domain ( DBD ) point , ΔRD and exon7 missense proteins are untagged and the C-terminal deletion mutants are Myc-tagged . ( I ) Anti-TBX3 IB of HEK293 cell lysates transfected with control or anti-TBX3 shRNA . ( J ) Anti-CAPERα IB of anti-CAPERα IP'd samples from HEK293 cells transfected with anti-TBX3 shRNA and expressing mouse Tbx3 proteins listed at top . Production and IP of endogenous CAPERα is not affected by production of mutant Tbx3 proteins . ( J′ ) Anti-Tbx3 IB of anti-CAPERα IP'd samples from HEK293 cells transfected with anti-TBX3 shRNA and expressing Tbx3 proteins as in J . The DBD point mutant proteins ( lanes 2 , 3 ) interact with CAPERα as efficiently as wild type Tbx3 ( lanes 1 , 4 ) . ( K ) Anti-Myc IB of anti-Myc IP'd samples from HEK293 cell lysates expressing Myc-tagged mouse Tbx3 C-terminal deletion mutants . The mutant proteins are expressed and efficiently IP'd . These cells were not treated with anti-TBX3 shRNA because the expression constructs produce a Myc- tagged mutants that can be IP'd independently of endogenous TBX3 . ( K′ ) anti-CAPERα IB of anti-Myc IP'd samples from HEK293 cell lysates expressing Myc-tagged mouse Tbx3 C-terminal deletion mutants . These cells were not treated with anti-TBX3 shRNA because the expression constructs produce a Myc- tagged mutants that can be IP'd independently of endogenous TBX3 . ( L ) Anti-Tbx3 IB of anti-Tbx3 IP'd samples from HEK293 cells transfected with anti-TBX3 shRNA and expressing wt or repressor domain deletion mutant ( ΔRD ) mouseTbx3 . The shRNA does not prevent production of the overexpression proteins . ( L′ ) Anti-CAPERα IB of HEK293 cells transfected with anti-TBX3 shRNA and expressing mouse wt or ΔRD Tbx3 proteins and IP'd with anti-Tbx3 or IgG . Loss of the repressor domain prevents interaction with CAPERα . Black arrowheads indicate IgG heavy chain and red indicate protein of interest ( CAPERα or TBX3 ) . TBX3 , CAPERα = human; Tbx3 , Caperα = mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 02805 . 00310 . 7554/eLife . 02805 . 004Figure 1—figure supplement 1 . Missense mutation of the C-terminus of Tbx3 disrupts interaction with CAPERα . ( A ) Anti-Tbx3 IB of exon 7 missense ( ex7 ) and wt proteins expressed in HEK293 cells also transfected with anti-TBX3 shRNA . The overexpressed proteins are produced ( red arrowhead ) . ( B ) anti-CAPERα and anti-TBX3 ( C ) IB of anti-CAPERα and negative control IP'd samples from HEK293 cells transfected with anti-TBX3 shRNA and overexpressing ex7 missense or wt Tbx3 . Production and IP of endogenous CAPERα is not affected by production of mutant Tbx3 proteins . ( C ) Anti-Tbx3 IB of anti-CAPERα and negative control IP'd samples from HEK293 cells transfected with anti-TBX3 shRNA and overexpressing ex7 missense or wt Tbx3 . The missense mutation disrupts interaction between Tbx3 and CAPERα . Black arrowheads indicate IgG heavy chain and red indicate protein of interest ( CAPERα or TBX3 ) . TBX3 , CAPERα = human; Tbx3 , Caperα = mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 02805 . 004 To determine if Tbx3 and Caperα interact in vivo , we IP'd endogenous Caperα from embryonic day ( e ) 10 . 5 mouse embryo lysates ( Figure 1B ) . Immunoblotting for Tbx3 confirmed its interaction with Caperα ( Figure 1C , lane 5 ) and in vitro pull down assays revealed that their interaction is direct ( Figure 1D , lane 6 ) . Caperα is very broadly expressed during mouse embryonic development ( Moon , unpublished ) , whereas Tbx3 expression is very tissue specific and dynamic . We thus questioned whether the endogenous proteins interact in mouse tissues relevant to malformations seen in humans with UMS . Immunohistochemistry on sectioned e10 . 5 embryos showed that Tbx3 and Caperα proteins are co-expressed and have distinct localization patterns in different tissues: Caperα is detected in all dorsal root ganglia nuclei ( Figure 1E ) , some of which contain co-localized Tbx3; in proximal limb mesenchyme , Tbx3 and Caperα co-localize in nuclei ( Figure 1F ) while in some distal cells and the ectoderm , Caperα is nuclear and Tbx3 is cytoplasmic ( Figure 1G , white arrowheads ) . Such tissue specificity suggests that functions of the Caperα/Tbx3 complex are context dependent . TBX3 DNA binding and repressor domains ( DBD , RD ) independently mediate interactions with partner proteins ( Carlson et al . , 2001; Coll et al . , 2002; Kumar et al . , 2014 ) . To identify domains required for CAPERα interaction , we used a series of overexpression plasmids encoding mouse Tbx3 proteins with different mutations and functional domains ( Figure 1H ) . The DBD , deleted repressor domain ( ΔRD ) and exon7 missense mutants are untagged proteins , whereas the C-terminal deletion mutants are Myc-tagged . To assay the interactions of the untagged exogenous proteins with endogenous CAPERα in HEK293 cells , we needed to knockdown endogenous TBX3 with shRNA ( Figure 1I ) . We previously demonstrated that mutant Tbx3 proteins produced from the overexpression plasmids are present in TBX3 knockdown HEK293 cells ( Figure 2 in Kumar et al . 2014 ) . CAPERα is present and can be IP'd in the context of knockdown of endogenous TBX3 and subsequent overexpression of mutant mouse Tbx3 proteins ( Figure 1J ) . Immunoblot of anti-CAPERα IP'd samples shows that the endogenous CAPERα interacts with Tbx3 DBD mutant proteins ( Figure 1J′ , lanes 2 and 3 are L143P and N227D , respectively ) . The Tbx3 deletion constructs encode Myc- tagged mutants that can be distinguished from endogenous TBX3 , so interactions were assayed in wild-type HEK293 cells . Myc-tagged deletion mutants are IP'd by the anti-Myc antibody ( Figure 1K ) , and probing anti-Myc IP'd material for CAPERα reveals that deletions more proximal than amino acid 655 disrupt the CAPERα/Tbx3 interaction ( Figure 1K′ ) . The observation that deletions of the Tbx3 C-terminus disrupt the CAPERα/Tbx3 interaction led us to test whether the C-terminal repressor domain , which is crucial for the ability of Tbx3 to function as a transcriptional repressor and immortalize fibroblasts ( Carlson et al . , 2001 ) , plays a role . Although the untagged ΔRD mutant is produced in TBX3 shRNA knockdown cells and IP'd by the anti-Tbx3 antibody ( Figure 1L and Kumar et al . , 2014 ) it does not interact with CAPERα ( Figure 1L′ ) . CAPERα also fails to interact with a C-terminal Tbx3 frameshift mutant similar to one identified in humans with UMS ( Bamshad et al . , 1999 ) ( Figure 1—figure supplement 1 ) . Roles for TBX3 in cell cycle regulation and senescence of primary cells have not been reported . We employed loss-of-function to test whether TBX3 is required for sustained proliferation of primary cultured human foreskin fibroblasts ( HFFs ) and to determine if CAPERα functions in this process . We tested two different CAPERα and TBX3 shRNAs ( please see ‘Materials and methods’ for sequences and location in target mRNAs ) . Both CAPERα and TBX3 shRNAs effectively decreased the amount of CAPERα mRNA ( Figure 2—figure supplements 1A and 2A , B ) . Knockdown of either protein resulted in a dramatic increase in senescence associated β-galatosidase activity ( SA-βgal , Figure 2A–D; Figure 2—figure supplements 1 and 2C–H ) . This effect is specific because it occurs with two different shRNAs and is rescued by overexpression of CAPERα ( Figure 2—figure supplement 1B , E , G , H ) and Tbx3 ( Figure2—figure supplement 2B , E , G , H ) . For all subsequent experiments , CAPERα shRNA 'A' and TBX3 shRNA 'A' were used to perform knockdown ( KD ) in HFFs ( protein knockdowns are shown in Figure 2—figure supplements 1 and 2 , I panels ) . 10 . 7554/eLife . 02805 . 005Figure 2 . Knockdown of endogenous CAPERα and TBX3 in primary human fibroblasts and mouse embryos induces premature senescence and disrupts expression of cell cycle and senescence regulators . ( A–C ) Representative bright field images of senescence associated β-galactosidase ( SA-βG ) assays of HFFs transduced with control , TBX3 shRNA A or CAPERα shRNA A . Only occasional cells in the control transduction have detectable lacZ staining ( blue ) whereas knockdown of either TBX3 or CAPERα results in marked changes in cell morphology and increased lacZ staining . ( D ) Bar graph quantitating % beta-galactosidase positive cells from four replicate plates of SA-βgal assays . * indicates p<0 . 001 compared to control . ( E and F ) 3T5 cell proliferation assay ( Lessnick et al . , 2002 ) of cumulative population doublings in HFFs transduced at passage 30 with control , TBX3 or CAPERα shRNAs . These are representative curves of duplicate experiments; each point on the curve is a measurement of cell count from a single plating followed over the course of the experiment as described in methods . ( G–J ) Immunohistochemical analysis of H3K9me3 immunoreactivity ( red ) and DAPI ( blue ) in HFFs after knockdown with control ( G and I ) , TBX3 ( H ) , or CAPERα ( J ) shRNAs . Individual channels are shown and the merged image is on the right . Note increased nuclear punctate staining consistent with Senescence-associated heterochromatin foci ( SAHFs ) in both channels and evidence of nuclear disruption ( white arrowheads in red channel ) after loss of either TBX3 or CAPERα . ( K–M ) Analysis of cell cycle and senescence marker transcript levels in HFFs transduced with control , TBX3 , or CAPERα shRNAs . ( K ) Relative transcript levels assessed by quantitative real time-PCR ( qPCR ) of cDNA . Values reflect fold change in knockdown HFFs relative to control after normalization to HPRT levels . Note general pattern of expression changes are similar in TBX3 ( blue ) and CAPERα ( red ) knockdowns . Data are plotted as fold change mean ± standard deviation . * indicates p<0 . 05 relative to control . ( L and M ) Agarose gel of PCR amplicons of cDNAs reverse transcribed from TBX3 ( L ) or CAPERα ( M ) shRNA knockdown HFF RNA reveals similar decreases in cell cycle promoting genes CDK2 and 4 in TBX3 and CAPERα knockdowns and increased p21 levels . ( N and O ) SA-βgal assay of wild type and Tbx3 null MEFS reveals that Tbx3 is required to prevent premature senescence of primary murine embryonic fibroblasts ( MEFs ) . ( P ) Quantitation of % beta-galactosidase positive cells from five replicate experiments exemplified in O , P . * indicates p<0 . 01 . ( Q ) 3T5 cell proliferation assay of cumulative population doublings in wild-type and Tbx3 null MEFs . These are representative curves from duplicate experiments; each point on the curve is a measurement of cell count from a single plating followed over the course of the experiment as described in 'Materials and methods' . ( R ) IBs to assay levels of cell cycle and senescence proteins in wild type and Tbx3 null embryo lysates . Tubulin loading control is at top left ( Tub ) . The changes at the protein level correlate with those observed at the RNA level ( K–M ) and RB is hypophosphorylated on multiple serine residues consistent with increased p16 and decreased CDK activity . TBX3 , CAPERα = human; Tbx3 , Caperα = mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 02805 . 00510 . 7554/eLife . 02805 . 006Figure 2—figure supplement 1 . Effective knockdown of endogenous CAPERα in primary human foreskin fibroblasts using viral shRNA transduction . ( A ) RT-PCR analysis of CAPERα and HPRT transcript levels in HFFs transduced with two different retroviruses producing anti-CAPERα shRNAs ( CAP sh A and B ) and control shRNA virus ( Ctl sh ) . Red arrowhead indicates CAPERα-specific amplicon . ( B ) RT-PCR analysis of CAPERα and HPRT transcript levels in HFFs transduced with retroviruses producing anti-CAPERα shRNA A and a CAPERα overexpression virus ( CAP OE ) . Note rescue of CAPERα expression by overexpression virus . ( C–G ) SA-βGal assays of HFFs transduced with control or CAPERα shRNAs A or B and rescue by CAPERα overexpression . ( H ) Quantitation of SA-βGal assays in C–G . * indicates p<0 . 01 compared to control shRNA . ( I ) Western blot showing depletion of endogenous CAPERα protein by CAP shRNA A . Anti-tubulin IB is loading control . This CAPERα shRNA ‘A’ was used for all subsequent CAPERα shRNA knockdown experiments . TBX3 , CAPERα = human; Tbx3 , Caperα = mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 02805 . 00610 . 7554/eLife . 02805 . 007Figure 2—figure supplement 2 . Effective knockdown of endogenous TBX3 in primary human foreskin fibroblasts using viral shRNA transduction . ( A ) RT-PCR analysis of TBX3 and HPRT transcript levels in primary human foreskin fibroblasts ( HFFs ) transduced with control ( Ctl sh ) or TBX3 ( TBX3 shA ) shRNA retrovirus . ( B ) RT-PCR analysis of TBX3 and HPRT transcript levels in primary human foreskin fibroblasts ( HFFs ) transduced with control ( Ctl sh ) or TBX3 ( TBX3 shB ) shRNA retrovirus . ( C–G ) SA-βGal assays of HFFs transduced with control or TBX3 shRNAs A or B and rescue by Tbx3 overexpression . ( H ) Quantitation of SA-βGal assays in C–G . * indicates p<0 . 01 compared to control shRNA . ( I ) Western blot showing depletion of endogenous TBX3 protein by TBX3 shRNA A . Anti-tubulin IB is loading control . This TBX3 shRNA ‘A’ was used for all subsequent TBX3 shRNA knockdown experiments . TBX3 , CAPERα = human; Tbx3 , Caperα = mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 02805 . 00710 . 7554/eLife . 02805 . 008Figure 2—figure supplement 3 . Tbx3 null murine embryonic fibroblasts ( MEFS ) have altered lamin β1 localization , nuclear disruption and mislocalized Caperα . . ( A–B′ ) Representative WT and Tbx3 null MEFs cells stained for laminβ1 at passage ( P ) 4 ( A and B ) and P1 ( A′ and B′ ) ; note nuclear distortion and rupture in senescing Tbx3 null MEFs as early as P1 . ( C ) Quantitation of % distorted nuclei in WT vs Tbx3 null MEFs . * indicates p<0 . 05 . ( D–F′ ) Immunohistochemistry for Caperα ( green ) and DNA ( DAPI , blue ) in control and Tbx3 null MEFs at P1 ( D and D′ ) and P2 ( E–F′ ) . In mutant cells , Caperα signal shifts to nucleus from cytoplasm at P1 , and large intranuclear Caperα+ foci are present by P2 . ( G ) qPCR quantitation of senescence marker genes in WT vs Tbx3 null MEFs . Data are displayed as mean fold change ± standard deviation relative to WT after normalization to HPRT levels . * indicates p<0 . 01 . # indicates p<0 . 05 . TBX3 , CAPERα = human; Tbx3 , Caperα = mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 02805 . 008 The effects of CAPERα and TBX3 KD on HFF cell growth , and SA-βgal activity suggest induction of premature senescence . Consistent with this , both KDs dramatically influenced nuclear structure , chromatin organization and formation of SAHFs ( Figure 2G–J ) . Expression of senescence mediators was increased and conversely , expression of cell growth and cell cycle promoting genes was similarly decreased by CAPERα and TBX3 KD ( Figure 2K–M ) . Increased expression of CDKN2A-p16INK ( henceforth referred to as p16INK ) and decreased PCNA , E2F1 and 2 , CDK2 , CDK4 , CDC2 transcripts indicate that CAPERα/TBX3 represses the p16/RB pathway in proliferating HFFs . PMAIP1 , CDKN1A-p21 , and other p53 pathway members were also increased . Collectively , these data indicate that CAPERα and TBX3 are required to prevent senescence of primary HFFs and act upstream of major cell cycle and senescence regulatory pathways . Tbx3 deficiency in mice causes lethal embryonic arrhythmias and limb defects however , these phenotypes are not due to increased apoptosis ( Frank et al . , 2012 and Emechebe and Moon , unpublished ) . We hypothesized that Tbx3 may prevent senescence of embryonic cells , and so examined murine embryonic fibroblasts ( MEFs ) from e13 . 5 wild type ( WT ) and Tbx3 null ( −/− ) embryos . WT MEFs undergo ∼10 passages with regular , 20 hr doubling times . In contrast , Tbx3−/− MEFs had increased SA-βgal activity and ceased proliferating after only four passages ( Figure 2N–Q ) . Most Tbx3−/− MEFs had distorted or ruptured nuclei ( Figure 2—figure supplement 3A–C ) and laminβ1 staining was already altered at passage 1 ( Figure 2—figure supplement 3B′ ) . Caperα null mutant embryos do not survive long enough to generate MEFs for complementary experiments ( Emechebe and Moon , unpublished ) however , Caperα localization is markedly abnormal in Tbx3−/− MEFS after only 1 passage ( Figure 2—figure supplement 3D–F′ ) . These data suggest that Tbx3 is required for preservation of nuclear architecture and to tether Caperα in its normal nuclear domains in proliferating cells . Consistent with premature senescence seen in Tbx3−/− MEFs , key pro-senescence pathways are activated after loss of Tbx3 in vivo: in protein lysates from Tbx3−/− embryos , RB was hypophosphorylated on multiple serine residues , consistent with increased p16 and decreased Cdk2 and Cdk4 protein levels relative to control ( Figure 2R ) . The levels of p21 and other senescence markers were increased , while numerous Cyclins and other Cdks were decreased ( Figure 2R , Figure 2—figure supplement 3G ) . All of these findings are consistent with a requirement for Tbx3 to prevent senescence in embryonic mice and MEFs . Previous studies have suggested that overexpression of TBX3 permits senescence bypass by directly repressing CDKN2A-p14ARF ( p14ARF ) to activate p53 ( Brummelkamp et al . , 2002 ) , but a role for TBX3 in regulating p16INK and the RB pathway has not been demonstrated . Thus , we expected that loss of p53 would rescue senescence resulting from TBX3 or CAPERα KD . To test this , we transduced TBX3 and CAPERα KD HFFs with shRNA to p53 ( Masutomi et al . , 2003 ) and assayed SA-βgal activity and growth . Surprisingly , although p53 shRNA effectively decreased p53 ( Figure 3—figure supplement 1A ) , it did not rescue SA-βgal activity or growth arrest due to absence of TBX3 or CAPERα ( Figure 3B , E , G , H ) . In contrast , shRNA-mediated KD of either RB ( Boehm et al . , 2005 ) or p16 ( Haga et al . , 2007 ) ( Figure 3—figure supplement 1B , C ) rescued these phenotypes in TBX3 and CAPERα KD cells ( Figure 3C , F–H , I–N ) . These rescue experiments demonstrate that the p16/RB pathway mediates senescence downstream of CAPERα and TBX3 loss-of-function in primary cells . 10 . 7554/eLife . 02805 . 009Figure 3 . RB and p16 mediate senescence after CAPERα/TBX3 loss of function and CAPERα/TBX3 regulates chromatin structure of CDKN2A-p16 . ( A–F ) SA-βgal assays of HFFs stably transduced with control ( Ctl ) or p53 ( Masutomi et al . , 2003 ) or RB ( Boehm et al . , 2005 ) shRNAs subsequently transduced with CAPERα or TBX3 shRNAs . ( G ) % Quantitation of A–F from three replicate experiments . * indicates p<0 . 05 relative to Control or p53 shRNAs . ( H ) Cell proliferation assayed by crystal violet incorporation ( OD units ) in HFFs treated as in A–F . * indicates p<0 . 001 relative to Ctl or p53 shRNAs . ( I–L ) SA-βgal assays of HFFs stably transduced with control or p16 ( Haga et al . , 2007 ) shRNAs subsequently transduced with CAPERα or TBX3 shRNAs . ( M ) % Quantitation of I-L from three replicate experiments . * indicates p<0 . 05 relative to Ctl shRNA . ( N ) Cell proliferation assayed by crystal violet incorporation ( OD units ) in HFFs treated as in I–L . * indicates p<0 . 01 relative to Ctl shRNA . ( O ) ChIP-PCR with antibodies listed at top on three regions upstream of the CDKN2A-p16 transcriptional start site ( TSS ) ; position relative to ( TSS ) is indicated in parentheses at left of panels . PCR of input material used for the ChIP is shown under ‘Input’ . The shRNA transduced is listed above each lane ( HFF Tx ) . TBX3 knockdown decreases binding of TBX3 ( lanes 8 ) and CAPERα ( lanes 11 ) to all three regions . CAPERα knockdown has minimal effect on TBX3 binding ( lanes 9 ) . Knockdown of either TBX3 or CAPERα decreases the repressive chromatin mark H3K9me3 ( lanes14 , 15 ) and increases the activating chromatin mark H3K4me3 ( lanes 17 , 18 ) . TBX3 , CAPERα = human; Tbx3 , Caperα = mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 02805 . 00910 . 7554/eLife . 02805 . 010Figure 3—figure supplement 1 . Effective knockdown of p53 , RB and p16 in HFFs . ( A–C ) RT-PCR analysis of p53 ( A ) , RB ( B ) and p16 ( C ) transcript levels relative to HPRT after shRNA-mediated KD in HFFs . The shRNAs employed for these knockdowns were obtained from Addgene and have been previously employed by numerous investigators ( Masutomi et al . , 2003; Boehm et al . , 2005; Haga et al . , 2007; Hong et al . , 2009; Elzi et al . , 2012 ) . TBX3 , CAPERα = human; Tbx3 , Caperα = mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 02805 . 01010 . 7554/eLife . 02805 . 011Figure 3—figure supplement 2 . UCSC Genome Browser view of the CDKN2A locus and 5′ regions screened for binding by CAPERα and TBX3 . Seven regions tested upstream of CDKN2A-p16 promoter by ChIP with anti-TBX3 and anti-CAPERα antibodies . Amplicons are numbered black boxes 1–7 ‘Your Seq’ at top superimposed on window from UCSC genome browser . Chromatin states in various cell types based are noted by colored bars below . Of these 7 regions , 3 were bound by both TBX3 and CAPERα: regions 3 , 4 and 5 ( data are presented in Figure 3O ) . TBX3 , CAPERα = human; Tbx3 , Caperα = mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 02805 . 01110 . 7554/eLife . 02805 . 012Figure 3—figure supplement 3 . CDKN2a-p16 H3K27 trimethylation markedly decreases in HFFS after knockdown of CAPERα or TBX3 consistent with activation of CDKN2a-p16 expression . ChIP-PCR of CDKN2A-p16 regulatory elements with anti-H3K27me3 in control , TBX3 or CAPERα shRNA-transduced HFFs . Locations of amplicons relative to transcription start site are noted in parentheses below each panel and correspond to regions 3 , 4 and 5 in Figure 3—figure supplement 2 . TBX3 , CAPERα = human; Tbx3 , Caperα = mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 02805 . 01210 . 7554/eLife . 02805 . 013Figure 3—figure supplement 4 . Testing CAPERα and TBX3 binding to p14 , p21 , CDK2 , CDK4 , and CDKN1B regulatory elements . ( A ) ChIP-PCR of CDKN2A-p14 promoter with antibodies listed at top in control ( C ) and TBX3 siRNA ( C′ ) transduced HFFs . Red arrowhead indicates loss of CAPERα binding after TBX3 knockdown . ( B–E ) ChIP/PCR of HFF chromatin showing lack of TBX3 and CAPERα binding to known regulatory elements ( Baksh et al . , 2002; Wang et al . , 2005; Louie et al . , 2010 ) of: ( B ) CDKN1A-p21 ( location relative to transcription start site is noted in parentheses at the bottom of the panels ( C ) CDK4 ( D ) CDK2 ( E ) CDKN1B . TBX3 , CAPERα = human; Tbx3 , Caperα = mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 02805 . 013 Increased p16 protein and RB hypophosphorylation in Tbx3−/− embryos and p16/RB-mediated senescence after CAPERα and TBX3 KD could result from loss of direct repression of p16INK by CAPERα/TBX3 in proliferating cells . We screened 7 amplicons spanning ∼6 kb upstream of p16INK by ChIP-PCR of HFF chromatin ( Figure 3—figure supplement 2 ) ; 3 amplicons were bound by CAPERα and TBX3 ( Figure 3O , lanes 7 , 10 ) . Loss of either protein decreased the heterochromatic marks H3K9me3 ( Figure 3O , lanes 14 , 15 ) and H3K27me3 ( Figure 3—figure supplement 3 ) and increased the euchromatic mark H3K4me3 ( Figure 3O , lanes 17 , 18 ) . Notably , less CAPERα occupied p16INK elements after TBX3 KD ( Figure 3O , lanes 11 ) while the amount of TBX3 bound post-CAPERα KD was comparable to control ( Figure 3O , lanes 9 vs 7 ) . This is consistent with the abnormal localization of CAPERα seen in Tbx3−/− MEFS ( Figure 2—figure supplement 3D′–F′ ) and indicates that CAPERα requires TBX3 to occupy p16INK regulatory chromatin . We examined whether CAPERα and/or TBX3 associate with promoters of other cell cycle genes that are transcriptionally dysregulated after CAPERα/TBX3 loss-of-function ( Figure 2K–M ) . Antibodies against TBX3 and CAPERα ChIP'd the p14ARF initiator ( Lingbeek et al . , 2002 ) ( Figure 3—figure supplement 4A ) ; here too , TBX3 KD disrupted CAPERα binding ( Figure 3—figure supplement 4A′ , red arrowhead ) . Neither CAPERα nor TBX3 associated with amplicons scanning 1 . 8 kb upstream of CDKN1A-p21 or elements reportedly bound by TBX2 or TBX3 in other cell types ( Figure 3—figure supplement 4B ) ( Prince et al . , 2004; Saramaki et al . , 2006; Hoogaars et al . , 2008 ) . Testing for association with known regulatory elements of CDK2 , CDK4 , CDKN1B was also negative ( Figure 3—figure supplement 4C–E ) ( Baksh et al . , 2002; Wang et al . , 2005; Louie et al . , 2010 ) . These data indicate that in proliferating primary cells , CAPERα/TBX3 specifically and directly repress the CDKN2A locus by binding multiple regulatory sequence elements and regulating chromatin marks . To identify novel genes repressed by CAPERα/TBX3 , we employed differential display to detect transcripts that increased in response to KD of TBX3 and CAPERα in HEK293 cells ( Figure 4A–C ) . Although most transcripts were unaffected by either KD , or changes were not shared ( Figure 4—figure supplement 1A ) , DUSP4 and UCA1 were upregulated ( Figure 4D , Figure 4—figure supplement 1B ) . DUSP4 is known to regulate cell survival and tumor progression , and overexpression induces senescence downstream of RB/E2F ( Torres et al . , 2003; Wang et al . , 2007 ) , thus placing CAPERα/TBX3 upstream of another p16/RB effector . Little is known about the function of the lncRNA UCA1 ( Wang et al . , 2006 , 2008 ) , so we investigated it further . 10 . 7554/eLife . 02805 . 014Figure 4 . CAPERα/TBX3 directly represses expression of the long noncoding RNA UCA1 . ( A–C ) Gel showing RT-PCR analysis of TBX3 , CAPERα , and HPRT expression in control , TBX3 and CAPERα siRNA-transfected HEK293 cells . The siRNAs effectively decreased transcript levels of their targets . ( D ) Differential display: representative PAGE gel of cDNAs derived from random primed , RT-PCR'd mRNAs from CAPERα , TBX3 and control siRNA transfected HEK293 cells . Blue arrowheads denote upregulated transcripts subsequently identified by sequencing as DUSP4 and UCA1 . ( E and F ) qPCR analysis of TBX3 and CAPERα transcript levels in control and TBX3 or CAPERα shRNA transduced HFFs ( repeat of experiment shown in Figure 2—figure supplements 1A and 2A ) . ( G ) RT-PCR analysis of UCA1 and HPRT gene expression in control , TBX3 or CAPERα shRNA-transduced HFFs . ( H ) qPCR analysis of UCA1 transcript levels in control , TBX3 or CAPERα shRNA transduced HFFs . Results confirm differential display result that KD of TBX3 or CAPERα results in increase in UCA1 transcript levels . ( I ) Schematic representation of the UCA1 locus with primer sets employed for ChIP-PCR amplification of denoted regions 5′ of gene ( A1 , A2 , A3 ) . ( J ) Anti-TBX3 ChIP-PCR of regions of the UCA1 promoter in HFFs; only A3 is ChIP'd by TBX3 ( lane 18 , red arrowhead ) . ( K ) Anti-CAPERα ChIP-PCR of regions of the UCA1 promoter in HFFs; only A3 chromatin is ChIP'd ( lane 18 , red arrowhead ) . ( L ) ChIP-PCR analysis of UCA1/A3 chromatin from in HFFs transduced with control ( C ) or TBX3 ( KD ) shRNA; ChIP antibodies are listed at top . Note decreased CAPERα binding after TBX3 KD ( lane 17 , red arrowhead ) , gain of activating mark H3K4me3 and loss of repressive marks H3K9me3 and H3K27me3 . ( M ) ChIP-PCR analysis of UCA1/A3 with antibodies listed at top of panel in HFFs transduced with control ( C ) or CAPERα shRNAs . Note continued TBX3 binding despite CAPERα KD ( lane 11 , red arrowhead ) and changes in chromatin marks parallel those seen in with TBX3 KD in panel L . TBX3 , CAPERα = human; Tbx3 , Caperα = mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 02805 . 01410 . 7554/eLife . 02805 . 015Figure 4—figure supplement 1 . Validation of differential display findings . ( A ) Additional representative differential display gels with transcripts unchanged or independently affected by knockdown of CAPERα or TBX3 in HEK293 cells . ( B ) RT-PCR validating differential display result of increased DUSP4 transcripts ( Figure 4D ) after CAPERα or TBX3 KD in HEK293 cells . TBX3 , CAPERα = human; Tbx3 , Caperα = mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 02805 . 015 We found that shRNA KD of CAPERα or TBX3 in primary HFFS recapitulated the increase in UCA1 transcripts seen in HEK293 cells ( Figure 4E–H ) . We then tested whether CAPERα/TBX3 directly control transcription of UCA1 by interacting with potential regulatory elements . Public ChIP data ( http://genome . ucsc . edu/ ) indicate that the 2 kb upstream of UCA1 may contain such elements . We assayed 3 amplicons in this region ( Figure 4I: A1 , A2 , A3 ) by ChIP-PCR of TBX3 and CAPERα: only region A3 was bound ( Figure 4J , K , lanes 18 , red arrowheads ) . We next determined whether increased UCA1 expression in response to KD of CAPERα or TBX3 was associated with altered chromatin structure ( as seen with p16INK , Figure 3O ) . UCA1/A3 is normally in a heterochromatin configuration in HFFs , with repressive marks H3K9me3 and H3K27me3 ( Figure 4L , lanes 12 , 14 ) and little H3K4me3 ( Figure 4L , lane 18 ) . After TBX3 KD , activating chromatin marks replaced repressive ones ( Figure 4L , lanes 13 , 15 and 19 ) and markedly less CAPERα was bound ( Figure 4L , lane 17 , red arrowhead ) . CAPERα KD also led to loss of repressive marks on UCA1/A3 ( Figure 4M lanes 9 , 16 ) , although TBX3 remained bound ( Figure 4M , lane 11 , red arrowhead ) . Combined with previous findings , we conclude that: ( 1 ) TBX3 recruits CAPERα to UCA1/A3 chromatin , ( 2 ) TBX3 alone is insufficient to repress UCA1 and , ( 3 ) the default state of UCA1 in proliferating HFFs is repression conferred by CAPERα/TBX3 . UCA1 modulates behavior of bladder cancer cell lines ( Wang et al . , 2008 ) , but there are no data on its function in primary cells; our results suggest that UCA1 may be involved in premature senescence . UCA1 transcripts are low in proliferating HFFs , but 4 days after overexpression of UCA1 ( Figure 5A ) , a robust SA-βgal response is evident ( Figure 5B–D ) . Cells constitutively expressing UCA1 ceased proliferating during selection and accumulated SAHFs ( Figure 5E , F ) . Cell proliferation decreased in a UCA1 dosage-sensitive manner ( Figure 5G–I ) , consistent with reduced levels of cell cycle promoting transcripts and increased levels of pro-senescence ones ( Figure 5J ) . These transcriptional changes were manifest at the protein level ( Figure 5—figure supplement 1 ) . Premature senescence resulting from overexpression of UCA1 in HFFs reveals that this lncRNA is a novel regulator of cell proliferation and may function as a tumor suppressor in some contexts . 10 . 7554/eLife . 02805 . 016Figure 5 . UCA1 expression is sufficient to induce senescence and required for normal execution of oncogene-induced senescence . ( A ) UCA1 and HPRT transcripts assessed by RT-PCR in control and UCA1-overexpressing HFFs . ( B and C ) Representative bright field images of SA-βgal assay of cultured HFFs transfected with control and UCA1 overexpression plasmids . ( D ) % quantitation of SA-βgal cells from five replicates in control and UCA1 overexpressing HFFs . * indicates p<0 . 05 . ( E and F ) Immunohistochemical analysis reveals co-localization of H3K9me3 and DAPI in SAHFs in HFFs transfected with UCA1 overexpression plasmid ( F ) but not control plasmid ( E ) . ( G ) Cell count of control and UCA1 overexpressing HFFs 3 days post transfection . Mean ± SD of 3 plates is shown at each time point . * indicates p<0 . 005 relative to control . ( H ) Crystal violet assay of cell growth in control and UCA1 overexpressing HFFs transfected with 2 μg of expression or control vector and assayed daily for 3 days post- transfection . * indicates p<0 . 01 relative to control . ( I ) Crystal violet assay of HFFs cultured for 3 days after transfecting 0 , 1 , 2 , or 4 μg of control or UCA1 overexpression plasmid . * indicates p<0 . 01 relative to control . ( J ) Transcript levels assessed by qPCR; values reflect fold change in UCA1-overexpressing HFFs relative to control after normalization to HPRT levels . * indicates p<0 . 05 relative to control . ( K ) qPCR analysis of UCA1 expression in untransduced , presenescent ( PS ) HFFs and HFFs transduced with constitutively active G12VRAS ( RAS ) . * indicates p<0 . 05 relative to PS . ( L ) Efficient knockdown of UCA1 transcripts in RAS HFFs with UCA1 shRNA ( quantitated in panel T ) . ( M–P ) SA-βgal assays of RAS HFFs transduced with either control or UCA1 shRNA at 3 ( M and O ) and 5 ( N and P ) days post transduction . ( Q ) % quantitation of SA-βgal cells from six replicate experiments as represented in panels M–P . * indicates p<0 . 001 relative to control . ( R ) % quantitation of Ki67 + cells from three replicates in control vs UCA1 shRNA transduced RAS HFFs . * indicates p<0 . 001 relative to control . ( S ) RT-PCR for UCA1 transcripts shows persistent knockdown of UCA1 in RAS shRNA cells with increasing passage ( P0–P2 ) . ( T ) qPCR analysis of fold changes in transcript levels of cell cycle and senescence genes after UCA1 shRNA knockdown in RAS HFFs . * indicates p<0 . 05 relative to control . ( U ) ChIP-PCR analysis of UCA1 region A3 with antibodies listed at top in PS and RAS HFFs . Note gain of activating ( H3K4me3 , H3K9ace , H4K5ace ) and loss of repressive marks ( H3K9me3 , H3K27me3 ) at the UCA1 locus after oncogene-induced senescence by RAS . TBX3 , CAPERα = human; Tbx3 , Caperα = mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 02805 . 01610 . 7554/eLife . 02805 . 017Figure 5—figure supplement 1 . Western blots showing changes in protein levels in response to UCA1 overexpression in HFFs . pcDNA3 . 1 are control transfected cells and UCA1 were transfected with UCA1 expression plasmid in pcDNA3 . 1 ( as in Figure 5A ) . Note increased p16 and p21 levels and hypophosphorylation of RB . TBX3 , CAPERα = human; Tbx3 , Caperα = mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 02805 . 01710 . 7554/eLife . 02805 . 018Figure 5—figure supplement 2 . ChIP-PCR assay for H3K9 acetylation of known regulatory elements of prosenescence and cell cycle genes whose expression is dyregulated after UCA1 overexpression . Input , rabbit IgG negative control ChIP , and H3K9acetylation ChIP in control “C” or UCA1 “U” transfected HFFs for gene regulatory regions as labeled at bottom ( primer sequences listed in ChIP primers section of methods ) . P16 a and b refer to amplicons – ( 2457–2040 ) and – ( 3107–2710 ) , respectively . No changes in H3K9ace levels were detected in response to UCA1 overexpression , suggesting that altered chromatin structure and subsequent increased transcription are not the cause of observed changes in transcript levels detected with UCA1 overexpression and shown in Figure 5J . TBX3 , CAPERα = human; Tbx3 , Caperα = mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 02805 . 018 We tested the hypothesis that UCA1 is required for induction of oncogene-induced senescence ( OIS ) in primary cells ( ‘RAS’: HFFs transduced with constitutively active G12VRAS [Serrano et al . , 1997] ) . There are markedly more UCA1 transcripts in RAS compared to presenescent ‘PS’ HFFs ( Figure 5K ) . Knockdown of UCA1 in RAS HFFs reduced SA-βgal activity ( Figure 5L–Q ) and improved RAS cell growth: the number of Ki67 + RAS cells was increased at days 3 and 6 after UCA1 KD ( Figure 5R , P0 and P1 ) . However , by passage 2 , the number of Ki67 + cells was not statistically different in UCA1 KD cells from control , despite persistently low levels of UCA1 ( Figure 5S ) and decreased levels of pro-senescence transcripts ( Figure 5T ) . Overall , this indicates that senescence can occur in the absence of high levels of UCA1 but that timely execution of the OIS program requires UCA1 . We next investigated whether increase in UCA1 transcripts in OIS is a manifestation of loss of CAPERα/TBX3 occupancy/repression of UCA1/A3 . Indeed , the repressor dissociates from UCA1/A3 in RAS HFFs and UCA1/A3 chromatin switches from heterochromatic to euchromatic marks ( Figure 5U ) . This is consistent with the senescence-inducing effects of CAPERα/TBX3 loss-of-function ( Figure 2 ) and resulting upregulation of UCA1 ( Figure 4 ) , and establishes CAPERα/TBX3 regulation of UCA1 in an independent model of senescence . Some lncRNAs influence transcription by recruiting chromatin modifiers to target genes ( Fatica and Bozzoni , 2014 ) . We tested whether the increased levels of prosenescence transcripts occurring in response to UCA1 ( Figure 5J ) were the result activating chromatin changes however , ChIP-PCR assay for H3K9 acetylation of the p16INK , p14ARF , CDKN1A-p21 ( and other ) promoters did not reveal changes in this activating mark in response to UCA1 ( Figure 5—figure supplement 2 ) . We thus tested whether altered mRNA stability contributed to the observed changes . HFFs were transfected with UCA1 expression or control plasmid and after 2 days , treated with Actinomycin D . Total RNA was collected at 0–4 hr post-treatment and mRNA levels assayed using RT-PCR . Remarkably , overexpression of UCA1 resulted in the stabilization of mature p16INK , p14ARF , E2F1 , and TGFβ1 mRNAs: in the time frame examined , p16INK , p14ARF , and E2F1 mRNAs do not decay and their t1/2 values are therefore denoted as ‘n’ ( no decay ) . The half-life estimates shown were calculated using linear regression; those best fit lines , their equations and R values are shown in Figure 6—figure supplement 1 . t1/2 of p16INK mRNA in control cells was 3 . 9 hr vs n in UCA1 overexpressing cells; p14ARF , 2 . 4 vs n; E2F1 , 7 . 2 vs n; TGFβ1 , 1 . 9 vs 2 . 9 . In marked contrast , MYC , CDKN1A-p21 , CDKN2D and RB mRNAs decayed at rates indistinguishable from control ( Figure 6A; Figure 6—figure supplement 1 ) . The effects of UCA1 overexpression on p16INK mRNA stability were confirmed by Northern blot ( Figure 6—figure supplement 2 ) . 10 . 7554/eLife . 02805 . 019Figure 6 . UCA1 stabilizes CDKN2A-p16 mRNA levels during senescence by sequestering hnRNP A1 . ( A ) Graphs of transcript levels assayed by RT-qPCR in HFFs transfected with control ( blue ) or UCA1 ( red ) expression plasmids and treated with Actinomycin ( D ) . Y axis shows % mRNA level relative to time zero and X axis shows time in hours assayed post treatment . The estimated half-lives ( t1/2 ) were obtained using linear regression; the best fit lines , their equations and R values are shown in Figure 6—figure supplement 1 . * indicates p<0 . 04 for p16INK and p<0 . 01 for all others . ( B ) Assay of mRNA levels in HFFs transfected with control or hnRNPA1 siRNA and treated with Actinomycin D . Axes and t1/2 calculations are as in panel A . * indicates p<0 . 05 . ( C–E ) Agarose gels of RT-PCR products assessing levels of CDKN2A-p16 ( p16 , panel C ) , UCA1 ( panel D ) , and negative control lncRNA TUG1 ( panel E ) transcripts in PS and RAS HFFs treated as labeled at top and subjected to RIP with anti-hnRNPA1 antibody . mIgG lanes are negative controls for RIP assays . Gels from left to right show: PS vs RAS; control vs UCA1 overexpression; control vs TBX3 or CAPERα knockdown; RAS vs RAS/UCA1 knockdown . ( C ) Lane 7 ( red arrowhead ) shows loss of p16INK /hnRNP A1 interaction in RAS . Lane 14 ( red arrowhead ) shows loss of p16INK /hnRNP A1 interaction with UCA1 overexpression . Lanes 23 and 24 show loss of p16INK /hnRNP A1 interaction after TBX3 or CAPERα knockdown . Lane 27 shows that UCA1 knockdown decreases the total amount of p16INK mRNA in RAS cells . Lane 31 shows that UCA1 knockdown increases p16INK mRNA/hnRNP A1 binding ( red arrowhead ) in RAS cells , even though there is less total p16INK ( lane 27 ) . ( F ) Panels show immunoblots to detect hnRNP A1 protein in input samples assayed in panels C–E . Lanes are numbered to correspond with panels above . TBX3 , CAPERα = human; Tbx3 , Caperα = mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 02805 . 01910 . 7554/eLife . 02805 . 020Figure 6—figure supplement 1 . Graphs showing best fit lines , their equations , and R values used to calculate estimated mRNA half-life values shown in Figure 6A . TBX3 , CAPERα = human; Tbx3 , Caperα = mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 02805 . 02010 . 7554/eLife . 02805 . 021Figure 6—figure supplement 2 . Northern blot assay of p16INK mRNA levels in the absence and presence of UCA1 . ( A ) Top panel shows Northern blot of HFF cells transfected with control plasmid pcDNA3 . 1 and treated with Actinomycin D for the times ( hr ) indicated at top . ( A′ ) The ethidum bromide stained gel prior to transfer is shown for loading control and RNA quality . ( A″ ) The signals obtained by probing for p16INK mRNA in A were subjected to densitometric quantitation . Note decrease in signal at 2 and 4 hr consistent with the decay/t1/2 obtained in Figure 6A . ( B ) Top panel shows Northern blot of HFF cells transfected with UCA1 expression plasmid and treated with Actinomycin D for the times ( hr ) indicated at top . ( B′ ) The ethidum bromide stained gel prior to transfer is shown for loading control and RNA quality . ( B″ ) The signals obtained by probing for p16INK mRNA in B were subjected to densitometric quantitation . Note that UCA1 expression results in minimal decrease in signal at 2 and 4 hr , consistent with UCA1-mediated mRNA stabilization observed in Figure 6A . TBX3 , CAPERα = human; Tbx3 , Caperα = mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 02805 . 02110 . 7554/eLife . 02805 . 022Figure 6—figure supplement 3 . Graphs showing best fit lines , their equations and R values used to calculate estimated half-life values after hnRNP A1 siRNA knockdown shown in Figure 6B . ( A ) Western blot assaying hnRNP A1 protein levels in HFFs after transfection of control or anti-hnRNP A1 siRNA . ( B ) Graphs of best fit lines , equations and R values for half-lives shown in Figure 6B . TBX3 , CAPERα = human; Tbx3 , Caperα = mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 02805 . 02210 . 7554/eLife . 02805 . 023Figure 6—figure supplement 4 . RNA Immunoprecipitation analysis of hnRNP A1 interactions with Myc and p14ARF mRNAs . RIP-PCR of MYC and CDKN2A-p14 mRNAs shows they are bound by hnRNP A1 but these interactions are unaffected by OIS/RAS , UCA1 overexpression , or knockdown of TBX3 or CAPERα . TBX3 , CAPERα = human; Tbx3 , Caperα = mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 02805 . 02310 . 7554/eLife . 02805 . 024Figure 6—figure supplement 5 . RIP-PCR of HFF lysates using antibodies listed at top . Only hnRNP A1 ( A ) and hnRNP D ( B ) bind UCA1 lncRNA , while TUG1 and H19 lncRNAs are bound by other hnRNPs . TBX3 , CAPERα = human; Tbx3 , Caperα = mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 02805 . 02410 . 7554/eLife . 02805 . 025Figure 6—figure supplement 6 . RIP-PCR indicates that RB , p21 , and CDK6 mRNAs do not interact with hnRNP A1 in PS or RAS HFFs . TBX3 , CAPERα = human; Tbx3 , Caperα = mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 02805 . 025 Regulation of p16INK transcript stability is a critical mechanism for growth control ( Wang et al . , 2005; Chang et al . , 2010; Zhang et al . , 2012 ) and hnRNP A1 has been postulated to stabilize p16INK mRNA ( Zhu et al . , 2002 ) , but this has not been tested . To this end , we treated HFFs with siRNA to hnRNP A1 and used Actinomycin D to assess stability of p16INK transcripts . Loss of hnRNP A1 ( Figure 6—figure supplement 3 ) stabilized both p16INK ( t1/2–2 . 1 in control vs 12 . 3 after HNRNP A1 knockdown ) and p14ARF mRNAs ( t1/2–1 . 5 in control vs 6 . 9 after hnRNP A1 knockdown ) but not those of E2F1 or MYC ( Figure 6B ) . Half-life estimates were obtained as described for panel A and the best fit lines , their equations and R values are shown in Figure 6—figure supplement 3B . The differences in control half-lives between Figure 6A , B are likely attributable to the different treatments used: in A , control cells were transfected with pcDNA3 . 1 plasmid , while in B , control cells were transfected with control siRNA . The half-life of an mRNA is cell/context specific ( as evident in the differences in control half-lives in 6A vs 6B ) and in general , cell cycle regulatory genes have short half-lives ( Sharova et al . , 2009 ) . The t1/2 of p16INK mRNA we observed in HFFs transfected with either control plasmid ( t1/2–3 . 9 ) or control siRNA ( t1/2–2 . 1 ) is similar to that reported in HeLa cells ( t1/2–2 . 9 ) ( Chang et al . , 2010 ) . The results we obtained were also similar to those reported for MYC mRNA ( Herrick and Ross , 1994; Sharova et al . , 2009 ) , CDKN1A mRNA in HT29-tsp53 cells ( Melanson et al . , 2011 ) and ES cells ( Sharova et al . , 2009 ) , and E2F1 mRNA in ES cells ( Sharova et al . , 2009 ) . The half- lives of Rb and TGFβ1 are mRNAs extremely variable and those we obtained in HFFs were shorter than reported in ES cells ( Sharova et al . , 2009 ) . We next used RNA-IP ( RIP ) to determine if hnRNP A1 binds p16INK and p14ARF mRNAs in proliferating cells and found that this was indeed the case ( Figure 6C , lane 6 and Figure 6—figure supplement 4 ) . Remarkably , hnRNP A1/p16INK binding was lost in RAS HFFs ( Figure 6C , lane 7 ) , despite an overall increase in the number of p16INK transcripts ( Figure 6C , lane 3 ) . As shown previously , UCA1 RNA levels also increase with RAS ( Figure 6D , lane 3 ) . UCA1 is bound by hnRNP A1 in PS cells ( Figure 6D , lanes 6 , 7; Figure 6—figure supplement 5 ) , but unlike p16INK , the hnRNP A1/UCA1 interaction increases in RAS cells ( Figure 6D , lane 7 ) . TUG1 lncRNA serves as a negative control ( Figure 6E ) . Protein levels for hnRNP A1 are shown in Figure 6F . The interaction between UCA1 and hnRNP A1 is specific , as UCA1 does not bind hnRNP K , C1/C2 , H , U , or D ( Figure 6—figure supplement 5 ) . Although hnRNP A1 binds MYC and p14ARF mRNAs ( Figure 6—figure supplement 4 ) , it does not bind RB , p21 or CDK6 mRNAs under the numerous conditions tested ( Figure 6—figure supplement 6 ) . The opposite binding properties of UCA1 and p16INK mRNA with hnRNP A1 in PS vs RAS HFFs led us to postulate that UCA1 stabilizes p16INK mRNA during OIS by disrupting the interaction between hnRNP A1 and p16INK mRNA . In control transfected proliferating cells , there is robust binding of p16INK to hnRNP A1 ( Figure 6C , lane13 ) , but direct overexpression of UCA1 ( Figure 6D , lane 10 ) or that resulting from TBX3 or CAPERα KD ( Figure 6D , lanes 17 , 18 ) disrupts the hnRNP A1/p16INK mRNA interaction ( Figure 6C , lanes14 , 23 , 24 , red arrowheads ) . These findings support the hypothesis that loss of hnRNP A1/p16INK mRNA interaction in OIS ( Figure 6C , lane 7 ) is the result of increased UCA1 expression and its binding and sequestration of hnRNP A1 ( Figure 6D , lane 7 ) . To further test this , we used shRNA to KD UCA1 in RAS HFFs ( Figure 6D , lane 27 ) . UCA1 KD restored the interaction between hnRNP A1 and p16INK mRNA ( Figure 6C , lane 31 ) and led to lower levels of total p16INK mRNA ( Figure 6C , lane 27 ) , a finding consistent with the negative effects of hnRNP A1/ p16INK interaction on stability of p16INK transcripts . The effects of UCA1 on p16INK mRNA stability are specific , because hnRNP A1 interactions with MYC or p14ARF mRNAs are unaffected by UCA1 ( Figure 6—figure supplement 1 ) . In total , these findings indicate that in proliferating cells , the very low quantity of UCA1 transcripts is insufficient to disrupt hnRNP A1/p16INK binding , and levels of p16INK mRNA are low due to: ( 1 ) direct repression by CAPERα/TBX3 and , ( 2 ) p16INK mRNA instability conferred by hnRNP A1 . When UCA1 levels increase during OIS , by UCA1 overexpression , or via KD of CAPERα/TBX3 , UCA1 binds and sequesters hnRNP A1 , preventing it from destabilizing p16INK mRNA . Increased p16 protein is required for RAS-induced senescence in MEFS and some human cell types ( Serrano et al . , 1997 ) , leading us to determine whether OIS affects CAPERα/TBX3 occupancy of p16INK chromatin . CDKN2A-p16INK genomic regulatory elements bound in PS HFFs ( Figure 4I ) were not occupied by either TBX3 or CAPERα in RAS HFFs ( Figure 7A ) . Chromatin marks on these regions switched from heterochromatic to euchromatic ( Figure 7B , Figure 7—figure supplement 1A ) . This was also observed with UCA1/A3 ( Figure 5U ) and DUSP4 chromatin ( Figure 7—figure supplement 1B ) . 10 . 7554/eLife . 02805 . 026Figure 7 . Disruption of the CAPERα/TBX3 repressor by OIS activates CDKN2A-p16 and UCA1 to trigger a senescence transcriptional response . ( A ) ChIP-PCR of regions upstream of the CDKN2A-p16 transcriptional start site ( position relative to TSS in parentheses ) in PS and RAS HFFs; the −-3706–3308 amplicon is a negative control . OIS disrupts binding of p16 regulatory elements ( initially identified in Figure 3O ) by TBX3 and CAPERα . ( B ) ChIP-PCR of p16 -4855 element shown in A . Decreased TBX3 and CAPERα binding in RAS correlates with loss of repressive chromatin marks and gain of activating marks . Evaluation of chromatin marks on the other CDKN2A-p16 CAPERα/TBX3- responsive regulatory elements is shown in Figure 7—figure supplement 1A . ( C ) IBs for TBX3 , CAPERα , and actin loading control show increased amount of both proteins in RAS compared to PS HFFs . ( D ) Anti-TBX3 and anti-CAPERα IBs of IP'd proteins from PS and RAS HFFs . ( F–M ) Immunocytochemical staining of PS ( F , G , J , K ) and RAS ( H , I , L , M ) HFFS for TBX3 ( F and H ) , Hoechst ( DNA; G and I ) , CAPERα ( J and L ) . Panels K and M are merged Hoechst/CAPERα . Scale bar for all panels is sown at lower right of panel I . ( N–O′ ) Functional analyses of genome wide transcriptional profiles of TBX3 KD , CAPERα KD , and control HFFs . All comparisons were statistically significant with p values <<<<0 . 0001; see Figure 7—source data 3 for hypergeometric test , as implemented in the R statistical language , used to test significance of the number of genes found to be co-regulated between samples . ( N ) Venn diagrams show highly significant number of CAPERα/TBX3 co-upregulated transcripts ( 446 total ) , especially in the GO biologic process ( BP ) category of transcriptional regulation ( 122 transcripts ) as assayed with DAVID . Pie chart shows KEGG pathway analysis of co-regulated genes . ( N′ ) Venn diagram showing 48 CAPERα/TBX3 co-upregulated transcripts also upregulated by RAS/OIS ( Loayza-Puch et al . , 2013 ) , especially in BP categories of transcriptional regulation and programmed cell ( pc ) death . qPCR validation of coregulated genes is in S . Figure 6A . Pie chart shows KEGG pathway analysis of OIS dataset . ( O and O′ ) As in N and N′ but for downregulated genes . Pie chart in O′ shows KEGG pathway analysis of OIS data set; note most pathways are the same as in TBX3/CAPERα . ( P and Q ) Models of CAPERα/TBX3 repressor and UCA1 function in proliferating ( PS ) HFFs vs RAS HFFs . In PS cell nuclei , CAPERα/TBX3 represses UCA1 , p16 , p14 , and DUSP4 promoters in heterochromatin which permits ongoing cell proliferation . RAS disrupts the CAPERα/TBX3 complex and CAPERα relocates to dense intranuclear foci . Pro-senescence genes including UCA1 and p16 are converted to euchromatin and their expression/products induce senescence . In the cytoplasm of PS cells , hnRNP A1 binds and destabilizes p16 mRNA , but activation of UCA1 expression in OIS allows UCA1 to sequester hnRNP A1 and stabilize p16 mRNA . TBX3 , CAPERα = human; Tbx3 , Caperα = mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 02805 . 02610 . 7554/eLife . 02805 . 027Figure 7—source data 1 . Differentially expressed genes after knockdown of CAPERα in HFFs detected by RNA-Seq . DOI: http://dx . doi . org/10 . 7554/eLife . 02805 . 02710 . 7554/eLife . 02805 . 028Figure 7—source data 2 . Differentially expressed genes after knockdown of TBX3 in HFFs detected by RNA-Seq . DOI: http://dx . doi . org/10 . 7554/eLife . 02805 . 02810 . 7554/eLife . 02805 . 029Figure 7—source data 3 . Determining the statistical significance of shared differentially expressed genes using the hypergeometric test , as implemented in the R statistical language ( phyper ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02805 . 02910 . 7554/eLife . 02805 . 030Figure 7—figure supplement 1 . Repression of CDKN2A-p16 and DUSP4 by CAPERα /TBX3 correlates with chromatin architecture and is relieved during oncogene induced senescence . ( A ) ChIP-PCR to assess chromatin marks on CDKN2A-p16 regulatory elements in PS and RAS HFFs; antibodies are listed at top . ( B ) ChIP-PCR of DUSP4 promoter in PS and RAS HFFs; antibodies are listed at top . TBX3 and CAPERα bind the DUSP4 promoter in PS ( lanes 6 , 8 ) but not RAS HFFs ( lanes 7 , 9 ) , and their occupancy correlates with altered chromatin marks consistent with de-repression in OIS/RAS cells ( lanes 10–15 ) . TBX3 , CAPERα = human; Tbx3 , Caperα = mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 02805 . 03010 . 7554/eLife . 02805 . 031Figure 7—figure supplement 2 . CAPERα relocalization due to oncogene-induced senescence is independent of PML bodies . Immunocytochemical assay for endogenous CAPERα ( green ) , PML ( red ) , and DNA ( DAPI , blue ) in PS and RAS HFFs . TBX3 , CAPERα = human; Tbx3 , Caperα = mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 02805 . 03110 . 7554/eLife . 02805 . 032Figure 7—figure supplement 3 . Validation of RNA-Seq identified expression changes induced by CAPERα and TBX3 KD . qPCR validation of a subset of transcripts with altered expression detected by genome wide RNA-Seq on cDNA prepared from CAPERα ( red ) and TBX3 ( blue ) KD , and RAS HFFs ( green ) . Downregulated transcripts are listed at left , upregulated at right . TBX3 , CAPERα = human; Tbx3 , Caperα = mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 02805 . 03210 . 7554/eLife . 02805 . 033Figure 7—figure supplement 4 . IL6 and HDAC9 are direct targets of CAPERα/TBX3 . ChIP-PCR with antibodies listed at top showing CAPERα/TBX3 directly binds IL6 ( and HDAC9 ) control elements . Effects of TBX3 or CAPER KD on chromatin marks are shown compared with control KD . ChIP-PCR examining CAPERα/TBX3 binding to IL6 and HDAC9 control elements in PS and RAS HFFs; loss of binding correlates with altered chromatin marks . TBX3 , CAPERα = human; Tbx3 , Caperα = mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 02805 . 033 We investigated the possibility that altered quantity of either CAPERα or TBX3 could disrupt the stoichiometry of their interaction and cause dissociation from p16INK and UCA1 regulatory elements in OIS . Surprisingly , both TBX3 and CAPERα protein levels were increased in RAS HFFs ( Figure 7C ) , but they no longer co-IP'd ( Figure 7D , red box ) . Immunocytochemistry of endogenous TBX3 and CAPERα in PS and RAS HFFs confirmed increased protein levels in OIS ( Figure 7F–M ) , and revealed dramatic changes in CAPERα localization: CAPERα immunoreactivity became concentrated in large intranuclear foci ( Figure 7L , M ) , as we previously observed in early passage Tbx3−/− MEFS ( Figure 2—figure supplement 2D–F′ ) . These foci are distinct from SAHFs and PML bodies ( Figure 7M and Figure 7—figure supplement 2 ) . To further investigate the molecular basis of senescence initiation after loss of CAPERα/TBX3 , we performed genome-wide transcriptional profiling 2 days post CAPERα , TBX3 and control KD in HFFs . More than half of the transcripts with expression altered 1 . 5-fold or more by CAPERα KD were similarly affected by loss of TBX3 ( N = 2375 CAPERα KD , 2188 TBX3 KD; 1157 co-regulated , p<<<<0 . 0001 , Figure 7—source data 1–3 , Figure 7N , O ) . Gene ontology-biologic process ( GO-BP ) analysis with DAVID ( Huang da et al . , 2009a , 2009b ) showed highly significant co-regulation of ‘transcription regulation’ ( increased expression ) and ‘cell-cycle’ ( decreased expression ) transcripts ( Figure 7N , O ) . We tested a subset of these with known roles in senescence by qPCR: 100% validated and were similarly altered by RAS ( Figure 7—figure supplement 3 ) . Further interrogation of this group revealed that IL6 and HDAC9 are CAPERα/TBX3 direct targets and their upregulation in RAS is associated with loss of CAPERα/TBX3 binding ( Figure 7—figure supplement 4 ) . We compared CAPERα/TBX3 co-regulated transcripts to a published data set comparing PS and G12VRAS fibroblasts ( Loayza-Puch et al . , 2013 ) . This revealed that 11% of CAPERα/TBX3 up-regulated transcripts were also increased by RAS ( Figure 7N′ ) ; among these , GO-BP ‘programmed cell death’ ( 31% ) and ‘transcription regulation’ ( 34% ) were highly overrepresented . 30% of CAPERα/TBX3 down-regulated transcripts were also in the RAS data set; >1/3 of these were cell cycle genes ( Figure 7O′ ) . In all comparisons , the number of transcripts common to both groups was greater than predicted by chance and highly statistically significant ( Figure 7—source data 3 ) . KEGG pathway analyses revealed overrepresented pathways that were common to both CAPERα/TBX3 and RAS data sets ( Figure 7N–O′ , pie charts ) , but notably fewer pathways were shared in the upregulated group: JAK/STAT , TLR and TGFβ signaling pathways were only significantly overrepresented in the CAPERα/TBX3 data set . Our knowledge of the regulatory mechanisms that govern the onset and maintenance of senescence in different contexts must be considered fragmentary ( Wang and Chang , 2011; Fatica and Bozzoni , 2014 ) . In this study , we provide compelling evidence for critical and novel functions of CAPERα , the lncRNA UCA1 and TBX3 in the regulation of cell proliferation and senescence . We have discovered a CAPERα/TBX3 complex that is required to prevent senescence of primary human and mouse cells in vivo and that functions as a master regulator of cell proliferation by directly repressing transcription of lncRNA UCA1 , p16INK and other tumor suppressor genes ( Figure 7P ) . Overexpression of UCA1 occurs after loss of TBX3/CAPERα and in OIS ( Figure 7Q ) , and is itself sufficient to induce senescence at least in part , by disrupting the interaction of p16INK mRNA with hnRNP A1 leading to increased p16INK mRNA stability ( Figure 7P , Q ) . Disrupting the CAPERα/TBX3 complex by decreasing the amount of either TBX3 or CAPERα , or by CAPERα mislocalization during OIS , coordinately increases activity of multiple pro-senescence targets at both the transcriptional and post-transcriptional levels in a reinforcing mechanism . Increased CAPERα has been reported in human breast cancers and a shift from cytoplasmic to nuclear localization correlates with transition from pre-malignant to malignant lesions ( Mercier et al . , 2009 ) . In contrast , CAPERα co-activates vRel mediated transcription but inhibits vREL transforming activity in vitro ( Dutta et al . , 2008 ) . It is likely that anti- or pro- oncogenic activity of CAPERα is determined by cell type and the interacting protein ( s ) present in a given context; our results suggest that CAPERα has oncogenic potential in primary cells since loss of CAPERα/TBX3 induces premature senescence , a vital tumor suppressor mechanism . CAPERα binds to regulatory chromatin domains via TBX3 but dissociates from these domains and becomes concentrated in large intranuclear foci prior to senescence induced by loss of TBX3 or during OIS . Future efforts will define the composition of CAPER + nuclear foci and the role of this nuclear subdomain during senescence induction . The TBX3 RD is required for TBX3 to interact with CAPERα ( this study ) , immortalize primary fibroblasts and confer senescence bypass ( Carlson et al . , 2001 ) . Since loss of CAPERα activates target gene transcription despite continued TBX3 occupancy , it is the CAPERα/TBX3 complex ( interacting via TBX3 RD ) that represses pro-senescence target loci . It will be important to determine if previously identified targets of TBX3 transcriptional repression are actually regulated by this complex . Additional studies are warranted to determine the precise mechanisms whereby histone status is regulated by CAPERα/TBX3: TBX3 is known to interact directly with HDACs ( Yarosh et al . , 2008 ) , but there are no reports of it or CAPERα interacting with histone methyltransferases or demethylases . Our recently published Mass Spec screen for Tbx3/TBX3 interactors did not identify such factors however , the screen cannot be considered exhaustive as we did not reproducibly detect HDACs or transcription factors previously reported to interact with Tbx3 . Future studies to specifically determine whether TBX3 and/or CAPERα interact with , recruit , or modify the function of EZH2 , SUV39 and other methyltransferases will be informative . Previous studies showed that TBX3 represses transcription of p14ARF ( upstream of p53 ) ( Bamshad et al . , 1997; Fan et al . , 2009; Kumar et al . , 2014 ) , yet embryonic lethality and mammary phenotypes of Tbx3 mutants are p53-independent ( Jerome-Majewska et al . , 2005 ) . Our findings reconcile these observations because CAPERα/TBX3 represses p16INK , the p16/RB pathway is activated in Tbx3−/− embryos , and knockdown of either RB or p16 ( but not p53 ) prevents senescence after loss of CAPERα/TBX3 . Furthermore , Tbx3−/− and Cdk2−/−;Cdk4−/− mutant embryos share multiple phenotypes including RB hypo-phosphorylation , reduced E2F-target gene expression , decreased proliferation and premature senescence of MEFs ( Berthet et al . , 2012; Frank et al . , 2012 , 2013 ) . Our discoveries of multiple CAPERα/TBX3 binding sites across the CDKN2A locus , and altered chromatin marks after TBX3 and CAPERα KD , indicate that the complex directly represses transcription by regulating chromatin structure . In total , the data conclusively demonstrate that p16 elevation , CDK2 and CDK4 downregulation , and RB hypophosphorylation mediate senescence downstream of CAPERα/TBX3 loss of function in primary human cells and Tbx3 null mutant embryos . When combined with the pleiotropic effects of CAPERα/TBX3 on UCA1 , DUSP4 , IL6 , HDAC9 and other pathways , it is clear why loss of this repressor induces senescence . TBX3 may function in nuclear organization and structure: severe changes in nuclear morphology and mislocalization of both CAPERα and laminβ1 are apparent in Tbx3−/− MEFs after just one passage , prior to other signs of senescence . Progeria is a rare disease in which LMNA mutations induce cellular and organismal senescence in part by altering stoichiometry and interactions of type A and B Lamins . Progeria fibroblasts have decreased expression of TBX3 , TBX3 interacting proteins , and TBX3 targets ( Csoka et al . , 2004 ) . LMNβ1 is a TBX3 interacting protein ( Kumar et al . , 2014 ) and expression of LMNA , LMNβ1 , and LMNβ2 is disrupted by TBX3/CAPERα KD ( Figure 7—source data 1–3 and Figure 7—figure supplement 3 ) . TBX3 may regulate LMN gene expression and physically interact with Lamins to influence nuclear homeostasis . There are many downregulated genes common to the senescence responses triggered by RASG12V and loss of CAPERα/TBX3 however , upregulated transcripts and pathways are largely distinct ( Figure 7N′ ) . This is likely attributable to the presence of direct targets of CAPERα/TBX3 repression in the upregulated data set . It will be informative to determine which Jak-STAT , TLR , and TGFβ pathway members ( Figure 7N ) are direct CAPERα/TBX3 targets , as the complex roles of these pathways in the senescence associated secretory phenotype , inducing or enforcing autocrine and paracrine senescence , and tumor progression are emerging ( Hubackova et al . , 2010; Senturk et al . , 2010; Hubackova et al . , 2012; Davalos et al . , 2013 ) . Recent discoveries of the pervasive functions of lncRNAs as ‘signals , decoys , guides and scaffolds’ ( Wang and Chang , 2011 ) , conferred by their ability to interact with other nucleic acids and as protein ligands , has added new layers of complexity to regulation of transcriptional and post-transcriptional gene expression and translation . Although there has been a logarithmic increase in studies exploring lncRNA expression and activity , potential senescence-regulating activities are still largely unexplored . LncRNA HOTAIR functions as a scaffold to regulate ubiquitination of Ataxin-1 and Snurportin-1 to prevent premature senescence ( Yoon et al . , 2013 ) . Global alterations in lncRNA expression have been reported in association with replicative senescence ( Abdelmohsen et al . , 2013 ) , and telomere-specific lncRNAs that regulate telomere function during this process have been identified ( Yu et al . , 2014 ) . As this manuscript was in revision , regulation of H4K20 trimethylation of rRNA genes by interaction of quiescence-induced lncRNAs PAPAS and Suv4-20h2 was reported ( Bierhoff et al . , 2014 ) . To our knowledge , UCA1 is the first lncRNA sufficient to induce senescence . UCA1 is expressed in bladder transitional cell carcinomas ( Wang et al . , 2006 ) and influences tumorigenic potential of bladder cancer cell lines ( Wang et al . , 2008; Yang et al . , 2012 ) . A very recent study identified hnRNP I as a UCA1 interacting protein that stabilizes UCA1 RNA; this interaction was postulated to decrease translation of p27 to support growth of the MCF7 breast cancer line ( Huang et al . , 2014 ) . In contrast , our results support a tumor suppressor/prosenescence function for UCA1 in primary cells . UCA1 increases stability of p16INK mRNA by sequestering hnRNP A1 , employing a decoy mechanism that is in some aspects reminiscent of lncRNA PANDA sequestering NF-YA transcription factor to prevent activation of proapoptotic p53 targets and promote cell cycle arrest in the DNA damage response ( Wang and Chang , 2011 ) . In the case of UCA1 and hnRNP A1 however , the sequestration has a very specific effect: even though UCA1 expression stabilizes ( and hnRNP A1 destabilizes ) both p16INK and p14ARF mRNAs ( Figure 6A , B ) , UCA1 only disrupts the association of hnRNP A1 with p16INK mRNA ( Figure 6C and Figure 6—figure supplement 4 ) . In proliferating cells , abundant hnRNP A1 binds with p16INK mRNA resulting in p16INK degradation . In senescing cells , p16INK mRNA levels increase via reinforcing mechanisms of increased transcription and stability: loss of CAPERα/TBX3 activates transcription of p16INK and UCA1 , in turn , UCA1 sequesters hnRNPA1 . We recognize that the systems we employed ( primary HFFs , mouse embryos and MEFs ) , while very informative models , provide limited information directly applicable to aging or tumorigenesis without further experimentation . Our data support an important role for CAPERα/TBX3 in regulation of senescence in developmental contexts and , since the CAPERα/TBX3 complex regulates known critical tumor suppressors and there is an increasing literature supporting roles for both TBX3 and CAPERα in tumor biology , this is another worthy area for future investigation . As noted above , expression of CDKN2A-p14ARF and CDKN1A-p21CIP are repressed by TBX2 and TBX3 and this is postulated to confer the ability of overexpressed TBX2 and TBX3 to permit senescence bypass of Bmi1−/− and SV40 transformed mouse embryonic fibroblasts , respectively ( Jacobs et al . , 2000; Brummelkamp et al . , 2002; Prince et al . , 2004 ) . Numerous overexpression studies have suggested a role for TBX3 in breast cancer ( ( Liu et al . , 2011 ) and references therein ) and recent papers have reported the tumorigenic and proinvasive effects of overexpressed TBX3 in melanoma cells ( Peres et al . , 2010; Peres and Prince , 2013 ) which may derive in part from TBX3 repression of E-cadherin expression ( Rodriguez et al . , 2008 ) . More relevant to our work on the importance of the CAPERα/TBX3 complex to prevent senescence and regulate cell proliferation are reports that Tbx3 improves the pluripotency of iPS cells ( Han et al . , 2010 ) and prevents differentiation of mouse ES cells ( Ivanova et al . , 2006 ) . In conclusion , CAPERα/TBX3 acts as a master regulator of cell growth and fate , exerting pleotropic effects by at least two modes of action: ( 1 ) regulating chromatin structure and transcription of both coding and non-coding genes and , ( 2 ) modulating mRNA stability by altering the association of RNA binding proteins with target transcripts via UCA1 . Further exploration will identify tissue-specific UCA1 targets and binding proteins , and determine whether the ability of TBX3 to confer senescence bypass in other contexts requires CAPERα interaction and/or UCA1 repression . Mining the pathways regulated by UCA1 and CAPERα/TBX3 will reveal factors that control cell proliferation and fate during development and disease and thus constitute novel cancer therapeutic targets . Mass spectroscopy as in Kumar et al . , ( 2014 ) Dignam lysates were prepared and incubated for 4 hr at 4°C with the appropriate antibody followed by 2 hr at 4°C with the pre equilibrated Dynabeads Protein G ( Invitrogen ) . Immune complexes were collected and washed three times with lysis buffer . Pelleted beads were resuspended in 6X Laemmli buffer and subjected to SDS-PAGE analysis followed by immunoblotting with specific antibodies . Input lanes contain 5% of protein lysate used for IP; the rest was used in the IP and of the IP'd material , 25% was loaded onto the gel for immunoblotting . Tbx3 ( Frank et al . , 2012 , 2013 ) , TBX3 ( SC-17871 , MAB10089 , A303-098A ) , CAPERα ( A300-291A ) , GST ( SC-33613 ) , LaminB1 ( SC-56144 ) , C-Myc ( SC-40 ) , R-IgG ( SC-2027 ) , m-IgG ( SC-2025 ) , Anti-Flag ( Sigma , F3165 ) , H3K9me3 ( Cell Signaling , 9754 ) , H3K4me3 ( Cell Signaling , 9751 ) , H3K27me3 ( Cell Signaling , 9733 ) , H3K9ace ( Cell Signaling , 9649 ) , H4K5ace ( Cell Signaling , 9672 ) , H3K14ace ( Cell Signaling , 4353 ) , p-RB -Ser 810--811 ( SC-16670 ) , p-RB -Ser 795 ( SC-7986 ) , p-RB -Ser 780 ( SC-12901 ) , Rb1 ( SC-73598 ) , H3S10P ( SC-8656 ) , H2A K119ub ( 8240S ) , p21 ( SC-756 ) , p53 ( Invitrogen 134100 ) , Cyclin D1 ( SC-753 ) , Cyclin D2 ( SC-754 ) , Cyclin D3 ( SC-755 ) , Cyclin E ( SC-20648 ) , CDK2 ( SC-6248 ) , CDK4 ( SC-601 ) CDK6 ( SC-177 ) , hnRNP K ( SC-53620 ) , hnRNP C1/C2 ( SC-32308 ) , hnRNP H ( SC-10042 ) , hnRNP U ( SC-32315 ) , hnRNP A2/B1 ( SC-53531 ) , hn RNP A1 ( SC-32301 ) , and hnRNP D1 ( AB-61193 ) . Amylose bound MBP and MBP-tagged TBX3 affinity columns were prepared as per the procedure ( E8022S , NEB ) described in the manufacturer's protocol . These beads were incubated with 5 and 10 μg of GST and GST-CAPER at 4°C for 8 hr . Bound proteins were eluted with reduced glutathione and analyzed by Western blotting with anti-CAPER antibody . Transfections were performed in HEK293 or EBNA-293 cells with Lipofectamine 2000 ( Invitrogen ) or in Human fibroblasts with X-tremeGENE HP DNA transfection Reagent ( Roche ) as per the manufacturer's recommendations . Wild-type Tbx3 and exon 7 missense , deleted repressor domain ( Tbx3ΔRD1 ) , and Tbx3ΔNLS were generated by PCR amplification and cloned into pcDNA3 . 1 . C-terminal deletion constructs Tbx3 1-655 , Tbx3 1-623 , Tbx3 1-565 , Tbx3 1-470 were generated by PCR amplification and cloned into pCS2 with an N-terminal Myc tag . Tbx3 L143P and N277D point mutants were kind gifts of Phil Barnett . UCA1 and CAPERα cDNAs were cloned into pCDN3 . 1 and PQCXIH for over- expression studies , respectively . Sequence of all plasmids was confirmed . Tbx3 L143P and N277D point mutants plasmids were kind gifts of Phil Barnett . Wild-type CAPERα was generated by PCR amplification and then cloned into pQCXIH retroviral vector; sequence was confirmed . Full length UCA1 was amplified by PCR and then cloned into pcDNA3 . 1 vector; sequence was confirmed . UCA1 Cloning FP: AGTTGCGGCCGCTGACATTCTTCTGGACAATGAGUCA1 Cloning RP: TCCTGCGGCCGCTTGGCATATTAGCTTTAATGTAGCAPERα Cloning FP: CATCGCGGCCGCATGGCAGACGATATTGATATTGCAPERα Cloning RP: ACGTGGATCCTCATCGTCTACTTGGAACCAGTAG E10 . 5 embryos were harvested in PBS followed by overnight fixation at 4°C in 4% paraformaldehyde and processed for 7 μm cryosections . For cell lines , human fibroblasts were cultured on 8-well chamber slides ( BD Flacon ) and processed for Immunohistochemistry . Immunohistochemistry was performed using primary antibodies listed above and detected using donkey anti-goat or anti-rabbit Alexa fluor 594 ( 1:500 ) and goat anti-mouse Alexa fluor 488 ( 1:500 ) from Invitrogen . Nuclei were stained with Hoechst or DAPI . Slides were imaged with a Nikon ARI inverted confocal microscope at the University of Utah Imaging Core . shRNA oligonucleotides ( see sequences below ) were annealed and cloned into the pGFP-B-RS , pRFP-C-RS ( Origen ) vector and PMK0 . 1 vector . shRNA against luciferase served as a negative control . High-titer retrovirus was produced by transfection of shRNA retroviral construct along with gag/pol and VSVG encoding plasmids into EBNA-293 cells by lipofectamine 2000 reagent as per the manufacturer's protocol . Virus containing supernatant was collected after 48 hr of transfection and filtered through 0 . 45-μM filters ( Fisher 09-720-4 ) . HEK293 or HFFs were incubated with DMEM containing polybrene ( 8 mM ) and 500 μl of TBX3 or CAPERα shRNA encoding retrovirus . 24 hr post infection , cells were split to lower densities and blasticidin or puromycin antibiotic selection applied for 2 days . Stably integrated colonies were selected and analyzed for knock down efficiency by western analysis using Tbx3 or CAPERα antibody . TBX3 shRNA A: targets TBX3 exon 7TBX3 shA FP: CCGG GACCATGGAGCCCGAAGAA ttcaagaga TTCTTCGGGCTCCATGGTC TTTTTGTBX3 shA RP: AATTCAAAAA GACCATGGAGCCCGAAGAA tctcttgaa TTCTTCGGGCTCCATGGTCTBX3 shRNA B: targets TBX3 exon 5TBX3 shB FP: CCGG CAGCTCACCCTGCAGTCCA ttcaagaga TGGACTGCAGGGTGAGCTG TTTTTGTBX3 shB RP: AATTCAAAAA CAGCTCACCCTGCAGTCCA tctcttgaa TGGACTGCAGGGTGAGCTGCAPERα shRNA A: targets CAPERα ( gene name RBM39 ) exon 5CAPERα shA FP: CCGG GACAGAAATTCAAGACGTTttcaagagaAACGTCTTGAATTTCTGTCTTTTTGCAPER shA RP: AATTCAAAAA GACAGAAATTCAAGACGTT tctcttgaa AACGTCTTGAATTTCTGTCCAPERα shRNA B: targets CAPERα exon 1CAPER shB P:CCGG AAAGCAAGAGCAGAAGTCGTAttcaagagaTACGACTTCTGCTCTTGCTTT TTTTTGCAPER shB RP: AATTCAAAAA AAAGCAAGAGCAGAAGTCGTA tctcttgaa TACGACTTCTGCTCTTGCTTT The pMKo . 1 puro RB and pMKo . 1 puro p53 shRNA vectors were a kind gift of William Hahn obtained via Addgene . pRB shRNA: Addgene #10670p53 shRNA: Addgene #10672p16 shRNA: Addgene #22271 Efficacy and specificity of the pRb , p53 , and p16 shRNAs was validated with second shRNAs , and these reagents have been used extensively by many investigators in the years since their initial publication ( Masutomi et al . , 2003; Stewart et al . , 2003; Boehm et al . , 2005; Haga et al . , 2007; Hong et al . , 2009; Elzi et al . , 2012 ) . UCA1 shRNA: targets UCA1 exon 3UCA1 shA FP: GATCCGTTAATCCAGGAGACAAAGAtcaagagTCTTTGTCTCCTGGATTAACTTTTTTGGAUCA1 shA RP: AGCTTCCAAAAAAGTTAATCCAGGAGACAAAGActcttgaTCTTTGTCTCCTGGATTAACGSenescence associated β-galactosidase assayPerformed as per the manufacturer's protocol ( 9860 , Cell Signaling ) . Population doubling assay/3T5 growth curves ( Figure 2E , F , R ) Primary HFFs were plated in a 10-cm dish and transduced with retrovirus . After 24 hr , cells were cultured with antibiotic selection ( puromycin or blasticidin ) for an additional 24–72 hr . On day 0 of the 3T5 growth curve , cells were trypsinized , counted and 500 , 000 cells were then plated per 10-cm dish . This procedure was repeated every 3 days for 15 days . Population doublings were calculated by ( logN1/log2 ) − ( logN0/log2 ) N1 = current cell count , N0 = Initial cell count . Curves shown in Figure 2 are representative of two independent experiments . Primary HFFs were plated in 6-well dishes and transfected at 70% confluence . At days noted in the figure , cells were trypsinized and counted using a hemocytometer . 5 × 105 cells were plated per well in 6-well tissue culture plates . At times indicated , medium was removed and cells were washed with PBS , and fixed for 10 min in 10% formalin solution . Cells were rinsed 5X with distilled water , and then stained with 100 μl 0 . 1% crystal violet solution for 30 min , rinsed 5X in water and dried . Cell-associated crystal violet dye was extracted with 500 μl of 10% acetic acid . Aliquots were collected and optical density at 590 nm measured . Each point on the curve shown represents three independent plates . Primary HFFs were incubated with TBX3 or CAPERα or Control shRNA encoding retrovirus medium with fresh virus added every 8 hr for 48 hr , followed by antibiotic selection for 6 days . 6 days after selection , floating cells were discarded and adherent cells were utilized for senescence associated β-gal assay or preparation of RNA . Total RNA was prepared using the RNeasy RNA isolation kit ( Qiagen ) or NucleoSpin RNA II Kit ( Clontech ) and cDNA was synthesized by cDNA EcoDry Premix Double Primed ( Clontech ) kit . Q-RT-PCR was performed with SoFast Evagreen Supermix ( Bio-Rad ) as per manufacturer's protocol . TBX3: TGAGGCCTTTGAAGACCATG , TCAGCAGCTATAATGTCCATCCAPERα: CGGAACAGGCGTTTAGAGAA , TGGCACTGCTCAACTTGTTCCDK2: GCTTTCTGCCATTCTCATCG , GTCCCCAGAGTCCGAAAGATCDK4: ACGGGTGTAAGTGCCATCTG , TGGTGTCGGTGCCTATGGGAP21: TCAGAGGAGGCGCCATGT , TGTCCACTGGGCCGAAGACDC2: GGGGATTCAGAAATTGATCA , TGTCAGAAAGCTACATCTTCMDM2: ACCTCACAGATTCCAGCTTCG , TTTCATAGTATAAGTGTCTTTTTMAPK14: TTCTGTTGATCCCACTTCACTGT , ACACACATGCACACACACTAACCDKN2C: CAATGGCTCAGTTTTGCTGAATAA , GTAAGATCTGCCTGCCAAAAGCCDKN2B: AACGGAGTCAACCGTTTCGG , TGTGCGCAGGTACCCTGCAP16: CAACGCACCGAATAGTTACG , AGCACCACCAGCGTGTCSerpinE1:CCGGAACAGCCTGAAGAAGTG , GTGTTTCAGCAGGTGGCGCP14ARF: CCCTCGTGCTGATGCTACTG , ACCTGGTCTTCTAGGAAGCGGMCM3: CCTTTCCCTCCAGCTCTGTC , CTCCTGGATGGTGATGGTCTTGFb: AAGGACCTCGGCTGGAAGTG , CCCGGGTTATGCTGGTTGTAEGR1: CCAGGAGCGATGAACGCAAGCGGCATACCAAG , GGAGTACGTGGTGGCCACCGACGGGGACCCE2F1: ATGTTTTCCTGTGCCCTGAG , ATCTGTGGTGAGGGATGAGGE2F2: GGCCAAGAACAACATCCAGT , TGTCCTCAGTCAGGTGCTTGIL6R: CATTGCCATTGTTCTGAGGTTC , AGTAGTCTGTATTGCTGATGTCGSK3b: ACTCCACCGGAGGCAATTG , GCACAAGCTTCCAGTGGTGTTUCA1:GAAATGGACAACAGTACACGCATATGGGGC , CCTGTTGCTAAGCCGATGATACATTACCCTHPRT: GCTGGTGAAAAGGACCTCT , CACAGGACTAGAACACCTGCPCNA: AAGAGAGTGGAGTGGCTTTTG , TGTCGATAAAGAGGAGGAAGCCHK2: CTTATGTGGAACCCCCACCTAC , CAGCACGGTTATACCCAGCAPMAIP1: GTTTTTGCCGAAGATTACCG , CAATGTGCTGAGTTGGCACTMYC: CTCCCTCCACTCGGAAGGA , GCATTTTCGGTTGTTGCTGATCDKN2D: CAACCGCTTCGGCAAGAC , CAGGGTGTCCAGGAATCCAP53: CCTCACCATCATCACACTGG , TCTGAGTCAGGCCCTTCTGTRB: TGTGAACATCGAATCATGGAA , TCAGTTGGTGGTTCTCGGTCCXCL10: GAAATTATTCCTGCAAGCCAATTT , TCACCCTTCTTTTTCATGTAGCAIFNB1: GAATGGGAGGCTTGAATACTGCCT , TAGCAAAGATGTTCTGGAGCATCTCATF3: GTTTGAGGATTTTGCTAACCTGAC , AGCTGCAATCTTATTTCTTTCTCGTDUSP2: GGCCTTTGACTTCGTTAAGC , CCACCTCAGTGACACAGCACCREB5: CGTGCCTCCTTGAAACAAGCCATT , ATGAAACACCAGCACCTGCCTAGAHDAC9: AGTGTGAGACGCAGACGCTTAG , TTTGCTGTCGCATTTGTTCTTTSP140: TGGGTCAGTTTCTTGTTTATCTGC , AGCAGGCTAGAAGCAAGCTCEGR2: TTGGTGCCTTGTGTGATGTAGAC , CTTTCCATAAGGCAACCCATTTHMGA2: GTCCCTCTAAAGCAGCTCAAAA , CTCCCTTCAAAAGATCCAACTGBIRC5: CATGGTAGGTGCAGGTGATG , CATGGTAGGTGCAGGTGATGASF1: GGTTCGAGATCAGCTTCGAG , CATGGTAGGTGCAGGTGATGWDR66: CCGAGAAGCAACAGGAGAAA , CTGTGTCTCCAAACGGATCACDC25C: GACACCCAGAAGAGAATAATCATC , CGACACCTCAGCAACTCAGCENPF: CGAAGAACAACCATGGCAACTCG , TTCTCGGAGGATGGTGCCTGAATLAMA2: AATTTACCTCCGCTCGCTAT , CCTCCAATGTACTTTCCACGLMNB1: AAGCAGCTGGAGTGGTTGTT , TTGGATGCTCTTGGGGTTCLMNB2: GCTCTGACCAGAACGACAAGG , CCAGCATCTTCCGGAACTTGCDC20: TCCAAGGTTCAGACCACTCC , GATCCAGGCCACAGACCATADUSP5: GCTCGCTCAACGTCAACCTCAACTCGGTG , AGTGGCGGCTGCCCTGGTCCAGCACCACCDUSP4: CCTGGCAGCCATCCCACCCCCGGTTCCCC , GCTGATGCCCAGGGCGTCCAGCATGTCTCTCmTbx3: TGAGGCCTCTGAAGACCATG , TCAGCAGCTATAATGTCCATCmSerpinE1: AGCCAACAAGAGCCAATCAC , GGATTCTCGGAGGGGTAAAGmIL6: GATGGATGCTACCAAACTGGA , CCAGGTAGCTATGGTACTCCAGAAmP21: TCCACAGCGATATCCAGACA , GGCACACTTTGCTCCTGTGmCdc2: CTGCAATTCGGGAAATCTCT , TCCATGGACAGGAACTCAAAmReprimo: CTTACGGACCTGGGACTTTG , CCAGCACTGAATTCATCACG All steps were performed under aseptic conditions . Pregnant female mice were euthanized and 13 . 5-day-old embryos were isolated from the uterus . Embryos were washed in sterile PBS in 60-mm tissue culture dish at room temperature and transferred into 15-ml sterile falcon tube containing 1 ml of 50% trypsin in DMEM medium . Embryos were minced using fine scissors followed by gentle pipetting with 1 ml pipette tips and dispersed into cell suspensions in 5 min . Suspensions were plated into 10-cm plates in 10 ml of DMEM with 5% FBS and penicillin/streptomycin and incubated for 8 hr in CO2 incubator . Culture medium was replaced with fresh medium every day for 3 days . Passage 0 refers to the stage when cell suspension from the embryos was put into cell culture and subsequent passages are numbered . Performed as per the manufacturer's protocol ( 9003S , Cell Signaling ) . For differential display ( Figure 4 ) , HEK293 cells were transfected with control siRNAs ( Sense; 5′-CAGCGACUAAACACAUCA-3′ Antisense; 5′-UUGAUGUGUUUAGUCGCUGTT-3′ ) or TBX3 specific siRNA A ( Sense: GACCAUGGAGCCCGAAGAA , Antisense: UUCUUCGGGCUCCAUGGU ) or CAPERα-specific siRNA ( Sense: GACAGAAAUUCAAGACGUU , Antisense: AACGUCUUGAAUUUCUGUC ) using lipofectamine 2000 ( Invitrogen ) or X-treme GENE HP DNA transfection reagent as per manufacturer's instructions . HNRNP A1 siRNA for knockdown in HFFs ( Figure 6 ) was obtained from Cell Signaling ( cat . #7668 ) . V12GRAS virus was produced with pBABE-V12GRAS as per the procedure described above . HFFs were transduced with RAS virus and incubated with antibiotic selection medium ( puromycin 2 μg/ml ) for 4–5 days . For RNA immunoprecipitation , 10 million cells were lysed in 1 ml of NP-40 lysis buffer ( 50 mM Tris HCl , ph7 . 4 , 150 mM NaCl , 1% NP-40 and Protease inhibitor cocktail ) . Lysate was cleared by centrifugation at 12 , 000 RCF for 15 min . Cleared lysate was immunoprecipitated independently with 5 μg of anti-hnRNP A1 , anti-hnRNP D , Anti-hnRNP A2/B1 , Anti-hnRNP C1/C2 , Anti-hnRNP K , mIgG and R-IgG antibodies . Immune complexes were incubated with 30 μl of pre-equilibrated Dynabeads for 4 hr at 4°C . Dynabead purified immune complexes were subjected to Proteinase K digestion at 37°C for 1 hr followed by NucleoSpin RNA II purification kit and cDNA was prepared by RNA-to-cDNA EcoDry Premix kit ( Clontech ) . cDNA was used as a template in PCR amplifications with gene specific primers . TBX3 , CAPERα , or Control shRNA KD , PS and RAS HFFs were cultured in 6-well culture dishes for 2 days to 80% confluence . Then Actinomycin D was added to a final concentration of 5 mg/ml to suppress transcription . At 0 , 1 , 2 , and 4 hr after addition of Actinomycin D , equal numbers of cells were harvested from each sample and mRNA was prepared by nucleoSpin RNA II purification kit and cDNA was prepared by RNA-to-cDNA EcoDry Premix kit ( Clontech ) followed by qRT-PCR for specific transcripts . HFFs were transfected with pcDNA3 . 1 control or UCA1 expression plasmids as described above , incubated +/− Actinomycin D , and total cellular RNA was harvested at 0 , 1 , 2 , and 4 hr post treatment . For northern blot analysis , 5 µg total RNA from each time point was electrophoresed through a 1% agarose gel . The RNA was blotted onto Hybond-N+ membrane ( Amersham Pharmacia ) , and membranes were UV crosslinked . Membranes were hybridized for 18 hr with ( Torres et al . , 2003 ) P-labeled probes . Probes were generated by end-labeling DNA oligonucleotides containing following sequences complementary to p16INK mRNA:5′ GAGGAGGTGCTATTAACTCCGAGCATTAGCGAATGTGGC5′ AATCCTCTGGAGGGACCGCGGTATCTTTCCAGGCAAGGGG5′AAGGCTCCATGCTGCTCCCCGCCGCCGGCTCCATGCTGCT End-labeling reactions were performed using T4 polynucleotide Kinase ( NEB ) according to the manufacturer's directions . The hybridized blots were washed , and autoradiographs were developed as per standard procedure . Band intensities were measured by Image J analysis , and densitometric vales were plotted as bar graphs . HFFs were incubated with TBX3 or CAPER α shRNA encoding retrovirus for 48 hr followed by incubation for an additional 48 hr in selection medium . Total RNA was isolated and purity was assessed . Poly-A RNA was purified , fragmented , primed with random hexamers and used to generate first strand cDNA using reverse transcriptase . Samples that passed quality control steps were used for Illumina library preparation using the Illumina TruSeq RNA Sample Prep protocol . All libraries were sequenced ( with barcoding ) on a single lane of an Illumina HiSeq instrument for 50 cycles from a single end . A total of 177 , 155 , 781 reads were produced in total for all 10 libraries ( median 17 , 348 , 374 reads ) . Base calling was performed using Illumina software . Sequence reads were aligned ( 98 . 5% mapped ) to the human genome build 37 . 2 with Tophat ( v2 . 0 . 8b ) using default parameters . Aligned reads were assembled into transcripts and their relative abundance was measured using Cufflinks ( v2 . 1 . 1 ) with fragment bias correction ( frag-bias-correct ) and multi-read correction ( multi-read-correct ) . Cufflinks transcript assemblies were based on transcripts of NCBI Homo sapiens annotation release 104 and miRBase release 19 as provided in the Illumina iGenomes data set . Cuffdiff was used to test for differential expression between samples and controls and expression differences were taken as significant if the FDR adjusted p-value was less than 0 . 05 ( Source Data Files 1 and 2 ) . Statistically overrepresented gene ontology/biologic process categories and KEGG pathways were determined using DAVID ( Huang da et al . , 2009a , 2009b ) . The hypergeometric test , as implemented in the R statistical language ( phyper ) , was used to test significance of the number of genes found to be co-regulated between samples ( Figure 7—source data 3 ) .	Cell division and growth are essential for survival . But it is equally important that cells can stop dividing , because failing to do so can lead to the uncontrolled tumor growth seen in cancer . One such quality control mechanism is called senescence , which stops the growth and multiplication of cells that are old , damaged or behaving in ways that may harm the organism . All cells eventually stop dividing and undergo senescence , but a number of factors may trigger the process early , such as DNA damage , stress or the appearance of cancer-causing proteins . Senescence can be harmful if it occurs too early in life and interferes with normal growth . Severe birth defects—including fatal heart problems and limb malformations—occur if senescence is inappropriately triggered early in development . Mutations in a gene encoding a protein called TBX3 have been linked to these severe birth defects . Normally , TBX3 stops the production of other proteins that trigger senescence in early development , and helps to maintain stable conditions in adult cells . Understanding how it does so could help scientists understand normal cell function and aging , and also help to find ways to trigger senescence in cancerous cells . Kumar et al . found that a protein called CAPERα—for short Coactivator of AP1 and Estrogen Receptor—forms a complex with TBX3 that stops cells dividing in living organisms in at least two different ways . One way is by altering how DNA is folded . The other way involves a non-coding strand of RNA from a gene called UCA1: this RNA prevents the degradation of proteins that stop cell division . In normal proliferating cells , the CAPERα/TBX3 protein complex prevents the production of UCA1 RNA . In contrast , in cells that received a cancer causing stimulus , TBX3 and CAPERα physically separate: this activates production of UCA1 RNA and causes senescence . Further studies will be required to establish exactly how the CAPERα/TBX3 protein complex interacts with DNA and RNA to control senescence and prevent cancer .	[ "Abstract", "Introduction", "Results", "Discussion", "Material", "and", "methods" ]	[ "developmental", "biology", "cell", "biology" ]	2014	Coordinated control of senescence by lncRNA and a novel T-box3 co-repressor complex
Every DNA segment in a eukaryotic genome normally replicates once and only once per cell cycle to maintain genome stability . We show here that this restriction can be bypassed through alternative transposition , a transposition reaction that utilizes the termini of two separate , nearby transposable elements ( TEs ) . Our results suggest that alternative transposition during S phase can induce re-replication of the TEs and their flanking sequences . The DNA re-replication can spontaneously abort to generate double-strand breaks , which can be repaired to generate Composite Insertions composed of transposon termini flanking segmental duplications of various lengths . These results show how alternative transposition coupled with DNA replication and repair can significantly alter genome structure and may have contributed to rapid genome evolution in maize and possibly other eukaryotes . Initiation of DNA replication in eukaryotic cells is controlled by the replication licensing system ( Blow , 1993; Blow and Dutta , 2005; Truong and Wu , 2011 ) , which ensures that each segment of the genome is replicated only once per cell cycle . The expression and activity of the replication licensing factors are precisely regulated , and misexpression or mutation of these factors can lead to DNA re-replication , genome instability , major chromosomal rearrangements , and tumorigenesis ( Melixetian et al . , 2004; Green and Li , 2005; Rice et al . , 2005; Hook et al . , 2007; Liontos et al . , 2007; Sugimoto et al . , 2009; Green et al . , 2010 ) . Misregulation of some histone methyltransferases can also result in DNA re-replication in plants and animals ( Jacob et al . , 2010; Tardat et al . , 2010; Fu et al . , 2013 ) . Although DNA replication is strictly controlled , some DNA segments can escape this restriction and replicate more than once in a single cell cycle in normal cells . For example , some Class II DNA transposons , including the maize Ac/Ds system , E . coli TN10 , and E . coli TN7 , are known to transpose during DNA replication ( Roberts et al . , 1985; Chen et al . , 1987; Peters and Craig , 2001 ) . If a replicated transposon excises and reinserts into an unreplicated site , the transposon can undergo one additional replication in the same S phase; the re-replication , however , is limited to the TE itself and does not extend into the TE-flanking regions . We and others have previously shown that a pair of Ac termini in reversed orientation can undergo transposition , generating major chromosomal rearrangements such as deletions , inversions , permutations , duplications , and reciprocal translocations ( Zhang and Peterson , 2004; Zhang et al . , 2006; Huang and Dooner , 2008; Zhang et al . , 2009 , 2013 ) ; this transposition reaction is termed reversed Ac ends transposition ( RET ) . All the RET-generated genome rearrangements described to date are fully explained by models in which the excised TE termini inserted into target sites that had completed DNA replication . However , it seems reasonable to expect that RET , like standard Ac/Ds transposition , may also occur during DNA replication , and that the excised reversed Ac termini could insert into unreplicated target sites . Here , we show that such events do occur , and that they can induce re-replication of the TE and its flanking sequences . This process generates novel structures termed Composite Insertions ( CIs ) that contain TE sequences and variable lengths of the flanking genomic DNA . The allele P1-ovov454 ( GenBank accession # KM013692 ) carries an intact Ac element and a fractured Ac ( fAc ) element inserted in the second intron of the maize p1 gene; the 5′ terminus of Ac and the 3′ terminus of fAc are present in reversed orientation with respect to each other and separated by an 822-bp inter-transposon segment ( Figure 1A ) ( Yu et al . , 2011 ) . Our recent work showed that the P1-ovov454 allele undergoes RET to generate derivative alleles containing either deletions or Tandem Direct Duplications ( TDDs; Figure 1—figure supplement 1 ) . These are formed as a direct consequence of transposition of the Ac/fAc termini into a replicated target site on the sister chromatid ( Zhang et al . , 2013 ) . The deletions and TDDs vary in size depending on the position of the insertion site ( green/black triangle in Figure 1—figure supplement 1 ) ; the TDDs previously characterized range in size from 8 kb to 5 . 3 Mb ( Zhang et al . , 2013 ) . 10 . 7554/eLife . 03724 . 003Figure 1 . Reversed Ac ends transposition ( RET ) during DNA replication generates Tandem Direct Duplication ( TDD ) and Composite Insertion ( CI ) . Lines indicate a replicating chromosome , hexagons indicate replicons . The blue boxes are exons 1 , 2 , and 3 ( right to left ) of the p1 gene , and the green/black triangles are the transposition target site . Red lines with arrow ( s ) indicate Ac/fAc insertions , and the open and solid arrowheads indicate Ac/fAc 3′ and 5′ ends , respectively . Two replication forks considered here are marked α and β . For animated version , see Video 1 . ( A ) The locus containing fAc/Ac is replicated . Vertical arrows indicate the sites of Ac transposase cuts at the fAc 3′ and Ac 5′ ends . ( B ) Transposase cleaves and the inter-transposon segment is ligated to form a circle . The excised transposon ends will insert into an unreplicated target site indicated as the green/black triangle . Like standard Ac/Ds transposition , insertion of the Ac/fAc termini into the target site generates an 8-bp target site duplication ( TSD; green/black triangle ) . ( C ) Insertion of the excised transposon termini places fAc and fAc-flanking DNA ahead of replication fork β ( upper chromatid ) , and Ac and Ac-flanking DNA ahead of replication fork α to generate a rolling circle replicon ( lower chromatid ) . DNA replication continues . ( D ) Following re-replication of fAc , Ac , and a portion of the flanking sequences , DNA replication forks α and β stall and abort , resulting in chromatids terminated by broken ends ( the red > or < symbol ) ( Michel et al . , 1997 ) . The dotted red line connects the two broken ends that will fuse together . ( E ) Chromatid fusion produces a chromosome with two unequal sister chromatids: The upper chromatid contains a deletion of the segment from fAc to the a/b target site . The lower chromatid contains a TDD ( left-hand loop ) , as well as a new CI ( right-hand loop ) . The TDD contains the DNA deleted from the upper chromatid; the CI contains the re-replicated Ac , fAc and flanking sequences . The junction where broken chromatid ends were joined is indicated by the red × . DOI: http://dx . doi . org/10 . 7554/eLife . 03724 . 00310 . 7554/eLife . 03724 . 004Figure 1—figure supplement 1 . Reversed Ac ends transposition after DNA replication generates Tandem Direct Duplications ( TDDs ) . The two lines indicate sister chromatids of fully replicated maize chromosome 1 , joined at the centromere ( black ) . The blue boxes are exons 3 , 2 , and 1 ( left to right ) of the p1 gene . Red lines with arrowhead ( s ) indicate Ac/fAc insertions , and the open and solid arrowheads indicate the 3′ and 5′ ends , respectively , of Ac/fAc . The short horizontal arrows show the orientations and approximate positions of PCR primers , and the numbers below are the primer names . The green/black triangles indicate the transposon target site sequences and target site duplications . ( A ) Ac transposase cleaves the lower chromatid at the 3′ end of fAc and the 5′ end of Ac ( arrows ) . ( B ) Following transposase cleavage , the internal p1 genomic sequences are joined to form a circle . Dotted lines indicate the insertion of the fAc and Ac termini into the a/b site on the sister chromatid . ( C ) Transposon ends insert into the upper sister chromatid at the a/b target site . ( D ) The Ac 5′ end joins to the distal side ( green ) of the target site and the fAc 3′ end joins to the proximal side ( black ) of the target site to generate a proximal deletion ( upper chromatid ) and a direct duplication ( lower chromatid ) . The shaded arrows encompass the duplicated segments . Note: this Figure is adopted from Figure 1 of Zhang et al . ( 2013 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03724 . 004 Here , we asked: what are the consequences of RET events that occur during DNA replication ? We developed and tested models in which replicated Ac/fAc termini are excised by RET and inserted into unreplicated target sites . As shown in Figure 1 ( See also the animation Video 1 ) , this type of transposition reaction places already-replicated DNA in front of a replication fork where it may undergo a second round of replication . We propose that the re-replication fork may spontaneously abort , yielding two chromatid fragments terminated by double-strand breaks ( DSBs ) ; fusion of the DSBs restores the chromosome linearity and generates CIs containing Ac/fAc and their flanking sequences at the duplication breakpoints . By comparing RET events involving insertion sites that are unreplicated ( Figure 1 ) vs replicated ( Figure 1—figure supplement 1 ) , we can see that both types of events generate TDDs whose sizes are determined by the transposon insertion site . However , only events with unreplicated insertion sites also generate CIs via re-replication of the Ac/fAc and their flanking sequences; the resulting products are termed TDDCI alleles . Because the formation of TDDs was described in detail previously ( Zhang et al . , 2013 ) , here we will focus on the origin and characterization of the CI of the TDDCI alleles . 10 . 7554/eLife . 03724 . 005Video 1 . Animation showing model for reversed Ac ends transposition during DNA replication . See Figure 1 legend for details . DOI: http://dx . doi . org/10 . 7554/eLife . 03724 . 005 Both TDD and TDDCI alleles contain similar duplication structures and should exhibit similar phenotypes . Therefore , we screened maize ears as described previously to visually identify putative TDD-containing alleles ( Zhang et al . , 2013 ) . We identified 25 candidate alleles , and cloned and sequenced the duplication/Ac junctions ( the green segment flanking the Ac 5′ end in Figure 1E and Figure 2A ) from 16 of the 25 TDD/TDDCI candidates via Ac casting ( Singh et al . , 2003; Wang and Peterson , 2013 ) or inverse PCR ( iPCR ) ( See Zhang et al . ( 2013 ) for detailed screening and cloning methods ) . To identify the TDDCI alleles , we designed PCR primers that flank the progenitor insertion target sites for each allele ( Figure 2A , primers 1 and 2 ) . Primers 1 + Ac5 can amplify a product from both TDD and TDDCI while primers 2 + Ac3 can amplify a product only from TDDCI since the latter contains an additional CI ( Figure 2A ) . As expected , PCR using primers 1 + Ac5 produced bands of the expected sizes in all the 16 alleles ( Figure 2B , upper panel; seven examples are shown here ) . Whereas , primers 2 + Ac3 produced bands with expected sizes from only seven alleles ( Figure 2B , lower panel ) . Sequencing of the PCR products obtained from primers 1 + Ac5 and 2 + Ac3 revealed that these seven TDDCI candidates have duplication/insertion breakpoints located from 13 , 392 bp to 1 . 7 Mb proximal to the p1 locus on chromosome 1 ( Table 1 ) . Importantly , the Ac termini are flanked by 8-bp target site duplications ( TSDs; green/black triangles in Figure 1E ) as predicted by the model in Figure 1 ( See Supplementary file 1 for sequences containing TSDs ) . 10 . 7554/eLife . 03724 . 006Figure 2 . PCR screening and DNA gel blotting of candidate TDDCI alleles . ( A ) Detailed structures of P1-ovov454 ( progenitor ) and RET-generated P1-rr-twin/p1-ww-twin ( TDDCI/Deletion ) alleles deduced from Figure 1 . The horizontal blue lines are p1 gene sequence while the green lines are p1 proximal sequences , including the p2 gene sequence ( a p1 paralog , ∼70 kb proximal to p1 ) ; the blue and green boxes are exons 1 , 2 , and 3 ( right to left ) of p1 and p2 , respectively . The small horizontal arrows indicate the orientation and the approximate position of the PCR primers . The gray boxes indicate probe 8B used in DNA gel blot analysis , the short vertical black lines are SacI sites , and the numbers between the SacI sites indicate the lengths of those fragments detected by probe 8B . The hatched boxes represent the distal ( black ) and proximal ( green ) 5248 bp repeats flanking the p1 locus . These repeats are identical except for six SNPs , indicated by short red vertical lines inside the green hatched box ( SNPs 3 and 4 are only 43 bp apart ) . Other symbols have the same meaning as in Figure 1 . ( B ) PCR products obtained using primers 1 + Ac5 ( upper ) or 2 + Ac3 ( lower ) . Lane 1 , 1 kb DNA ladder; lane 2 , P1-ovov454; lane 3 , P1-rr-T22; 4 , p1-ww-T22; lane 5 , P1-ovov454; lane 6 , P1-rr-T24; 7 , p1-ww-T24; lane 8 , P1-ovov454; lane 9 , P1-rr-E17; lane10 , P1-ovov454; lane 11 , P1-rr-E340; lane 12 , P1-ovov454; lane 13 , P1-rr-T21; 14 , p1-ww-T21; lane 15 , P1-ovov454; lane 16 , P1-rr-E5; lane 17 , P1-ovov454; lane 18 , P1-rr-E311 . Note: the sequences of primers 1 and 2 are specific for each allele . ( C ) DNA gel blot analysis of the TDDCI/deletion alleles . Genomic DNA was digested with SacI and the blot was hybridized with probe 8B ( see Figure 2A for the position of the probe ) . Lane 1: p1-ww[4Co63] , lane 2: P1-ovov454/p1-ww[4Co63] , lane 3: P1-rr-T22/p1-ww[4Co63] , lane 4: p1-ww-T22/p1-ww[4Co63] , lane 5: P1-rr-T24/p1-ww[4Co63] , lane 6: p1-ww-T24/p1-ww[4Co63] , lane 7: P1-rr-E17/p1-ww[4Co63] , lane 8: P1-rr-E340/p1-ww[4Co63] , lane 9: P1-rr-T21/p1-ww[4Co63] , lane 10: p1-ww-T21/p1-ww[4Co63] , lane 11: P1-rr-E311/p1-ww[4Co63] , lane 12: P1-rr-E5/p1-ww[4Co63] . DOI: http://dx . doi . org/10 . 7554/eLife . 03724 . 00610 . 7554/eLife . 03724 . 007Table 1 . Features of alleles generated by RET-induced DNA re-replicationDOI: http://dx . doi . org/10 . 7554/eLife . 03724 . 007Allele numberAllele typeDistance from donor locus to CIP1-rr-T21Solo-CI13 , 392 bpP1-rr-E5Solo-CI16 , 497 bpP1-rr-T22TDDCI70 kbP1-rr-T24TDDCI80 kbP1-rr-E340TDDCI447 kbP1-rr-E311Solo-CI563 kbP1-rr-E17TDDCI1 . 7 MbDistance given is from the 5′ end of Ac in the progenitor P1-ovov454 allele , to the point of insertion of the CI; that is , the distance between the TDD and CI insertion points in Figure 2A . In TDDCI alleles , this distance is also the length of the duplicated segment . Except for the fully sequenced alleles P1-rr-T21 and P1-rr-E5 , the values given are based on the B73 reference genome sequence ( Schnable et al . , 2009 ) , which likely differs from the genotype used in these experiments . Of particular importance are the results derived from three red/white twinned sectors , in which a sector of red kernel pericarp ( seed coat ) is twinned with an adjacent white pericarp sector ( Figure 3 ) . From each red pericarp sector , we isolated P1-rr alleles ( P1-rr-T21 , P1-rr-T22 and P1-rr-T24 ) , and from each white twin sector , we isolated corresponding p1-ww alleles ( p1-ww-T21 , p1-ww-T22 , and p1-ww-T24 ) . Similar types of twinned pericarp sectors have been shown to arise from the reciprocal products of standard Ac transposition events ( Greenblatt and Brink , 1962; Chen et al . , 1992 ) . Here , we propose that each pair of red/white twinned alleles are derived from the reciprocal TDDCI/deletion products of RET ( sister chromatids shown in Figure 1E ) . This was tested by PCR using primers 2 + Ac3; as shown in Figure 2B ( lower panel ) , these primers produced bands of the same size for each set of twinned alleles . Moreover , for each pair of red/white co-twins , the sequences of the PCR products obtained using primers 2 + Ac3 are identical ( Supplementary file 1 ) . Together these results are consistent with the model of RET during DNA replication as shown in Figure 1 . 10 . 7554/eLife . 03724 . 008Figure 3 . An ear with twinned sectors . The photo shows two sides of the same ear . Left-side view has a large area with parental P1-ovov454 phenotype ( orange pericarp with frequent colorless sectors ) , while the right-side view shows a large area with typical P1-rr-Twin phenotype ( dark red pericarp with few colorless sectors ) . A single large p1-ww-Twin sector ( kernels with mostly colorless pericarp ) is visible in both views . The solid purple kernels present in all the sectors result from an independent germinal reversion of the r1-m3::Ds allele and can be ignored . DOI: http://dx . doi . org/10 . 7554/eLife . 03724 . 008 Because PCR only provides information on rearrangement junctions , we further analyzed the structures of the candidate TDDCI alleles by DNA gel blot . Genomic DNA was digested with SacI and the blot was hybridized with probe 8B ( gray boxes in Figure 2A ) . This probe detects the p1 gene ( 12 . 7 kb band ) , the paralogous p2 gene ( 4 . 7 kb band ) , and the p1-ww[4Co63] allele ( 5 . 0 kb band ) ( Goettel and Messing , 2010 ) on the homologous chromosome . First , the 12 . 7 kb p1 band is absent in the three twinned p1-ww alleles ( p1-ww-T22 , p1-ww-T24 , and p1-ww-T21; Figure 2C , lanes 4 , 6 and 10 , respectively ) . This result confirms the presence of a deletion as predicted by the model shown in Figure 1 . Second , the alleles P1-rr-T24 , P1-rr-E17 , and P1-rr-E340 show a more intense 4 . 7 kb p2 band in comparison with the 5 . 0 kb band ( Figure 2C , lanes 5 , 7 , 8 ) . This result is also expected because these three alleles have duplications of >70 kb ( Table 1 ) that generate additional copies of the p2 gene located ∼70 kb proximal to p1 . Third , alleles P1-rr-T22 , P1-rr-T21 , and P1-rr-E5 ( Figure 2C , lanes 3 , 9 and 12 , respectively ) exhibit one or two new bands hybridizing with probe 8B . This is consistent with the presence of a CI that contains a newly-generated copy of the 8B sequence ( Figure 2A ) . In P1-rr-T22 , the duplication/insertion breakpoint occurred in the p2 band containing probe 8B , resulting in a shift of the 4 . 7 kb band to ∼8 kb ( Figure 2C , lane 3 ) . Moreover , this ∼8 kb band is more intense than the 5 . 0 kb p1-ww[4Co63] band and the 12 . 7 kb p1 band in P1-rr-T22 ( lane 3 in Figure 2C ) . The model in Figure 1 and our analyses indicate that the intense ∼8 kb band is actually a triplet containing two copies of a new 8461 bp p1 fragment ( one from the TDD , and a second from the CI , see below ) and one copy of a 8127 bp p2 fragment from the rearrangement junction . Further DNA gel blot analyses with a different p1 probe ( not shown ) confirm that P1-rr-T22 contains a TDD . All together , these results indicate that these four alleles—P1-rr-T22 , P1-rr-T24 , P1-rr-E17 , and P1-rr-E340—contain the TDDCI structure . We then characterized the structures of the CIs in the four TDDCI alleles . The model in Figure 1 predicts that the insertion size and structure are determined by where re-replication aborts and how the resulting DSBs are repaired ( Figure 1D ) . The structures of the CIs were determined by PCR using a series of divergent primer pairs flanking the Ac/fAc insertions ( δ and π , the blue arrows in Figure 2A ) . These primers will not amplify products from the progenitor P1-ovov454 allele because they point away from each other ( Figure 2A ) . However , if the CI is formed by re-replication and the Ac/fAc flanking segments are fused as shown in Figure 1E and Figure 2A , then these primers will be oriented towards each other and can amplify the internal sequence of the insertion . In this way , we obtained the internal sequences carried by the CIs in P1-rr-T22 and P1-rr-E17 . The CI in P1-rr-T22 is 23 , 238 bp in length ( GenBank accession # KM013690 ) , consisting of 14 , 484 bp of fAc and its distal flanking sequence and 8754 bp of Ac and its proximal flanking sequence ( Figure 4 ) ; these two fragments are joined at a 4-bp microhomology sequence consistent with DSB repair via non-homologous end joining ( NHEJ ) . In addition to the CI , the P1-rr-T22 allele carries a 70-kb TDD ( Table 1 ) , and its white co-twin p1-ww-T22 carries a reciprocal 70-kb deletion; moreover , the breakpoints of both the P1-rr-T22 duplication and p1-ww-T22 deletion contain 8-bp target site duplications . All of these features are predicted by the RET/re-replication model shown in Figure 1 . 10 . 7554/eLife . 03724 . 009Figure 4 . The structures and sizes of Composite Insertions ( CIs ) . The double-headed arrows ( left side ) indicate Ac elements , while the single-headed arrows ( right side ) indicate fAc . The red × symbol indicates the junction of the two re-replicated segments in the insertion . Other symbols have the same meaning as in Figure 1 and Figure 2A . DOI: http://dx . doi . org/10 . 7554/eLife . 03724 . 009 The CI in P1-rr-E17 ( GenBank accession # KM013689 ) is 19 , 341 bp in length ( Figure 4 ) ; its structure suggests that the DSBs predicted in Figure 1D were repaired via homologous recombination ( HR ) between two direct repeat sequences that flank the p1 gene in P1-ovov454 ( Lechelt et al . , 1989 ) . These repeats ( hatched boxes in Figure 2A ) are 5248 bp in length; the proximal copy is 4555 bp from the Ac element while the distal copy is 2934 bp from fAc ( Figure 2A ) . If re-replication continued beyond the Ac and fAc segments and into the flanking 5248 bp repeats before aborting , then the DSBs could be repaired via HR to generate the observed structures ( Figure 5 ) . The two repeat copies flanking P1-ovov454 differ at six SNPs in the distal half of the repeats ( Figure 2A , red vertical short lines in the hatched box ) . Sequences of the P1-rr-E17 allele show that the repeat in the CI is identical to the proximal copy . These results suggest that the HR crossover occurred between the proximal halves of the two repeats ( Figure 5 ) . 10 . 7554/eLife . 03724 . 010Figure 5 . RET followed by homologous recombination generates identical 19 , 341 bp Composite Insertions in P1-rr-E17 and P1-rr-E5 . ( A ) Structure of the chromosome 1S segment containing the progenitor P1-ovov454 allele , prior to RET . ( B ) Drawing shows the RET stage corresponding to Figure 1D . Recombination between the 5248 bp repeats near the two DSBs ( marked by > or < ) generates a Composite Insertion . ( C ) Structure of P1-rr-E17 containing TDD ( left-hand triangle ) and Composite Insertion ( right-hand triangle ) . All the symbols have the same meaning as in Figure 2 . Note: P1-rr-E5 contains the 19 , 341 bp CI but does not contain the TDD . See text for details . DOI: http://dx . doi . org/10 . 7554/eLife . 03724 . 010 For P1-rr-T24 , no product could be amplified using the divergent primer strategy described above . However , a band of ∼5 . 0 kb could be amplified using primers 1 + 2 which flank the insertion site . This band was sequenced and found to contain an intact Ac element ( Figure 4 ) . It seems very unlikely that this Ac was inserted through a simple transposition event , because the insertion site is located precisely at the duplication junction that is generated by RET , and an independent Ac transposition would not be expected to insert into precisely the same site . We suggest that the Ac insertion in P1-rr-T24 was produced by HR between the re-replicated Ac and fAc segments as they share 2039 bp of sequence identity ( Figure 6 and Video 2 ) . Finally , the structure of the CI in P1-rr-E340 is still unknown; DNA gel blotting ( not shown ) indicated that the Ac-proximal fragment is in the range of 18–90 kb and the fAc-distal fragment is greater than 18 kb , resulting in a CI of at least 36 kb in length . 10 . 7554/eLife . 03724 . 011Figure 6 . RET followed by homologous recombination generates a simple Ac insertion in P1-rr-T24 . ( A ) , ( B ) , ( C ) , and ( D ) are the same as in Figure 1 . ( E ) Homologous recombination occurs between the re-replicated Ac and fAc . ( F ) Two new chromatids are formed: the lower chromatid contains a Tandem Direct Duplication and an Ac insertion , and the upper chromatid carries a reciprocal deletion . For animated version , see Video 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 03724 . 01110 . 7554/eLife . 03724 . 012Video 2 . Animation showing model for RET followed by homologous recombination and generation of a simple Ac insertion in P1-rr-T24 . See text for details . DOI: http://dx . doi . org/10 . 7554/eLife . 03724 . 012 PCR results show that the P1-rr-E311 , P1-rr-T21 , and P1-rr-E5 alleles contain junctions consistent with the presence of CI ( Figure 2B , lanes 12–18 ) . However , DNA gel blot analysis suggests that these same alleles do not contain TDDs ( Figure 2C , lanes 9–12 ) . Importantly , the CI in P1-rr-T21 is flanked by a target site duplication , and the CI insertion site is identical to the deletion breakpoint in the co-twin p1-ww-T21; these results strongly suggest that these twinned alleles were generated as the reciprocal products of an alternative transposition mechanism . We propose that the solo-CI alleles were formed by a mechanism similar to that shown in Figure 1 , except that the termination of replication ( Figure 1C ) resulted in release and loss of the rolling circle . Because the TDD originates from the DNA included in the rolling circle , release of the rolling circle and subsequent DSB repair will result in a chromatid that carries only the CI ( Figure 7 and Video 3 ) . The CI structures of these three alleles were characterized via PCR using primers δ and π as described above and are diagrammed in Figure 4 . 10 . 7554/eLife . 03724 . 013Figure 7 . Generation of a Composite Insertion in the absence of a duplication . ( A ) , ( B ) , and ( C ) are the same as in Figure 1 . ( D ) Upper chromatid contains deletion; in lower chromatid stalling and abortion of rolling circle replication fork releases the circle . ( E ) The two chromatids fuse to form a new chromatid containing a Composite Insertion . For animated version , see Video 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 03724 . 01310 . 7554/eLife . 03724 . 014Video 3 . Animation showing model for RET followed by NHEJ repair and generation of a CI in P1-rr-E311 and P1-rr-T21 . The CI in P1-rr-E5 was generated via a similar mechanism ( i . e . the rolling circle was released when forming a DSB ) , but the DSBs were repaired by homologous recombination as shown in Figure 5 ( without the TDD ) . See text for details . DOI: http://dx . doi . org/10 . 7554/eLife . 03724 . 014 In P1-rr-T21 , the CI is 14 , 287 bp in length and contains a 3-bp microhomology region at the internal junction ( GenBank accession # KM013688 ) , consistent with DSB repair via NHEJ . For P1-rr-E311 , the CI is 23 , 647 bp in length and has no apparent microhomology sequence at the internal junction ( GenBank accession # KM013691 ) , which is not uncommon for NHEJ-mediated repair ( Kramer et al . , 1994; Wu et al . , 1999; Lloyd et al . , 2012 ) . P1-rr-E311 does not contain a TDD , and its CI does not include fragment 8B; therefore the DNA gel blotting pattern in P1-rr-E311 is the same as its progenitor P1-ovov454 ( lane 2 and lane 11 in Figure 2C ) . Finally , the CI in P1-rr-E5 is 19 , 341 bp; its structure is identical to that in P1-rr-E17 ( Figure 4 ) , even though these alleles arose independently and have the CI in different positions ( 16 , 497 bp and 1 . 7 Mb proximal to the Ac element in P1-ovov454 , respectively; Table 1 ) . We propose that both cases were produced via HR between the 5248 bp p1-flanking repeat sequences as described above and shown in Figure 5 . In addition to the above alleles , we identified another allele ( P1P2-3 , Figure 8A ) that contains a CI but which was derived from a different progenitor allele ( p1-vv-D103 ) . The structure of p1-vv-D103 is similar to that of P1-ovov454 , except that the fAc element is shorter ( 779 bp vs 2039 bp in P1-ovov454 ) and the sequence distal to fAc has been replaced by chromosome 10 due to a chromosome 1–10 reciprocal translocation ( in preparation ) . Like the examples described above , the P1P2-3 allele arose in a single generation from p1-vv-D103; it contains a TDD of 80 kb , and a CI of 10 , 191 bp composed of 5017 bp of Ac and Ac-proximal flanking sequence and 5174 bp of fAc and fAc-distal flanking sequence . This structure is the same as that predicted by the model in Figure 1 . The internal breakpoint junction of the CI contains a 9-bp homologous sequence , consistent with DSB repair via a microhomology-mediated end joining ( MMEJ ) mechanism ( Ma et al . , 2003; McVey and Lee , 2008 ) . 10 . 7554/eLife . 03724 . 015Figure 8 . Two additional maize alleles likely generated by RET and re-replication . ( A ) Structure of progenitor allele p1-vv-D103 ( upper ) and TDDCI allele P1P2-3 ( lower ) . The p1-vv-D103 allele is carried on a chromosome 1–10 translocation; the brown line indicates DNA segment from chromosome 10 . See text for details . Other symbols have the same meaning as in previous figures . ( B ) TDDCI structure of bz1-m4-D6856 . The bronze-colored boxes indicate exons 1 and 2 ( right to left ) of the bronze1 gene on maize chromosome 9 . The baseline shows the predicted structure of the progenitor of bz1-m4-6856 . The dashed box encloses a hypothetical Ds element proposed to have been involved in the generation of bz1-m4-D6856 via RET . For animation , see Video 4 . Other symbols as in previous figures . The structure of bz1-m4-D6856 is deduced from Dowe et al . ( 1990 ) and Klein et al . ( 1988 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03724 . 015 If alternative Ac/Ds transposition can induce DNA re-replication and the formation of linked duplications and Composite Insertions , one may be able to detect these products at other loci . Interestingly , Barbara McClintock isolated an allele of the maize bronze1 gene ( bz1-m4-D6856 ) ( McClintock , 1956 ) that has a complex structure consisting of three TDDs of bz1 and its flanking sequence , separated by Ds elements ( Figure 8B ) ( Klein et al . , 1988; Dowe et al . , 1990 ) . The third repeat is not complete; its proximal side ( including the bz1 coding sequence ) is truncated and joined to a truncated Ds sequence . This structure is similar to that of P1-rr-T22 , P1-rr-E17 , and P1P2-3 described above: two intact Tandem Direct Duplications ( p1 vs bz1 sequence ) , separated by TEs ( Ac vs Ds ) , adjacent to a CI . In the case of bz1-m4-D6856 , the CI contains the truncated copy of the tandem duplication and the truncated Ds and is flanked by 8 bp target site duplications . We propose that bz1-m4-D6856 originated via a mechanism very similar to that shown in Figure 1: RET of two Ds elements located distal to the bz1 gene , followed by insertion of the excised Ds termini into an unreplicated target site in the bz1 5′ UTR region . The three tandem repeats would have been formed by rolling circle replication; one replication fork would have dissociated from the circle distal to the bz1 coding region to generate the incomplete repeat , while the other fork would have dissociated from the Ds element to generate a truncated Ds ( Video 4 ) . This model presupposes the existence of a Ds element ( the leftmost element in Figure 8B ) distal to the tandem repeats in bz1-m4-D6856 and its progenitor allele . No such element was reported on the original bz1-m4-D6856 genomic clones ( Klein et al . , 1988; Dowe et al . , 1990 ) . Efforts in our lab to identify a Ds element in this position in bz1-m4-D6856 and related stocks have been unsuccessful . However , McClintock's description of the origin of bz1-m4-D6856 ( As reported in Klein et al . , 1988 ) indicates that the bz1-m4 progenitor produced a high frequency of dicentric chromosomes , while the bz1-m4-D6856 derivative exhibited low dicentric frequency . Dicentric chromosome formation is a characteristic feature of alternative transposition reactions , such as RET , involving two nearby Ac/Ds elements ( Huang and Dooner , 2008; Yu et al . , 2010 ) . The switch from high to low dicentric frequency observed by McClintock would be consistent with excision of the ‘missing’ Ds shortly after the formation of the bz1-m4-D6856 allele . 10 . 7554/eLife . 03724 . 016Video 4 . Animation showing model for generation of TDDCI structure of bz1-m4-D6856 via rolling circle replication . See Figure 8B legend for details . DOI: http://dx . doi . org/10 . 7554/eLife . 03724 . 016 We have identified a new pathway leading to re-replication of specific chromosome segments in maize . This pathway is initiated by transposase-induced excision of the replicated termini of nearby transposons , followed by insertion of the excised transposon ends into an unreplicated target site . Re-replication begins when chromosomal replication forks reach the transposon and may continue for considerable distances into the flanking DNA before aborting . The two resulting chromatid ends are joined together to restore chromosome linearity . This re-replication pathway is localized to the transposons and their flanking sequences and does not require origin re-initiation . In contrast , deregulating licensing factor activity results in re-firing of replication origin ( s ) , leading to re-replication at multiple dispersed origins ( Green et al . , 2006 ) . Although little is known about termination of eukaryotic DNA replication , studies in yeast indicate that termination does not require specific terminator sites , but occurs wherever two replication forks converge ( McGuffee et al . , 2013 ) . Here , we propose that alternative transposition reactions can interrupt normal fork convergence . For example , Figure 1 shows that converging replication forks α and β are separated from each other by alternative transposition ( Figure 1C ) ; if not terminated by other factors , replication fork β could in principle continue until the end of the chromosome , which is ∼48 Mb from the p1 locus . However , our results suggest that DNA re-replication tends to abort after relatively short distances . The re-replicated segments generated from a single replication fork range in size from 4781 bp to 18 , 866 bp; the structure of the insertion in P1-rr-E340 is unknown , but DNA gel blotting analysis suggests a size of at least 36 kb . Thus the total extent of DNA re-replication is less than 19 kb in eight of nine alleles examined . In contrast , break-induced replication in yeast is capable of replicating from the site of a DSB to the end of the chromosome ( Kraus et al . , 2001 ) . What causes termination of re-replication following alternative transposition in maize ? One possibility is fork chasing and head-to-tail fork collision ( rear-ending ) , which has been shown to cause fork collapse and termination of DNA re-replication in Xenopus ( Davidson et al . , 2006 ) . Alternatively , re-replication may spontaneously stall and abort due to compromised fork progression as reported in yeast ( Green et al . , 2010 ) . Our model proposes that DNA re-replication aborts to produce chromatids terminated by broken ends , which are joined together to restore chromosome linearity ( Figures 1 , 6 , 7 , and 8 ) . If the chromatid DSBs were not repaired , the cell would die and that event would not be recovered in our screen . From a population of ∼2000 plants , we isolated 16 alleles that carry a duplication and/or insertion structure . Nine of these 16 alleles ( 56% ) have only a duplication ( Zhang et al . , 2013 ) , which indicates that the target site was replicated at the time of RET ( Figure 1—figure supplement 1 ) ; whereas seven alleles have an insertion , which indicates that the target site was unreplicated ( Figures 1 , 6 and 7 ) . The frequency of insertion into an unreplicated target site is 7/16 ( 44% ) , which is similar to a previous estimate of Ac insertion into unreplicated sites ( Greenblatt and Brink , 1962 ) . Thus the products of insertion into unreplicated target sites are not significantly under-represented in our sample , suggesting that repair of re-replication-generated DSBs is quite efficient in mitotic S phase cells . DNA lesions caused by replication fork stalling and collapse can be repaired by HR , NHEJ , MMEJ , replication slippage , FoSTeS ( fork stalling and template switching ) , BIR ( break-induced replication ) , MMBIR ( microhomology-mediated break-induced replication ) , MMIR ( microhomology/microsatellite-induced replication ) , and other mechanisms ( Kraus et al . , 2001; Ma et al . , 2003; Lee et al . , 2007; McVey and Lee , 2008; Payen et al . , 2008; Hastings et al . , 2009a , 2009b ) . In mammalian cells , replication fork-associated DSBs are predominantly repaired via HR ( Arnaudeau et al . , 2001 ) . Among the six CI alleles sequenced here , three were repaired by HR and three by NHEJ , indicating that these two repair pathways have relatively similar activities during the S phase of mitosis in maize . An important advantage of the maize system is the ability to identify genetically twinned sectors and to propagate and analyze their corresponding alleles . Because twinned alleles are the reciprocal products of a single event ( Greenblatt and Brink , 1962 ) , their structures should reflect a single parsimonious mechanism of origin . This allows us to distinguish among a variety of possible mechanisms for formation of segmental duplications . For example , non-allelic homologous recombination ( NAHR ) could generate a TDD joined and flanked by Ac as observed in P1-rr-T24 if there were a p1-proximal Ac element in the progenitor allele P1-ovov454 ( Figure 9 ) ; however , such an NAHR event cannot explain the observed structure of the white co-twin p1-ww-T24 ( compare upper chromatids of Figures 6F and 9C ) . Similarly , re-replication-induced gene amplifications ( RRIGA , a mechanism that couples NAHR and DNA re-replication ) can also generate chromosome structures very similar to that of P1-rr-T24 ( Green et al . , 2010; Finn and Li , 2013 ) . Like NAHR , RRIGA would also require a p1-proximal Ac element as in Figure 9B . However , the reciprocal product of an RRIGA-generated TDD would be a chromosomal fragment that lacks a centromere and telomeres , which would be lost in subsequent cell divisions . Therefore , neither NAHR nor RRIGA can generate the white co-twin p1-ww-T24 . In contrast , the actual structure of the white co-twin p1-ww-T24 is exactly as predicted by the RET re-replication model shown in Figure 6 . The structures of the other TDDCI alleles are also inconsistent with NAHR and RRIGA: the duplicated segments are flanked by Ac on the left side and a CI on the right side ( Figure 1E , lower chromatid ) , while NAHR/RRIGA-induced duplications would be flanked by identical Ac copies ( Figure 9C , lower chromatid ) . Finally , NAHR and RRIGA generate TDDs of the same structure recurrently . In contrast , all of the TDDCI alleles we have isolated to date have different duplication breakpoints . This is consistent with their origin via alternative transposition , because the duplication endpoints are determined by the position of the transposon insertion site , which is expected to differ for each transposition event . Moreover , the RET reinsertion sites have the same characteristic features as for standard Ac/Ds transposition , including preferential insertion into nearby , hypomethylated , gene-rich regions ( Greenblatt and Brink , 1962; Chen et al . , 1992; Vollbrecht et al . , 2010 ) , and formation of 8-bp Target Site Duplications lacking sequence specificity ( Vollbrecht et al . , 2010 ) . Taken together , our results consistently support the proposed mechanism of alternative transposition , re-replication , and repair . 10 . 7554/eLife . 03724 . 017Figure 9 . NAHR generates Tandem Direct Duplications . All the symbols have the same meanings as in Figure 1 . ( A ) Ac transposes to a site between a and b . ( B ) Homologous recombination between two non-allelic Ac elements on sister chromatids generates a deletion ( upper chromatid ) and a TDD ( lower chromatid ) in ( C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03724 . 017 In summary , we show here that reversed Ac ends transposition can generate TDDs and CIs . The TDDs range in size from several kb to >1 Mb and thus can increase the copy number of multiple linked genes and their regulatory sequences . The CIs we have discovered may be 20 kb or more in length . These are produced as a consequence of Ac transposition during DNA replication , and they exhibit a number of interesting features . First , the internal portions contain sequences that were originally flanking the donor Ac/fAc elements; the relative positions of these sequences are now switched , and they are fused together at a new junction . Because Ac/Ds elements are commonly inserted within or near genic sequences in plants , CI formation may shuffle the coding and/or regulatory sequences of the formerly flanking genes to create novel products . Moreover , the CIs are bordered by transposition-competent Ac/fAc 5′ and 3′ termini; hence , the entire CI has the structure of a macrotransposon ( Huang and Dooner , 2008; Yu et al . , 2010 ) that could subsequently transpose to new sites and increase in copy number . Eukaryotic genomes contain significant portions of Tandem Direct Duplications , dispersed segmental duplications , and tandem multi-copy arrays ( Bailey et al . , 2003 , 2004; Rizzon et al . , 2006; Shoja and Zhang , 2006; Bailey et al . , 2008; Dujon , 2010; Tremblay Savard et al . , 2011 ) ; our results suggest that transposition-induced DNA re-replication may have played an important role in generating these segmental expansions during genome evolution . The maize p1 gene encodes an R2R3-Myb transcription factor that regulates kernel pericarp ( seed coat ) and cob coloration . The phenotype conferred by each p1 allele is indicated by the particular suffix: P1-rr specifies red pericarp and red cob , p1-ww specifies white ( colorless ) pericarp white ( colorless ) cob , and P1-ovov specifies orange variegated pericarp and orange variegated cob . P1-ovov454 confers orange/red pericarp with frequent colorless sectors attributed to alternative transposition events that abolish p1 function ( Yu et al . , 2011 ) . The p1-ww[4Co63] allele is from the maize inbred line 4Co63 ( Goettel and Messing , 2010 ) . Ears of plants of genotype P1-ovov454/p1-ww[4Co63] were fertilized with pollen from plants of genotype C1 , r1-m3::Ds , p1-ww[4Co63] . The r1-m3::Ds allele is an Ac reporter allele: Ac-encoded transposase excises Ds from r1-m3::Ds , resulting in r1 reversion and purple aleurone sectors . Changes in Ac copy number can be inferred by the negative Ac dosage effect: increased copy number of Ac delays the developmental timing of Ac/Ds transposition and reduces the frequency of early transposition events , generally producing variegated patterns with fewer , later transposition events ( McClintock , 1948 , 1951 ) . Reversed Ac ends transposition ( Figure 1 ) can generate two non-identical sister chromatids: one carries a TDDCI , and the other a reciprocal deletion ( Figure 1E ) . At mitosis these chromatids will segregate into adjacent daughter cells , forming an incipient twinned sector . The sector with the deletion chromosome has lost Ac and exons 1 and 2 of the p1 gene; loss of Ac and p1 functions will specify kernels with colorless pericarp and no purple aleurone sectors . The sector with the duplication chromosome retains a functional P1-ovov454 gene and three copies of Ac; the predicted kernel phenotype will be orange/red pericarp with fewer colorless pericarp sectors , and fewer/smaller kernel aleurone sectors . Similar twinned sectors can also be formed via the mechanism in Figure 6 or 7 . Mature ears were screened for multi-kernel twinned sectors with these characteristics; kernels from selected sectors were grown and analyzed . Alleles derived from twinned sectors or whole ears are indicated by a ‘T’ or ‘E’ , respectively , prior to the allele number . Total genomic DNA was extracted using a modified cetyltrimethylammonium bromide ( CTAB ) extraction protocol ( Porebski et al . , 1997 ) . Restriction enzyme digestions and agarose gel electrophoresis were performed according to manufacturers' protocols and Sambrook et al . ( 1989 ) . DNA gel blots and hybridizations were performed as described ( Sambrook et al . , 1989 ) , except hybridization buffers contained 250 mM NaHPO4 , pH 7 . 2 , 7% SDS , and wash buffers contained 20 mM NaHPO4 , pH 7 . 2 , 1% SDS . Sequences of oligonucleotide primers are shown in Table 2; note that primers 1 and 2 are specific to each allele , depending upon the flanking sequences . PCR was performed using HotMaster Taq polymerase from 5 PRIME ( Hamburg , Germany ) . Reactions were heated at 94°C for 2 min , and then cycled 35 times at 94°C for 20 s , 60°C for 10 s , and 65°C for 1 min per 1 kb length of expected PCR product , then 65°C for 8 min . In some reactions 0 . 5–1 M betaine and 4–8% DMSO were added to improve yield . PCR products were separated on agarose gels , purified and sequenced directly by the DNA Synthesis and Sequencing Facility , Iowa State University , Ames , Iowa , United States . Ac casting and inverse PCR were used to isolate sequences flanking Ac insertions; these were performed as described previously ( Zhang et al . , 2009 ) . 10 . 7554/eLife . 03724 . 018Table 2 . Primer sequencesDOI: http://dx . doi . org/10 . 7554/eLife . 03724 . 018Primer 1P1-rr-T22CTGTGGTCGTCCTGCTCCGP1-rr-E17AGATTTGACAGAACAGCCCGCACP1-rr-T24GGTCACGCCCATAATAAAACAATACP1-rr-E340AACCCGTCTCATCATCATCAGTGTP1-rr-T21GGTTTGTTTGTGCTGCCTCCP1-rr-E311TCGTTCTCTGGTTGGTCGTCGTP1-rr-E5ATTGGTCCCTCCCTCTCCCTPrimer 2P1-rr-T22AGAACTACTGGAACTCGCACCTCAP1-rr-E17CCAGAGTATAGGGTCATGGAGCCP1-rr-T24GCGTCCTCTATCCATTCACTTTCAP1-rr-E340TTTATGAGCCGCTGAATCGCP1-rr-T21CCGATGCTCTTTTCCTTCTCTTCCP1-rr-E311GCGATGCTATCAGTTAGACCAGGCP1-rr-E5CGCCGAACTTTCACTGCTCTGCTAAc3GATTACCGTATTTATCCCGTTCGTTTTCAc5CCCGTTTCCGTTCCGTTTTCGT	To make accurate copies of its genome , a cell takes precautions to make sure each section of DNA is only duplicated once in every round of copying . However , there are some sections of DNA called transposons that can avoid these restrictions and be duplicated more often . Transposons are mobile pieces of DNA: they can be ‘cut’ from one section of the genome and are able to ‘paste’ back in somewhere else . The amount of mobile DNA in a genome varies a great deal between species , and in the crop plant maize , it makes up nearly 85% of the genome . Some transposons can move while the genome is being duplicated . If a transposon is cut out of a section of DNA that has already been copied and is pasted into a site that is yet to be copied , the transposon can be copied again . The transposon may now be present in two different places in the genome . If two transposons are close together on a section of DNA , both transposons can move at the same time . As they move , they can carry along pieces of the genome , transferring them from one site to another . These transferred pieces can include sections of , or even entire , genes . This is called alternative transposition , but it is not clear whether this process can happen when the genome is actively being duplicated . Here , Zhang et al . studied transposons in maize . The experiments found that alternative transposition can take place between a site that has already been copied and another site that is still waiting to be copied . Therefore , after the round of copying is completed , both transposons and the flanking DNA can be present in two places in the genome . When single transposons move , or alternative transposition takes place , sections of the genome can be rearranged and genes can be deleted or new ones can be created . Therefore , transposons may have contributed to rapid evolution in maize and possibly other species .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "plant", "biology" ]	2014	Transposition-mediated DNA re-replication in maize
Heat shock factor ( Hsf1 ) regulates the expression of molecular chaperones to maintain protein homeostasis . Despite its central role in stress resistance , disease and aging , the mechanisms that control Hsf1 activity remain unresolved . Here we show that in budding yeast , Hsf1 basally associates with the chaperone Hsp70 and this association is transiently disrupted by heat shock , providing the first evidence that a chaperone repressor directly regulates Hsf1 activity . We develop and experimentally validate a mathematical model of Hsf1 activation by heat shock in which unfolded proteins compete with Hsf1 for binding to Hsp70 . Surprisingly , we find that Hsf1 phosphorylation , previously thought to be required for activation , in fact only positively tunes Hsf1 and does so without affecting Hsp70 binding . Our work reveals two uncoupled forms of regulation - an ON/OFF chaperone switch and a tunable phosphorylation gain - that allow Hsf1 to flexibly integrate signals from the proteostasis network and cell signaling pathways . The heat shock response is an ancient and conserved signaling pathway in cells that regulates the expression of molecular chaperones in the presence of thermal and other environmental stresses ( Lindquist , 1986; Richter et al . , 2010 ) . Chaperones function to maintain protein homeostasis ( proteostasis ) by enabling de novo protein folding in the crowded intracellular environment and targeting proteins for degradation ( Dobson , 2003; Labbadia and Morimoto , 2015 ) . Although the heat shock response has been extensively studied , key aspects of the pathway remain a mystery including the mechanisms governing its activation and regulation . In eukaryotes , the master transcriptional regulator of the heat shock response is heat shock factor 1 ( Hsf1 ) ( Anckar and Sistonen , 2011 ) . Hsf1 and its cognate DNA binding site , the heat shock element ( HSE ) , represent one of the most conserved protein•DNA interactions known , having been maintained since the last common ancestor of the eukaryotic lineage ( Wu , 1995 ) . The depth of functional conservation is underscored by the observation that an active form of human Hsf1 can carry out the essential function of Hsf1 in yeast ( Liu et al . , 1997 ) . Indeed , both mammalian and yeast Hsf1 drive a compact set of genes dedicated to proteostasis that forms a densely connected network centered around Hsp70 , Hsp40 and Hsp90 ( Mahat et al . , 2016; Solís et al . , 2016 ) . The small size of the Hsf1 regulon belies its outsize importance in cellular viability . In addition to maintaining proteostasis at the cellular level , Hsf1 plays important roles in organismal health and disease . Critically , Hsf1 is frequently activated in cancer cells: it has been shown to be required for cancer progression in animal models ( Dai et al . , 2007 ) , its activation is associated with poor prognosis in many human cancer patients ( Santagata et al . , 2011 ) , and it drives cancer-specific gene expression programs in both tumor cells and the supporting stroma ( Mendillo et al . , 2012; Scherz-Shouval et al . , 2014 ) . By contrast , a lack of Hsf1 activity has been suggested to contribute to neurodegenerative diseases with hallmark protein aggregates , and activation of Hsf1 has been proposed as a therapeutic avenue ( Labbadia and Morimoto , 2015; Neef et al . , 2011 ) . Moreover , Hsf1 contributes to organismal lifespan ( Hsu et al . , 2003 ) and protects against obesity ( Ma et al . , 2015 ) . Despite the deep conservation of Hsf1 and its physiological and clinical importance , the mechanisms regulating Hsf1 activity during stress remain poorly defined , and thus to date it is unclear how Hsf1 controls the heat shock response in cells . Some aspects of Hsf1 regulation are organism- and cell type-specific , such as trimerization , which is a regulated event in mammalian cells but constitutive in yeast ( Sorger et al . , 1987; Sorger and Nelson , 1989; Westwood et al . , 1991 ) . However , two common features are thought to contribute to Hsf1 regulation in all organisms: chaperone titration and phosphorylation ( Anckar and Sistonen , 2011 ) . The chaperone titration model suggests that Hsf1 is bound in an inhibitory complex by chaperones in basal conditions ( Voellmy and Boellmann , 2007 ) . Upon heat shock , the chaperones are titrated away by unfolded or misfolded proteins , leaving Hsf1 free to activate transcription of chaperone genes . Once proteostasis is restored , client-free chaperones again bind to Hsf1 and deactivate it . There is biochemical , pharmacological and genetic evidence to support roles for the Hsp70 and Hsp90 chaperones , their co-chaperones and the TRiC/CCT chaperonin complex in regulating Hsf1 ( Abravaya et al . , 1992; Baler et al . , 1992 , 1996; Duina et al . , 1998; Guo et al . , 2001; Neef et al . , 2014; Ohama et al . , 2016; Shi et al . , 1998; Zou et al . , 1998 ) . However , direct , unequivocal evidence for this model – i . e . , a complete cycle of Hsf1 ‘switching’ by dynamic dissociation and re-association with specific chaperone ( s ) during heat shock – is lacking . As a result , though widely invoked , the details of the chaperone titration model remain unclear . The second putative Hsf1 regulatory mechanism common across organisms is heat shock-dependent phosphorylation . Multiple phosphorylation sites have been mapped on Hsf1 ( Anckar and Sistonen , 2011; Guettouche et al . , 2005 ) , and mutational analysis has suggested activating , repressing , fine-tuning and condition-specific roles for individual sites of phosphorylation in yeast and mammalian cells ( Budzynski et al . , 2015; Cho et al . , 2014; Dai et al . , 2015; Hahn and Thiele , 2004; Hashikawa et al . , 2006; Hashikawa and Sakurai , 2004; Hietakangas et al . , 2003; Høj and Jakobsen , 1994; Holmberg et al . , 2001; Kline and Morimoto , 1997; Knauf et al . , 1996; Lee et al . , 2013; Soncin et al . , 2003; Sorger and Pelham , 1988; Tang et al . , 2015; Wang et al . , 2003; Yamamoto et al . , 2007 ) . Recent work in which 15 phosphorylation sites were simultaneously mutated in human Hsf1 failed to disrupt Hsf1 activation during heat shock ( Budzynski et al . , 2015 ) , and a genome-wide RNAi-based screen for modulators of Hsf1 activity found no evidence for kinase regulation ( Raychaudhuri et al . , 2014 ) . Thus , despite being a hallmark of the heat shock response , no clear role for Hsf1 phosphorylation has yet emerged . Here , we combine experimental and theoretical approaches to elucidate the mechanism of Hsf1-mediated activation and control of the heat shock response in budding yeast . Specifically , we combine mass spectrometry , biochemistry , mathematical modeling , genetics and synthetic biology to propose and validate that Hsp70 dynamically interacts with Hsf1 to form the basis of a bona fide activation ‘switch’ and feedback loop that regulates Hsf1 activity during heat shock . Based on this finding , we then investigate the role and quantitative contribution of Hsf1 phosphorylation in regulating the output of the heat shock transcriptional response . We use en masse mutational analysis , combined with transcriptomic and genome-wide ChIP-seq measurements , to systematically define the function of Hsf1 phosphorylation . We find that phosphorylation is fully dispensable for Hsf1 activation during heat shock , but contributes by enhancing transcriptional output levels as a 'positive gain' . Phosphorylation does not control the interaction between Hsf1 and Hsp70 , but rather enhances transcription independent of whether Hsp70 is bound or not . Our findings reveal that Hsf1 uses these two modes of regulation – a chaperone switch and phosphorylation fine-tuning – in a largely uncoupled fashion to dynamically control the heat shock response . We propose that this allows Hsf1 to integrate diverse signaling information without disrupting its direct readout of the proteostasis network . To identify proteins involved in the dynamic regulation of Hsf1 during heat shock , we performed immunoprecipitation ( IP ) of Hsf1 from cells harvested over time following a shift from 25°C to 39°C and analyzed the IP samples by mass spectrometry ( MS ) . We expressed dual epitope-tagged Hsf1-3xFLAG-V5 as the only copy of Hsf1 from its endogenous promoter and performed serial affinity purifications to reduce non-specific binding ( Figure 1—figure supplement 1A , see Materials and methods ) . MS analysis revealed that the only proteins that co-precipitated with Hsf1 in basal conditions in each of three independent replicates were the cytosolic Hsp70 chaperones Ssa1 and Ssa2 ( jointly Ssa1/2 ) which share 98% sequence identity ( Figure 1—figure supplement 1B , C , Figure 1—source data 1 ) . No peptides derived from Ssa1/2 were identified in an untagged control or when we expressed YFP-3xFLAG-V5 and purified it following the same protocol ( Figure 1—source data 1 ) . Notably , no peptides belonging to Hsp90 or any other chaperones were identified in any of the samples . Intriguingly , following a five-minute heat shock , Ssa1/2 were absent in two IP replicates and were greatly diminished from the third replicate ( Figure 1—figure supplement 1B , C , Figure 1—source data 1 ) . No proteins co-precipitated with Hsf1 exclusively in heat shock conditions . Western blot analysis of Hsf1 IP samples collected over a heat shock time course revealed a transient decrease in the relative amount of Ssa1/2 followed by restoration of Ssa1/2 levels at the later time points ( Figure 1A ) . 10 . 7554/eLife . 18638 . 003Figure 1 . In vivo , in vitro and in silico evidence for an Hsp70•Hsf1 dissociation switch as the core mechanism regulating the heat shock response . ( A ) IP/Western blot showing Hsp70 transiently dissociating from Hsf1 during heat shock ( upper panel ) . Western blots were probed with antisera recognizing Ssa1/2 ( top ) and an anti-FLAG antibody to recognize Hsf1 ( bottom ) . IP of recombinant proteins were performed with rHsf1-3xFLAG as bait and analyzed by Western blot ( lower panel ) . Blots were probed with an anti-HIS antibody to recognize recombinant Ssa2 ( rSsa2 , top ) and with an anti-FLAG antibody to recognize recombinant Hsf1 ( rHsf1 , bottom ) . The IPs were also performed in the presence of 1 mM ATP or five-fold molar excess of Aβ42 peptide . The numbers below the blots indicate the normalized ratio of Ssa2/Hsf1 . ( B ) Cartoon schematic of the mathematical model of Hsf1 regulation illustrating the network connections and the feedback loop . UP is an abbreviation for 'unfolded proteins' . See Figure 1—figure supplement 2 and Materials and methods for details , equations and parameters . ( C ) Quantification of the top three peptides derived from Hsp70 proteins Ssa1 or Ssa2 ( Ssa1/2 are grouped due to 98% identity ) relative to the top three peptides from Hsf1 as determined by IP/MS ( left panel ) . The IP experiments were performed in triplicate at the indicated time points following a shift to 39°C . See Figure 1—figure supplement 1 . The values are the average of the three replicates and error bars depict the standard deviation . Source data are included as Figure 1—source data 1 . Simulation of the levels of the Hsf1•Hsp70 complex over time following a shift from 25°C to 39°C ( right panel ) . ( D ) Simulation of the levels of the HSE-YFP reporter over time following upshift from 25°C to the indicated temperatures ( left panel ) . Flow cytometry measurements of cells expressing the HSE-YFP reporter following upshift from 25°C to the indicated temperatures ( right panel ) . See Materials and methods for assay and analysis details . DOI: http://dx . doi . org/10 . 7554/eLife . 18638 . 00310 . 7554/eLife . 18638 . 004Figure 1—source data 1 . Table of peptide counts from proteins identified in Hsf1-3xFLAG-V5 IP/MS experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 18638 . 00410 . 7554/eLife . 18638 . 005Figure 1—figure supplement 1 . Ssa2 co-precipitates with Hsf1 in basal conditions but is greatly reduced immediately following heat shock . ( A ) Schematic of the Hsf1-3xFLAG-V5 construct and the two-step serial affinity purification strategy . ( B ) Sequence coverage of Hsf1 and Ssa2 as determined by mass spectrometry in basal conditions at 25°C . Three IP replicates were performed in each condition . The sequences of identified peptides are highlighted in yellow . ( C ) Sequence coverage of Hsf1 and Ssa2 as determined by mass spectrometry following five minutes at 39°C . Three IP replicates were performed in each condition . The sequences of identified peptides are highlighted in yellow . ( D ) IP of recombinant proteins were performed with FLAG-rSsa2 as bait and analyzed by Western blot . Blots were probed with an anti-FLAG antibody to recognize recombinant Ssa2 ( rSsa2 , top ) and with an anti-HIS antibody to recognize recombinant Hsf1 ( rHsf1 , bottom ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18638 . 00510 . 7554/eLife . 18638 . 006Figure 1—figure supplement 2 . Description and parameterization of mathematical model of the heat shock response . ( A ) Model schematic depicting the kinetic parameters and transcriptional Hill function . ( B ) The function used to relate the concentration of unfolded proteins ( UP ) to temperature to stimulate the heat shock response . ( C ) The model was simulated with a wide range of values for each kinetic parameter . ‘Successful’ parameter sets fulfilled specific criteria observed in the experimental data ( see Materials and methods ) . The percentages of successful parameter sets that include the indicated values of each parameter are plotted . DOI: http://dx . doi . org/10 . 7554/eLife . 18638 . 006 To validate that Hsf1 and Hsp70 directly interact , we tested their ability to bind to each other in vitro . We purified recombinant Hsf1-6xHIS and 6xHIS-3xFLAG-Ssa2 from E . coli , mixed the recombinant proteins together at an equimolar ratio and performed an anti-FLAG IP . We precipitated Hsf1 only in the presence of Ssa2 , demonstrating that they can specifically and directly bind to each other ( Figure 1—figure supplement 1D ) . We reversed the affinity tags and likewise precipitated Ssa2 in the presence of 3xFLAG-Hsf1 ( Figure 1A ) . Addition of ATP neither enhanced nor disrupted the interaction between Hsf1 and Ssa2 ( Figure 1A ) . By contrast , addition of a five-fold molar excess of the aggregation-prone , Alzheimer’s disease-associated Aβ42 peptide reduced the interaction between Hsf1 and Ssa2 ( Figure 1A ) . This suggests that hydrophobic peptides can titrate Hsp70 away from Hsf1 . Taken together , these in vivo and in vitro results confirm that Hsp70 dynamically dissociates and re-associates with Hsf1 during heat shock . Based on our finding that Hsp70 transiently dissociates from Hsf1 during heat shock , we next sought to develop a simple mathematical model of the heat shock response . The goal of the model was two-fold: ( 1 ) to quantitatively explore if Hsp70 interaction ‘switching’ could serve as a core , minimal regulatory mechanism that recapitulates the dynamics of heat shock response; ( 2 ) to generate quantitative predictions to further test this molecular mechanism . Our model consisted of a system of six coupled ordinary differential equations , describing a feedback loop in which free Hsf1 induces production of Hsp70 , and free Hsp70 in turn binds to Hsf1 in a transcriptionally inactive complex ( Figure 1B , Figure 1—figure supplement 2A , see Materials and methods ) . In addition to binding to Hsf1 , Hsp70 can also bind to an unfolded protein ( UP ) , but cannot bind to both Hsf1 and UP simultaneously . UPs are introduced at a level that is based on the temperature in accordance with biophysical measurements ( Figure 1—figure supplement 2B , see Materials and methods ) ( Lepock et al . , 1993; Scheff et al . , 2015 ) . In this manner , temperature upshifts increase the level of UPs , simulating protein folding stress caused by heat shock . UPs titrate Hsp70 away from Hsf1 , leaving Hsf1 free to activate expression of more Hsp70 , thus generating the feedback loop . This minimal model ignores any potential role for phosphorylation or other post-translational modifications in regulating Hsf1 activity . We used the apparent kinetics of Hsp70 dissociation from Hsf1 ( Figure 1C ) to constrain the model by computationally screening for parameter sets that captured key features of these data ( Figure 1—figure supplement 2C , see Materials and methods ) ( Ma et al . , 2009 ) . This parameter screening approach revealed that a broad range of values satisfied the experimental constraints ( Figure 1—figure supplement 2C ) . We settled on parameter values that appeared most frequently in simulations that fulfilled the experimental constraints ( Figure 1C ) . We tested the mathematical model by simulating time courses of temperature upshifts ( 25°C to 35°C , 25°C to 39°C , and 25°C to 43°C ) . As an output of the model , we tracked transcription of a ‘reporter’ whose levels are dependent on Hsf1 activity . This in silico reporter can be directly compared to experimental measurements of an Hsf1-dependent HSE-YFP reporter that we integrated into the yeast genome ( Brandman et al . , 2012 ) ( Figure 1B ) . Simulations of the HSE-YFP reporter generated induction curves that reached temperature-dependent plateaus ( Figure 1D ) . Since YFP is a long-lived protein – in cells and in the model – the plateaus indicate that no more YFP is being produced in the simulation and Hsf1 has deactivated . We performed the same temperature step time course experiments in cells and measured the HSE-YFP reporter at discrete time points by flow cytometry . As predicted by the model , we observed induction leading to temperature-dependent maximal responses ( Figure 1D ) . However , following the initial rapid accumulation of YFP that was predicted by the model , the cells continued to accumulate signal slowly through the later time points ( this later phase of transcriptional output is addressed further below ) . These results suggest that a chaperone titration model is sufficient to account for the immediate Hsf1 activation dynamics in response to temperature steps . To rigorously test our model of the heat shock response , we used the mathematical framework to predict the Hsf1 activation response for synthetic perturbations of the feedback loop . For example , in addition to activation by UPs ( via increases in temperature ) , the model predicted that Hsf1 transcriptional activity can be driven by Hsf1 overexpression in the absence of temperature upshifts ( Figure 2A , C ) . Less intuitively , the model also predicted that overexpression of an Hsf1 ‘decoy’ , which lacks the ability to trimerize or bind DNA , would activate endogenous Hsf1 by titrating Hsp70 ( Figure 2B , C ) . We next used synthetic biology approaches to experimentally test the model predictions . We constructed a decoy mutant by removing the DNA binding and trimerization domains from Hsf1 and replacing them with the well-folded fluorescent protein mKate2 ( Figure 2B ) . We then placed either a wild type allele of HSF1 or the decoy under the synthetic control of a β-estradiol ( estradiol ) inducible system ( Pincus et al . , 2014 ) ( Figure 2A , B , Figure 2—figure supplement 1A , see Materials and methods ) . In agreement with the model , both full length Hsf1 and the decoy activated the HSE-YFP reporter in dose-dependent manners , with full length Hsf1 serving as a more potent activator than the decoy ( Figure 2D ) . By contrast , expression of mKate2 alone led to only a modest increase in HSE-YFP levels as a function of estradiol ( Figure 2D ) . Importantly , the decoy does not form aggregates when overexpressed , remaining diffusely localized in the nucleus in both basal and heat shock conditions , but does interact with the Hsp70 chaperones Ssa1/2 and disrupts the interaction between endogenous Hsf1 and Ssa1/2 in co-IP assays ( Figure 2—figure supplement 1B , Figure 2—source data 1 ) . Domain truncation analysis of the decoy revealed that the C-terminal activation domain of Hsf1 is both necessary and sufficient to activate endogenous Hsf1 , while the N-terminal activation domain is dispensable ( Figure 2—figure supplement 1C , D ) . Thus , the decoy is not merely a UP , but rather functions as a specific activator of Hsf1 , likely by titrating away Hsp70 via its C-terminal activation domain . 10 . 7554/eLife . 18638 . 007Figure 2 . Prediction and validation of synthetic perturbations to the Hsf1-Hsp70 feedback loop . ( A ) Cartoon schematic of activation by overexpressing full length Hsf1 . Hsf1 can be expressed at many different levels by titrating the concentration of estradiol in the media ( See Figure 2—figure supplement 1 and Materials and methods ) . The Hsf1 domain architecture is displayed below . The DNA binding domain ( DBD ) is between N- and C-terminal activation domains ( NTA and CTA ) . ( B ) Cartoon schematic of activation via overexpression of the Hsf1 decoy . The decoy domain architecture is displayed below . ( C ) Simulation of the HSE-YFP reporter as a function of the expression level of full length Hsf1 or the decoy . ( D ) Experimental measurement of the HSE-YFP reporter by flow cytometry in cells expressing full length Hsf1 , the decoy or mKate alone across a dose response of estradiol . Cells were monitored following growth in the presence of the indicated concentrations of estradiol for 18 hr . Data points are the average of median YFP values for three biological replicates , and error bars are the standard deviation . See Materials and methods for assay and analysis details . DOI: http://dx . doi . org/10 . 7554/eLife . 18638 . 00710 . 7554/eLife . 18638 . 008Figure 2—source data 1 . Table of peptide counts from proteins identified in decoy IP/MS experiments with decoy-3xFLAG-V5 and Hsf1-3xFLAG-V5 as bait . DOI: http://dx . doi . org/10 . 7554/eLife . 18638 . 00810 . 7554/eLife . 18638 . 009Figure 2—figure supplement 1 . Overexpression of a decoy of Hsf1 activates endogenous Hsf1 . ( A ) Anti-FLAG western blot of cells expressing Hsf1-FLAG-V5 under the control of an estradiol-inducible promoter . Cells were incubated for four hours across a two-fold dilution series of estradiol ( from 512 nM down to 1 nM ) and the Hsf1 expression level was compared to expression from the endogenous promoter . ( B ) Cells expressing YFP-Ubc9ts were induced to express the mKate2-labeled decoy for four hours with 512 nM estradiol and were either left at 25°C or shifted to 39°C for 15 min and imaged by spinning disc confocal microscopy . YFP-Ubc9ts forms aggregates during heat shock , while the decoy remains apparently soluble . ( C ) Schematic of decoy constructs for domain analysis . ( D ) Measurement of the HSE-YFP reporter as a function of estradiol for the constructs depicted in ( C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18638 . 009 While overexpression of either full length Hsf1 or the decoy activated the HSE-YFP reporter , neither was innocuous: both inhibited cell growth in a dose-dependent manner . Full length Hsf1 impaired growth 20-fold more than the decoy , and the decoy impaired growth three-fold more than mKate2 alone at the highest dose of estradiol ( Figure 3A ) . The growth impairment caused by Hsf1 overexpression was not the result of a specific cell cycle arrest , as the remaining cells displayed asynchronous cell cycle stages ( Figure 3—figure supplement 1A , B ) . 10 . 7554/eLife . 18638 . 010Figure 3 . Hsp70 and Hsp40 suppress Hsf1 overexpression . ( A ) Cells expressing full length Hsf1 , the decoy or mKate alone were assayed for growth by flow cytometry following 18 hr of incubation with the indicated doses of estradiol . Data points are the average of normalized cell count values for three biological replicates , and error bars are the standard deviation . See Materials and methods for assay and analysis details . ( B ) Dilution series spot assays in the absence and presence of galactose to monitor growth of cells expressing full length Hsf1 from the GAL1 promoter . Ydj1 ( Hsp40 ) , Ssa2 ( Hsp70 ) , Hsc82 ( Hsp90 ) and combinations thereof were expressed from strong Hsf1-independent promoters and assayed for their ability to rescue the growth defect caused by Hsf1 overexpression . ( C ) Cells expressing an extra copy of Hsp70 and Hsp40 were assayed for growth ( left panel ) and for induction of the HSE-YFP reporter ( right panel ) as a function of the expression level of full length Hsf1 , the decoy or mKate by flow cytometry following 18 hr of incubation with the indicated doses of estradiol . The thick lines in the background are the reference curves for cells lacking extra chaperone expression ( taken from Figures 3A and 2D ) . ( D ) Cells expressing an extra copy of Hsp90 were assayed for growth ( left panel ) and for induction of the HSE-YFP reporter ( right panel ) . Reference curves are depicted as above . DOI: http://dx . doi . org/10 . 7554/eLife . 18638 . 01010 . 7554/eLife . 18638 . 011Figure 3—figure supplement 1 . Hsf1 overexpression does not lead to a specific cell cycle arrest and cannot be explained by induction of a gratuitous transcriptional program . ( A ) Flow cytometry analysis of DNA content of wild type cells and cells overexpressing Hsf1 from a galactose-inducible promoter following release from arrest with alpha factor for the indicated times . Wild type cells arrest in G1 and progress normally through the cell cycle . GAL-HSF1 cells fail to synchronize an show a broad distribution of DNA contents , indicating that no specific cell cycle stage is enriched . ( B ) Anti-tubulin immuno-fluorescence images of cycling wild type cells and cells GAL-HSF1 cells that had been grown in the presence of galactose for 12 hr . Wild type cells show G1 , S and G2 phases , while GAL-HSF1 cells show non-standard tubulin staining . ( C ) Cells expressing wild type Hsf1 or a chimera of Hsf1’s DNA binding domain and the VP16 activation domain ( DBDHsf1-VP16 ) were assayed for HSE-YFP induction by flow cytometry following 18 hr of incubation with the indicated doses of estradiol . Wild type Hsf1 data are from Figure 3D . ( D ) Relative growth of cells expressing wild type Hsf1 or DBDHsf1-VP16 across an estradiol dose response . DOI: http://dx . doi . org/10 . 7554/eLife . 18638 . 01110 . 7554/eLife . 18638 . 012Figure 3—figure supplement 2 . The Hsp40 Ydj1 is not required for the interaction between Hsf1 and Hsp70 . ( A ) IP/Western blot showing that Hsp70 binds to Hsf1 under basal conditions and transiently dissociates from Hsf1 during heat shock in both wild type and ydj1∆ cells . Western blots were probed with antisera recognizing Ssa1/2 ( top ) and an anti-FLAG antibody to recognize Hsf1 ( bottom ) . The numbers below the blots indicate the ratio of Ssa1/2:Hsf1 normalized to the ratio in wild type cells under basal conditions . The relative level of Hsp70 binding to Hsf1 and the dissociation dynamics during heat shock are altered in the ydj1∆ cells , but this is difficult to interpret due to the increased basal Hsf1 activity and reduced fold change in activity during heat shock in these cells ( see below ) . ( B ) Wild type and ydj1∆ cells expressing the HSE-YFP reporter were assayed for Hsf1 transcriptional activity in control and heat shock conditions by flow cytometry . Bars are the average of median YFP values for three biological replicates , and error bars are the standard deviation . Compared to wild type cells , ydj1∆ cells show increased basal Hsf1 activity and reduced fold change in activity during heat shock . DOI: http://dx . doi . org/10 . 7554/eLife . 18638 . 012 Since both full length Hsf1 and the decoy bound to Hsp70 and induced the transcriptional response , the growth inhibition could be due to either Hsp70 sequestration or the transcriptional induction itself through ‘squelching’ of the transcriptional machinery ( Gill and Ptashne , 1988 ) and/or gratuitous gene expression . To isolate the consequences of inducing the transcriptional program in the absence of stress , we constructed a synthetic fusion of the Hsf1 DNA binding and trimerization domains with a transcriptional activation domain derived from the herpes simplex virus protein 16 ( DBDHsf1-VP16 ) ( Sadowski et al . , 1988 ) , and placed this fusion under estradiol control . DBDHsf1-VP16 was a more potent inducer of the HSE-YFP reporter than full length Hsf1 but impaired growth equally to full length Hsf1 across the estradiol dose response ( Figure 3—figure supplement 1C , D ) . These data suggest that over-activating the Hsf1 transcriptional program impairs growth . Since Hsf1 overexpression impairs growth , inhibitors of Hsf1 activity should be genetic suppressors of the growth phenotype . To test if chaperones would behave as Hsf1 inhibitors , we placed Ssa2 ( Hsp70 ) , Hsc82 ( Hsp90 ) and Ydj1 ( Hsp40 ) under the control of strong Hsf1-independent promoters ( see Materials and methods ) ( Solís et al . , 2016 ) and assayed for their ability to suppress the growth defect of cells overexpressing Hsf1 . Interestingly , both Hsp70 and Hsp40 partially suppressed the growth inhibition caused by Hsf1 overexpression , while Hsp90 failed to provide any growth rescue ( Figure 3B ) . These data are consistent with prior reports in mammalian cells showing that overexpression of Hsp70 and Hsp40 attenuate Hsf1 activity ( Shi et al . , 1998 ) and in C . elegans showing that loss of Hsp70 results in >10 fold more Hsf1 activation than loss of Hsp90 ( Guisbert et al . , 2013 ) . Since Hsp40 chaperones deliver substrates and stimulate the ATPase activity of Hsp70 chaperones ( Kampinga and Craig , 2010 ) , Hsp40 may be enhancing the activity of endogenous Hsp70 to suppress Hsf1 overexpression . To determine if Hsp40 is required for efficient Hsp70 binding to Hsf1 , we deleted YDJ1 and performed a co-IP heat shock time course . We observed that Hsp70 was still able to robustly bind to Hsf1 under basal conditions and transiently dissociate during heat shock ( Figure 3—figure supplement 2A ) . However , basal Hsf1 activity was increased >3 fold as measured by the HSE-YFP reporter in ydj1∆ cells ( Figure 3—figure supplement 2B ) , indicating that there is likely to be greater total Hsp70 in these cells and complicating direct comparison of ydj1∆ and wild type cells . Thus , overexpression of Hsp40 may directly increase the ability for Hsp70 to bind to Hsf1 or generally improve global proteostasis such that there is more unoccupied Hsp70 available to repress Hsf1 . Co-expression of Hsp70 and Hsp40 afforded more rescue than either chaperone alone , but addition of Hsp90 did not enhance the suppression provided by Hsp70 alone ( Figure 3B ) . Hsp70 and Hsp40 together diminished the ability of Hsf1 and the decoy to inhibit growth and induce the HSE-YFP reporter across the estradiol dose response ( Figure 3C ) . By contrast , Hsp90 failed to rescue growth or reduce HSE-YFP induction ( Figure 3D ) . Taken together , these data support a model in which Hsp70 , assisted by Hsp40 , represses Hsf1 with little contribution from Hsp90 . The mathematical model along with the biochemical and genetic data implicates Hsp70 as the predominant regulator of Hsf1 . Given these results , we wondered if we could identify a role for phosphorylation in regulating Hsf1 activity . To address this question , we mapped Hsf1 phosphorylation sites in basal and heat shock conditions , performed site-directed mutagenesis of identified sites and assayed for changes in HSE-YFP levels ( see Materials and methods ) . Across 11 IP/MS experiments , we observed phosphorylation of 73 out of 153 total serine and threonine ( S/T ) residues in Hsf1 in at least one condition ( Figure 4—source data 1 ) . Strikingly , none of the 40 single point mutations we tested – whether mutated to alanine to remove the ability to be phosphorylated or to aspartate to mimic phosphorylation – showed significant differences in HSE-YFP levels compared to wild type Hsf1 in basal or heat shock conditions , even when we created sextuple mutants of clustered residues ( Figure 4—figure supplement 1A ) . Since Hsf1 activity was robust to so many mutations , we opted to remove phosphorylation altogether . We generated a synthetic HSF1 allele with all 153 S/T codons mutated to alanine . We chose to mutate all S/T codons rather than just the 73 identified phosphorylation sites in order to prevent utilization of alternative sites , ensuring that we completely removed the ability to be phosphorylated . HSF1 is an essential gene in yeast , and when we expressed this mutant as the only copy of HSF1 , unsurprisingly the cells were inviable . However , restoration of a single conserved serine ( S225 ) , which we found was required for Hsf1 to bind DNA in vitro , was sufficient to restore growth ( Figure 4—figure supplement 1B , C ) . While wild type Hsf1 robustly incorporated 32P during heat shock , this 152 alanine-substituted mutant ( termed Hsf1∆po4 ) showed no 32P incorporation during heat shock though it was stably expressed ( Figure 4A ) . Despite its 152 mutations , Hsf1∆po4 showed normal subcellular localization and unaltered DNA binding across the genome ( Figure 4—figure supplement 1D , E ) . 10 . 7554/eLife . 18638 . 013Figure 4 . Phosphorylation is dispensable for Hsf1 function but tunes the gain of its transcriptional activity . ( A ) 32P incorporated into wild type Hsf1-3xFLAG-V5 and Hsf1∆po4-3xFLAG-V5 during 30 min of heat shock ( upper panel ) . Hsf1-3xFLAG-V5 and Hsf1∆po4-3xFLAG-V5 were affinity purified by anti-FLAG IP , resolved by SDS-PAGE and phosphor-imaged . Western blot of total lysate from wild type Hsf1-FLAG-V5 and Hsf1∆po4-FLAG-V5 cells was probed with an anti-FLAG antibody; Hsf1∆po4-FLAG migrates faster than wild type Hsf1-FLAG ( lower panel ) . Schematics of the domain architecture and color code for wild type Hsf1 , Hsf1∆po4 and Hsf1PO4* . ( B ) Wild type , Hsf1∆po4 and Hsf1PO4* cells were monitored for growth by dilution series spot assays . Cells were incubated at the indicated temperatures for two days . ( C ) Wild type , Hsf1∆po4 and Hsf1PO4* cells expressing the HSE-YFP reporter were assayed for Hsf1 transcriptional activity in control and heat shock conditions by flow cytometry . Bars are the average of median YFP values for three biological replicates , and error bars are the standard deviation . See Materials and methods for assay and analysis details . ( D ) Genome-wide mRNA levels were quantified in basal conditions in wild type and Hsf1∆po4 cells by RNA-seq . Within each sample , relative expression levels for each mRNA ( gray dots ) are plotted as fragments per kilobase per million mapped reads ( FPKM ) . Hsf1-dependent genes ( HDGs ) are highlighted in purple . Source data are included as Figure 4—source data 2 . ( E ) Fold changes of each mRNA in heat shock conditions compared to basal conditions were calculated for wild type and Hsf1∆po4 cells and plotted against each other . Hsf1-dependent genes ( HDGs ) are highlighted in purple . Source data are included as Figure 4—source data 2 . ( F ) Genome-wide mRNA levels were quantified in basal conditions in wild type and Hsf1PO4* cells by RNA-seq ( gray dots ) . Hsf1-dependent genes ( HDGs ) are highlighted in orange . Source data are included as Figure 4—source data 2 . ( G ) Fold changes of each mRNA in heat shock conditions compared to basal conditions were calculated for wild type and Hsf1PO4* cells and plotted against each other . Hsf1-dependent genes ( HDGs ) are highlighted in orange . Source data are included as Figure 4—source data 2 . ( H ) Schematic of mutants with different numbers of aspartate ( D ) residues . 33 , 49 or 82 D residues were introduced in the CTA in the ∆po4 background . ( I ) Mutants depicted in ( H ) expressing the HSE-YFP reporter were assayed for Hsf1 transcriptional activity in control and heat shock conditions by flow cytometry as above . DOI: http://dx . doi . org/10 . 7554/eLife . 18638 . 01310 . 7554/eLife . 18638 . 014Figure 4—source data 1 . Table of Hsf1 phosphorylation sites identified in Hsf1-3xFLAG-V5 IP/MS IP/MS experiments in various conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 18638 . 01410 . 7554/eLife . 18638 . 015Figure 4—source data 2 . Table of genome wide transcript levels as measured by RNA-seq under basal ( 30°C ) and heat shock conditions ( 30 min at 39°C ) in wild type , Hsf1∆po4 and Hsf1PO4* cells . Values are FPKM . DOI: http://dx . doi . org/10 . 7554/eLife . 18638 . 01510 . 7554/eLife . 18638 . 016Figure 4—figure supplement 1 . Mutational analysis of Hsf1 phosphorylation sites reveals a single essential serine and allows for generation of a phospho-mimetic . ( A ) Hsf1 mutants lacking or mimicking single or clustered phosphorylation sites were assayed for activity in basal and heat shock conditions by measuring the HSE-YFP reporter by flow cytometry . The average of the median of the YFP distribution of three replicates of each mutant is plotted and the error bars represent the standard deviation . The ∆NTA and ∆CTA mutants are known to be hyperactive and impaired , respectively ( Sorger , 1990 ) , and serve as positive controls for altered activity . ( B ) Electrophoretic mobility shift assay showing that recombinant full-length wild type Hsf1 efficiently binds to and shifts HSE-containing DNA . The shift can be reverted with competition with excess unlabeled HSE . However , mutations S225A , which removes the hydroxyl group , and S225D , which partially mimics a phosphate group , reduce DNA binding and thus diminish the shift . The same amount of total Hsf1 was loaded in each lane , and the percent of labeled HSE shifted was quantified . ( C ) S225 is the only essential serine in Hsf1 . Mutation of S225 to alanine renders cells inviable ( top right plate ) . Restoration of S225 as the only serine in a mutant with all the other 152 S/T residues mutated to alanine rescues growth . hsf1∆ cells bearing a URA3-marked copy of wild type HSF1 on a plasmid ( pRS316-HSF1 ) and transformed with the indicated Hsf1 mutant were streaked on 5-FOA plates and incubated at 30°C for two days . ( D ) Fluorescent microscopy images of wild type Hsf1 , Hsf1∆po4 and Hsf1PO4* under basal conditions tagged at their C-termini with YFP showing that all localize to the nucleus . ( E ) ChIP-seq data ( reads per million mapped reads , RPM ) for Hsf1 and hsf1∆PO4 under basal conditions plotted along the first 150 kb of chromosome XII . DOI: http://dx . doi . org/10 . 7554/eLife . 18638 . 016 In addition to supporting growth in basal conditions , Hsf1∆po4 allowed cells to grow at elevated temperature , albeit with a deficit compared to wild type cells ( Figure 4B ) . In accordance , Hsf1∆po4 induced the HSE-YFP reporter in response to heat shock to 75% of the level induced by wild type Hsf1 ( Figure 4C ) . Global analysis of mRNA levels by deep sequencing ( RNA-seq ) revealed that wild type and Hsf1∆po4 cells were highly correlated across the transcriptome in basal conditions ( R2 = 0 . 98 , Figure 4D , Figure 4—source data 2 ) . However , while the correlation remained robust in response to heat shock ( R2 = 0 . 86 ) , the subset of genes that most strongly depend on Hsf1 for their transcription ( HDGs , for Hsf1-dependent genes ) ( Solís et al . , 2016 ) fell below the correlation axis ( Figure 4E , Figure 4—source data 2 ) . These data suggest that Hsf1 phosphorylation is required for its full potency as a transcriptional activator . Since phosphorylation appeared to be necessary to fully activate Hsf1 during heat shock , we wondered if it would be sufficient to drive increased transcription in the absence of stress . To test this , we generated a synthetic mutant , termed Hsf1PO4* , in which all 116 S/T residues that fall outside of the DNA binding and trimerization domains were replaced with aspartate to mimic constitutive hyper-phosphorylation ( Figure 4A ) . Indeed , when expressed as the only copy of Hsf1 in the cell , Hsf1PO4* activated the HSE-YFP reporter to a level seven-fold higher than wild type Hsf1 in basal conditions ( Figure 4C ) . Despite its high basal activity , Hsf1PO4* was able to induce the HSE-YFP even further during heat shock , suggesting its activity is still restrained by Hsp70 binding in basal conditions ( Figure 4C ) . RNA-seq analysis corroborated the hyperactivity of Hsf1PO4* in basal conditions as well as its heat shock inducibility ( Figure 4F , G ) . However , while Hsf1PO4* cells grew comparably to wild type cells at 30°C , they showed marked growth inhibition at 37°C ( Figure 4B ) . Thus , rather than protecting cells against proteotoxic stress , hyper-activation of Hsf1 impaired growth during stress . Since Hsf1PO4* contained 116 aspartate substitutions , we wondered if bulk negative charge was driving enhanced transcriptional activation . To test if the number of negatively charged residues would be proportional to transcriptional activity , we generated mutants with 33 , 49 or 82 aspartate substitutions in an otherwise Hsf1∆po4 background ( Figure 4H ) . Thus , there is no opportunity for heat shock-induced phosphorylation . Remarkably , we observed a direct correspondence between total negative charge and the transcriptional output ( Figure 4I ) . These data suggest that increasing Hsf1 phosphorylation positively tunes its transcriptional activity . The observation that Hsf1PO4* retained its heat shock inducibility suggested that phosphorylation plays little or no role in its regulation by Hsp70 . To test this , we deployed the decoy activation assay described above ( Figure 2B ) . Since activation of Hsf1 by the decoy depends on titration of Hsp70 away from endogenous Hsf1 , we wondered if the phosphorylation state of the decoy would affect its ability to induce the HSE-YFP reporter . We generated additional decoy mutants derived from Hsf1∆po4 and Hsf1PO4* to remove or mimic constitutive phosphorylation ( Figure 5A ) . Like the wild type decoy , these constructs contained mKate2 in place of the DNA binding and trimerization domains and were expressed across a dose response of estradiol . The wild type , ∆po4 and PO4* decoys displayed superimposable HSE-YFP induction profiles with matching slopes as a function of absolute expression level ( Figure 5B ) . Thus , neither removing nor mimicking phosphorylation altered the activity of the decoy , suggesting that phosphorylation does not affect the ability of Hsf1 to bind to Hsp70 . 10 . 7554/eLife . 18638 . 017Figure 5 . Hsp70 binding and phosphorylation are uncoupled Hsf1 regulatory mechanisms . ( A ) Schematic cartoon of decoy constructs based on wild type Hsf1 ( WT , black ) , Hsf1∆po4 ( ∆po4 , purple ) and Hsf1PO4* ( PO4* , orange ) . The various decoys will activate endogenous Hsf1 in proportion to their affinity for Hsp70 . ( B ) Measurement of the HSE-YFP reporter by flow cytometry in cells expressing decoy constructs derived from wild type Hsf1 , Hsf1∆po4 or Hsf1PO4* as a function of the expression level of each decoy ( mKate fluorescence ) . Data points are the average of median YFP and mKate values for three biological replicates , and error bars are the standard deviation . See Materials and methods for assay and analysis details . The slope of the input-output curves are plotted ( inset ) . ( C ) IPs of recombinant proteins were performed with 3xFLAG-rSsa2 as bait and analyzed by Western blot . rHsf1 was pre-incubated with ATP alone or in the presence of ATP and either active casein kinase II ( CKII ) or boiled CKII . Blots were probed with an anti-FLAG antibody to recognize recombinant rSsa2 ( top ) and with an anti-HIS antibody to recognize recombinant rHsf1 ( bottom ) . The numbers below the blots indicate the normalized ratio of Hsf1/Ssa2 . ( D ) Schematic cartoon of full-length overexpression constructs for wild type Hsf1 ( WT , black ) , Hsf1∆po4 ( ∆po4 , purple ) and Hsf1PO4* ( PO4* , orange ) , each with mKate2 fused to its C-terminus . The full-length constructs will activate the HSE-YFP reporter in proportion to their transcriptional activity . ( E ) Measurement of the HSE-YFP reporter by flow cytometry in cells expressing full length constructs of wild type Hsf1 , Hsf1∆po4 or Hsf1PO4* tagged at their C-termini with mKate2 as a function of expression level as in B . ( F ) ChIP-seq for Med4-3xFLAG-V5 , a component of the Mediator complex , in basal conditions in wild type Hsf1 , Hsf1∆po4 , and Hsf1PO4* cells at the HSP82 locus . Wild type Hsf1-3xFLAG-V5 ChIP-seq was also performed in basal conditions ( gray filled curve ) . See Figure 5—figure supplement 1 for more loci . DOI: http://dx . doi . org/10 . 7554/eLife . 18638 . 01710 . 7554/eLife . 18638 . 018Figure 5—figure supplement 1 . Hsf1PO4* recruits Mediator more efficiently than wild type Hsf1 or Hsf1∆po4 . ( A ) DNA binding profiles were determined by ChIP-seq for Med4-FLAG-V5 , a component of the Mediator complex , in basal conditions in wild type , Hsf1∆po4 , and Hsf1PO4* cells . Wild type Hsf1-FLAG-V5 ChIP-seq was also performed in basal conditions . Med4 enrichment in Hsf1 , Hsf1∆po4 , and Hsf1PO4* cells was plotted as reads per million mapped reads ( RPM ) at the SSA1 locus ( left y-axis ) . Hsf1 enrichment was plotted as RPM ( right y-axis ) . The lower track shows the position of the SSA1 open reading frame . ( B ) As in ( A ) , but at the SSA4 locus . DOI: http://dx . doi . org/10 . 7554/eLife . 18638 . 018 To directly test whether Hsf1 phosphorylation modulates its interaction with Hsp70 , we phosphorylated Hsf1 in vitro with purified casein kinase II ( CKII ) and monitored its binding to 3xFLAG-Ssa2 . CKII phosphorylated Hsf1as evidenced by its mobility shift , yet 3xFLAG-Ssa2 retained the ability to pull down Hsf1 in an anti-FLAG IP comparably to non-phosphorylated controls ( Figure 5C ) . These data indicate that phosphorylation does not preclude binding between Hsf1 and Hsp70 . Since phosphorylation does not disrupt the interaction with Hsp70 , we suspected that it serves to increase the ability of Hsf1 to directly activate transcription . To test this , we tagged full-length wild type Hsf1 , Hsf1∆po4 and Hsf1PO4* with mKate2 and monitored their ability to activate the HSE-YFP reporter when expressed as the only versions of Hsf1 in the cell as a function of estradiol in the absence of heat shock ( Figure 5D ) . In contrast to the indistinguishable transcriptional responses of the decoys , these three versions of full length Hsf1 displayed distinct induction curves as a function of their absolute expression level ( Figure 5E ) . Hsf1∆po4 was a slightly weaker activator than wild type Hsf1 , while Hsf1PO4* was by far the strongest activator with the steepest slope ( Figure 5E ) . We hypothesized that Hsf1PO4* exerts its increased activity by more efficiently recruiting the transcriptional machinery . Since Hsf1 is known to engage RNA polymerase II via interaction with the Mediator complex ( Kim and Gross , 2013 ) , we performed ChIP-seq of the Mediator subunit Med4 in wild type , Hsf1∆po4 and Hsf1PO4* cells . Indeed , recruitment of Med4 to the promoters of HDGs , such as HSP82 , SSA4 and SSA1 , was very clearly a function of phosphorylation state , with Hsf1PO4* recruiting the most Med4 , wild type Hsf1 recruiting less , and finally Hsf1∆PO4 showing very little recruitment ( Figure 5F , Figure 5—figure supplement 1 ) . The mathematical model , which ignored phosphorylation , failed to account for the persistent Hsf1 activity that followed Hsp70 re-association during the heat shock time course ( Figure 1D ) . To determine if Hsf1 phosphorylation could explain this sustained activity , we monitored Hsf1 phosphorylation kinetics via mobility shift of Hsf1-3xFLAG-V5 throughout a heat shock time course by Western blot . Rather than coinciding with the dissociation of Hsp70 , which occurs within the first five minutes following temperature upshift ( Figure 1A ) , Hsf1 phosphorylation peaked after 20 min and was maintained out to at least two hours ( Figure 6A ) . Thus , the timing of Hsf1 phosphorylation matches the second phase of Hsf1 transcriptional activity ( Figure 6B ) . This result suggests that Hsf1 phosphorylation can drive increased transcription even when Hsf1 is bound to Hsp70 . Consistent with phosphorylation driving the second phase of Hsf1 transcriptional activity , Hsf1∆po4 completely lacked sustained activity at the later time points ( Figure 6B ) . Incorporating phosphorylation into the mathematical model as a 'positive gain' according to the experimentally determined kinetics ( Figure 6—figure supplement 1A , see Materials and methods ) allowed the model to recapitulate the activation dynamics of both wild type Hsf1 and Hsf1∆po4 throughout the heat shock time course ( Figure 6—figure supplement 1B ) . Taken together , the data presented here support a model in which Hsf1 integrates negative feedback regulation by Hsp70 and positive fine-tuning by phosphorylation to dynamically control the heat shock response ( Figure 6D ) . 10 . 7554/eLife . 18638 . 019Figure 6 . Hsf1 phosphorylation accounts for sustained Hsf1 transcriptional activity during heat shock . ( A ) Western blot of Hsf1 phosphorylation , indicated by its mobility shift , over time following temperature upshift . ( B ) The HSE-YFP reporter was measured by flow cytometry in wild type Hsf1 and Hsf1∆po4 cells over time following upshift to 39°C . Data points are the average of median YFP values for three biological replicates , and error bars are the standard deviation . See Materials and methods for assay and analysis details . The sustained activation attributable to phosphorylation is depicted as the orange segment of the wild type curve . ( C ) Cartoon schematic of the integrated phosphorylation/chaperone titration model of Hsf1 regulation . Phosphorylation ( PO4 ) increases the transcriptional activity of Hsf1 in a manner uncoupled from the Hsp70 feedback loop . DOI: http://dx . doi . org/10 . 7554/eLife . 18638 . 01910 . 7554/eLife . 18638 . 020Figure 6—figure supplement 1 . Inclusion of the role of phosphorylation in the mathematical model of Hsf1 regulation . ( A ) Phosphorylation gain ( β ) is incorporated dynamically according to the kinetics determined experimentally . ( B ) Simulation of the HSE-YFP reporter over a heat shock time course in wild type Hsf1 and Hsf1∆po4 cells using a model that includes both the Hsp70 negative feedback loop and uncoupled activation gain control by phosphorylation ( see Materials and methods for details . ) DOI: http://dx . doi . org/10 . 7554/eLife . 18638 . 020 It has been suggested that the heat shock response operates as a feedback loop , in which Hsf1 activity is determined by the abundance of free chaperones ( Voellmy and Boellmann , 2007 ) . However , the elegance of this model has perhaps overshadowed the lack of data to support it . Here we provide multiple lines of evidence for a chaperone titration model in budding yeast . Specifically , we showed that the Hsp70 chaperone binds to Hsf1 in basal conditions , dissociates during the acute phase of heat shock , and subsequently re-associates at later time points , thus providing the first direct evidence of a dynamic Hsf1 ‘switch’ ( Figure 1A ) . Furthermore , we reconstituted the interaction between Hsp70 and Hsf1 in vitro with recombinant proteins and partially disrupted the complex with a hydrophobic peptide ( Figure 1A ) . We then constructed a minimal mathematical model of the Hsp70 feedback loop , and showed that it recapitulated the dynamics of Hsf1 transcriptional activity during heat shock ( Figure 1B–D ) . The model also correctly predicted that synthetic perturbations to the feedback loop , such as adding a ‘decoy’ of Hsf1 , would activate the endogenous Hsf1 response via Hsp70 titration ( Figure 2 ) . Finally , we provided independent genetic support for the model by showing that increased expression of Hsp70 and Hsp40 ( an Hsp70 co-factor ) suppresses the growth impairment caused by Hsf1 overexpression ( Figure 3 ) . Thus , biochemical , genetic and computational approaches converged to support a model in which Hsp70 and Hsf1 form a feedback loop that controls heat shock response activation . While our results are consistent with studies reporting biochemical and genetic interactions between Hsf1 and Hsp70 ( Abravaya et al . , 1992; Baler et al . , 1992 , 1996; Brandman et al . , 2012; Guisbert et al . , 2013; Ohama et al . , 2016; Shi et al . , 1998 ) , they are inconsistent with other reports that implicate Hsp90 as a major repressor of Hsf1 activity ( Brandman et al . , 2012; Duina et al . , 1998; Guo et al . , 2001; Zou et al . , 1998 ) . Biochemically , our inability to detect Hsp90 binding to Hsf1 could be due to the serial affinity purification strategy that we employed: if Hsp90 weakly associates with Hsf1 , we would likely lose the interaction during the two-step purification . However , the combination of the lack of biochemical evidence with the genetic result that Hsp90 overexpression is unable to suppress Hsf1 overexpression suggests a parsimonious explanation that Hsp90 simply does not repress yeast Hsf1 . Using our Hsp70-centric model , we can explain the genetic studies in which loss of Hsp90 function activates Hsf1 ( Brandman et al . , 2012; Duina et al . , 1998 ) by supposing that reduced Hsp90 leads to increased levels of unfolded proteins that titrate Hsp70 away from Hsf1 . Phosphorylation of Hsf1 , which is a hallmark of the heat shock response , is a second longstanding mechanism proposed to regulate Hsf1 ( Sorger and Pelham , 1988 ) . We identified 73 sites of phosphorylation on Hsf1 across various conditions ( Figure 4—source data 1 ) . Remarkably , however , Hsf1 retained its essential basal functionality and heat shock-induced activity in the complete absence of phosphorylation ( Figure 4 ) . Consistent with this observation , human Hsf1 also remained heat shock-inducible following mutation of a subset of its phosphorylation sites ( Budzynski et al . , 2015 ) . Despite its qualitative functionality in the absence of phosphorylation , Hsf1∆po4 was quantitatively impaired in its ability to induce its target genes during heat shock ( Figure 4E ) . Conversely , mimicking hyper-phosphorylation increased basal expression of the Hsf1 target regulon without disrupting its heat shock inducibility ( Figure 4F , G ) . Moreover , the number of phospho-mimetic residues correlated with transcriptional output ( Figure 4I ) . Thus , in contrast to prevailing models suggesting that phosphorylation is required for activation , we conclude that phosphorylation is not the switch that activates Hsf1; rather , phosphorylation is a positive fine-tuner that amplifies the transcriptional activity of Hsf1 . We propose that increased negative charge in the transcriptional activation domains of Hsf1 , endowed by phosphorylation or phospho-mimetic mutations , increases the ability of Hsf1 to recruit the Mediator complex and initiate transcription ( Figure 5F ) . In this manner , phosphorylation renders the activation domains into the 'acid blobs' that have long been associated with potent transcriptional activators ( Sigler , 1988 ) . Although this work yields a coherent synthesized model for Hsf1 regulation during heat shock , a number of interesting questions remain . These include identifying the molecular determinants of the Hsf1•Hsp70 interaction , defining the mechanism by which Hsp70 represses Hsf1 under basal conditions and whether this mechanism also applies to Hsf1 deactivation , determining potential roles for other chaperones ( such as Hsp40 , Hsp90 and chaperonins ) in Hsf1 regulation – particularly in different evolutionary lineages – and identifying the kinases that phosphorylate Hsf1 . On the latter , many kinases have been shown to regulate Hsf1 in various conditions , but the pathways that converge on Hsf1 have yet to be systematically defined . Given the low level of conservation of the Hsf1 activation domains ( Anckar and Sistonen , 2011 ) and the preponderance of S/T residues in putatively unstructured , solvent exposed regions of the protein , many kinases could potentially find a substrate site on Hsf1 . Defining the cohort of active kinases present in the nucleus in various conditions would provide a useful starting point to identify the kinases responsible for phosphorylating Hsf1 . Although both Hsp70 binding and phosphorylation contribute to regulating Hsf1 activity , they are independent events that exert orthogonal control . Hsf1 phosphorylation does not interfere with Hsp70 binding ( Figure 5B , C ) , and the kinetics of Hsf1 phosphorylation are delayed with respect to Hsp70 dissociation , with phosphorylation peaking after Hsp70 has re-associated with Hsf1 ( Figures 1A and 6A ) . These two distinct regulatory modes correlate with an immediate surge in Hsf1 transcriptional activity followed by a sustained moderate level of activity ( Figure 6B ) . Uncoupled regulation by chaperone binding and phosphorylation allows Hsf1 to function as an integration hub . Hsf1 can directly link to the proteostasis network by sensing the availability of Hsp70 as a proxy for protein folding conditions as well as respond to signals from multiple kinase pathways that convey information about other intracellular and extracellular conditions , such as oxidative stress and nutrient availability ( Hahn and Thiele , 2004; Yamamoto et al . , 2007 ) . In this manner , proteotoxic stress could activate Hsf1 without phosphorylation , and kinases could activate Hsf1 without chaperone dissociation . In this study , we employed a suite of approaches that allowed us to converge on a simple model of Hsf1 regulation and the ensuing dynamics of the heat shock response in budding yeast . However , given the poor conservation of the regulatory domains of Hsf1 , combined with the promiscuity of chaperone protein interactions and the ease with which phosphorylation sites are gained and lost through evolution , Hsf1 regulation has the potential to be rewired in different organisms and perhaps even in different cell types within the same organism ( Guisbert et al . , 2013 ) . Nevertheless , this work for the first time defines a regulatory scheme that synthesizes the roles of both canonical Hsf1 regulatory mechanisms , and as such can serve as both a precedent and template for the dissection of Hsf1 regulation in other cellular models . With a quantitative understanding of how Hsf1 is regulated when it is functioning properly , we can begin to unravel how it breaks down in neurodegenerative disorders and is usurped in cancer . Yeast strains and plasmids used in this work are described in Supplementary files 1 and 2 , respectively . All strains are in the W303 genetic background . PCR-mediated gene deletion and gene tagging was carried out as described ( Longtine et al . , 1998 ) . The serial immunoprecipitation procedure is described in more detail at Bio-protocol ( Zheng and Pincus , 2017 ) . 250 ml of cells were grown to OD600 = 0 . 8 in YPD media at 25°C with shaking . Basal condition samples were collected by filtration and filters were snap frozen in liquid N2 and stored at −80°C . For heat shocked samples , 250 ml of YPD pre-warmed to 53°C was added to the 250 ml culture to immediately raise the temperature to 39°C and cultures were incubated with shaking for the indicated times ( 5 , 15 , 30 or 60 min ) before being collected as above . Cells were lysed frozen on the filters in a coffee grinder with dry ice . After the dry ice was evaporated , lysate was resuspended in 1 ml IP buffer ( 50 mM Hepes pH 7 . 5 , 140 mM NaCl , 1 mM EDTA , 1% triton x-100 , 0 . 1% DOC , complete protease inhibitors ) , transferred to a 1 . 5 ml tube and spun to remove cell debris . Clarified lysate was transferred to a fresh tube and serial IP was performed . First , 50 µl of anti-FLAG magnetic beads ( 50% slurry , Sigma ) were added , and the mixture was incubated for 2 hr at 4°C on a rotator . Beads were separated with a magnet and the supernatant was removed . Beads were washed three times with 1 ml IP buffer and bound material eluted with 1 ml of 1 mg/ml 3xFLAG peptide ( Sigma-Aldrich , St . Louis , MO ) in IP buffer by incubating at room temperature for 10 min . Beads were separated with a magnet and eluate was transferred to a fresh tube . Next , 25 µl of anti-V5 magnetic beads ( 50% slurry , MBL International ) were added and the mixture was incubated for 2 hr at 4°C on a rotator . Beads were separated with a magnet and the supernatant was removed . Beads were washed five times with 1 ml IP buffer . Bound material was eluted by adding 75 µl of SDS-PAGE sample buffer and incubating at 95°C for 10 min . Beads were separated with a magnet and sample was transferred to a fresh tube for analysis . Control samples included an untagged strain and a strain expressing GFP-3xFLAG-V5 . Mass spectrometry analysis was performed at the Whitehead Proteomics Core Facility . IP eluates were digested with trypsin and analyzed by liquid chromatography ( NanoAcuity UPLC ) followed by tandem mass spectrometry ( Thermo Fisher LTQ ) . Mass spectra were extracted and analyzed by MASCOT searching the yeast proteome modified to include the 3xFLAG-V5 tagged bait proteins with a fragment ion mass tolerance of 0 . 8 Da and a parent ion tolerance of 20 PPM . Proteins were identified by Scaffold v4 . 4 . 1 to validate peptide and protein IDs . Peptides were accepted with a confidence of >95% and protein IDs were accepted only if they could be established at99% confidence and contained at least two peptides . Proteins that contain indistinguishable peptides were clustered . Quantification was performed for Hsf1 and Ssa1/2 using the 'top three' peptide total ion current method ( Grossmann et al . , 2010 ) . 15 µl of each IP sample was loaded into 4–15% gradient SDS-PAGE gels ( Bio-Rad ) . The gels were run at 25 mA for 2 hr , and blotted to PVDF membrane . After 1 hr blocking in Li-Cor blocking buffer , the membrane was incubated with anti-FLAG primary antibody ( SIGMA , F3165 ) for 1 hr , anti-Ssa1/2 ( gift from V . Denic ) or anti-HIS antibody ( all 1:1000 dilutions ) . The membranes were washed three times with TBST . The proteins were probed by anti-mouse-800 IgG ( Li-Cor , 926–32352 , 1:10000 dilution ) . The fluorescent signal scanned with the Li-Cor/Odyssey system . For the heat shock time course , cells expressing Hsf1-3xFLAG-V5 were grown to OD6000 . 8 in 25 ml YPD at 25°C . At time t = 0 , 25 ml of 53°C media was added to instantly bring the culture to 39°C and then the culture was incubated with shaking at 39°C . 5 ml samples were collected at each time point by centrifugation . Pellets were boiled with 2X SDS loading buffer for 10 min . Total protein concentration was measured by NanoDrop and an equal amount of each sample was loaded into 7 . 5% SDS-PAGE gel and otherwise processed as above . Full-length wild type HSF1 was cloned into pET32b with a C-terminal 6x-HIS tag and sequenced . Site-directed mutagenesis was performed to introduce the S225A and S225D mutations . Full length SSA2 was cloned into pET32b with an N-terminal 6xHIS-3xFLAG tag . The plasmids were transformed into BL21 ( DE3 ) cells ( Invitrogen ) . One liter of cells at OD600 = 0 . 4 were induced with 1 mM IPTG for 3 at 37°C . Cells were lysed by sonication , protein was purified with Ni-NTA agarose ( Qiagen ) and eluted with imidazole . 6xHIS-3xFLAG-Ssa2 and Hsf1-6xHIS were mixed with each at a final concentration of 5 µM in 100 µl of IP buffer alone , in the presence of 1 mM ATP or in the presence of 25 µM Aβ42 . Reactions were incubated at room temperature for 10 min . 25 µl of anti-FLAG magnetic beads were added and incubated for 15 min before magnetic separation . The unbound fraction was removed , beads were washed three times with 1 ml IP buffer and proteins were eluted by incubating at 95°C in SDS-PAGE sample buffer . All heat shock reporter assays were performed with untagged Hsf1 and mutants . For time course reporter inductions , 500 µl of OD600 = 0 . 1 cells were incubated at 39 with shaking on a thermo-mixer in 1 . 5 ml tubes . At designated time points , 50 µl samples were taken and cycloheximide was added at 50 µg/ml to arrest translation . Arrested cells were incubated at 30°C for 2 hr to allow fluorophores to mature . Samples were measured by flow cytometry , and population medians were computed with FlowJo . Each data point is the mean of three or four biological replicates . Error bars are the standard deviation . For basal versus heat shock experiments , we developed a protocol in which samples were pulsed with repeated 15 min heat shocks at 39°C followed by recovery at 25 for 45 min . As a control , a sample was kept at 25°C during the same time as heat-shocked pulses experiment . All experiments were performed using C1000 Touch Thermal Cycler ( Bio-Rad ) . Cells had been serially diluted five times ( 1:5 ) in SDC and grown overnight at room temperature . Cells in logarithmic phase were chosen the next morning for the experiment and 50 µl of each strain was transferred to two sets of PCR tubes and thermal cycled as described above . After that samples were transferred to 96-well plates with 150 μl of 1xPBS . HSE activity was measured using flow cytometry ( BD LSRFortessa ) and data analyzed using Flowjo as above . Cells bearing the HSE-YFP reporter and a chimeric transcription factor , GEM , consisting of the Gal4 DNA binding domain , the human estrogen receptor and the Msn2 activation domain ( Pincus et al . , 2014 ) were transformed with either the Hsf1 decoy , phospho-mutant decoys or mKate alone expressed from the GAL1 promoter and integrated as single copies in the genome . Full-length wild type Hsf1 and full-length C-terminally mKate tagged wild type Hsf1 and phospho-mutant constructs were also expressed under the control of the GAL1 promoter and integrated as single copies into the genome , but in a strain background that additionally contained a genomic deletion of HSF1 and a CEN/ARS URA3-marked plasmid bearing HSF1 . Upon integration of these constructs , the plasmid was counter-selected on 5-FOA . The GEM construct makes the GAL1 promoter ‘leaky’ enough that the cells are viable in the absence of estradiol and the presence of glucose . Cells were first grown to saturation overnight in synthetic media with dextrose and complete amino acids ( SDC ) . To assay for growth impairment and transcriptional activity as a function of expression level , cells in 10 different concentrations of estradiol ranging from 512 nM to 1 nM in SDC across a two-fold serial dilution series in deep well 96 well plates ( 5 µl of saturated culture diluted into 1 ml of each estradiol concentration ) . Each dose of estradiol was performed in triplicate . To prevent saturation of the cultures , each estradiol concentration was serially diluted 1:4 into media with the same concentration of estradiol . Following 18 hr of growth , cell counts , HSE-YFP levels and mKate levels were measured by flow cytometry by sampling 10 µl of each culture ( BD LSRFortessa equipped with a 96-well plate high-throughput sampler ) and the data were analyzed in FlowJo . Relative growth rates were calculated by dividing the number of cells in each concentration of estradiol by the 1 nM estradiol counts . Cells expressing YFP-Ubc9ts and the Hsf1 decoy were grown overnight in complete media with raffinose . Cells were induced with 2% galactose for 4 hr to induce expression of YFP-Ubc9ts and the decoy . One sample was untreated and the other was heat shocked for 15 at 39°C . 96 well glass bottom plates were coated with 100 µg/ml concanavalin A in water for 1 hr , washed three times with water and dried at room temperature . 80 µl of low-density cells were added to a coated well . Cells were allowed to settle and attach for 15 min , and unattached cells were removed and replaced with 80 µl SD media . Imaging was performed at the W . M Keck Microscopy Facility at the Whitehead Institute using a Nikon Ti microscope equipped with a 100× , 1 . 49 NA objective lens , an Andor Revolution spinning disc confocal setup and an Andor EMCCD camera . Yeast strains containing wild type or mutated versions of HSF1 as the only copy of the gene in the genome were grown overnight in YPD . They were diluted to an identical final OD600 = 0 . 3 in phosphate buffered saline ( 1xPBS ) and serially diluted 1:5 in 1xPBS . 3 . 5 µl of each diluted yeast culture was spotted on the appropriate plate . Photographs were taken after two days of growth at 30 or 37°C . Tubulin immunofluorescence was performed in the presence or absence of GAL-overexpressed Hsf1 following release from alpha factor arrest as described ( Kilmartin and Adams , 1984 ) . DNA content analysis was performed in the presence or absence of GAL-overexpressed Hsf1 following release from alpha factor arrest as described ( Hochwagen et al . , 2005 ) . Hsf1-3xFLAG-V5 was immunoprecipitated as described above following the appropriate treatment through the first ( anti-FLAG ) purification step . Rather than eluting with 3xFLAG peptide , samples were incubated at 95°C in SDS-PAGE sample buffer to elute Hsf1 . The samples were run on SDS-PAGE and stained with coomassie . All phosphorylation site identification was outsourced to Applied Biomics ( Hayward , CA ) . We sent them samples ( cut bands from coomassie-stained gels ) and received excel sheets with phospho-peptides identified , called sites , coverage stats , and neutral loss spectra . Site-directed mutagenesis was performed with QuickChange according to the manufacturer’s directions ( Agilent ) . En masse mutational analysis was possible because of gene synthesis . We ordered gBlocks from IDT containing regions of Hsf1 with all S/T codons mutated to alanine . The C-terminal portion required codon optimization in order to remove repetitive sequence to allow synthesis . Originally , restriction sites were introduced at the boundaries of the regions to enable cut-and-paste combinatorial cloning . Finally , all restriction sites were removed by assembling the fragments via Gibson assembly . Gibson assembly was performed as directed by the manufacturer ( NEB ) . Strains bearing Hsf1-3xFLAG-V5 or Hsf1∆po4-3xFLAG-V5 expressed under estradiol control were grown in YPD liquid media to OD600 = 0 . 5 . Then protein expression was induced with 1 µM estradiol for 2 . Cells were pelleted and washed with 50 ml SDC media without phosphate . Cells were finally resuspended in 15 ml SDC media without phosphate , and incubated at room temperature for 30 min . 50 μCi of 32P-orthophosphate was added into each culture and the cells were incubated for 15 min . The samples were heat shocked at 39°C for 30 min , harvested , and Hsf1 was IP’ed as above . All the protein was loaded into an SDS-PAGE gel . After blotting the proteins to the PVDF membrane , the signal was detected by FujiFilm BAS-2500 system . The same membranes were then blotted for total Hsf1 as described above . 5 ml of cells were grown to OD600 = 0 . 5 and treated with the designated condition . Cells were spun and pellets were snap frozen and stored at −80°C . Pellets were thawed on ice , and total RNA was purified via phenol/chloroform separation using phase lock tubes ( five prime ) followed by ethanol precipitation ( Pincus et al . , 2010 ) . Total RNA samples were submitted to the Whitehead Genome Technology Core where polyA + RNA was purified , fragmented and sequencing libraries were prepared with barcoding . 12 samples were multiplexed in each lane of an Illumina Hi-Seq 2500 and deep sequencing was performed . Reads were assigned by the barcode to the appropriate sample . Data was processed using a local version of the Galaxy suite of next-generation sequencing tools . Reads were groomed and aligned to the S . cerevisiae orf_coding reference genome ( Feb . 2011 ) using Tophat , transcripts were assembled and quantified using Cufflinks and fold changes were computed using Cuffdiff ( Trapnell et al . , 2012 ) . An oligo containing four repeats of the HSE was synthesized with and without a 3’ fluorescent probe ( IRDye800 ) by IDT . The reverse complement was also synthesized and the oligos were annealed by heating to 95 followed by room temperature cooling . Labeled dsDNA was prepared at 5 µM and unlabeled at 50 µM . 1 . 5 µg of each Hsf1 prep was added to 1 µl of labeled DNA or 1 µl of labeled DNA plus 1 µl of unlabeled DNA in 10 µl total volume . Reactions were incubated at room temperature for 5 min . 2 µl 6x DNA loading dye were added and samples were loaded into 4–20% TBE gels ( Bio Rad ) . Gels were run at 30 mA for 1 hr and scanned on the LiCor . Images were analyzed and % shifted oligo was quantified in ImageJ Casein kinase II ( CKII ) was used to phosphorylate rHsf1 in vitro as described by the manufacturer ( NEB ) . As a control CKII was boiled to denature and deactivate it prior to incubation with rHsf1 . 50 ml of cells were fixed with addition of 1% formaldehyde for 20 min at room temperature followed by quenching with 125 mM glycine for 10 min . Cells were pelleted and frozen in liquid N2 and stored at −80°C . Cells were lysed frozen in a coffee grinder with dry ice . After the dry ice was sublimated , lysate was resuspended in 2 ml ChIP buffer ( 50 mM Hepes pH 7 . 5 , 140 mM NaCl , 1 mM EDTA , 1% triton x–100 , 0 . 1% DOC ) and sonicated on ice 10 times using a probe sonicator ( 18W , 30 s on , one minute off ) . 1 ml was transferred to a 1 . 5 ml tube and spun to remove cell debris . Input was set aside , and a serial IP was performed . First , 25 µl of anti-FLAG magnetic beads ( 50% slurry , Sigma ) were added the mixture was incubated for 2 hr at 4°C on a rotator . Beads were separated with a magnet and the supernatant was removed . Beads were washed five times with 1 ml ChIP buffer ( 5 min incubations at 4°C between each wash ) and bound material eluted with 1 ml of 1 mg/ml 3xFLAG peptide ( Sigma ) in ChIP buffer by incubating at room temperature for 10 min . Beads were separated with a magnet and eluate was transferred to a fresh tube . Next , 25 µl of anti-V5 magnetic beads ( 50% slurry , MBL International ) were added and the mixture was incubated for 2 hr at 4°C on a rotator . Beads were separated with a magnet and the supernatant was removed . Beads were washed three times with ChIP buffer , followed by a high salt wash ( ChIP buffer +500 mM NaCl ) and a final wash in TE . Bound material was eluted with 250 µl TE +1% SDS by incubating at 65°C for 15 min . Beads were separated with a magnet and eluate was transferred to a fresh tube and incubated overnight at 65°C to reverse crosslinks . Protein was degraded by adding 250 µl 40 µg/ml proteinase K in TE ( supplemented with GlycoBlue ) and incubating at 37°C for 2 hr . DNA fragments were separated from protein by adding 500 µl phenol/chloroform/isoamyl alcohol ( 25:24:1 ) , and the aqueous layer was added to a fresh tube . 55 µl of 4M LiCl was added along with 1 ml of 100% EtOH , and DNA was precipitated at −80°C overnight . DNA was pelleted by spinning for 30 min at 4°C and resuspended in 50 µl TE . Sequencing libraries were prepared by the WIGTC , and sequenced on the Illumina Hi-Seq 2500 .	Proteins are strings of amino acids that carry out crucial activities inside cells , such as harvesting energy and generating the building blocks that cells need to grow . In order to carry out their specific roles inside the cell , the proteins need to “fold” into precise three-dimensional shapes . Protein folding is critical for life , and cells don’t leave it up to chance . Cells employ “molecular chaperones” to help proteins to fold properly . However , under some conditions – such as high temperature – proteins are more difficult to fold and the chaperones can become overwhelmed . In these cases , unfolded proteins can pile up in the cell . This leads not only to the cell being unable to work properly , but also to the formation of toxic “aggregates” . These aggregates are tangles of unfolded proteins that are hallmarks of many neurodegenerative diseases such as Alzheimer’s , Parkinson’s and amyotrophic lateral sclerosis ( ALS ) . Protein aggregates can be triggered by high temperature in a condition termed “heat shock” . A sensor named heat shock factor 1 ( Hsf1 for short ) increases the amount of chaperones following heat shock . But what controls the activity of Hsf1 ? To answer this question , Zheng , Krakowiak et al . combined mathematical modelling and experiments in yeast cells . The most important finding is that the ‘on/off switch’ that controls Hsf1 is based on whether Hsf1 is itself bound to a chaperone . When bound to the chaperone , Hsf1 is turned ‘off’; when the chaperone falls off , Hsf1 turns ‘on’ and makes more chaperones; when there are enough chaperones , they once again bind to Hsf1 and turn it back ‘off’ . In this way , Hsf1 and the chaperones form a feedback loop that ensures that there are always enough chaperones to keep the cell’s proteins folded . Now that we know how Hsf1 is controlled , can we harness this understanding to tune the activity of Hsf1 without disrupting how the chaperones work ? If we can activate Hsf1 , we can provide cells with more chaperones . This could be a therapeutic strategy to combat neurodegenerative diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "computational", "and", "systems", "biology" ]	2016	Dynamic control of Hsf1 during heat shock by a chaperone switch and phosphorylation
The neurophysiological processes underlying non-invasive brain activity measurements are incompletely understood . Here , we developed a connectome-based brain network model that integrates individual structural and functional data with neural population dynamics to support multi-scale neurophysiological inference . Simulated populations were linked by structural connectivity and , as a novelty , driven by electroencephalography ( EEG ) source activity . Simulations not only predicted subjects' individual resting-state functional magnetic resonance imaging ( fMRI ) time series and spatial network topologies over 20 minutes of activity , but more importantly , they also revealed precise neurophysiological mechanisms that underlie and link six empirical observations from different scales and modalities: ( 1 ) resting-state fMRI oscillations , ( 2 ) functional connectivity networks , ( 3 ) excitation-inhibition balance , ( 4 , 5 ) inverse relationships between α-rhythms , spike-firing and fMRI on short and long time scales , and ( 6 ) fMRI power-law scaling . These findings underscore the potential of this new modelling framework for general inference and integration of neurophysiological knowledge to complement empirical studies . Empirical approaches to characterizing the mechanisms that govern brain dynamics often rely on the simultaneous use of different acquisition modalities . These data can be merged using statistical models , but the inferences are constrained by information contained in the different signals , rendering a mechanistic understanding of neurophysiological processes elusive . Brain simulation is a complementary technique that enables inference on model parameters that reflect mechanisms that underlie emergent behavior , but that are hidden from direct observation ( Breakspear , 2017 ) . Brain network models are dynamical systems of coupled neural mass models for simulating large-scale brain activity; coupling is often mediated by estimations of the strengths of anatomical connections based on diffusion-weighted MRI data ( so-called structural connectivity or ‘connectomes’ ) . Here , we develop a novel type of brain network model , dubbed ‘hybrid model’ , where each subject’s EEG data is used to drive neural mass dynamics ( Figure 1 ) . In brief , we were able to use the resulting hybrid models to reproduce ongoing subject-specific fMRI time series over a period of 20 min and a variety of other empirical phenomena ( Figure 2 ) . In contrast to previous brain network models that used noise as input , hybrid models are driven by EEG source activity ( i . e . EEG sensor activity mapped onto cortical locations ) and therefore simultaneously incorporate structural and functional information from individual subjects ( Figure 1 ) . The injected EEG source activity serves as approximation of excitatory synaptic input currents ( EPSCs ) , which helps to increase the biological plausibility of generated model activity ( Buzsáki et al . , 2012; Haider et al . , 2016; Isaacson and Scanziani , 2011; Nunez and Srinivasan , 2006 ) . Individualized hybrid models yield predictions of ongoing empirical subject-specific resting-state fMRI time series ( Figure 3 ) . Additionally , several empirical phenomena from different modalities and temporal scales are reproduced: spatial topologies of fMRI functional connectivity networks ( Figure 4 ) , excitation-inhibition ( E/I ) balance of synaptic input currents , the inverse relationship between α-rhythm phase and spike-firing on short time scales ( Figure 5 ) , and the inverse relationship between α-band power oscillations and spike-firing , respectively fMRI oscillations , on long time scales ( Figure 6 ) , and fMRI power-law scaling ( Figure 7 ) . More importantly , our subsequent analysis of intrinsic model activity reveals neurophysiological processes that could explain how brain networks produce the aforementioned signal patterns ( Figures 5–7 ) . That is , simulation results not only predict ongoing subject-specific resting-state fMRI time series and several empirical phenomena observed with invasive electrophysiology methods , but more importantly , they also show how the network interaction of neural populations leads to the emergence of these phenomena and how they are connected across multiple temporal scales in a time scale hierarchy . Resting-state fMRI studies identified so-called ‘resting-state networks’ ( RSNs ) , which are widespread networks of coherent activity that spontaneously emerge across a variety of species in the absence of an explicit task ( Biswal et al . , 1995; Fox and Raichle , 2007; Raichle et al . , 2001 ) . Despite correlations between fMRI and intracortical recordings ( He et al . , 2008; Logothetis et al . , 2001 ) , EEG ( Becker et al . , 2011; Goldman et al . , 2002; Mantini et al . , 2007; Moosmann et al . , 2003; Ritter et al . , 2009 ) and magnetoencephalography ( Brookes et al . , 2011; de Pasquale et al . , 2010 ) the link between RSNs and electrical neural activity is not fully understood . A prominent feature of electrical neural activity are oscillations in the α-band , which is rhythmic activity in the 8 to 12 Hz frequency range first discovered by Hans Berger ( Berger , 1929 ) . A growing body of research indicates that changes in information processing , attention , perceptual awareness , and cognitive performance are accompanied by rhythmic modulation of α-power and phase ( Busch et al . , 2009; Klimesch , 1999; Mathewson et al . , 2009 ) . The observed inverse relationship between α-band activity and neural firing is central to hypotheses on its functional significance termed ‘gating by inhibition’ and ‘pulsed inhibition’ ( Jensen and Mazaheri , 2010; Klimesch et al . , 2007 ) . Interestingly , intracellular recordings showed that inhibitory events are inseparable from excitatory events , resulting in an ongoing excitation-inhibition balance ( E/I balance ) ( Isaacson and Scanziani , 2011; Okun and Lampl , 2008 ) . The significance of the α-rhythm is underscored by strong negative correlations between ongoing α-band power fluctuation and resting-state fMRI amplitude fluctuation ( de Munck et al . , 2008; Feige et al . , 2005; Goldman et al . , 2002; Moosmann et al . , 2003 ) . Lastly , despite wide-spread interest in critical dynamics ( Bak , 2013 ) , the key determinants of emergent power-law scaling , a signal pattern that is ubiquitous in nature and commonly observed in neural activity , are unclear ( Beggs and Timme , 2012; Marković and Gros , 2014 ) . To illustrate the potential of this framework for inference of neurophysiological processes , we show inferred mechanisms for three different empirical phenomena and how they relate to other well-established neural signal patterns ( Figure 2 ) . Upon finding that the hybrid model predicts fMRI activity , we first sought to identify how injected EEG drove the prediction of subject-specific fMRI time series . Analysis led us to a mechanism that transformed α-power fluctuations of injected EEG source activity into fMRI oscillations . The identified mechanism may explain the empirically observed correlation between EEG and fMRI on the longer time scale of slow fMRI oscillations . Consequently , we asked how the inhibitory effect of increased α-band power was created on the faster time scale of α-phase fluctuations . Analysis led us to the identification of an inhibitory effect resulting from the interaction of postsynaptic current oscillations and local population circuitry . Interestingly , parameter space exploration showed that prediction quality decreased when long-range coupling was deactivated ( i . e . when the nodes of the long-range network were isolated from each other ) . Therefore , we interrogated the model for the influence of structural coupling on the emergence of fMRI oscillations and found that global coupling amplified brain oscillations in a frequency-dependent manner , amplifying slower oscillations more than faster oscillations , which facilitated the emergence of power-law scaling . Starting with fast time scale effects , our first model outcome accounts for the invasively observed inverse relationship between spike-firing and α-rhythm phase by identifying a mechanism that relates this phenomenon to ongoing E/I balance . The second model outcome posits a neural origin of fMRI RSN oscillations by identifying an explicit mechanism that transforms ongoing α-power fluctuations into slow fMRI oscillations , which also explains the empirically observed anti-correlations between α-power and fMRI time series . Our third model outcome indicates that scale invariance of fMRI power spectra results from self-reinforcing feedback excitation via long-range structural connectivity , which leads to frequency-dependent amplification of neural oscillations . In summary , our biophysically grounded brain model has the potential to test mechanistic hypotheses about emergent phenomena such as scale-free dynamics , the crucial role of excitation-inhibition balance and the haemodynamic correlates of α-activity . However , there is another perspective on this form of hybrid modeling . Because it uses empirical EEG data to generate predictions of fMRI responses , it can be regarded as a form of multimodal fusion under a generative model that is both physiologically and anatomically grounded . In addition , because we use connectivity constraints based on tractography , it also serves to fuse structural with functional data . The used brain network models are dynamical systems where individual brain areas are simulated by coupled neural mass models . Long-range coupling was weighted by heterogeneous strength estimates obtained from white-matter tractography , a method that estimates neural tracts from diffusion-weighted MRI data . The used neural mass models approximate the average ensemble behaviour of networks of spiking neuron models and were derived in a previous study ( Deco et al . , 2013 ) using a dynamic mean-field technique ( Deco et al . , 2008; Wong and Wang , 2006 ) . In contrast to previous brain network models that used noise as input , the neural mass models of our ‘hybrid’ model are driven by EEG source activity that was simultaneously acquired with fMRI ( Figure 1 ) . Simulation results predicted a considerable part of the variance of ongoing subject-specific resting-state fMRI time series ( Figure 3 ) and spatial network topologies , that is , fMRI functional connectivity ( Figure 4; functional connectivity is here defined as the pair-wise correlation matrix between region time series ) . Furthermore , fitted models reproduced a variety of empirical phenomena observed with EEG and invasive electrophysiology ( Figure 2 ) and , more importantly , simulation results revealed mechanistic explanations for the emergence of these phenomena ( Figures 5 , 6 and 7 ) . We constructed individual hybrid brain network models for 15 human adult subjects using each subject’s own structural connectomes and injected each with their own region-wise EEG source activity time courses that were acquired simultaneously with the fMRI data subsequently predicted . Using exhaustive searches , we tuned three global parameters for each of the 15 individual hybrid brain network models to produce the highest fit between each of the subject’s empirical region-average fMRI time series and corresponding simulated time series ( Figure 3—figure supplement 1 ) . Our motivation for choosing that parameter set that produced the highest correlation between simulated and empirical fMRI time series is based on our goal to infer the underlying ( but unobservable ) dynamics and parameters of the real system . This idea is based on the assumption that when the model optimally fits observable brain activity , then also the underlying unobservable brain activity is faithfully reproduced . The first parameter scales the global strength of long-range coupling between regions . The second and third parameters scale the strengths of EEG source activity inputs injected into excitatory and inhibitory populations , respectively . To better assess the quality of fMRI predictions , we compared hybrid model results with three control scenarios: ( i ) the original noise-driven brain network model , ( ii ) a variant of the hybrid model where the time steps of the injected EEG source activity time series were randomly permuted and ( iii ) a statistical model where the ongoing α-band power fluctuation of injected EEG source activity was convoluted with the canonical hemodynamic response function ( henceforth called α-regressor ) . The first two controls are brain network models and the third is inspired by traditional analyses of empirical EEG-fMRI data . The controls serve to exclude that the obtained correlations between simulated and empirical fMRI is a trivial outcome that would also be produced by the original noise-driven model or with random input time series . Visual inspection of example time series showed good reproduction of characteristic slow ( <0 . 1 Hz ) RSN oscillations by the hybrid model and the α-regressor ( albeit inverted for the latter ) , but poor reproduction of temporal dynamics in the case of noise and random permutations models ( Figure 3 ) . We compared the average correlation coefficients between all simulated and empirical fMRI time series between all four scenarios ( i . e . hybrid model and the three control setups ) . Predictions from the hybrid model correlated significantly better with empirical fMRI time series than predictions from the two random models and the α-regressor ( Figure 3b ) . For the hybrid model , five-fold cross-validation showed no significant difference of prediction quality between training and validation data sets ( two-tailed Wilcoxon rank sum test , p=0 . 71 , t = 0 . 54 , Cliff’s delta d = 0 . 0044 ) and between validation data sets and prediction quality for the full time series ( two-tailed Wilcoxon rank sum test , p=0 . 42 , t = −0 . 2 , Cliff’s delta d = −0 . 067 ) . To estimate the ability of the four scenarios to predict the time courses of different commonly observed RSNs we performed a group-level spatial independent component analysis ( ICA ) of the empirical fMRI data . Next , we computed average correlation coefficients between each subject-specific RSN time course and the model regions at the position of the respective RSN . As in the case of region-wise fMRI ( Figure 3b ) , correlation coefficients of the hybrid model were significantly larger than the control network models for most RSNs ( Figure 3d ) . The sliding-window analyses showed that prediction quality varied over time , regions and subjects: window-wise prediction quality was highly correlated with the standard deviation of RSN temporal modes ( Figure 3c , d ) . That is , the higher the variance contributed to overall fMRI activity by an RSN in a given subject and time window , the better the prediction of empirical fMRI , which might reflect increased synchrony of electrical activity ( see Discussion ) . As a consequence , epochs in the upper quartile of RSN s . d . s were significantly better predicted than epochs in the lower quartile ( Figure 3d ) . In order to assess the subject-specificity of fMRI time series predictions , we correlated all simulation results ( i . e . for every subject and every tested parameter combination ) also with the empirical fMRI activity of all other subjects . We found that the maximum correlation coefficients over all tested parameters were significantly larger when empirical and simulated data sets belonged to the same subject compared to when they came from different subjects ( p<<0 . 01 , Wilcoxon rank sum test ) . Next , we compared the ability of all four setups to predict the spatial topology of empirical fMRI networks . In contrast to time series prediction , the α-regressor showed low correlations with empirical functional connectivity ( FC ) . Compared to the α-regressor , all three model-based approaches provided significantly better predictions of subjects’ individual long-epoch FC and short-epoch FC ( Figure 4 ) . Furthermore , hybrid model simulation results correlated significantly better with empirical network topology than predictions obtained from the noise-driven model ( Figure 4a , b ) . Interestingly , correlations for hybrid and random permutation models were effectively the same , likely because the long-range network dynamics , which drive the emergence of FC by structural coupling , would be relatively preserved when permuting injected activity . Prediction of group-average FC ( all pairwise FC values averaged over all subjects ) was better for the hybrid model compared to the α-regressor ( Figure 4c ) . After fitting the individual hybrid models for each of the 15 subjects , we analyzed the local population activity to infer neurodynamic mechanisms underlying predicted fMRI time series . We found that on the fast time scale of individual α-cycles ( ~100 ms ) the optimized hybrid model reproduced the inverse relationship between α-phase and firing rates observed in invasive recordings ( Haegens et al . , 2011 ) ( Figure 5a ) . To investigate these fast-acting dynamics related to α-phase , we computed grand average waveforms of modeled synaptic inputs , population firing rates , and synaptic gating time-locked to the zero-crossings of α-cycles . Resulting waveforms illustrate how the ongoing balancing of excitatory and inhibitory inputs generated the inverse relation between α-oscillations and neural firing ( Figure 5b ) . In hybrid models , individual subject’s ( Figure 5 , column I ) source activity ( column II ) is used as an approximation of EPSCs ( column III ) . As a result of optimizing the three model parameters , EPSCs dominated the sum of synaptic input current to inhibitory populations ( column IV ) . Consequently , inhibitory populations’ ( column V ) firing rates ( column VI ) and synaptic gating ( column VII ) closely followed the shape of EPSCs . Because of the monotonic relationship between input currents and output firing rates ( defined by Equation 3 and 4 ) , the waveform of inhibitory firing rates and synaptic gating also closely followed injected EPSCs . As increased input to inhibitory populations leads to increased inhibitory effect and vice versa , resulting feedback inhibition waveforms ( IPSC , column III ) were inverted to EPSCs . Furthermore , the amplitude fluctuation of EPSCs and IPSCs was proportional . That is , stronger EPSCs preceded and helped to generate stronger IPSCs . In other words , excitation and inhibition were balanced during each cycle , which is in accordance with published electrophysiology results ( Haider et al . , 2016; Isaacson and Scanziani , 2011; Okun and Lampl , 2008; Xue et al . , 2014 ) . Consequently , IPSCs peaked during the trough of the α-phase and were lowest during the peak of the α-phase . Fitting the models to fMRI activity resulted in a biologically plausible ratio of EPSCs to IPSCs ( Xue et al . , 2014 ) , with IPSC amplitudes being about three times larger than EPSC amplitudes ( compare left axes of EPSC and IPSC plots ) . Because IPSCs have dominated excitatory population inputs , excitatory populations’ firing rates showed a similar shape as IPSCs , that is , they peaked during the trough of the α-cycle and fell to their minimum during the peak of the α-cycle , thereby reproducing the empirical relationship between α-cycle and firing rate ( Haegens et al . , 2011 ) . Columns IV , VI and VII refer to Equations 1 , 3 and 5 ( excitatory population ) and 2 , 4 and 6 ( inhibitory population ) , respectively . In summary , the fast population activity underlying fMRI predictions showed a rhythmic modulation of firing rates on the fast time scale of individual α-cycles in accordance with empirical observations ( Haegens et al . , 2011 ) . Analyses revealed that periodically alternating states of excitation and inhibition resulted from the ongoing balancing of EPSCs by feedback IPSCs , which explains α-phase-related neural firing . Similar to intracranial recordings in monkey ( Haegens et al . , 2011 ) , we found that increased α-power of injected EEG source activity was accompanied by decreased firing rates ( Figure 6—figure supplement 1 ) . Furthermore , we also observed the empirically observed inverse relationship between α-power and fMRI amplitude ( Goldman et al . , 2002; Moosmann et al . , 2003 ) in our empirical data in the form of negative correlations between the α-regressor and fMRI activity ( Figure 3 ) . Our findings raised the question what physiological mechanism led to this inverse relationship between α-power and firing rate , respectively , fMRI amplitude . We therefore analyzed model activity on the longer time scale of α-power fluctuations . To isolate the effects of α-waves from other EEG rhythms , we replaced the injected EEG-source activity in the 15 individual hybrid models with artificial α-activity ( Figure 6a , column I ) and simulated all 15 hybrid models using the single parameter set that previously generated the highest average fMRI time series prediction quality ( Figure 3—figure supplement 1 ) . Injected activity consisted of a 10 Hz sine wave that contained a single brief high-power burst in its center in order to allow for model activity to stabilize for sufficiently long phases before and after the high-power burst . After simulation , we computed grand average waveforms of model state variables over all simulated region time series and found that input currents , firing rates , synaptic activity and fMRI activity of excitatory populations decreased in response to the α-burst ( Figure 6a ) . Notably , this behavior emerged despite the fact that injected activity ( column I ) was centered at zero , that is , positive and negative deflections of input currents were balanced . The reason for the observed asymmetric response to increasing input α-power levels originated from inhibitory population dynamics: while positive deflections of α-cycles generated large peaks in ongoing firing rates of inhibitory populations , negative deflections were bounded by 0 Hz ( column V ) . Because of this rectification of high-amplitude negative half-cycles , average per-cycle firing rates of inhibitory populations increased with increasing α-power . As a result , also feedback inhibition ( IPSC , column II ) had increased for increasing α-power , which in turn led to increased inhibition of excitatory populations , decreased average firing rates , synaptic gating variables ( column VI ) and ultimately fMRI amplitudes ( column VII ) . We next analyzed the relationship between α-power fluctuations and fMRI oscillations . We generated artificial α-activity consisting of a 10 Hz sine wave that was amplitude modulated by slow oscillations ( cycle frequencies between 0 . 01 and 0 . 03 Hz ) and injected it into the hybrid models of all subjects ( Figure 6b , column I ) . As in the previous example , inhibitory populations filtered negative α-deflections during epochs of increased power ( column V ) . This half-wave rectification led to a modulation of average per-cycle firing rates in proportion to α-power . Consequently , the power modulation of the injected α-oscillation was introduced as a new slow frequency component into the resulting time series . The activity of inhibitory populations can be compared to envelope detection used in radio communication for AM signal demodulation . The new frequency component introduced by half-wave rectification of α-activity modulated feedback inhibition ( IPSC , column II ) , which in turn modulated excitatory population firing rates ( column V ) . Furthermore , the resulting oscillation of firing rates was propagated to synaptic dynamics ( column VI ) where the large time constant of NMDAergic synaptic gating ( τNMDANMDA100 ms vs . τGABAGABA10 ms ) led to an attenuation of higher frequencies . The low-pass filtering property of the hemodynamic response additionally attenuated higher frequencies such that in fMRI signals ( column VII ) only the slow frequency components remained , based on the assumption that neurovascular coupling was mediated exclusively by excitatory synaptic activity . To restate: α-power fluctuation introduced an inverted slow modulation of firing rates and synaptic activity; the low-pass filtering properties of synaptic gating and hemodynamic responses attenuated higher frequencies such that only the slow oscillation remained in fMRI signals . To check whether this mechanism is robust to the choice of the frequency of the injected α-rhythm ( 10 Hz ) we simulated otherwise identical models for artificial α-waves at 9 Hz and 11 Hz frequencies and found qualitatively identical results: simulated fMRI and moving average firing rate time series of the 9 Hz and the 11 Hz model had correlation coefficients r > 0 . 99 with the respective time series of the 10 Hz model . In summary , we found that increased α-power led to increased feedback inhibition of excitatory populations introducing a slow modulation of population firing , which can explain the empirically observed anticorrelation between α-power and fMRI . Empirical fMRI power spectra follow a power-law distribution P ∝ f β , where P is power , f is frequency and β the power-law exponent . In accordance with systematic analyses of empirical data ( He , 2011 ) , average power spectra of our empirical fMRI data obeyed power-law distributions with exponent βemp = −0 . 82 ( Figure 7a and Figure 7—figure supplement 1 ) . We tested for the existence of power-law scaling in the time domain by using rigorous model selection criteria that overcome the limitations of simple straight-line fits to power spectra ( see Materials and methods; for illustration purposes straight-line fits are shown in Figure 7a and Figure 7—figure supplement 1 ) . Our previous results associated resting-state fMRI oscillations with electrical neural activity by identifying a neural mechanism that transforms α-band power fluctuations into fMRI oscillations ( Figure 6 ) . This mechanism suggests that EEG α-band power fluctuations are transformed into fMRI amplitude fluctuations . Therefore , it is surprising that the power spectra of wide-band and α-band EEG have considerably smaller negative exponents than empirical fMRI ( βα-band = −0 . 53 for α-power and βwide-band = −0 . 47 for wide-band power ) . However , in agreement with empirical fMRI , our simulated fMRI had a larger negative exponent ( βsim = −0 . 73 ) than the α-band power of the injected EEG source activity ( βα-band = −0 . 53 ) . This result implies that the power-law slope increased during the process that transformed electrical band-power fluctuations into fMRI amplitude fluctuations . Indeed , comparison of power spectra indicated that simulated fMRI had a higher negative exponent than EEG source-activity , because the power of slower oscillations increased relative to the power of faster oscillations ( Figure 7a and Figure 7—figure supplement 1 ) . That is , model dynamics transformed synaptic input activity such that the amplitude of output oscillations increased inversely proportional to their frequency . Interestingly , when long-range coupling was deactivated in simulations that used EEG source activity as input , the power-law exponent of simulated fMRI ( βsim_Gzero = −0 . 54 ) was close to the exponent of the α-band power time course of the injected EEG source activity ( βα-band = −0 . 53 ) . The effect was also visible when comparing our previous model simulations that used artificial α-activity ( Figure 6b ) , with simulations where long-range coupling was deactivated ( Figure 7b ) . When long-range coupling was deactivated , the amplitudes of fMRI oscillations were equally large for all oscillation frequencies ( Figure 7b , column VII ) . In contrast , when long-range coupling was activated , with everything else being identical , the amplitudes of slower fMRI oscillations were larger than the amplitudes of faster oscillations ( Figure 6b , column VII ) , although the amplitudes of the injected artificial α-band power oscillations were equally large for all oscillations ( Figures 6b and 7b , column I ) . When long-range coupling was present , the amplitudes of slow oscillations increased and the relationship between power and frequency of oscillations approximated the power-law exponent found in empirical fMRI power spectra ( Figure 7a ) . With everything else being identical , we concluded that long-range coupling was responsible for increasing the power of slower oscillations relative to faster oscillation . Comparison of the individual components of population inputs for activated ( Figure 6b , column II ) vs . deactivated ( Figure 7b , column II ) long-range coupling reinforced that the only difference in population inputs between both setups was the shape of long-range input . The amplitudes of long-range input oscillations ( Figure 6b , column II , green trace ) were inversely proportional to the band-power oscillation of injected artificial α-activity . In accordance with the effect of α-band power on population activity that we described earlier , long-range input increased when α-band power decreased , while during epochs of increased α-activity long-range coupling decreased . Consequently , this fluctuation of long-range input was coherent with the fluctuation of IPSCs that resulted from the fluctuation of α-band power , which further amplified the effect of α-band power on population activity . During epochs of low α-activity long-range coupling conveyed feedforward excitation that further reinforced the increasing of firing and synaptic gating . Because of this consensual modulation of input currents , total input currents were increased when α-band power was decreased , which resulted in larger amplitudes of firing rates , synaptic activity and fMRI . Due to the large time constant of excitatory synaptic gating ( τNMDA = 100 ms ) , long-range excitation decayed relatively slowly , which enabled excitatory activity to accumulate and perpetually reinforce within the long-range network . The period of time for which this feedforward excitation persisted was longer during slower oscillations than during faster oscillations . Consequently , synaptic activity ( column VI ) had more time to accumulate and was therefore larger during slower oscillations compared to faster oscillations . As a result , the amplitudes of excitatory population output ( columns V , VI , and VII ) reached higher values during slower oscillations than during faster oscillations when long-range coupling was activated ( Figure 6b ) . Accordingly , the power of slower oscillations , and therefore the slope of the power spectrum , increases in the case of long-range coupling . Note that this effect ( i . e . that slower oscillations reach higher amplitudes ) can already be observed in firing rates and synaptic gating time series , which excludes an influence of the hemodynamic forward model . In contrast , in the case of deactivated long-range coupling ( Figure 7b ) all amplitude peaks are approximately equal , which was the expected result , since the amplitude-peaks of the power modulation of injected α-activity were equally high by construction ( column I , orange trace ) . We asked how the relative strengths of white-matter excitation and feedback inhibition influence power-law scaling . In order to test how E/I balance affects power-law scaling , we varied the strength of long-range coupling and , also globally , the strength of feedback inhibition . That is , in contrast to our previous simulations , the strength of feedback inhibition was controlled by a single parameter for all inhibitory populations . The other parameters , that is the strengths of EEG source activity injected into excitatory and inhibitory populations , were kept fixated . Screening of individual parameter spaces showed that the power-law exponent of simulated fMRI depended on the balance of long-range excitation and local inhibition: the 2D distribution of the prediction quality of fMRI time series , functional connectivity and the power-law exponent showed a characteristic diagonal pattern . That is , increased long-range coupling required increased local feedback inhibition for producing best predictions of fMRI , FC and power-law exponents , which demonstrated the crucial role of E/I balance for the emergence of scale invariance and long-range correlations ( Figure 7—figure supplement 1 and Figure 7—figure supplement 2 ) . In this work , we describe a biophysically based brain network model that predicts a considerable part of subject-specific fMRI resting-state time series on the basis of concurrently measured EEG . Importantly , we show how this novel modelling approach can be used to infer the neurophysiological mechanisms underlying neuroimaging signals . Instead of mere reproduction of empirical observations , our central aim was to provide an integrative framework that unifies empirical data with theory of the nervous system in order to derive mechanisms of brain function underlying empirical observations across many scales . Clearly , the sequence of analyses and implicit hypothesis testing presented in this paper represents one of many lines of enquiry . The more general point made by this report is that our hybrid model can be used to both test hypotheses and to build hypotheses . In other words , many of the questions ( for which we offer answers ) only emerged during application of the model , which allowed us to pursue a particular narrative in understanding the genesis of different empirical phenomena . A key point of consideration is that the brain model was built from networks of generic neural population models that were constrained by empirical data , but not explicitly constructed to address specific reproduced phenomena . This is mirrored by the emergence of processes at considerably faster time scales than the subject-specific fMRI time series that were the target of the model fitting . It is important to point out that the inferred mechanisms constitute candidate hypotheses that require empirical falsification . The model-derived mechanisms make concrete predictions on the waveforms of different input currents , output firing rates , synaptic activities and fMRI signals , which can be empirically tested . Through ongoing integration of biological knowledge , falsification with empirical data and subsequent refinement , hybrid brain network models are intended to represent a comprehensive and increasingly accurate theory about large-scale brain function . The construction of hybrid brain network models and our major results are visualized in Video 1 . Hybrid models draw on empirically estimated EEG source activity to constrain synaptic input current dynamics . This approach is motivated by the need for a model that not only reproduces static features of brain activity , like functional connectivity , but that produces these features on the basis of biologically plausible time series dynamics . Underlying the approach is the consideration that commonly used fitting targets of BNMs , like FC or power spectral features ( e . g . slow BOLD oscillations , EEG α-peak ) , can in principle be generated by time series that are , except from the fitted features , not necessarily biologically plausible . For example , a wide range of waveforms can produce FC-like correlation patterns without necessarily having a biological underpinning . The goal was not to have an abstract converter that simply transforms EEG into an fMRI modality such as time series . Rather , EEG source activity serves as an approximation of ongoing subject-specific synaptic currents and parameter fitting is performed to tune the model to optimally explain empirical fMRI time series . In contrast to a simple ‘converter’ , our biophysical model is able to additionally capture other features of functional brain data not used for model fitting . We show that in fact the parameter space converges for different metrics of brain activity toward a single optimal subspace indicating validity of our model . In our approach , both functional datasets , EEG and fMRI , are fused within the framework of a biophysically grounded and structurally constrained model in order to optimally approximate the underlying ( but unobservable ) behavior and parameters of the real system . Models , by definition , omit features of the modeled system for the sake of simplicity , generality and efficiency . Adding degrees of freedom renders parameter spaces increasingly intractable and increases the risk of over-fitting . Injection of source activity is a way to systematically probe sufficiently abstract neural systems while maintaining biologically realistic behavior . Thereby , the approach aims to balance a level of abstraction that is sufficient to provide relevant insights , with being detailed enough to guide subsequent empirical study . It is not the goal of this approach to attain the highest possible fit between different imaging modalities at the cost of biological plausibility , which would be the case for abstract statistical models that do not relate to biological entities and therefore preclude the inference of neurophysiological knowledge . Here , imperfect reproduction of neural activity directly points to deficits in our understanding and conceptualization of large-scale brain structure and function , which to iteratively improve is the goal of this approach . We note that our comparison of prediction qualities of the hybrid model and the three control scenarios is not a result in the sense of formal model comparison where goodness of fit is assessed in light of model complexity . Rather , the informal comparison serves to better assess the hybrid model’s prediction quality in relation to the original model and the α-regressor . Although it was a priori clearly unlikely that the noise-driven model or the injection of time permuted EEG would correlate with the empirical time series , these controls serve to exclude that hybrid model correlations were obtained by a trivial mechanism potentially also present in noise models . Furthermore , to test whether it is the specific temporal sequence of time points in the injected activity that enabled fMRI prediction , we simulated the hybrid model’s response to permuted input time series . More importantly , these correlations enable us to show that although noise and permuted input do not produce noteworthy time series correlations , like the α-regressor , they nevertheless predict FC , while the hybrid model predicts both , time series and FC . Although the α-regressor makes noteworthy fMRI time series predictions , it yields low correlations with FC and , importantly , it is unable to predict the electric neuronal phenomena that have been reproduced with the hybrid model as it is not based on state variables that correspond to biological entities like the hybrid model . Hence , if during formal model comparison model complexity is penalized without accounting for the accuracy of the model to predict diverse data sets that originate from different modalities and that involve different kinds of metrics ( as the hybrid model does ) , then it is likely that the α-regressor is favoured , because it relies on zero free parameters while achieving similar time series prediction , despite the fact that it clearly has less power to concurrently explain the different sorts of neuronal phenomena explained by the hybrid model . In order to better estimate the relative quality of this kind of models , we are working on a theoretical framework that extends existing Bayesian system identification frameworks ( Friston et al . , 2003 ) to account for the concurrent prediction of the dynamics of different biological phenomena , data sets and metrics which goes beyond the scope of this study and shall be the subject of an additional publication . The idea of the hybrid approach is to test how biophysically based and structurally constrained models respond to biologically plausible synaptic input currents , comparable to in vivo or in vitro electrophysiology current injection experiments . However , it must be noted that the hybrid model is clearly limited by the fact that it is not an autonomous ( self-contained ) model of the brain , but depends on externally injected activity . Furthermore , EEG-based approximation of local EPSCs is limited by the coarse spatial resolution of EEG and the inability to disentangle local EPSCs from other currents that contribute to EEG as all currents in the brain superimpose at any given point in space to generate a potential at that location ( Buzsáki et al . , 2012 ) . This limitation would become apparent when the hybrid model is coupled with a forward model to predict EEG on the basis of the entire sum of input currents ( Equations 1 and 2 ) . However , when predicting EEG on the basis of local EPSCs only by application of the forward model , this would again yield the original EEG . Notably , EEG source activity can only be viewed as an approximation of EPSCs and it is unclear how EEG exactly relates to EPSCs , that is , to which extend this approximation reflects biological reality . Although theoretical considerations suggest that excitatory postsynaptic potentials dominate current source density ( CSD ) amplitudes ( Mitzdorf , 1985 ) , empirical observations repeatedly showed exceptions to this proposition . For example , CSD profiles of neuronal oscillations that were entrained to rhythmic stimulus streams showed a temporal alternation of states dominated by net ensemble depolarization and hyperpolarization , indicating the contribution of IPSCs to CSD profiles ( Lakatos et al . , 2008; Lakatos et al . , 2013 ) . Despite these limitations several empirical phenomena were reproduced and the input injection approach opens up avenues for future research to investigate the neural mechanisms underlying a wide range of different phenomena . For example , an important feature of the α-rhythm is its characteristic bistable jumping between low-power and high-power modes and a ‘dwelling’ in each state that follows a stretched-exponential ( Freyer et al . , 2009a ) . This behavior was remarkably closely reproduced by a multistable corticothalamic model that identified the underlying mechanism as a multistable switching between a fixed point and a limit cycle attractor that is driven by noise ( Freyer et al . , 2011 ) . Importantly , the closest reproduction of EEG α-switching in the model of Freyer et al . ( 2011 ) emerged only when the uncorrelated Gaussian noise term ( injected into mean membrane potentials ) was replaced by a state-dependent ( autoregressive ) noise term , which made the injected stochastic fluctuations effectively autocorrelated . This result is interesting in the context of the present study as our simulations identified the switching between high- and low-power modes of the α-rhythm as a potential generative mechanism underlying fMRI resting-state oscillations . Extending from these results , future BNM studies could systematically investigate the role of autocorrelated compared to Gaussian inputs and their impact on emerging fMRI dynamics ( like FC dynamics ) , especially since inputs like the EEG source activity used in our hybrid model better capture the autocorrelation structure of biological source currents , which are different from white noise ( Haider et al . , 2016; Okun et al . , 2010 ) . In line with our results , cellular-level studies indicate that rhythmic GABAergic input from the interneuronal network is associated with E/I balance ( Dehghani et al . , 2016 ) and α-related firing ( Jensen and Mazaheri , 2010; Lorincz et al . , 2009; Osipova et al . , 2008 ) . However , the identification of an exact physiological mechanism that explains how α-rhythms can produce an inhibitory effect remained elusive ( Jensen and Mazaheri , 2010; Klimesch , 2012 ) . Mazaheri and Jensen ( 2010 ) suggest that α-related inhibition occurs due to an observed amplitude asymmetry of ongoing oscillations , also termed baseline-shift . Our results suggest , in accordance with the model from Mazaheri and Jensen ( 2010 ) , that a symmetrically oscillating driving signal in the α-range leads to asymmetric firing rates and synaptic currents , but we extend this scheme with an explicit explanation of the generation of inhibitory pulses from oscillating input currents . Furthermore , our results with artificial α-activity may help to shed new light on the ‘gating by inhibition’ hypothesis , which posits that information is routed through the brain network by functionally blocking off task-irrelevant pathways and that this inhibition is reflected by α-activity ( Jensen and Mazaheri , 2010 ) . In agreement with this hypothesis , we found that long-range input decreased during states of high α-power and increased again when α-power decreased , but further studies are required to examine the effect of α-power on long-range communication and its interaction with other frequency bands . It is unclear to which degree non-neuronal processes affect the fMRI signal , as different physiological signals such as respiration and cardiac pulse rate were shown to be correlated with resting-state oscillations ( Biswal et al . , 1996; Power et al . , 2017 ) , which raised concerns that RSN oscillations may be unrelated to neuronal information processing , but rather constitute an epiphenomenon ( Birn et al . , 2006; de Munck et al . , 2008; Shmueli et al . , 2007; Yuan et al . , 2013 ) . The interpretation and handling of these signal modulations is therefore hotly debated and they are often considered as artefactual and removed from fMRI studies ( Birn et al . , 2006; Chang and Glover , 2009 ) . Importantly , however , low-frequency BOLD fluctuations are also strongly correlated with electrical neural activity , which was shown by studies that analysed fMRI jointly with EEG ( Goldman et al . , 2002; Laufs et al . , 2003; Moosmann et al . , 2003 ) , intracortical recordings ( He et al . , 2008; Logothetis et al . , 2001 ) or MEG ( Brookes et al . , 2011; de Pasquale et al . , 2010 ) . Similarly , strong temporal correlations and spatially similar correlation maps of EEG α-power , respiration and BOLD ( Yuan et al . , 2013 ) , as well as of EEG α-power , heart rate variations and BOLD ( de Munck et al . , 2008 ) suggest that these fMRI fluctuations are not unrelated to neural activity , but may be of neural origin . Our results extend the current understanding by showing an explicit mechanism for a neural origin of fMRI RSN oscillations that explains a large part of their variance by a chain of neurophysiological interactions . That is , our simulated activity not only reproduces the negative correlation between α-power fluctuations and BOLD signal , but also reveals a mechanism that transforms ongoing α-power fluctuation into fMRI oscillations . The hybrid approach therefore constitutes a multimodal data fusion approach ( Friston , 2009; Valdes-Sosa et al . , 2009 ) that enables the direct characterization of the previously reported temporal correlations between BOLD and EEG signals in terms of the underlying neural activity and explicit forward models . In addition to fMRI time series , the hybrid model also reproduces the spatial topology of fMRI networks , which are not predicted by the α-power regressor . These findings thereby add to accumulating evidence suggesting that RSNs originate from neuronal activity ( Brookes et al . , 2011; de Pasquale et al . , 2010; Goldman et al . , 2002; He et al . , 2008; Logothetis et al . , 2001; Mantini et al . , 2007; Moosmann et al . , 2003 ) rather than being a purely hemodynamic phenomenon that is only correlated , but not caused by it ( Birn et al . , 2006; de Munck et al . , 2008; Shmueli et al . , 2007 ) . The conclusions from these results have important implications for future fMRI studies , as they implicate that low-frequency fMRI oscillations may be attributed to a neural process that has a considerable state-dependent effect on neural information processing as indicated by the large modulations of neuronal firing and synaptic activity . Methods for physiological noise correction might remove variance from fMRI experiments that is related to neuronal activity and may therefore exclude relevant information for the interpretation of fMRI data . Parameter space exploration shows that structural coupling is critical for fMRI prediction , as prediction quality decreases for sub-optimal global coupling strengths or when global coupling is deactivated altogether ( Figure 3—figure supplement 1 , Figure 7—figure supplement 2 ) . In this study , we did not address the effect of coupling time delays , as they were non-essential for the emergence of the described phenomena . Our initial application of the novel hybrid model aimed to study the effect of input injection while minimizing the degrees of freedom of the simulation and the set of parameters to be varied . Further studies are required to determine the effect of coupling delays , as previous studies demonstrated their important role for emerging large-scale dynamics ( Deco et al . , 2011; Jirsa , 2008 , 2009 ) . We observed that the prediction quality of resting-state network activation time courses fluctuates over time and is highest during epochs of highest variance of the respective temporal mode . During these , time windows resting-state networks contribute the largest variance to whole-brain fMRI , that is , they are the most active . A possible explanation may be that during states of asynchronous neural activity ( i . e . when the variance of RSN temporal modes is low ) volume conduction and cancellation of electromagnetic waves decreases the ability of source imaging methods to reconstruct source activity . It is important to note that the observed processes may not be specific to α-oscillations , but may apply also to other frequencies or non-oscillatory signal components , for example , phase-locked discharge of neurons occurs over a range of frequency bands and is not limited to the α-rhythm ( Buzsaki , 2006 ) . Furthermore , the α-rhythm , though prominent , is certainly not superior to other rhythms with respect to neuronal computation and cognition ( Fries , 2015 ) . In fact , it may be best thought of as one of several modes of brain operation , even during the so-called resting-state ( Engel et al . , 2013 ) . Additional empirical and theoretical studies will be needed to address these limitations more comprehensively . Although our analysis revolved around α-oscillations , the hybrid modeling approach is not restricted to α-activity , as the injected EEG source activity was not limited to the α-band . The hybrid modeling approach itself does not set any requirements on the frequency spectrum of the injected source activity . Importantly , our focus on α-rhythms was not ‘by construction’ , but , as outlined in the introduction , emerged from a sequence of analyses that we performed to understand how the hybrid model generated the correlation with empirical fMRI time series . In this regard , it is interesting to note that the time series correlations obtained by the α-regressor and the hybrid model are comparable , which indicates that the α-rhythm was the main driver for the hybrid model’s fMRI time series prediction . Despite the ubiquity of scale invariant dynamics , models that generate power-law distributions are often rather generic and detached from the details of the modeled systems ( Bak et al . , 1987; Marković and Gros , 2014 ) . Furthermore , the precise mechanisms that lead to the emergence of fMRI power spectrum power-law scaling or the relationship between brain network interaction and fMRI power-law scaling are unclear ( He , 2011 ) . Our simulation results indicate that fMRI spectra power-law scaling is due to the observed frequency-dependent amplification of oscillatory activity in networks that contain self-reinforcing feedback excitation together with slow decay of activity . Central to theories on the emergence of criticality is the tuning of a control parameter ( e . g . connection strengths ) that leads the system to a sharp change in one or more order parameters ( e . g . firing rates ) when the control parameter is moved over a critical point that marks the boundary of a phase transition . It is important to point out that the existence of power-laws alone does not prove criticality . Rather , criticality requires the existence of a control parameter that can be adjusted to move the system through a phase transition at a critical point ( Beggs and Timme , 2012 ) . In vivo , in vitro and in silico results show that the dynamical balance between excitation and inhibition was found to be essential to move the system towards or away from criticality , for example , by pharmacologically altering the excitation-inhibition balance in anesthetized rats ( Osorio et al . , 2010 ) , acute slices ( Beggs and Plenz , 2003 ) or by changing parameters that control global excitation and inhibition in computational models ( Deco et al . , 2014 ) . However , the exact role played by excitation-inhibition balance is unclear . In line with these results , we found that power-law scaling varied as a function of the relative levels of global excitation and inhibition , further emphasizing the need for a proportional relationship between these control parameters ( Figure 7—figure supplement 2 ) . Extending from that , our simulation results indicate that E/I balance may cause a tuning of the relative strengths of local and long-range inputs to neural populations that supports constructive interference between the different input currents , which in turn amplifies slower oscillations more than faster oscillation . These results address an open question on whether power-laws in neural networks result from power-law behavior on the cellular level or from a global network-level process ( Beggs and Timme , 2012 ) , by giving an explanation for scale-free fMRI power spectra as an emergent property of long-range brain network interaction that does not require small-scale decentralized processes like the constant active retuning of microscopic parameters as proposed in some theories of self-organized criticality ( Bak et al . , 1987; Hesse and Gross , 2014 ) . Furthermore , these results explicitly address the effect of input activity , while in vitro and in silico studies have so far focused on systems without or considerably decreased input ( Hesse and Gross , 2014 ) . The observed co-emergence of spatial long-range correlations ( i . e . functional connectivity networks ) and power-law scaling may point to a unifying explanation within the theory of self-organized criticality , as previously proposed by others ( Linkenkaer-Hansen et al . , 2001 ) . Note that we have used the hybrid model not simply to establish the prevalence of scale invariant dynamics , but to use the power law scaling in a quantitative sense to understand the mechanisms leading to particular power law exponents; for example , the importance of extrinsic ( between node ) connections in explaining the differences between power law scaling at the electrophysiological and haemodynamic level . This is an important point because scale-free behavior per se would be difficult to avoid in simulations of this sort . A wide range of disorders like autism , schizophrenia , intellectual disabilities , Alzheimer’s disease , multiple sclerosis or epilepsy have been linked to disruption of E/I balance ( Marín , 2012 ) and altered structural and functional network connectivity ( Stam , 2014 ) . The presented modelling approach may therefore play a key role for identifying the precise mechanisms underlying the pathophysiology of different disorders and assist in developing novel therapies that restore altered E/I balance or brain connectivity , for example , by identifying the targets for neural stimulation therapies or by guiding individually customized therapy . The ability of the hybrid model to infer precise neurophysiological mechanisms that give rise to empirical phenomena and to link the involved mechanisms and signal patterns across different scales and neuroimaging modalities makes it a potentially valuable tool for neuroscience research . The model used in this study is based on the large-scale dynamical mean field model used by Deco and colleagues ( Deco et al . , 2014; Wong and Wang , 2006 ) . Brain activity is modeled as the network interaction of local population models that represent cortical areas . Cortical regions are modelled by interconnected excitatory and inhibitory neural mass models . In contrast to the original model , excitatory connections were replaced by injected EEG source activity . The dynamic mean field model faithfully approximates the time evolution of average synaptic activities and firing rates of a network of spiking neurons by a system of coupled non-linear differential equations for each node i: ( 1 ) Ii ( E ) =WEI0+G∑jCijSj ( E ) −JiSi ( I ) +wBG ( E ) IBG ( 2 ) Ii ( I ) =WII0−Si ( I ) +wBG ( I ) IBG ( 3 ) ri ( E ) =aEIi ( E ) −bE1−exp ( −dE ( aEIi ( E ) −bE ) ) ( 4 ) ri ( I ) =aIIi ( I ) −bI1−exp ( −dI ( aIIi ( I ) −bI ) ) ( 5 ) dSi ( E ) ( t ) dt=−Si ( E ) τE+ ( 1−Si ( E ) ) γEri ( E ) ( 6 ) dSi ( I ) ( t ) dt=−Si ( I ) τI+γIri ( I ) Here , ri ( E , I ) denotes the population firing rate of the excitatory ( E ) and inhibitory ( I ) population of brain area i . Si ( E , I ) identifies the average excitatory or inhibitory synaptic gating variables of each brain area , while their input currents are given by Ii ( E , I ) . In contrast to the model used by Deco et al . ( 2014 ) that has recurrent and feedforward excitatory coupling , we approximate excitatory postsynaptic currents IBG using region-wise aggregated EEG source activity that is added to the sum of input currents Ii ( E , I ) . This approach is based on intracortical recordings that suggest that EPSCs are non-random , but strongly correlated with electric fields in their vicinity , while IPSCs are anticorrelated with EPSCs ( Haider et al . , 2016 ) . The weight parameters ωBG ( E , I ) rescale the z-score normalized EEG source activity independently for excitatory and inhibitory populations . G denotes the long-range coupling strength scaling factor that rescales the structural connectivity matrix Cij that denotes the strength of interaction for each region pair i and j . All three scaling parameters are estimated by fitting simulation results to empirical fMRI data by exhaustive search . Initially , parameter space ( n-dimensional real space with n being the number of optimized parameters ) was constrained such that the strength of inhibition was larger than the strength of excitation , satisfying a biological constraint . Furthermore , for each tested parameter set ( containing the three scaling parameters mentioned above ) , the region-wise parameters Ji that describe the strength of the local feedback inhibitory synaptic coupling for each area i ( expressed in nA ) are fitted with the algorithm described below such that the average firing rate of each excitatory population in the model was close to 3 . 06 Hz ( i . e . the cost function for tuning parameters Ji was solely based on average firing rates and not on prediction quality ) . The overall effective external input I0 = 0 . 382 nA is scaled by WE and WI , for the excitatory and inhibitory pools , respectively . ri ( E , I ) denotes the neuronal input-output functions ( f-I curves ) of the excitatory and inhibitory pools , respectively . All parameters except those that are tuned during parameter estimation are set as in Deco et al . ( 2014 ) . Please refer to Table 1 for a specification of state variables and parameters . BOLD activity was simulated on the basis of the excitatory synaptic activity S ( E ) using the Balloon-Windkessel hemodynamic model ( Friston et al . , 2003 ) , which is a dynamical model that describes the transduction of neuronal activity into perfusion changes and the coupling of perfusion to BOLD signal . The model is based on the assumption that the BOLD signal is a static non-linear function of the normalized total deoxyhemoglobin voxel content , normalized venous volume , resting net oxygen extraction fraction by the capillary bed , and resting blood volume fraction . Please refer to Deco et al . ( 2013 ) for the specific set of Ballon-Windkessel model equations that we used in this study . For each brain network model , three parameters were varied to maximize the fit between empirical and simulated fMRI: the scaling of excitatory white-matter coupling and the strengths of the inputs injected into excitatory and inhibitory populations ( please refer to Table 2 for an overview over the obtained parameter values ) . Following in vivo observations ( Xue et al . , 2014 ) , we ensured that at excitatory populations EPSC amplitudes are smaller than IPSC amplitudes by constraining the range of values for the ratio ωBG ( I ) / ωBG ( E ) between 5 and 200 , which we found through initial pilot simulations . Note that the ratio ωBG ( I ) / ωBG ( E ) is not identical to the amplitude ratio of IPSCs vs . EPSCs , but depends also on the specific settings of all other varied parameters . For example , a large ratio ωBG ( I ) / ωBG ( E ) can still lead to a small ratio of IPSCs vs . EPSCs amplitudes if the local feedback inhibition parameter Ji is small . Apart from these initial pilot simulations to restrict the ratio of postsynaptic currents to a biologically plausible range , the specific combination of all varied parameters was exclusively found through fitting simulated to empirical fMRI time series under the constraint of plausible firing rates . That is , besides tuning these three global parameters using the sole optimization criterion of maximizing the fit between simulated and empirical fMRI time series , we adjusted local inhibitory coupling strengths in order to obtain biologically plausible firing rates in excitatory populations . For this second form of tuning , termed feedback inhibition control ( FIC ) , average population firing rates were the sole optimization criterion , without any consideration of prediction quality , which was only dependent on the three global parameters . FIC modulates the strengths of inhibitory connections that is required to compensate for excess or lack of excitation resulting from the large variability in white-matter coupling strengths obtained by MRI tractography , which is a prerequisite to obtain plausible ranges of population activity that is relevant for some results ( Figure 5 and Figure 6 ) . Prediction quality was measured as the average correlation coefficient between all simulated and empirical region-wise fMRI time series of a complete cortical parcellation over 20 . 7 min length ( TR = 1 . 94 s , 640 data points ) thereby quantifying the ability of the model to predict the activity of 68 parcellated cortical regions . Accounting for the large-scale nature of fMRI resting-state networks , the chosen parcellation size provides a parsimonious trade-off between model complexity and the desired level of explanation . What this parcellation may lack in spatial detail , it gains in providing a full-brain coverage that can reliably reproduce ubiquitous large-scale features of empirical data , which we further present below . To exclude overfitting and limited generalizability , a five-fold cross-validation scheme was performed on the hybrid model simulation results . Therefore , the data was randomly divided into two subsets: 80% as training subset and 20% as testing subset . Prediction quality was estimated using the training set , before trained models were asked to predict the testing set . Resulting prediction quality was compared between training and test data set and between test data set and the data obtained from fitting the full time series . Furthermore , despite the large range of possible parameters , the search converged to a global maximum ( Figure 3—figure supplement 1 ) . Therefore , we ensured that when the model has been fit to a subset of empirical data , that it was able to generalize to new or unseen data . In contrast to model selection approaches , where the predictive power of different models and their complexity are compared against each other , we here use only a single type of model . The excitatory populations of isolated nodes of the original model described in Deco et al . ( 2014 ) have an average firing rate of 3 . 06 Hz . That is , without long-range coupling G∑jCijSj ( E ) and without injected activity wBG ( E ) IBG and wBG ( I ) IBG ( cf . Equations 1 and 2 ) , the used excitatory populations have an average firing rate of 3 . 06 Hz . This value conforms to the empirically measured Poisson-like cortical in vivo activity of ~3 Hz ( Softky and Koch , 1993; Wilson et al . , 1994 ) and results from the dynamic mean field approximation of the average ensemble behaviour of a large-scale spiking neuron model used in Deco et al . ( 2014 ) . In contrast to isolated nodes , the firing rate of coupled nodes change in dependence of the employed structural connectivity matrix and the injected input . To compensate for a resulting excess or lack of excitation , a local regulation mechanism , called feedback inhibition control ( FIC ) , was used . The approach was previously successfully used to significantly improve FC prediction as well as for increasing the dynamical repertoire of evoked activity and the accuracy of external stimulus encoding ( Deco et al . , 2014 ) . Despite the mentioned advantages of FIC tuning , it has the disadvantage of increasing the number of open parameters of the model . To prove that prediction quality is not due to FIC , but solely due to the three global parameters and to exclude concerns about over-parameterization or that FIC may be a potentially necessary condition for the emergence of scale-freeness , we devised a control model that did not implement FIC , but used a single global parameter for inhibitory coupling strength . Instead of tuning the 68 individual local coupling weights individually , only a single global value for all inhibitory coupling weights Ji was varied . We compared the effect of FIC on time series prediction quality and found no significant difference in prediction quality to simulations that used only a single value for all local coupling weights Ji per subject ( one-tailed Wilcoxon rank sum test , p=0 . 36 , z = −0 . 37 , Cliffs’s delta d = −0 . 15 ) . In contrast to simulations that are driven by noise ( Deco et al . , 2014 ) , FIC parameters for injected input must be estimated for the entire simulated time series , since the non-stationarity of stimulation time series leads to considerable fluctuations of firing rates . Therefore , we developed a local greedy search algorithm for fast FIC parameter estimation based on the algorithm in Deco et al . ( 2014 ) . To exert FIC , local inhibitory synaptic strength is iteratively adjusted until all excitatory populations attained a firing rate close to the desired mean firing rates for the entire ~20 min of activity . During each iteration , the algorithm performs a simulation of the entire time series . Then , it computes the mean firing activity over the entire time series for each excitatory population and adapts Ji values accordingly , that is , it increases local Ji values if the average firing rate over all excitatory populations during the k-th iteration r̂k is larger than 3 . 06 Hz and vice versa . In order to reduce the number of iterations the value by which Ji is changed is , in contrast to the algorithm by Deco et al . ( 2014 ) , dynamically adapted in dependence of the firing rate obtained during the current iteration ( 7 ) Jik+1=Jik+ ( r^k−3 . 06 ) τkwhere Jik denotes the value of feedback inhibition strength of node i and τk denotes the adaptive tuning factor during the k-th iteration . In the first iteration , all Ji values are initialized with one and τk is initialized with 0 . 005 . The adaptive tuning factor is dynamically changed during each iteration based on the result of the previous iteration: ( 8 ) τk+1= ( ∑i ( Jik−1−Jk ) ) / ( r^k−1−r^k ) . For the case that the result did not improve during the current iteration , that is , ( 9 ) \|r^−3 . 06\|≥\|r^k−1−3 . 06\| , the adaptive tuning factor is decreased by multiplying it with 0 . 5 and the algorithm continues with the next iteration . After 12 iterations , all Ji values are set to the values they had during the iteration k where \|r̂k – 3 . 06\| was minimal . Structural and functional connectomes from 15 healthy human subjects ( age range: 18–31 years , eight female ) were extracted from full data sets ( diffusion-weighted MRI , T1-weighted MRI , EEG-fMRI ) using a local installation of a pipeline for automatic processing of functional and diffusion-weighted MRI data ( Schirner et al . , 2015 ) . From a local database of 49 subjects ( age range 18–80 years , 30 female ) that was acquired for a previous study ( Schirner et al . , 2015 ) , we selected the 15 youngest subjects that fulfilled highest EEG quality standards after applying MR artefact correction routines . EEG quality was assessed by standards that were defined prior to the experimental design and that are routinely used in the field ( Becker et al . , 2011; Freyer et al . , 2009b; Ritter et al . , 2010; Ritter et al . , 2007 ) : occurrence of spikes in frequencies > 20 Hz in power spectral densities , excessive head motion and cardio-ballistic artefacts . Research was performed in compliance with the Code of Ethics of the World Medical Association ( Declaration of Helsinki ) . Written informed consent was provided by all subjects with an understanding of the study prior to data collection , and was approved by the local ethics committee in accordance with the institutional guidelines at Charité Hospital Berlin . Subjects with a self-reported history of neurological , cognitive , or psychiatric conditions were excluded from the experiment . Structural ( T1-weighted high-resolution three-dimensional MP-RAGE sequence; TR = 1 , 900 ms , TE = 2 . 52 ms , TI = 900 ms , flip angle = 9° , field of view ( FOV ) = 256 mm x 256 mm x 192 mm , 256 × 256 × 192 Matrix , 1 . 0 mm isotropic voxel resolution ) , diffusion-weighted ( T2-weighted sequence; TR = 7500 ms , TE = 86 ms , FOV = 192 mm x 192 mm , 96 × 96 Matrix , 61 slices , 2 . 3 mm isotropic voxel resolution , 64 diffusion directions ) , and fMRI data ( two-dimensional T2-weighted gradient echo planar imaging blood oxygen level-dependent contrast sequence; TR = 1 , 940 ms , TE = 30 ms , flip angle = 78° , FOV = 192 mm x 192 mm , 3 mm x 3 mm voxel resolution , 3 mm slice thickness , 64 × 64 matrix , 33 slices , 0 . 51 ms echo spacing , 668 TRs , 7 initial images were acquired and discarded to allow magnetization to reach equilibrium; eyes-closed resting-state ) were acquired on a 12-channel Siemens 3 Tesla Trio MRI scanner at the Berlin Center for Advanced Neuroimaging , Berlin , Germany . Extracted structural connectivity matrices intend to give an aggregated representation of the strengths of interaction between regions as mediated by white matter fiber tracts . As in the original model by Deco et al . ( 2014 ) , conduction delays were neglected in this study as they were non-essential for the described features . Strength matrices Cij were divided by their respective maximum value for normalization . In short , the pipeline proceeds as follows: for each subject a three-dimensional high-resolution T1-weighted image image was used to divide cortical gray matter into 68 regions according to the Desikan-Killiany atlas using FreeSurfer’s ( Fischl , 2012 ) automatic anatomical segmentation and registered to diffusion data . The gyral-based brain parcellation is generated by an automated probabilistic labeling algorithm that has been shown to achieve a high level of anatomical accuracy for identification of regions while accounting for a wide range of inter-subject anatomical variability ( Desikan et al . , 2006 ) . The atlas was successfully used in previous modelling studies and provided highly significant structure-function relationships ( Honey et al . , 2009; Ritter et al . , 2013; Schirner et al . , 2015 ) . Details on diffusion-weighted and fMRI preprocessing can be found in Schirner et al . ( Schirner et al . , 2015 ) Briefly , probabilistic white matter tractography and track aggregation between each region-pair was performed as implemented in the automatic pipeline and the implemented distinct connection metric extracted . This metric weights the raw track count between two regions according to the minimum of the gray matter/white matter interface areas of both regions used to connect these regions in distinction to other metrics that use the unweighted raw track count , which was shown to be biased by subject-specific anatomical features ( see Schirner et al . ( 2015 ) for a discussion ) . After preprocessing , the cortical parcellation mask was registered to fMRI resting-state data of subjects and average fMRI signals for each region were extracted . The first five images of each scanning run were discarded to allow the MRI signal to reach steady state . To identify RSN activity a spatial Group ICA decomposition was performed for the fMRI data of all subjects using FSL MELODIC ( Beckmann and Smith , 2004 ) ( MELODIC v4 . 0; FMRIB Oxford University , UK ) with the following parameters: high pass filter cut off: 100 s , MCFLIRT motion correction , BET brain extraction , spatial smoothing 5 mm FWHM , normalization to MNI152 , temporal concatenation , dimensionality restriction to 30 output components . ICs that correspond to RSNs were automatically identified by spatial correlation with the 9 out of the 10 well-matched pairs of networks of the 29 , 671-subject BrainMap activation database as described in Smith et al . ( 2009 ) ( excluding the cerebellum network ) . All image processing were performed in the native subject space of the different modalities and the brain atlas was transformed from T1-space of the subject into the respective spaces of the different modalities . Details of EEG preprocessing are described in supplementary material of Schirner et al . ( Schirner et al . , 2015 ) . First , to account for slow drifts in EEG channels and to improve template construction during subsequent MR imaging acquisition artefact ( IAA ) correction all channels were high-pass filtered at 1 . 0 Hz ( standard FIR filter ) . IAA correction was performed using Analyser 2 . 0 ( v2 . 0 . 2 . 5859 , Brain Products , Gilching , Germany ) . The onset of each MRI scan interval was detected using a gradient trigger level of 300 µV/ms . Incorrectly detected markers , for example due to shimming events or heavy movement , were manually rejected . To assure the correct detection of the resulting scan start markers each inter-scan interval was controlled for its precise length of 1940 ms ( TR ) . For each channel , a template of the IAA was computed using a sliding average approach ( window length: 11 intervals ) and subsequently subtracted from each scan interval . For further processing , the data were down sampled to 200 Hz , imported to EEGLAB and low-pass filtered at 60 Hz . ECG traces were used to detect and mark each instance of the QRS complex in order to identify ballistocardiogram ( BCG ) artefacts . The reasonable position and spacing of those ECG markers was controlled by visual inspection and corrected if necessary . To correct for BCG and artefacts induced by muscle activity , especially movement of the eyes , a temporal ICA was computed using the extended Infomax algorithm as implemented in EEGLAB . To identify independent components ( ICs ) that contain BCG artefacts the topography plot , activation time series , power spectra and heartbeat triggered average potentials of the resulting ICs were used as indication . Based on established characteristics , all components representing the BCG were identified and rejected , that is , the components were excluded from back-projection . The remaining artificial , non-BCG components , accounting for primarily movement events especially eye movement , were identified by their localization , activation , power spectral properties and ERPs . Detailed descriptions of EEG and fMRI preprocessing have been published elsewhere ( Becker et al . , 2011; Freyer et al . , 2009a; Ritter et al . , 2010; Ritter et al . , 2007 ) . EEG source imaging was performed with the freely available MATLAB toolbox Brainstorm using default settings and standard procedure for resting-state EEG data as described in the software documentation ( Tadel et al . , 2011 ) . Source space models were based on the individual cortical mesh triangulations as extracted by FreeSurfer from each subject’s T1-weighted MRI data and downsampled by Brainstorm . From the same MRI data , head surface triangulations were computed by Brainstorm . Standard positions of the used EEG caps ( Easy-cap; 64 channels , MR compatible ) were aligned by the fiducial points used in Brainstorm and projected onto the nearest point of the head surface . Forward models are based on Boundary Element Method head models computed using the open-source software OpenMEEG and 15002 perpendicular dipole generator models located at the vertices of the cortical surface triangulation . The sLORETA inverse solution was used to estimate the distributed neuronal current density underlying the measured sensor data since it has zero localization error ( Pascual-Marqui , 2002 ) . EEG data were low-pass filtered at 30 Hz and imported into Brainstorm . There , the epochs before the first and after the last fMRI scan were discarded and the EEG signal was time-locked to fMRI scan start markers . Using brainstorm routines , EEG data were projected onto the cortical surface using the obtained inversion kernel and averaged according to the Desikan-Killiany parcellation that was also used for the extraction of structural and functional connectomes and region-averaged fMRI signals . The resulting 68 region-wise source time series were imported to MATLAB , z-score normalized and upsampled to 1000 Hz using spline interpolation as implemented by the Octave function interp1 . To enable efficient simulations , the sampling rate of the injected activity was ten times lower than model sampling rate . Hence , during simulation identical values have been injected during each sequence of 10 integration steps . Simulations were performed with a highly optimized C implementation of the previously described model on the JURECA supercomputer at the Juelich Supercomputing Center . Simulation and analyses code and used data is open source and available from online repositories ( Schirner et al . , 2017a , see ‘Data and code availability’ ) . An exhaustive brute-force parameter space scan using 3888 combinations of the parameters G and ωBG ( E , I ) was performed for each subject . Each of these combinations was computed 12 times to iteratively tune Ji values . As control setup , further simulations were performed with random permutations of the input time series . Therefore , the individual time points of each source activity time series were randomly permuted ( individually for each region and subject ) using the Octave function randperm ( ) and injected into simulations using all parameter combinations that were previously used . As an additional control situation the original dynamic mean field model as described in Deco et al . ( 2014 ) was simulated for the 15 SCs . Here , the parameters G and JNMDA were varied and FIC tuning was performed using the same algorithm as used for the source activity injection model . The simulation and FIC optimization process was identical for all three models . The length of the simulated time series for each subject was 21 . 6 min . Simulations were performed at a model sampling rate of 10 , 000 Hz . BOLD time series were computed for every 10th time step of excitatory synaptic gating activity using the Balloon-Windkessel model ( Friston et al . , 2003 ) . Since the Balloon-Windkessel model acts like a low-pass filter that attenuates frequencies above ~0 . 15 Hz ( Robinson et al . , 2006 ) , additional low-pass filtering was unnecessary for downsampling of simulated fMRI time series . Hence , from the resulting time series every 1940th step was stored in order to obtain a sampling rate of simulated fMRI that conforms to the empirical fMRI TR of 1 . 94 s . The first 11 scans ( 21 . 34 s ) of activity were discarded to allow model activity and simulated fMRI signal to stabilize . For each subject and modelling approach the simulation result that yielded the highest average correlation between all 68 empirical and simulated region time series for all tested parameters was used for all analyses . To ensure region-specificity of simulation results only corresponding simulated and empirical region time series were correlated in the case of raw fMRI , respectively , for resting-state networks only simulated regions that overlap with the spatial activation pattern of the respective network were used for estimating prediction quality . Specifically , for RSN analysis , only those regions were compared with the temporal modes of RSNs that had a spatial overlap of at least 40% of all voxels belonging to the respective region . To assess time-varying prediction quality , a correlation analysis was performed in which a window with a length of 100 scans ( 194 s ) was slid over the 68 pairs of empirical and simulated time series and the average correlation over all 68 regions was computed for each window . For the estimation of signal correlation , the computation of entries of FC matrices and as a measure of similarity of FC matrices Pearson’s linear correlation coefficient was used . FC matrices were compared by stacking all elements below the main diagonal into vectors and computing the correlation coefficient of these vectors . Short-epoch FC prediction quality was estimated by computing the mean correlation obtained for all window-wise correlations of a sliding window analysis of empirical and simulated time series ( window-size: 100 scans = 194 s ) . To ensure scale-freeness of empirical and simulated signals , region time series were tested using rigorous model selection criteria; on average 79% of all 1020 region-wise time series ( 15 subjects x 68 regions ) for the seven analyzed signal types ( empirical fMRI , simulated fMRI , simulated fMRI without global coupling , simulated fMRI without FIC , simulated fMRI without FIC and without global coupling , α-power , α-regressor ) tested as scale-free; for every signal type every subject had at least five regions to test as scale-free . PSDs were computed using the Welch method as implemented in Octave , normalized by their total power and averaged . Resulting average power spectra were fitted with a power-law function f ( x ) =axβ using least-squares estimation in the frequency range 0 . 01 Hz and 0 . 17 Hz which is identical to the range for which the test for scale invariance was performed . Frequencies below were excluded in order to reduce the impact of low-frequency signal confounds and scanner drift , frequencies above that limit were excluded to avoid aliasing artefacts in higher frequency ranges ( TR = 1 . 94 s , hence Nyquist frequency is around 0 . 25 Hz ) . In order to compare the scale invariance of our empirical fMRI data with results from previous publications ( He , 2011 ) , we also computed power spectra in a range that only included frequencies < 0 . 1 Hz . In order to adequately determine the existence of scale invariance we applied rigorous model selection to every time series to identify power-law scaling and excluded all time series from analyses that were described better by a model other than a power-law . Nevertheless , we compared the obtained results from this strict regime with results obtained when all time series were included and found them to be qualitatively identical . To test for the existence of scale invariance we used a method that combines a modified version of the well-established detrended fluctuation analysis ( DFA ) with Bayesian model comparison ( Ton and Daffertshofer , 2016 ) . DFA is , in contrast to PSD analyses , robust to both stationary and nonstationary data in the presence of confounding ( weakly non-linear ) trends . It is important to note , that a simple linear fit of the detrended fluctuation curve without proper comparison of the obtained goodness of fit with that of other models would entirely ignore alternative representations of the data different than a power law . For quantification of the goodness of fit with simple regression its corresponding coefficient of determination , R2 , is ill-suited as it measures only the strength of a linear relationship and is inadequate for nonlinear regression ( Ton and Daffertshofer , 2016 ) . It is important to note that with this method the assessment of power-law scaling is based on maximum likelihood estimation , which overcomes the limitations of a minimal least-squares estimate obtained from linear regression in the conventional DFA approach . Details of the used method are described elsewhere ( Ton and Daffertshofer , 2016 ) . For the different signals the majority of time series were tested as being scale free: 83% for empirical fMRI , 69% for simulated fMRI , 71% for simulated fMRI with deactivated FIC , 83% for simulated fMRI with deactivated global coupling , 86% for simulated fMRI with deactivated global coupling and FIC , 90% for α-power and 70% for the α-regressor . To compute grand average waveforms , state-variables were averaged over all 15 subjects and 68 regions ( N = 1020 region time series ) time-locked to the zero crossing of the α-amplitude , which was obtained by band-pass filtering source activity time series between 8 and 12 Hz; to obtain sharp average waveforms , all α-cycle epochs with a cycle length between 95 and 105 ms were used ( N = 4 , 137 , 994 α-cycle ) . For computing ongoing α-power time courses , instantaneous power time series were computed by taking the absolute value of the analytical signal ( obtained by the Hilbert transform ) of band-pass filtered source activity in the 8–10 Hz frequency range; the first and last ~50 s were discarded to control for edge effects . To compute the α-regressor , power time series were convolved with the canonical hemodynamic response function , downsampled to fMRI sampling rate and shifted relative to fMRI time series to account for the lag of hemodynamic response . The highest negative average correlation over all 68 region-pairs obtained within a range of ±3 scans shift was used for comparison with simulation results . All statistical analyses were performed using MATLAB ( The MathWorks , Inc . , Natick , Massachusetts , United States ) . Data are represented as box-and-whisker plots . As normality was not achieved for the majority of data sets ( assessed by Lilliefors test at significance level of 0 . 05 ) , differences between groups were compared by non-parametric statistical tests , using either two-tailed Wilcoxon rank sum test or , in case of directional prediction , one-tailed Wilcoxon rank sum test; a value p<0 . 05 was considered significant . Brain network models are implemented in the open source neuroinformatics platform The Virtual Brain ( Ritter et al . , 2013; Sanz-Leon et al . , 2015 , Sanz Leon et al . , 2013 ) that can be downloaded from thevirtualbrain . org . Code and data that support the findings of this study can be obtained from https://github . com/BrainModes/The-Hybrid-Virtual-Brain ( Schirner et al . , 2017b; copy archived at https://github . com/elifesciences-publications/The-Hybrid-Virtual-Brain ) and https://osf . io/mndt8/ ( Schirner et al . , 2017a ) .	Neuroscientists can use various techniques to measure activity within the brain without opening up the skull . One of the most common is electroencephalography , or EEG for short . A net of electrodes is attached to the scalp and reveals the patterns of electrical activity occurring in brain tissue . But while EEG is good at revealing electrical activity across the surface of the scalp , it is less effective at linking the observed activity to specific locations in the brain . Another widely used technique is functional magnetic resonance imaging , or fMRI . A patient , or healthy volunteer , lies inside a scanner containing a large magnet . The scanner tracks changes in the level of oxygen at different regions of the brain to provide a measure of how the activity of these regions changes over time . In contrast to EEG , fMRI is good at pinpointing the location of brain activity , but it is an indirect measure of brain activity as it depends on blood flow and several other factors . In terms of understanding how the brain works , EEG and fMRI thus provide different pieces of the puzzle . But there is no easy way to fit these pieces together . Other areas of science have used computer models to merge different sources of data to obtain new insights into complex processes . Schirner et al . now adopt this approach to reveal the workings of the brain that underly signals like EEG and fMRI . After recording structural MRI data from healthy volunteers , Schirner et al . built a computer model of each person’s brain . They then ran simulations with each individual model stimulating it with the person’s EEG to predict the fMRI activity of the same individual . Comparing these predictions with real fMRI data collected at the same time as the EEG confirmed that the predictions were accurate . Importantly , the brain models also displayed many features of neural activity that previously could only be measured by implanting electrodes into the brain . This new approach provides a way of combining experimental data with theories about how the nervous system works . The resulting models can help generate and test ideas about the mechanisms underlying brain activity . Building models of different brains based on data from individual people could also help reveal the biological basis of differences between individuals . This could in turn provide insights into why some individuals are more vulnerable to certain brain diseases and open up new ways to treat these diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "neuroscience" ]	2018	Inferring multi-scale neural mechanisms with brain network modelling
N-myristoylation is a ubiquitous class of protein lipidation across eukaryotes and N-myristoyl transferase ( NMT ) has been proposed as an attractive drug target in several pathogens . Myristoylation often primes for subsequent palmitoylation and stable membrane attachment , however , growing evidence suggests additional regulatory roles for myristoylation on proteins . Here we describe the myristoylated proteome of Toxoplasma gondii using chemoproteomic methods and show that a small-molecule NMT inhibitor developed against related Plasmodium spp . is also functional in Toxoplasma . We identify myristoylation on a transmembrane protein , the microneme protein 7 ( MIC7 ) , which enters the secretory pathway in an unconventional fashion with the myristoylated N-terminus facing the lumen of the micronemes . MIC7 and its myristoylation play a crucial role in the initial steps of invasion , likely during the interaction with and penetration of the host cell . Myristoylation of secreted eukaryotic proteins represents a substantial expansion of the functional repertoire of this co-translational modification . Toxoplasmosis currently affects approximately one third of the world’s population ( Robert-Gangneux and Dardé , 2012 ) . It is caused by the obligate protozoan parasite Toxoplasma gondii originating from the phylum Apicomplexa . While the majority of human infections are asymptomatic , the disease manifests its severity in immunocompromised individuals , such as those receiving chemotherapy and transplants or in HIV/AIDS patients ( Montoya and Liesenfeld , 2004 ) . Key steps in the successful propagation of Toxoplasma infection in the acute phase are orchestrated cycles of invasion and egress of tachyzoites from host cells ( Black and Boothroyd , 2000 ) . These crucial processes are regulated by several post-translational modifications ( PTMs ) , such as phosphorylation ( Gaji et al . , 2015; Jacot and Soldati-Favre , 2012; Lourido et al . , 2010; Treeck et al . , 2014 ) , ubiquitination ( Silmon de Monerri et al . , 2015 ) , and also protein lipidation , such as palmitoylation and myristoylation ( Alonso et al . , 2012; Frénal et al . , 2014 ) . While the extent of protein palmitoylation in Toxoplasma has been investigated ( Caballero et al . , 2016; Foe et al . , 2015 ) , the myristoylated proteome remains largely uncharacterised . N-myristoylation is an irreversible , predominantly co-translational covalent addition of myristic acid to an N-terminal glycine ( Boutin , 1997; Gordon et al . , 1991 ) . Functionally , myristoylation often primes proteins for subsequent palmitoylation and a stable protein-membrane association ( Martin et al . , 2011; Wright et al . , 2010 ) , however , it has also been shown to facilitate protein-protein interactions ( PPIs ) ( Chow et al . , 1987; Mousnier et al . , 2018 ) , affect protein activity ( Zhu et al . , 2019 ) as well as structure and stability ( Zheng et al . , 1993 ) . It is catalysed by N-myristoyl transferase ( NMT ) , which is conserved across many organisms , including Toxoplasma , and has been reported to be a prominent drug target in fungal ( Devadas et al . , 1995; Nagarajan et al . , 1997 ) , Trypanosome ( Frearson et al . , 2010; Wright et al . , 2016 ) and Leishmania infections ( Hutton et al . , 2014; Wright et al . , 2015 ) . In Plasmodium falciparum ( the major causative agent of malaria ) , inhibition of NMT leads to severe pleiotropic consequences affecting parasite development ( Schlott et al . , 2019; Wright et al . , 2014 ) , highlighting the importance of myristoylation for parasite survival and progression . An N-terminal glycine ( MG motif ) is a requirement , but not a predictor of myristoylation . Approximately 6% of all gene products in Toxoplasma contain an N-terminal glycine and an in silico prediction of myristoylation suggests that ~ 1 . 8% of all T . gondii gene products are modified ( Alonso et al . , 2019 ) . The functional significance of myristoylation has been described for only a few T . gondii proteins , and mainly in conjunction with adjacent palmitoylation that promotes stable membrane attachment . These proteins include key signal mediators in parasite egress and invasion , for example CDPK3 ( Garrison et al . , 2012; McCoy et al . , 2012 ) , PKG ( Brown et al . , 2017 ) , PKAr ( Jia et al . , 2017b; Uboldi et al . , 2018 ) ; proteins involved in invasion , for example IMP1 ( Jia et al . , 2017a ) ; parasite gliding , for example GAP45 and GAP70 ( Frénal et al . , 2010 ) ; division , for example F-box protein 1 and ISP1 , 2 , 3 ( Baptista et al . , 2019; Beck et al . , 2010 ) ; and correct rhoptry positioning required for invasion , for example ARO ( Cabrera et al . , 2012; Mueller et al . , 2013 ) . Collectively , these studies show key roles for myristoylation throughout the parasite’s lytic cycle , but the function of myristoylation in the absence of palmitoylation and its relationship to other PTMs remains poorly described . By combining several chemoproteomic tools for substrate identification with a small-molecule NMT inhibitor , we provide experimentally-validated libraries of myristoylated as well as glycosylphosphatidylinositol ( GPI ) anchored proteins in T . gondii . We identify all the previously reported myristoylated proteins , as well as novel substrates with heterogeneous localisations and variable functions across the lytic cycle . Furthermore , by analysing substrate orthology in other Apicomplexans we provide new clues to the identity of previously uncharacterized myristoylated proteomes across the phylum . We validate the presence and elucidate the functional importance of myristoylation for the microneme protein MIC7 , a predicted type I transmembrane protein . Utilizing conditional substrate depletion and complementation with wild-type ( cWT ) and myristoylation mutant ( cMut ) versions , we demonstrate that myristoylation of MIC7 is functionally important in host cell invasion . Taken together , our study identifies a large proportion of the Toxoplasma myristoylated proteome and points to unexpected and novel functions of myristoylation in Toxoplasma that extend beyond priming for palmitoylation and stable membrane attachment . To visualise the extent of myristoylation in Toxoplasma , we adapted a metabolic labelling approach that has previously been applied to mammalian cells ( Broncel et al . , 2015; Thinon et al . , 2014 ) and protozoan parasites ( Wright et al . , 2014; Wright et al . , 2016; Wright et al . , 2015 ) . In this workflow , a myristic acid ( Myr ) analogue containing a terminal alkyne group ( YnMyr ) is added to cell culture upon infection with Toxoplasma tachyzoites ( Figure 1A ) . The hydrophobic nature of YnMyr allows for optimal biomimicry of myristate and conversion into the active co-substrate YnMyr-CoA in situ , while the alkyne tag allows for NMT-mediated metabolic labelling of both host and parasite substrate proteins . Upon cell lysis , labelled proteins are liberated and conjugated to azide-bearing multifunctional capture reagents by a click reaction ( Heal et al . , 2012 ) . The conjugation process introduces secondary labels , like biotin and fluorophores , allowing for substrate enrichment on streptavidin beads and visualisation via in-gel fluorescence ( igFL ) , respectively . To investigate the extent of YnMyr incorporation , intracellular tachyzoites were treated with either Myr or increasing concentrations of YnMyr for 16 hr . Within this timeframe protein labelling in vivo did not appear to exert any toxic effects on Toxoplasma parasites . Labelled proteins were then conjugated to a capture reagent and resolved by SDS-PAGE . As visualised by igFL , the labelling was concentration-dependent with only negligible background ( Figure 1—figure supplement 1A ) . In addition , the extent of labelling did not seem to depend on parasite localisation inside or outside the host cell , and was efficiently out-competed by excess myristate , indicating that YnMyr is an effective mimic of myristate in Toxoplasma parasites ( Figure 1—figure supplement 1B ) . To estimate the efficiency of substrate enrichment , we took advantage of the biotin moiety that enables a streptavidin-based pull down . Using igFL as readout we observed robust enrichment of protein substrates in a YnMyr-dependent manner , and detected very little background in controls ( Figure 1—figure supplement 1C ) . It has been reported that in Plasmodium parasites , YnMyr can be incorporated not only at N-terminal glycines via amide bonds , but also through ester-linked incorporation of myristate into GPI anchors ( Wright et al . , 2014 ) . These two distinct types of labelling can be readily distinguished by their different sensitivity to base treatment; amide bonds are stable in basic conditions , whereas ester bonds are hydrolysed . To visualise the extent of YnMyr incorporation into GPI anchors in Toxoplasma , we performed base treatment prior to enrichment of substrate proteins and observed a reduction of igFL signal for selected enriched bands ( Figure 1B ) . To further validate the base treatment approach , we probed known N-myristoylated and GPI-anchored Toxoplasma proteins , GAP45 and SAG1 , for their ability to be enriched in a base-dependent manner . In the absence of treatment , both proteins were robustly pulled down with YnMyr , while upon base treatment , only GAP45 remained enriched , confirming that it is a true myristoylation substrate ( Figure 1C ) . Collectively , we confirmed that YnMyr is a robust and high-fidelity myristate analogue and demonstrated that it can be applied to profile both N-myristoylated and GPI-anchored proteins in live T . gondii . To confidently identify YnMyr-labelled proteins in Toxoplasma , we applied state-of-the-art mass spectrometry ( MS ) -based proteomics combined with validated chemical tools ( Figure 2—figure supplement 1A ) ; ( Broncel et al . , 2015; Speers and Cravatt , 2005; Thinon et al . , 2014; Wright et al . , 2014 ) . We started with a small-scale pilot experiment to test our workflow and differentiate between N-myristoylation-based enrichment and GPI-anchored substrates . We metabolically labelled intracellular tachyzoites of the RH strain ( Huynh and Carruthers , 2009 ) with either YnMyr or Myr each at 25 µM for 16 hr . We then lysed the infected cell monolayers and performed the click reaction with the azido biotin capture reagent ( reagent 1 ) to facilitate YnMyr-dependent enrichment of labelled proteins . To distinguish myristoylated from GPI-anchored substrates , we applied base treatment prior to the streptavidin-based pull down . Following trypsin digestion , we analysed samples by LC-MS/MS and performed label free quantification ( LFQ ) of enriched proteins . We quantified 2363 human and Toxoplasma proteins , 349 of which were parasite proteins with YnMyr intensities irrespective of base treatment ( Supplementary file 1 ) . To identify GPI-anchored proteins , we calculated log2 fold changes between base-treated and untreated samples . To threshold we utilised the least extreme negative value ( log2 fold change < −1 ) quantified from all Surface Antigen Proteins ( SAGs ) detected in our study , which are known to be GPI-anchored ( Figure 2—figure supplement 1B ) . This selection strategy yielded 52 substrates , that included known and predicted GPI-anchored proteins ( Supplementary file 1 ) . To identify myristoylated proteins we utilised a stringent selection method based on three criteria: a ) robust YnMyr/Myr enrichment ( log2 fold change > 2 ) with threshold selected based on known myristoylated proteins ( Figure 2—figure supplement 1C ) , b ) the presence of an MG motif and c ) insensitivity to base treatment . 56 proteins met these criteria , including those previously reported as myristoylated ( Supplementary file 1 ) . Analysis of post enrichment supernatants did not reveal any substantial changes between proteomes of the YnMyr- and Myr-treated samples , confirming that the observed enrichment is not due to globally altered protein abundance ( Figure 2—figure supplement 1D and Supplementary file 1 ) . After successful testing of the metabolic labelling workflow , we performed a more elaborate MS experiment using cleavable capture reagents bearing either trypsin ( reagent 2 ) or TEV ( reagent 3 ) cleavable linkers ( Figure 2—figure supplement 1A ) . In contrast to a non-cleavable reagent ( e . g . reagent 1 ) that provides only indirect proof of substrate myristoylation , cleavable reagents allow for detection of myristoylated peptides in addition to peptides that originate from the enriched proteins ( Figure 2A ) . This additional layer of confidence in MS-based substrate identification is especially important given the high level of non-myristoylation dependent background reported for metabolic labelling with YnMyr ( Broncel et al . , 2015; Wright et al . , 2016; Wright et al . , 2015 ) . While reagent 2 has been validated as a tool for myristoylated protein and peptide discovery ( Broncel et al . , 2015 ) , reagent 3 ( Speers and Cravatt , 2005 ) , which is expected to produce less background and improve myristoylated peptide discovery , has not previously been applied to study protein myristoylation . We therefore first tested reagent 3 in terms of YnMyr-dependent protein enrichment and observed robust pull down of potential NMT substrates ( Figure 2—figure supplement 1E ) . We next generated samples for the MS workflow as described above but , instead of conjugating reagent 1 , we conjugated either 2 or 3 , each in biological triplicate , to labelled proteins via click reaction to enable myristoylation-dependent pull down . As depicted in Figure 2A , reagent 2 requires only a single trypsin digestion step to liberate both unmodified and myristoylated peptides in one pool . By contrast , reagent 3 requires both trypsin and TEV protease digestion and , depending on the enzyme combination , releases unmodified and myristoylated peptides in either one ( TEV I ) or two ( TEV II ) separate fractions ( Figure 2A ) . In the TEV I strategy TEV protease is used to cleave proteins from beads followed by trypsin digestion of proteins into peptides . The cleavage will only occur for proteins bound via the TEV linker and not for the non-specifically bound ones , which should significantly reduce background . In the TEV II strategy , trypsin is used first to remove most proteins from the beads , only retaining the captured myristoylated peptides . These are then specifically released using TEV protease cleavage resulting in much reduced sample complexity , which increases the myristoylated peptide discovery by MS . Following digestion , all samples were subjected to LC-MS/MS , and LFQ was performed to identify proteins robustly enriched in YnMyr-dependent manner . This yielded 206 human and 117 T . gondii proteins bearing an N-terminal MG motif ( Supplementary file 2 ) . Within the parasite protein pool , we obtained statistically significant ( FDR 1% , log2 fold change >2 ) enrichment in YnMyr over Myr controls for 72 potential substrates using reagent 2 ( Supplementary file 2 ) . For reagent 3 , which was used in two different scenarios ( TEV I and TEV II ) resulting in larger variability between replicates , we utilised a fold change based threshold ( log2 fold change >2 ) and obtained 48 robustly enriched proteins ( Supplementary file 2 ) . Reassuringly , we observed a ~ 5 and~8 fold reduction in background in TEV I vs TEV II and TEV I vs reagent 2 , respectively , as shown by the number of proteins quantified in Myr controls ( Supplementary file 2 ) . Collectively we identified 76 significantly YnMyr-enriched proteins utilizing reagents 2 and 3 with an overlap of 60% ( Figure 2B , Supplementary file 2 ) which provides substantial confidence to the accuracy of our results . Application of the same selection criteria to 206 human MG proteins identified in our study yielded 102 potential substrates . 84 of these proteins have previously been reported as myristoylated ( Broncel et al . , 2015; Castrec et al . , 2018; Thinon et al . , 2014 ) , which further strengthens our substrate identification strategy . We next focused on the identification of myristoylated peptides in samples processed with reagents 2 and 3 . Using stringent criteria for the unbiased identification of the myristoylation adduct , as well as manual validation of the acquired MS/MS spectra , we identified 31 myristoylated peptides ( Supplementary file 2 ) , 24 of which were detected using reagent 2 , and 20 using reagent 3 ( Figure 2C ) . None of these peptides were detected in Myr controls , and the myristoylation adduct was not identified on cysteine residues . Despite almost equal numbers of peptides detected by the two reagents , the overlap was only 40% ( Figure 2C ) , confirming the added value of orthogonal methods for modified peptide detection . As envisioned in our design strategy , we obtained an increase in myristoylated peptide discovery in TEV II ( 17 ) vs TEV I ( 12 ) workflow ( Figure 2—figure supplement 1F ) . Finally , to summarise our global proteomic study , we combined our results on both protein enrichment and the modified peptide levels . We filtered for proteins identified with at least two of three capture reagents or proteins for which we detected a lipid modified peptide . This resulted in 65 proteins , of which 48% have direct MS/MS evidence for protein myristoylation ( Supplementary file 2 , Figure 2D ) . Given that our global proteomic screen provided direct proof for substrate myristoylation for approximately 50% of selected proteins , we sought for an alternative strategy for substrate validation using NMT inhibitors ( NMTi ) . Here , parasites are treated with NMTi to specifically reduce the incorporation of YnMyr into nascent proteins , which can be quantified by MS ( Thinon et al . , 2014; Wright et al . , 2014; Wright et al . , 2016; Wright et al . , 2015 ) . In the absence of a dedicated TgNMTi , we used IMP-1002 , a compound recently shown to inhibit NMT of Plasmodium falciparum ( Schlott et al . , 2019 ) which is related to Toxoplasma . Homology modelling ( SWISS-MODEL , [Waterhouse et al . , 2018] ) of the TgNMT sequence onto the available Plasmodium vivax ( another important malaria causing Plasmodium spp . ) NMT crystal structure with bound IMP-1002 ( PDB: 6MB1 , [Schlott et al . , 2019] ) revealed high sequence identity ( 57% ) and showed that all residues directly involved in compound binding are conserved within the TgNMT active site and therefore predicted to adopt an identical structural arrangement ( Figure 3A ) . We therefore reasoned that IMP-1002 should also inhibit TgNMT . To test this , we co-treated intracellular parasites 16 hr post invasion with YnMyr and increasing concentrations of the inhibitor for 5 hr and analysed the effects on YnMyr labelling of Toxoplasma proteins . A dose-dependent drop in igFL labelling of most protein bands was observed and further confirmed by specifically probing for CDPK1 , a substrate identified herein ( Figure 3B ) . Consistent with TgNMT inhibition , CDPK1 pull down was reduced with increasing inhibitor concentrations . This was not a general reduction of protein levels as shown by anti-Toxoplasma antibodies . Plaque assays in the presence of inhibitor showed dose-dependent killing of parasites , suggesting that treatment with IMP-1002 has severe consequences for the in vitro expansion of the tachyzoite population but not for the host cells which appeared unaffected ( Figure 3C ) . Having confirmed target engagement with IMP-1002 , we next performed a large-scale MS-based inhibitor response analysis . Intracellular parasites were fed with 25 µM YnMyr alone or co-incubated with 0 . 05 µM and 0 . 5 µM NMTi in biological triplicates . Samples were then clicked , pulled down and the level of protein myristoylation in response to IMP-1002 quantified by MS and LFQ . No major effect was observed for the lower concentration of the inhibitor , therefore we performed statistical analysis between triplicate samples treated with either only YnMyr , or YnMyr + 0 . 5 µM NMTi . We identified a statistically significant ( FDR 5% , log2 fold change > 0 . 5 ) response for 56 proteins ( Figure 3D , Supplementary file 3 ) . Analysis of total proteomes from inhibited samples confirmed that the observed substrate response was not due to the altered protein abundance ( Figure 3—figure supplement 1 , Supplementary file 3 ) . 49 significant responders contained the MG motif with 47 of these being significantly enriched in our previous experiment , while two proteins were not quantified . Specific dose-responses were plotted for selected proteins identified previously as significantly YnMyr-enriched ( Figure 3E ) . While most showed a robust response to the highest concentration of inhibitor , CDPK3 and PKG did not , despite substantial literature evidence for myristoylation ( Brown et al . , 2017; Garrison et al . , 2012; McCoy et al . , 2012 ) , including the presence of a myristoylated peptide in case of CDPK3 ( Supplementary file 2 ) . In fact , a total of 7 proteins for which the myristoylated peptide was detected did not respond robustly to NMT inhibition ( Supplementary file 3 ) . This behaviour is surprising , however a similar observation has been described for other organisms ( Thinon et al . , 2014; Wright et al . , 2014; Wright et al . , 2016; Wright et al . , 2015 ) and could be due to low protein turnover , higher affinity for NMT , or potential interference from other modifications , such as protein S-acylation for example . A further seven proteins for which we obtained myristoylated peptides were not quantified at the highest concentration of IMP-1002 ( Supplementary file 3 ) , suggesting that their myristoylation state , and therefore enrichment , is most affected by NMT inhibition . A significant response to NMTi was also observed for seven proteins that did not contain the MG motif ( Figure 3D , Supplementary file 3 ) . This could be due to post-translational myristoylation where proteolysis results in formation of N-terminal glycine ( Martin et al . , 2011; Thinon et al . , 2014 ) , a tight association in complex with an NMT substrate ( Thinon et al . , 2014 ) , potential protein mis-annotation or an off-target effect of the inhibitor which was originally designed for PfNMT . Importantly , all non MG proteins previously assigned as sensitive to base treatment , including all SAG proteins , showed no significant response to inhibitor ( Supplementary file 3 , Figure 3F ) thus validating base sensitivity as a means to distinguish YnMyr incorporation at N-terminal glycines or GPI anchors . Finally , despite no apparent effect on host cells in plaque assays ( Figure 3C ) , we also observed a certain level of response to IMP-1002 for host proteins ( Supplementary file 3 ) . This suggests that human NMT is also targeted by this compound , however , without visible impact on the integrity of the monolayer of host cells . To generate a comprehensive list of myristoylated proteins in Toxoplasma , we combined YnMyr enrichment with the substrate response to NMT inhibition and the myristoylated peptide discovery . This stringent selection strategy yielded 65 substrates that were further split based on the confidence level ( Supplementary file 4 ) . 42 substrates were classified as high confidence ( the presence of myristoylated peptide and/or robust response to NMT inhibition with YnMyr enrichment ≥ two capture reagents ) while 19 were classified as medium confidence ( the presence of myristoylated peptide or response to NMT inhibition with YnMyr enrichment with one capture reagent ) . Finally , PKG and PPM5C that did not pass our confidence criteria , yet have been reported as myristoylated by others ( Brown et al . , 2017; Yang et al . , 2019 ) , as well as two proteins that responded to NMT inhibition , yet were not present in our global enrichment analysis , were classified as lower confidence hits ( Supplementary file 4 ) . Our substrate list includes all proteins previously reported as myristoylated , which validates our approach and indicates that this analysis covers a large fraction of the myristoylated proteome in Toxoplasma . Notably , in silico prediction for myristoylation ( Alonso et al . , 2019 ) disagreed with 11 of our high and medium confidence substrates ( Supplementary file 4 ) . This is not necessarily surprising , given that the prediction was based on a consensus myristoylation sequence derived from other organisms ( Martin et al . , 2011 ) , and highlights the importance of experimental validation of NMT substrates . 90% of our substrate pool represent novel substrates of TgNMT . Several of these proteins have previously been shown to play important functions across the lytic cycle , for example CDPK1 ( egress/invasion; [Lourido et al . , 2010] ) ; PPM5C ( attachment; [Yang et al . , 2019] ) ; ARF1 and Rab5B ( trafficking; [Kremer et al . , 2013; Liendo et al . , 2001] ) . Others , for which the precise function has yet to be discovered , were assigned by gene ontology into key functional classes , like kinases , phosphatases , hydrolases and protein binding ( Supplementary file 4 ) . We did not obtain any evidence for myristoylation on known secreted Toxoplasma proteins , such as rhoptry or dense granule proteins , indicating that these are not substrates of host NMT after secretion . Approximately one third of the reported substrates are uncharacterized proteins , indicating that a large amount of myristoylation-related biology is still to be uncovered . As expected , the identified substrates showed heterogeneous localisation ( Figure 4A ) . Utilizing the localisation of organelle proteins by isotope tagging ( LOPIT ) prediction ( Barylyuk et al . , 2020 ) within ToxoDB ( Gajria et al . , 2008 ) , we found proteins from key cellular organelles , including the nucleus , mitochondrion , proteasome and micronemes . In agreement with the functional relevance of myristoylation , we found 50% substrates with known or predicted localisation at the plasma membrane ( PM ) , as well as membrane-bound compartments ( e . g . inner membrane complex ( IMC ) , endoplasmic reticulum ( ER ) , and Golgi apparatus ) . Stable attachment at membranes may require a double acylation , that is both myristoylation and palmitoylation ( Wright et al . , 2010 ) , however , only 30% of our substrates were previously reported to be palmitoylated ( Caballero et al . , 2016; Foe et al . , 2015 ) and Supplementary file 4 ) . Since palmitoylation is frequently enriched at the protein N-terminus , in close proximity to the myristate , we analysed the first 20 amino acid sequences of our substrates ( Figure 4—figure supplement 1A ) and found that approximately half possessed cysteine residues ( sites of palmitoylation ) and , hence , the potential for double acylation . This number correlated well with the 54% palmitoylation prediction ( Ren et al . , 2008 ) for our substrate pool ( Supplementary file 4 ) . The reported and predicted palmitoylation data suggested that 12 of the 18 PM substrates likely utilise double acylation for stable membrane attachment while the remaining six may be targeted to the PM via alternative mechanisms . Of the 9 IMC localised substrates , 8 are reported or predicted as palmitoylated as well as 4 of the 5 Golgi-localised ones . This indicates that double acylation is a strong predictor for membrane targeting , albeit to different localisations within the cell , suggesting that further signals are required for their definitive subcellular localisation . Although the absence of palmitoylation cannot exclude the presence of other secondary signals , such as polybasic regions and PPI sites , which could still aid in PM attachment , we predict that about half of the substrates we identified are likely only myristoylated at the N-terminus . Consistent with this , all cytosolic and proteasome localised substrates were deprived of any palmitoylation and only 3 of the 12 nuclear proteins were shown to be palmitoylated . Within this varied group were CDPK1 ( Ojo et al . , 2010; Pomel et al . , 2008 ) , the two phosphatases PPM2A and PPM2B ( Yang et al . , 2019 ) and , surprisingly , the microneme protein MIC7 ( Meissner et al . , 2002 ) . For these proteins myristoylation is likely to serve a distinct function beyond just a simple PM anchor . The availability of the myristoylated proteomes of Toxoplasma and the related P . falciparum ( Wright et al . , 2014 ) allowed us to investigate conserved and non-conserved features of myristoylation across the Apicomplexa . First , we compared both myristoylated proteomes by converting Plasmodium myristoylated proteins into Toxoplasma orthologues using EuPathDB ( Aurrecoechea et al . , 2017 ) and compared the overlap of both species . This yielded 24 shared substrates , which corresponds to 37% of the Toxoplasma and 63% of the P . falciparum experimentally validated myristoylated proteome ( Figure 4—figure supplement 1B ) . 39 substrates from the Toxoplasma dataset have orthologues in P . falciparum and 30 of them contain the MG motif , hinting towards potentially unexplored PfNMT substrates ( Supplementary file 4 ) . We also investigated substrate orthology with other Apicomplexans ( Figure 4B , Supplementary file 4 ) . This analysis showed that the lowest level of substrate conservation is present in Babesia and Cryptosporidium ( 28 orthologues ) , followed by Plasmodium ( 39 ) , Cyclospora ( 62 ) , Eimeria and Neospora ( 69 ) and finally Hammondia ( 72 ) . Probing these species-specific orthologues for the presence of the MG motif indicates that between 14 ( Babesia ) and 66 ( Hammondia ) proteins could be potential substrates of NMT ( Figure 4B , Supplementary file 4 ) and therefore could also be myristoylated in these species . 13 proteins , including CDPK3 , PKG , PKAr and ARO , were present in all analysed species , suggesting that their myristoylation may be essential across the phylum ( Supplementary file 4 ) . Within our substrate list three proteins were classified as micronemal by LOPIT prediction ( Figure 4A ) . TGGT1_249970 was recently described as a protein on the microneme surface where dual acylation is important for its anchoring into the membrane ( Bullen et al . , 2016 ) . The second protein ( TGGT1_309990 ) is annotated as a multi-pass transmembrane protein of unknown function . The third , and perhaps the most interesting , was the microneme protein MIC7 ( TGGT1_261780 ) . MIC7 has been reported to be a putative type I transmembrane protein , comprising an N-terminal signal peptide , five EGF-like domains , a membrane-spanning region , and a short cytoplasmic tail ( Meissner et al . , 2002 ) . As MIC signal peptides are typically co-translationally cleaved upon entry into the ER ( Soldati et al . , 2001 ) , the presence of a myristate within the classical signal sequence of MIC7 was unusual . In addition , MIC7 has been suggested to be predominantly expressed in bradyzoites ( Meissner et al . , 2002 ) , the lifecycle stage responsible for the chronic phase of T . gondii infection . As our experiments were performed exclusively in tachyzoites , the stage responsible for acute infection , the presence of MIC7 within our dataset could represent a potential false positive identification . To exclude this possibility , we mined MS-based quantification data from an experiment comparing bradyzoite and tachyzoite proteomes ( Young et al . , 2020; PXD019729 ) . The log2 fold changes in protein abundance for MIC7 and the bradyzoite-specific marker MAG1 ( Tu et al . , 2019; Figure 5A , Supplementary file 5 ) revealed that in contrast to MAG1 , MIC7 is expressed in tachyzoites , supporting the MS and transcriptional evidence in ToxoDB . We next aimed to directly validate protein myristoylation using ectopically expressed HA-tagged MIC7 WT and myristoylation mutant ( Mut , G2G3 > KA ) under control of either the endogenous or the strong tubulin promoter . We metabolically labelled parasites with YnMyr and performed a myristoylation-dependent pull down on lysates . Only WT but not the Mut was enriched in this manner ( Figure 5B ) , confirming that MIC7 is indeed myristoylated . To investigate the functional relevance of MIC7 and its myristoylation , we created an inducible knock-out ( iKO ) line using the DiCre/loxP system ( Andenmatten et al . , 2013 ) that we recently optimised in RHΔku80 parasites ( Hunt et al . , 2019 ) . The Mic7 coding sequence was replaced with a floxed , HA-tagged copy of the gene , hereafter called MIC7HA , that could be excised upon rapamycin ( RAPA ) treatment ( Figure 5C ) . We verified correct integration at the endogenous locus and confirmed RAPA-induced excision by PCR ( Figure 5—figure supplement 1 ) . At the protein level , MIC7HA was efficiently depleted 24 hr post RAPA treatment ( Figure 5D ) . Correct trafficking of MIC7HA to micronemes was verified by the co-localisation with the micronemal marker MIC2 ( Figure 5E ) . Upon deletion of Mic7 , parasites no longer formed detectable plaques in host cell monolayers after 5 days in culture , but we could observe very small plaques emerging after 7 days ( Figure 5F ) . Collectively these results demonstrate an important , but non-essential , role for MIC7 in the lytic cycle . To investigate where in the lytic cycle MIC7 plays a role , and test the functional relevance of N-terminal myristoylation , we complemented the iKO line by introducing Ty1-tagged WT or myristoylation defective mutant ( hereafter called cWT and cMut , respectively ) copies of Mic7 into the Uprt locus ( Figure 6A ) . Both inserts were correctly integrated and both complemented lines retained efficient RAPA-induced Mic7 excision ( Figure 6—figure supplement 1 ) and depletion of MIC7HA ( Figure 6B ) . After confirming equivalent and RAPA-insensitive expression of cWT and cMut ( Figure 6B ) , we validated both lines in terms of their myristoylation-dependent enrichment and showed that only the cWT was selectively pulled down after metabolic labelling with YnMyr ( Figure 6C ) . In the next step , we investigated the co-localisation of MIC7HA with cWT and cMut ( Figure 6D ) . Both complementation isoforms localised to the micronemes , indicating that the myristate is not required for the trafficking of MIC7 to this organelle . We next sought to evaluate the role of MIC7 myristoylation in the parasite lytic cycle . While cWT rescued the iKO phenotype upon RAPA treatment , cMut parasites formed substantially smaller plaques under equivalent conditions ( Figure 6E ) . This demonstrates that myristoylation indeed plays a key role in MIC7 function . Given the well-established role of microneme proteins in facilitating host cell penetration , we explored whether myristoylation of MIC7 may be important for invasion . We treated iKO , cWT and cMut parasites with RAPA and performed a red/green assay ( Huynh et al . , 2003 ) which can distinguish invaded from attached parasites . As shown in Figure 6F , we observed efficient invasion of host cells by the cWT parasites . This was not the case in the iKO and cMut lines , where invasion was reduced by 57% and 32% , respectively . Compared to the cWT line , we also observed a consistent 61% drop in the total number of iKO parasites ( Figure 6G ) , which suggests a defect in the attachment to host cells . A modest but non-significant reduction of 15% in attachment was observed in the cMut strain . Collectively , these results indicate that MIC7 plays an important role in Toxoplasma propagation by facilitating parasite attachment and subsequent entry into host cells . Furthermore , myristoylation is not required for sorting MIC7 to the micronemes but appears to be important for its function in invasion of host cells . To monitor MIC7 N-terminus , and thus the fate of the myristate , we generated double tagged MIC7 variants bearing a Myc tag in the ectodomain and a Ty1 tag at the C-terminus . Placing a Myc tag in the region between the MIC7 transmembrane ( TM ) domain and the last predicted EGF domain ( EGF5 ) yielded non-functional protein ( Figure 7—figure supplement 1A ) . To select a likely suitable position for the tag , we resorted to structural predictions . The region between EGF5 and the TM domain ( EGF5-TM , residues 230–284 ) possesses two pairs of cysteine residues and may represent either an extension of EGF5 , or a non-canonical/truncated EGF6 . This could explain the inability to place an epitope tag in this region and yield functional MIC7 . We also excluded EGF3 and EGF4 , that possess the strongest signatures for calcium-binding motifs ( PFAM database domain entry PF07645 ) , which normally imparts a rigid domain arrangement with its neighbours . Finally , we considered possible locations ( Figure 7A ) where some degree of structural flexibility would be most likely . The loop between the last two cysteine residues is variable amongst EGF domains , therefore locations were chosen between C38 and C53 of EGF1 and between C86 and C97 of EGF2 as well as within the linker between EGF1 and EGF2 . Using ectopic expression , we tested all these positions in terms of protein expression and localisation . All three double tagged protein variants localised to micronemes ( Figure 7—figure supplement 1B ) , but the most abundant protein levels were observed when the Myc tag was placed within the EGF2 ( Figure 7—figure supplement 1C ) . This indicated that the tag is well tolerated in this position , readily detected by western blot and could be used for further experiments . Myc tagged cWT and cMut ( hereafter called MyccWT and MyccMut , respectively ) were then inserted into the Uprt locus of the iKO line . Both inserts correctly integrated , the new lines retained efficient RAPA-induced Mic7 excision and depletion of MIC7HA ( Figure 7—figure supplement 1D–F ) . After verifying equivalent and RAPA independent expression of the MyccWT and MyccMut ( Figure 7—figure supplement 1F ) , we confirmed their micronemal localisation by IFA ( Figure 7—figure supplement 1G ) . Several microneme proteins have been shown to dimerise ( Cérède et al . , 2002 ) , which harbours the potential that any additional copy of a protein in a merodiploid strain , even when lacking trafficking information , could piggy back as a heterodimer into the micronemes . Indeed , when testing co-IPs in the absence of RAPA , we observed that both complements can co-IP with MIC7HA ( Figure 7B ) , which is independent of the myristate . We therefore repeated the localisation experiments in the presence of RAPA to delete MIC7HA . As shown in Figure 7—figure supplement 1H , both MyccWT and MyccMut localise to the micronemes in RAPA-treated parasites , confirming that the myristate is not necessary for MIC7 sorting . Plaque assays of RAPA-treated complementation lines showed the expected defect in plaque formation for MyccMut , however , we also observed a small reduction of plaque size for MyccWT expressing parasites as compared to the DMSO control ( Figure 7C ) . This slightly reduced ability to form plaques is likely due to the effect of Myc tag insertion on the MIC7 function . However , since the tag is present in both complemented lines and the ability to form plaques was substantially more impeded in the MyccMut parasites , these lines are still suitable to investigate the specific function of MIC7 myristoylation . In order to shed some light on MIC7 topology within the micronemes as well as the fate of its N-terminal myristate , we performed proteinase K protection assays ( Figure 7D ) . In these experiments , proteins/domains that are accessible to proteinase K after digitonin-mediated plasma membrane permeabilisation are digested , while those retained within organelles , such as the micronemes , are protected from this proteolytic digest . We treated tachyzoites from both MyccWT and MyccMut lines with RAPA to deplete MIC7HA and fed with YnMyr to allow for myristoylation-dependent pull down of the complements . We then subjected parasites to digitonin and proteinase K treatment , followed by detection of Myc and Ty1 tags . As a control we used antibodies against the ectodomain of MIC2 , as it should be protected ( Bullen et al . , 2016 ) . Under these conditions the MIC7 C-terminus was digested , while the N-terminus was protected as visualised by the Ty1 and the Myc antibodies , respectively . As the parasites were treated with YnMyr prior to the experiment , we could use a YnMyr-dependent pull down to demonstrate that the protected N-terminus of MIC7 remains myristoylated ( Figure 7D ) . These results strongly suggest that MIC7 is indeed a transmembrane protein with a myristoylated N-terminus facing the microneme lumen and a short C-terminal cytoplasmic tail that faces the parasite cytoplasm . Having established the presence of MIC7 N-terminal myristoylation within the micronemes , we aimed to perform a more detailed characterisation of its function . First , we repeated invasion assays in large scale as described in Touquet et al . , 2018 . We performed three independent experiments with a total of 15 replicates per tested parasite line ( Figure 7E ) . Taking into account the phenotypic effect for the Myc tag insertion in complemented parasite lines we used untreated iKO parasites as a parental control . Parasites that no longer expressed MIC7HA displayed a 78% decrease in invasiveness when compared to the control . Complementation with MyccWT copy can restore the invasiveness to 61% , while MyccMut reach only 30% invasion capacity . These results are largely consistent with our previous observations ( Figure 6F ) and further confirm the critical role for MIC7 and its myristoylation in the invasion process . To gain a better understanding of the function of MIC7 during host cell penetration , we filmed invasion of the DMSO and RAPA-treated iKO and the complemented lines into GFP-GPI expressing host cells ( Figure 7—videos 1–4 ) . We calculated times of successful and failed invasions for each genetic background and observed that despite different success rates between lines , the parasites that did enter the host cell proceed with a similar speed as the control ( Figure 7F , Figure 7—videos 1–4 ) . This suggests a potential failure to initiate invasion . In several failed invasion events we also observed a notable membrane invagination ( Figure 7—figure supplement 2A , Figure 7—videos 2 and 4 ) , which normally occurs after secretion of the rhoptry neck components required for the formation of the tachyzoite-host cell junction ( Bichet et al . , 2016 ) . It is worth noting , that the immunofluorescent analysis of likely aborted invasion events in the RAPA treated iKO line revealed more than half of extracellular parasites ( 45 of 68 ) show a strong arch-shaped morphology while associated with a host cell ( Figure 7G ) . This is indicative of MIC7 KO parasites exerting force during the attempt to invade , possibly leading to the zoite deformation . The inability to initiate invasion in the iKO and MyccMut lines could be caused by a general defect in gliding motility or a failure to secrete micronemes . However , both circular and helical trails were detected with no obvious differences between parasites ( Figure 7—videos 1–4 , Figure 7—figure supplement 2B ) . We also observed no reduction in microneme secretion , as shown by efficient MIC2 processing in MIC7 KO parasites , even in the presence of 5-benzyl-3-isopropyl-1H-pyrazolo[4 , 3-d]pyrimidin-7 ( 6H ) -one ( BIPPO ) , a phosphodiesterase inhibitor that triggers signalling pathways leading to increased parasite motility and microneme secretion ( Figure 7—figure supplement 2C ) . To get a better understanding of MIC7 distribution in parasites before and during invasion , we co-localised MIC7 with SAG1 , a well characterized surface marker , which can be used to delineate the intra- and extracellular part of invading parasites . This revealed that MIC7 partially redistributes around the periphery of extracellular and invading parasites , but appears absent from the tachyzoite-host cell junction ( Figure 7H ) , a constriction through which the parasite moves as it enters the host cell ( Pavlou et al . , 2018; Tyler et al . , 2011 ) . This redistribution suggests , but is not proof , that MIC7 , as many other microneme proteins , is secreted onto the parasite surface during or after egress from the host cell . Most transmembrane microneme proteins undergo proteolytic maturation near or within the transmembrane domain after egress and during invasion ( Soldati et al . , 2001 ) . This process is facilitated by subtilisin or rhomboid proteases and is thought to relieve the high affinity interactions between the parasite and host cell receptors ( Carruthers , 2006; Dowse and Soldati , 2005 ) . We analysed if MIC7 could undergo similar processing and found that while MIC2 was efficiently cleaved and released into the culture supernatant , this was not the case for MIC7 , even in the presence of BIPPO ( Figure 7I ) . We also tested whether MIC7 could be shed during invasion by analysing the culture supernatant as well as parasite and host material of freshly invaded cells . Again , no MIC7 shedding was observed ( Figure 7—figure supplement 2D ) . These results are in agreement with previously reported observations ( Meissner et al . , 2002 ) , and confirm that MIC7 is not subject to a proteolytic cleavage . The live imaging data of the iKO parasites suggested that MIC7 plays an important role in the onset of invasion , potentially after the establishment of the tachyzoite-host cell junction . In one instance we could observe a membrane swirl formation at the apex of the parasite attempting invasion , suggesting that rhoptry contents may have been secreted ( Figure 7—video 2 ) . Rhoptries are apical organelles which contain a number of kinases the parasite injects into the host cell upon invasion . These effector kinases modulate many important functions in host-microbe interaction ( Boothroyd and Dubremetz , 2008 ) . One of these kinases , the rhoptry kinase 16 ( ROP16 ) acts as a JAK mimetic leading to rapid and direct phosphorylation of STAT6 ( Ong et al . , 2010; Saeij et al . , 2007 ) . Accordingly , we used STAT6 phosphorylation as a reporter for efficient rhoptry secretion and ROP16 translocation into the host cell ( Figure 7J ) . RAPA-treated MyccWT parasites showed 45% pSTAT6 positive nuclei . RAPA-treated iKO and MyccMut parasites showed a significant reduction of pSTAT6 positive nuclei ( 11% and 21% , respectively ) . No significant differences in pSTAT6 positive cells between tested parasite lines were observed when treated with DMSO . This indicates that iKO and MyccMut parasites can still secrete ROP16 into the host cell and induce STAT6 phosphorylation , although at significantly reduced levels . Our understanding of myristoylation and its functional consequences in Toxoplasma is hampered by the limited knowledge of NMT substrates . Using an integrated MS-approach we describe here the first experimentally validated myristoylated proteome in T . gondii . We combine two orthogonal chemoproteomic techniques , that is quantitative response to NMT inhibition with direct MS/MS evidence for substrate modification , which allows for high confidence in substrate identification as well as substantial substrate coverage . Despite the complex nature of our samples , consisting of both human and parasite proteins , our discovery includes all proteins previously reported to be myristoylated in Toxoplasma as well as novel and unexpected TgNMT substrates . The fact that these proteins are functionally diverse , and involved in all steps of the lytic cycle highlights the importance of myristoylation in Toxoplasma biology . Consistent with this , treatment with NMTi resulted in dose dependent parasite killing . Although we cannot exclude small off-target effects of IMP-1002 that was originally designed for Plasmodium spp . , we predict that this severe phenotype was largely due to pleiotropic effects of TgNMT inhibition on the Toxoplasma lytic cycle . Potent and selective TgNMT inhibitors are yet to be reported , however extensive work in other protozoan parasites ( Ritzefeld et al . , 2018 ) demonstrates that selective NMT inhibition could provide an attractive strategy to combat infection . Moreover , results presented here indicate that related parasites , with high structural conservation of their NMTs , could be inhibited by a single compound , which may allow for the development of a pan-parasite inhibitor in the future . Our substrates showed heterogeneous localisation with ca . 50% localised to PM or membrane bound compartments . Palmitoylation analysis confirmed that for the majority of these substrates stable attachment to membranes is likely driven by double acylation . Although we cannot exclude the presence of other secondary signals which could aid in PM targeting within our substrate pool , such as polybasic regions and PPI sites , our analysis and the predicted localisation suggest that many of our substrates may be myristoylated only , indicating that their myristoylation can serve more discrete functions than just a priming site for the palmitate . Such alternate functions could include reversible membrane binding by the conformation regulated exposure of the myristate as shown for mammalian ARF1 ( Goldberg , 1998 ) , regulation of protein activity as demonstrated for GPAT4 during glycerolipid synthesis ( Zhu et al . , 2019 ) , or involvement in PPIs , as shown for the viral capsid assembly ( Chow et al . , 1987; Mousnier et al . , 2018 ) . Here we identified unexpected myristoylation of the Toxoplasma microneme protein 7 ( MIC7 ) . Microneme proteins are key factors in Toxoplasma propagation involved in parasite egress , motility and host cell invasion . They are trafficked into the secretory pathway by virtue of an N-terminal signal peptide that is cleaved during ER import ( Soldati et al . , 2001 ) . We show that MIC7 is an important , yet not essential protein and we unequivocally demonstrate that MIC7 is myristoylated at its N-terminus , excluding the existence of an N-terminal signal sequence . While it is known that proteins can enter the secretory pathway by virtue of a recessed signal or leader peptide , this has not been reported for microneme proteins . Furthermore , many known microneme proteins are type I transmembrane proteins , where an N-terminal PPI or carbohydrate binding domain faces the microneme lumen and a short cytoplasmic domain faces the parasite cytosol . Upon microneme secretion the protein is transferred to the parasite PM , with the interacting domain exposed to bind to host cell receptors . Because MIC7 lacks a signal peptide but otherwise mimics this domain structure , it most likely enters the secretory pathway as a type III transmembrane protein ( Goder and Spiess , 2001 ) . Although we have not specifically tested this here , this is supported by several features in the MIC7 primary amino acid sequence . That is: a transmembrane domain of > 20 amino acids in length ( 22 in MIC7 ) and positive charges on the C-terminal side of the transmembrane domain . We demonstrated that MIC7 and its myristoylation are important in host cell invasion . Live imaging of the invasion process further revealed a potential defect in initiating invasion in the MIC7 KO and myristoylation defective parasites . This phenomenon was not caused by a fault in microneme secretion itself or parasite gliding . The live video analysis suggests that the parasites may be able to initiate the tachyzoite-host cell junction ( supported by some rhoptry content secretion as measured by STAT6 phosphorylation ) , but then fail to progress beyond the initiation . This phenotype is similar to Toxoplasma Myosin A knockout parasites , that are also able to initiate invasion , but then fail to invade given the lack of a functional actin-myosin system . However , the strong arch-shaped deformation of extracellular MIC7 KO parasites in live invasion assays is substantially different from the MyoA knockouts ( Bichet et al . , 2016 ) . While we could not unequivocally show that MIC7 is secreted to the zoite surface to engage with host cell receptors via its EGF domains , the fact that it redistributes around the periphery of the cell shortly after egress and during invasion supports this scenario . When secreted , the myristate could contribute to specific PPIs with other surface proteins or potentially host ligands to support attachment and initiate invasion . It is also tempting to speculate that alternatively , the myristate could be inserted into the lipid bilayer of the host cell . It has been shown that HBV viruses can utilise a myristoylated protein on their surface to enter host cells ( Maurer-Stroh and Eisenhaber , 2004 ) . In Toxoplasma this mechanism could contribute to host cell attachment , for which we observed a measurable defect upon MIC7 deletion , but also to the earliest events of invasion , that is rhoptry secretion or host-cell penetration . This is reminiscent of MIC8 , another microneme protein , that has been implicated in a signalling cascade leading to rhoptry discharge ( Kessler et al . , 2008 ) . Our experiments to probe secretion of ROP16 , which rapidly phosphorylates STAT6 ( Ong et al . , 2010; Saeij et al . , 2007 ) , indicate that rhoptry secretion per se is not affected upon MIC7 deletion , but reduced ROP16 mediated STAT6 phosphorylation is observed . We cannot currently distinguish whether the decrease in ROP16 injected cells is due to decreased invasion efficiency of MIC7 KO and MyccMut lines , or whether equal amounts of cells have been attempted to be invaded but ROP16 has not been efficiently transferred . However , the attachment phenotype observed for MIC7 myristoylation mutants is less profound than the reduction of pSTAT6 induction , suggesting that the major cause of the rhoptry secretion phenotype is independent of attachment and most likely due to an inefficient translocation of rhoptry contents into the host cell . Whether the failure to progress beyond the initiation of the tachyzoite-host cell junction lies in a failure to properly secrete rhoptry contents , or to form the junction itself remains to be clarified . In summary we provide a useful resource of experimentally validated myristoylated and GPI-anchored proteins , as well as first clues to the identity of so far uncharacterized myristoylated proteomes across the phylum . We show that an NMT inhibitor that was generated against Plasmodium spp . also inhibits Toxoplasma growth . The presence of several essential N-myristoylated proteins conserved across many Apicomplexa indicates that NMT inhibition by a single compound may be a viable strategy to target several pathogens . We have identified N-terminal myristoylation on a Toxoplasma protein that uses an unconventional mode of trafficking to the parasite secretory pathway and the micronemes , and displays a novel use of myristoylation in parasite biology . This unexpected discovery , which can likely be found in other organisms , demonstrates how our dataset can serve as a tool in target-specific investigations that can ultimately help to unravel the exciting biology of host-microbe or more broadly , cell-cell interactions . Reagents: CuSO4 , TCEP , TBTA , buffer salts , DTT , iodoacetamide , DMSO , BSA , Triton-X100 and Tween-20 were from Sigma-Aldrich . Azide-PEG3-biotin was from Sigma-Aldrich . Peptide synthesis coupling reagents HATU and HCTU were from Fluorochem and Merck , respectively . MS-grade water , acetonitrile , methanol , TFA and formic acid were from Thermo Scientific . IMP-1002 was synthesised as described in Schlott et al . , 2019 . BIPPO was synthesised as described in Howard et al . , 2015 . Primers used throughout this study are listed in Supplementary file 6 . Plasmid sequences were confirmed by Sanger sequencing ( Eurofins Genomics ) . To generate the Mic7 iKO plasmid , pG140_MIC7_HA_iKO_loxP100 , the Mic7 5’UTR with a loxP site inserted 100 bp upstream of the Mic7 start codon , and a recodonised Mic7 cDNA-HA sequence , were synthesized ( GeneArt strings , Life Technologies ) . These DNA fragments were Gibson cloned into the ApaI/PacI digested parental vector p5RT70loxPKillerRedloxPYFP-HX ( Andenmatten et al . , 2013 ) to generate an intermediate plasmid . The Mic7 3’UTR was subsequently amplified from genomic DNA using primers 1 and 2 , while mCherry flanked by Gra gene UTRs was amplified from pTKO2C ( Caffaro et al . , 2013 ) using primer pair 3/4 . The resulting fragments were Gibson cloned into the SacI-digested intermediate plasmid to generate pG140_MIC7_HA_iKO_loxP100 . To generate the complementation construct pUPRT_MIC7_Ty1 , the Mic7 sequence flanked by its 5’UTR was amplified from genomic DNA using primer pair 5/6 . In parallel , the Uprt targeting vector pUPRT_HA ( Reese et al . , 2011 ) was amplified by inverse PCR using primers 7 and 8 . The resulting PCR amplicons were Gibson cloned to generate pUPRT_MIC7_Ty1 . Primers 5 and 8 comprise overhangs to facilitate introduction of a Ty1 tag 3’ of the Mic7 sequence . To generate the complementation construct pUPRT_MIC7 ( G2K/G3A ) _Ty1 , the Mic7 5’UTR and Mic7 endogenous sequence were amplified using primer pairs 9/10 and 5/11 , respectively . In parallel , the Uprt targeting vector pUPRT_HA ( Reese et al . , 2011 ) was amplified by inverse PCR using primers 7 and 8 . The resulting PCR amplicons were Gibson cloned to generate pUPRT_MIC7 ( G2K/G3A ) _Ty1 . Primers 9 and 11 comprise overhangs that introduce point mutations G2K and G3A , while primers 5 and 8 introduce a Ty1 tag 3’ of the Mic7 sequence . To generate the complementation construct pUPRT_Myc_MIC7_Ty1 , the Myc tag coding sequence was introduced within pUPRT_MIC7_Ty1 plasmid by inverse PCR using primers 12 and 13 . The resulting linear fragment was circularized using KLD reaction buffer ( NEB ) as per manufacturer’s instructions . To generate the complementation construct pUPRT_Myc_MIC7 ( G2K/G3A ) _Ty1 , the Myc tag coding sequence was introduced within pUPRT_MIC7 ( G2K/G3A ) _Ty1 plasmid by inverse PCR using primers 12 and 13 . The resulting linear fragment was circularized using KLD reaction buffer ( NEB ) as per manufacturer’s instructions . To generate pSag1_Cas9-U6_sgMIC7 , the pSag1_Cas9-U6_sgUPRT ( Shen et al . , 2014; Addgene plasmid # 54467 ) vector was amplified by inverse PCR using primers 14 and 15 . Primer 15 comprises a sequence extension that replaces the Uprt-targeting sgRNA with a sgRNA sequence targeting Mic7 . The resulting linear fragment was circularized using KLD reaction buffer ( NEB ) as per manufacturer’s instructions . To generate pGra_5’UTRMIC7_MIC7_HA , the 5’UTR of Mic7 was amplified from gDNA using primer pair 30/31 , and recodonised Mic7 sequence was amplified from pG140_MIC7_HA_iKO_loxP100 using primers 32 and 33 . In parallel , the vector pGra_ApiAT5-3_HA ( Wallbank et al . , 2019 ) was amplified by inverse PCR using primer pair 34/35 . The three resulting PCR amplicons were Gibson assembled to generate pGra_5’UTRMIC7_MIC7_HA . To generate pGra_5’UTRMIC7_MIC7 ( G2K/G3A ) _HA , the 5’UTR of Mic7 was amplified from gDNA using primer pair 30/31 , and recodonised Mic7 ( G2K/G3A ) sequence was amplified from pG140_MIC7_HA_iKO_loxP100 using primers 36 and 33 . Primer 36 was used to introduce the point mutations G2K and G3A into the Mic7 recodonised sequence . In parallel , the vector pGra_ApiAT5-3_HA ( Wallbank et al . , 2019 ) was amplified by inverse PCR using primer pair 34/35 . The three resulting PCR amplicons were Gibson assembled to generate pGra_5’UTRMIC7_MIC7 ( G2K/G3A ) _HA . To generate pGra_5’UTRTUB_MIC7_HA , the Tub 5’UTR was amplified from gDNA using primer pair 37/38 , and recodonised Mic7 sequence was amplified from pG140_MIC7_HA_iKO_loxP100 using primers 39 and 33 . In parallel , the vector pGra_ApiAT5-3_HA ( Wallbank et al . , 2019 ) sequence was amplified by inverse PCR using primer pair 34/40 . The three resulting PCR amplicons were Gibson assembled to generate pGra_5’UTRTUB_MIC7 _HA . To generate pGra_5’UTRTUB_MIC7 ( G2K/G3A ) _HA , the Tub 5’UTR was amplified from gDNA using primer pair 37/38 , and recodonised Mic7 ( G2K/G3A ) sequence was amplified from pG140_MIC7_HA_iKO_loxP100 using primers 41 and 33 . Primer 41 was used to introduce the point mutations G2K and G3A into the Mic7 recodonised sequence . In parallel , the vector pGra_ApiAT5-3_HA ( Wallbank et al . , 2019 ) was amplified by inverse PCR using primer pair 34/40 . The three resulting PCR amplicons were Gibson assembled to generate pGra_5’UTRTUB_MIC7 ( G2K/G3A ) _HA . To generate pGra_5’UTRMIC7_Myc1_MIC7_HA , pGra_5’UTRMIC7_Myc1/2_MIC7_HA and pGra_5’UTRMIC7_Myc2_MIC7_HA , the Myc tag coding sequence was introduced at different positions within pGra_5’UTRMIC7_MIC7_HA plasmid by inverse PCR using primers 42 and 43 ( Myc1 ) , 44 and 45 ( Myc1/2 ) , and 46 and 47 ( Myc2 ) . The resulting linear fragments were circularized using KLD reaction buffer ( NEB ) as per manufacturer’s instructions . Freshly harvested parasites were transfected by electroporation ( 1500 V ) using the Gene Pulser Xcell system ( Bio-Rad ) as previously described ( Soldati and Boothroyd , 1993 ) . To generate the inducible MIC7 knock-out line ( RH DiCre∆ku80∆hxgprt_loxPMIC7_HA , referred to here as iKO MIC7 ) , the plasmid pG140_MIC7_HA_iKO_loxP100 was linearized using PciI and co-transfected with pSag1_Cas9-U6_sgMIC7 into the RH DiCre∆ku80∆hxgprt line ( Hunt et al . , 2019 ) . Recombinant parasites were selected 24 hr post transfection by addition of mycophenolic acid ( MPA; 25 µg/mL ) and xanthine ( XAN; 50 µg/mL ) to culture medium . Resistant non-fluorescent parasites were cloned , and successful 5’ and 3’ integration at the Mic7 locus was confirmed using primer pairs 16/17 and 18/19 , respectively . Absence of the endogenous Mic7 locus was confirmed using primers 24 and 25 . Rapamycin-induced excision of the loxPMic7 sequence was confirmed using primer pair 26/27 . To complement the iKO MIC7 line with MIC7-expressing constructs , pUPRT_MIC7_Ty1 and pUPRT_MIC7 ( G2K/G3A ) _Ty1 or pUPRT_Myc_MIC7_Ty1 and pUPRT_Myc_MIC7 ( G2K/G3A ) _Ty1 plasmids were linearized with ScaI and individually co-transfected with pSAG1_Cas9-U6_sgUPRT . Transgenic parasites were subjected to 5'-fluo-2'-deoxyuridine ( FUDR ) selection ( 5 µM ) 24 hr post transfection . Resistant parasites were cloned , and successful 5’ and 3’ integration was confirmed using primer pairs 20/21 and 22/23 . Disruption of the endogenous Uprt locus was confirmed using primer pair 28/29 . To generate lines that express WT and myristoylation mutant ( G2K/G3A ) MIC7 ectopically , plasmids pGra_5’UTRMIC7_MIC7_HA , pGra_5’UTRMIC7_MIC7 ( G2K/G3A ) _HA , pGra_5’UTRTUB_MIC7_HA , and pGra_5’UTRTUB_MIC7 ( G2K/G3A ) _HA were linearized using NotI and individually transfected into the RH ∆hxgprt strain . Recombinant parasites were selected 24 hr post transfection by addition of mycophenolic acid ( MPA; 25 µg/mL ) and xanthine ( XAN; 50 µg/mL ) to culture medium . To generate lines that ectopically express MIC7 with Myc tag within the EGF1 , EGF1/2 or EGF2 domains , plasmids pGra_5’UTRMIC7_Myc1_MIC7_HA , pGra_5’UTRMIC7_Myc1/2_MIC7_HA and pGra_5’UTRMIC7_Myc2_MIC7_HA were linearized using NotI and individually transfected into the RH ∆hxgprt strain . Recombinant parasites were selected 24 hr post transfection by addition of mycophenolic acid ( MPA; 25 µg/mL ) and xanthine ( XAN; 50 µg/mL ) to culture medium . Parasites of the RH strain were cultured in Human foreskin fibroblasts ( HFFs ) monolayers in Dulbecco's Modified Eagle Medium ( DMEM ) , GlutaMAX ( Thermo Fisher ) supplemented with 10% heat-inactivated foetal bovine serum ( FBS; Life technologies ) , at 37°C and 5% CO2 . All strains and host cell lines tested negative for the presence of mycoplasma . Upon infection of HFF monolayers the medium was removed and replaced by fresh culture medium supplemented with 25 µM YnMyr ( Iris Biotech ) or Myr ( Tokyo Chemical Industry ) . The parasites were then incubated for 16 hr , washed with PBS ( 2x ) and lysed on ice using a lysis buffer ( PBS , 0 . 1% SDS , 1% Triton X-100 , EDTA-free complete protease inhibitor ( Roche Diagnostics ) ) . Lysates were kept on ice for 20 min and centrifuged at 17 , 000 × g for 20 min to remove insoluble material . Supernatants were collected and protein concentration was determined using a BCA protein assay kit ( Pierce ) . Lysates were thawed on ice . Proteins ( 100–300 µg ) were taken and diluted to 1 mg/mL using the lysis buffer . A click mixture was prepared by adding reagents in the following order and by vortexing between the addition of each reagent: a capture reagent ( stock solution 10 mM in water , final concentration 0 . 1 mM ) , CuSO4 ( stock solution 50 mM in water , final concentration 1 mM ) , TCEP ( stock solution 50 mM in water , final concentration 1 mM ) , TBTA ( stock solution 10 mM in DMSO , final concentration 0 . 1 mM ) . Following the addition of the click mixture the samples were vortexed ( room temperature , 1 hr ) , and the reaction was stopped by addition of EDTA ( final concentration 10 mM ) . Subsequently , proteins were precipitated ( chloroform/methanol , 0 . 25:1 , relative to the sample volume ) , the precipitates isolated by centrifugation ( 17 , 000 x g , 10 min ) , washed with methanol ( 1 × 400 µL ) and air dried ( 10 min ) . The pellets were then resuspended ( final concentration 1 mg/mL , PBS , 0 . 4% SDS ) and the precipitation step was repeated to remove excess of the capture reagent . Next , samples were added to 15 µL of pre-washed ( 0 . 2% SDS in PBS ( 3 × 500 µL ) ) Dynabeads MyOne Streptavidin C1 ( Invitrogen ) and gently vortexed for 90 min . The supernatant was removed and the beads were washed with 0 . 2% SDS in PBS ( 3 × 500 µL ) . Beads were supplemented with 2% SDS in PBS ( 20 µL ) and 4x SLB ( Invitrogen ) , boiled ( 95°C , 10 min ) , centrifuged ( 1000 x g , 2 min ) and loaded on 10% or 4–20% SDS-PAGE gel ( Bio-Rad ) . Following electrophoresis ( 60 min , 160V ) , gels were washed with water ( 3x ) . In-gel fluorescence was detected using a Pharos FX Plus Imager ( Bio-Rad ) and the protein loading was checked by Coomassie staining . For western blotting proteins were transferred ( 25 V , 1 . 3 A , 7 min ) onto nitrocellulose membranes ( Bio-Rad ) using Bio-Rad Trans Blot Turbo transfer system . After a brief wash with PBS-T ( PBS , 0 . 1% Tween-20 ) membranes were blocked ( 5% milk , TBS-T , 1 hr ) and incubated with primary antibodies ( 5% milk , TBS-T , overnight , 4°C ) at the following dilutions: rat anti-HA ( 1:1000; Roche Diagnostics ) , mouse anti-Myc ( 1:1000; Millipore ) , mouse anti-Ty1 ( 1:2000; Thermo Fisher ) , rabbit anti-Gra29 ( 1:1000; [Young et al . , 2020] ) , rabbit anti-SFP1 ( 1:1000; [Young et al . , 2020] ) , mouse anti-Toxoplasma ( 1:1000; Abcam ) , mouse anti-CDPK1 ( 1:3000; Matt Bogyo Lab ) , rabbit anti-SAG1 ( 1:10 , 000; John Boothroyd Lab ) , rabbit anti-GAP45 ( 1:1000; Peter Bradley Lab ) , rabbit anti-TgCAP ( 1:2000; [Hunt et al . , 2019] ) , rabbit anti-MIC2 ( 1:500; Vernon Carruthers Lab ) , and mouse anti-MIC2 ( 6D10 ) ( 1:1000; Vernon Carruthers Lab ) . Following washing ( TBS-T , 3x ) membranes were incubated with IR dye-conjugated secondary antibodies from LI-COR Biosciences ( 1:10 , 000 , 5% milk , TBS-T , 1 hr ) , and after a final washing step imaged on a LiCOR Odyssey imaging system ( LI-COR Biosciences ) . In case of a biotin western blot , membranes were blocked with 3% BSA and incubated with Streptavidin-HRP ( 1:4000; Thermo Scientific ) in 0 . 3% BSA , PBS-T for 1 hr . ECL western blotting Detection Reagent ( GE Healthcare ) was then used for chemiluminescence based imaging on a ChemiDoc MP Imaging System ( Bio-Rad ) . TEV reagent: Solid phase synthesis took place on a CF peptide synthesizer ( Intavis ) using a Rink Amide LL resin ( 100 µmol; Merck ) and N ( α ) -Fmoc amino acids , including Fmoc-Lys ( N3 ) -OH ( Fluorochem ) and Fmoc-Gly- ( Dmb ) Gly-OH ( Merck ) . HCTU was used as the coupling reagent with 5-fold excess of amino acids . Fmoc-Lys ( Biotin ) -OH ( four eq; Merck ) in 6 mL DMSO:NMP ( 1:1 ) was coupled manually after automated assembly of the rest of the chain . DIPEA ( four eq ) was added , followed by HOBt ( 1 M , four eq ) in NMP . After 3 min DIC ( four eq ) was added , then after 30 min the solution was added to the resin and allowed to react overnight . The resin was washed with DCM and DMF prior to manual Fmoc removal and acetylation . The peptide was cleaved from the resin and protecting groups removed by addition of a cleavage solution ( 95% TFA , 2 . 5% H2O , 2 . 5% TIS ) . After 2 hr , the resin was removed by filtration and peptides were precipitated with diethyl ether on ice . The peptide was isolated by centrifugation , then dissolved in H2O and freeze dried overnight . After dissolving in methanol , portions of the peptide were purified on a C8 reverse phase HPLC column ( Agilent PrepHT Zorbax 300 SB-C8 , 21 . 2 × 250 mm , 7 m ) using a linear solvent gradient of 13–50% MeCN ( 0 . 08% TFA ) in H2O ( 0 . 08% TFA ) over 40 min at a flow rate of 8 mL/min . The peak fraction was analysed by LC–MS on an Agilent 1100 LC-MSD . The calculated molecular weight of the peptide was in agreement with the mass found . Calculated MW: 1804 . 08 , actual mass: 1803 . 87 . Trypsin reagent: Solid phase synthesis took place on a CF peptide synthesizer ( Intavis ) using a Fmoc-PEG-Biotin NovaTag resin ( 100 µmol; Merck ) , 2-Azidoacetic acid ( Fluorochem ) and N ( α ) -Fmoc amino acids , including Fmoc-Lys ( MMT ) -OH ( Merck ) . HATU was used as the coupling reagent with 5-fold excess of amino acids . Following chain assembly , the MMT protecting group was removed from the peptidyl-resin by treatment with 1% TFA in DCM ( 10 mL for 2 min x 8 ) and the resin washed with DCM and DMF . Next 5-TAMRA ( four eq; Anaspec ) was dissolved in 1 mL DMSO:NMP ( 1:1 ) . DIPEA ( four eq ) was added , followed by HOBt ( 1 M , four eq ) in NMP . After 3 min DIC ( four eq ) was added , then after 30 min the solution was added to the resin and allowed to react overnight . After washing the resin with DMF and DCM , the peptide was cleaved from the resin and protecting groups removed by addition of a cleavage solution ( 95% TFA , 2 . 5% H2O , 2 . 5% TIS ) . After 2 hr , the resin was removed by filtration and peptides were precipitated with diethyl ether on ice . The peptide was isolated by centrifugation , then dissolved in H2O and freeze dried overnight . After dissolving in MeCN:H2O ( 1:1 ) , portions of the peptide were purified on a C8 reverse phase HPLC column ( Agilent PrepHT Zorbax 300 SB-C8 , 21 . 2 × 250 mm , 7 m ) using a linear solvent gradient of 10–50% MeCN ( 0 . 08% TFA ) in H2O ( 0 . 08% TFA ) over 40 min at a flow rate of 8 mL/min . The peak fraction was analysed by LC–MS on an Agilent 1100 LC-MSD . The calculated molecular weight of the peptide was in agreement with the mass found . Calculated MW: 1396 . 31 , actual mass: 1395 . 60 . Click reaction - Reagent 1 and 2: lysates were thawed on ice and the click reaction was carried out with 1 mg of proteins at 2 mg/mL . Proteins were captured by adding a mixture of respective capture reagent ( final concentration 0 . 1 mM ) , CuSO4 ( final concentration 1 mM ) , TCEP ( final concentration 1 mM ) and TBTA ( final concentration 0 . 1 mM ) . The samples were vortex-mixed ( room temperature , 1 hr ) before the addition of EDTA ( final concentration 10 mM ) , methanol ( four volumes ) , chloroform ( 1 vol ) , and water ( three volumes ) . The samples were vortex-mixed briefly , centrifuged ( 10 , 000 × g , 20 min ) and the resulting pellets were either washed with methanol ( four volumes ) and dried ( reagent 1 ) or resuspended ( at 2 mg/mL , 1% SDS in PBS ) after which the precipitation step was repeated and the resulting pellets washed with methanol ( four volumes ) and dried ( reagent 2 ) . Reagent 3: lysates were thawed on ice and the click reaction was carried out with 1 mg of proteins at 2 mg/mL . Proteins were captured by sequential addition of the capture reagent ( final concentration 0 . 1 mM ) , TCEP ( final concentration 1 mM ) , TBTA ( stock in DMSO:t-Butanol 1:4 , final concentration 0 . 1 mM ) and CuSO4 ( final concentration 1 mM ) with mixing between each step . The samples were incubated at room temperature for 1 hr before the addition of EDTA ( final concentration 10 mM ) , methanol ( four volumes ) , chloroform ( 1 vol ) , and water ( three volumes ) . The samples were vortex-mixed briefly , centrifuged ( 10 , 000 × g , 20 min ) and the resulting pellets were washed with methanol ( four volumes ) and dried . Subsequently , the dried pellets were resuspended in 2% SDS in PBS and , once completely dissolved , PBS was added ( final concentration 0 . 8% SDS , 2 mg/mL ) . For samples treated with base , NaOH was added ( final concentration 0 . 2 M , 1 hr ) followed by neutralisation with equivalent amount of HCl . Base-treated and untreated samples were then diluted ( 1 mg/mL , 0 . 4% SDS , 1 mM DTT ) before pull down . Pull down , reduction and alkylation - NeutrAvidin agarose resin ( Thermo Scientific ) was washed with 0 . 2% SDS in PBS ( 3x ) . Typically , 50 µL of bead slurry was used for 1 mg of lysate . The samples were added to beads and the enrichment was carried out with gentle mixing ( 2 hr , room temperature ) . Following the removal of supernatants , the beads were sequentially washed with 1% SDS in PBS ( 3x ) , 4 M urea in PBS ( 2x ) and 50 mM ammonium bicarbonate ( 3x ) . The samples were reduced ( 5 mM DTT , 56°C , 30 min ) and cysteines alkylated ( 10 mM iodoacetamide , room temperature , 30 min ) in the dark with washing the beads ( 2x , 50 mM ammonium bicarbonate ) after each step . Protein digestion - for samples processed with reagent 1 and 2 as well as for supernatants ( proteomes ) MS grade trypsin ( Promega ) was used at 1:1000 w/w protease:protein , and samples were incubated overnight at 37°C . For reagent 3 two digestion strategies were used . TEV I: beads were washed ( 2x ) with water followed by TEV buffer ( 50 mM TrisHCl , 0 . 5 mM EDTA , 1 mM DTT , pH 8 . 0 ) and the TEV protease ( 50 units; Invitrogen ) was added . Samples were incubated overnight at 30°C . Supernatant was then removed and beads washed with TEV buffer ( 1x , 50 µL ) . The wash fraction was combined with the supernatant and stored at 4°C . A fresh portion of TEV protease ( 20 units ) was then added to beads which were incubated for additional 6 hr at 30°C . The supernatant and wash were combined with the first TEV elution . MS grade trypsin was subsequently added at 1:1000 w/w protease:protein , and samples were incubated overnight at 37°C . TEV II: samples were incubated overnight at 37°C with MS grade trypsin at 1:1000 w/w protease:protein . The supernatant was removed ( fraction 1 ) and beads washed with water and TEV buffer ( 2x each ) . The TEV protease was then added ( 50 units ) and beads incubated overnight at 30°C ( fraction 2 ) . Stage tip - samples were desalted prior to LC-MS/MS using Empore C18 discs ( 3M ) . Each stage tip was packed with one C18 disc , conditioned with 100 µL of 100% methanol , followed by 200 µL of 1% TFA . The samples were loaded in 1% TFA , washed 3 times with 200 µL of 1% TFA and eluted with 50 µL of 50% acetonitrile , 5% TFA . Desalted peptides were vacuum dried in preparation for LC-MS/MS analysis . HFFs were infected with Toxoplasma and cultured for 16 hr . The medium was then replaced and intracellular parasites co-treated with 25 µM YnMyr and the indicated concentrations of IMP-1002 for 5 hr . Following PBS wash ( 2x ) the cells were lysed on ice using the lysis buffer and further processed exactly as described above . The click reaction , pull down on NeutrAvidin beads and the MS sample prep were performed as described above for reagent 1 and 2 . Samples were resuspended in 0 . 1% TFA and loaded on a 50 cm Easy Spray PepMap column ( 75 μm inner diameter , 2 μm particle size , Thermo Fisher Scientific ) equipped with an integrated electrospray emitter . Reverse phase chromatography was performed using the RSLC nano U3000 ( Thermo Fisher Scientific ) with a binary buffer system ( solvent A: 0 . 1% formic acid , 5% DMSO; solvent B: 80% acetonitrile , 0 . 1% formic acid , 5% DMSO ) at a flow rate of 250 nL/min . Samples processed with reagent 1 were run on a linear gradient of 2–35% B in 90 min with a total run time of 120 min including column conditioning . Samples processed with reagents 2 and 3 were run on a linear gradient of 2–40% B or 2–55% B ( TEV II myristoylated peptide fraction ) in 155 min with a total run time of 180 min including column conditioning . The nanoLC was coupled to a Q Exactive mass spectrometer using an EasySpray nano source ( both Thermo Fisher Scientific ) . The Q Exactive was operated in data-dependent mode , acquiring HCD MS/MS scans ( R = 17 , 500 ) after an MS1 survey scan ( R = 70 , 000 ) on the 10 most abundant ions using MS1 target of 1E6 ions , and MS2 target of 5E4 ions . The maximum ion injection time utilised for MS2 scans was 120 ms , the HCD normalised collision energy was set at 28 and the dynamic exclusion was set at 30 s . The peptide match and isotope exclusion functions were enabled . NMTi samples were run on a linear gradient of 2–20% B in 55 min , followed by 20–40% B in 35 min and 40–60% B in 5 min with a total run time of 120 min including column conditioning . The nanoLC was coupled to a Orbitrap Lumos mass spectrometer using an EasySpray nano source ( both Thermo Fisher Scientific ) . The Orbitrap Lumos was operated in data-dependent mode ( 3 s cycle time ) , acquiring IT HCD MS/MS scans in rapid mode after an MS1 survey scan ( R = 120 , 000 ) . The MS1 target was 4E5 ions whereas the MS2 target was 2E3 ions . The maximum ion injection time utilised for MS2 scans was 300 ms , the HCD normalised collision energy was set at 32 and the dynamic exclusion was set at 30 s . Acquired raw files were processed with MaxQuant , versions 1 . 5 . 0 . 25 and 1 . 5 . 2 . 8 ( Cox and Mann , 2008 ) . Peptides were identified from the MS/MS spectra searched against Toxoplasma gondii ( combined TG1 , ME49 and VEG proteomes , ToxoDB ) and Homo sapiens ( UniProt ) proteomes using Andromeda ( Cox et al . , 2011 ) search engine . Cysteine carbamidomethylation was selected as a fixed modification and methionine oxidation was selected as a variable modification . The enzyme specificity was set to trypsin with a maximum of 2 missed cleavages . The precursor mass tolerance was set to 20 ppm for the first search ( used for mass re-calibration ) and to 4 . 5 ppm for the main search . The datasets were filtered on posterior error probability ( PEP ) to achieve a 1% false discovery rate on protein , peptide and site level . Other parameters were used as pre-set in the software . ‘Unique and razor peptides’ mode was selected to allow identification and quantification of proteins in groups ( razor peptides are uniquely assigned to protein groups and not to individual proteins ) . Label-free quantification ( LFQ ) in MaxQuant was performed using a built-in label-free quantification algorithm ( Cox and Mann , 2008 ) enabling the ‘Match between runs’ option ( time window 0 . 7 min ) within replicates . Each experiment comprised of replicates treated with YnMyr and the same number of replicates treated with Myr control or NMTi . The LFQ is based on intensities of proteins calculated by MaxQuant from peak intensities and based on the ion currents carried by peptides whose sequences match a specific protein or a protein group to provide an approximation of abundance . Myristoylated peptide search in MaxQuant was performed as described above applying the following variable modifications: cysteine carbamidomethylation , +463 . 2907 ( reagent 2 ) and +491 . 3220 ( reagent 3 ) at any peptide N-terminus and cysteine residues . In addition , the minimum peptide length was reduced to six amino acids and the ‘Match between runs’ option was disabled . MaxQuant utilises a scoring algorithm when matching experimental MS/MS spectra with a library of theoretical spectra generated from the in silico digestion of proteins within databases selected for the search . The algorithm is used to evaluate the quality of peptide-spectrum matches ( PSMs ) . To each PSM , MaxQuant also attributes a delta score , which is a difference between scores associated with the match to the best peptide candidate and the second best match within the database . The higher the score and the delta score , the more reliable the identification . In order to reduce a possibility for a false peptide sequence assignment even further , we applied relatively high delta score thresholds ( 20 vs 6 pre-set as default ) for all myristoylated peptides in our analysis . MaxQuant output files were processed with Perseus , version 1 . 5 . 0 . 9 ( Tyanova et al . , 2016 ) as described in the Results section and in Supplementary files 1–3 . The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE ( Perez-Riverol et al . , 2019 ) partner repository with the dataset identifier PXD019677 . The model of TgNMT was generated using SWISS-MODEL ( Waterhouse et al . , 2018 ) , and aligned with a crystal structure of PvNMT bound to NMT inhibitor IMP-1002 ( PDB: 6MB1 ) . The structural image was generated using PyMOL ( Schrodinger LLC ( 2010 ) , The PyMOL Molecular Graphics System , Version 1 . 3r1 ) . HFF monolayers were infected with Pru ∆hxgprt parasites in triplicate . For tachyzoite samples an MOI of 1 was used for a 27 hr infection . For bradyzoite samples monolayers were infected at an MOI of 0 . 8 for 3 . 5 hr , washed and grown in switch conditions ( RPMI , 1% FBS , pH 8 . 1 , ambient CO2 ) for 3 days . Triplicate samples were lysed in 2 mL ice cold lysis buffer ( 50 mM Tris-HCl , 75 mM NaCl , 8 M urea , pH 8 . 2 ) , supplemented with protease ( Roche Diagnostics ) and phosphatase ( Phos Stop , Roche Diagnostics ) inhibitors . Lysis was followed by sonication to reduce sample viscosity ( 30% duty cycle , 3 × 30 s bursts , on ice ) . Protein concentration was measured using a BCA protein assay kit ( Pierce ) . Lysates ( 1 mg per condition ) were subsequently processed for mass spectrometry as described ( Young et al . , 2020 ) and data analysis performed as explained in Supplementary file 5 . For full dataset please see Young et al . , 2020 and PXD019729 . Parasites were allowed to invade HFFs for 2 hr and then treated with 50 nM rapamycin ( Sigma-Aldrich ) or an equivalent volume of vehicle ( DMSO ) for 4 hr . The medium was then replaced and the parasites allowed to grow for at least 24 hr prior to PCR and western blot analysis . Parasite preparation for large scale invasion assay/live microscopy/gliding: HFF monolayers in T25 flasks were infected in culture conditions ( 37°C and 5% CO2 ) with recently egressed tachyzoites to achieve a one to two-per-cell parasite infection . Non-internalized parasites were removed with PBS , and the infected monolayers were cultivated for about 2 hr in complete culture medium ( DMEM supplemented with 10% FCS , 10 mM HEPES , 100 units/mL penicillin , and 100 mg/mL streptomycin ) . After 2 hr incubation the parasites were treated with 50 nM rapamycin or vehicle . Following a 14 hr incubation the medium was replaced with the complete medium and tachyzoites were used within 2 to 5 hr post-egress . Parasites were harvested by syringe lysis , counted , and 400 parasites were seeded on confluent HFF monolayers grown in 24-well plates ( Falcon ) . Parasites were allowed to invade overnight prior to treatment with 50 nM rapamycin or vehicle ( DMSO ) for 4 hr . Following medium replacement to standard culture medium , plaques were allowed to form for 5 days . iKO MIC7 line: Parasites were harvested by syringe lysis , counted , and 100 parasites were seeded on confluent HFF monolayers grown in 24-well plates ( Falcon ) . Parasites were allowed to invade overnight prior to treatment with 50 nM rapamycin or vehicle ( DMSO ) for 4 hr . Following medium replacement to standard culture medium , plaques were allowed to form for 7 days . NMTi: Parasites were harvested by syringe lysis , counted , and 200 parasites were seeded on confluent HFF monolayers grown in the presence of IMP-1002 for 5 days . Plaque formation was assessed by inspecting the methanol fixed and 0 . 1% crystal violet stained HFF monolayers . Parasite-infected HFF monolayers grown on glass coverslips were fixed with 3% formaldehyde for 15 min prior to washing with PBS . Fixed cells were then permeabilised ( 0 . 2% Triton X-100/PBS , 10 min ) , blocked ( 3% BSA/PBS , 1 hr ) and stained for 1 hr with primary antibodies at the following dilutions: rat anti-HA ( 1:1000; Roche ) , mouse anti-Myc ( 1:1000; Millipore ) , mouse anti-Ty1 ( 1:500; Thermo Fisher ) , rabbit anti-MIC2 ( 1:5000; Vernon Carruthers Lab ) . Labelled proteins were visualised with Alexa Fluor-conjugated secondary goat antibodies ( 1:2000 , 1 hr; Life Technologies ) . Nuclei were visualised with the DNA stain DAPI ( 5 µg/mL; Sigma ) supplemented with the secondary antibody . Stained coverslips were mounted on glass slides with Slowfade ( Life Technologies ) and imaged on a Nikon Eclipse Ti-U inverted fluorescent microscope using 100x oil objective . Images were analysed using Nikon NIS-Elements imaging software . Parasites were treated with 50 nM rapamycin for 4 hr and after replacing the medium allowed to grow for 24 hr . Red/green invasion assays were then performed . Parasites were lysed in an invasion non-permissive buffer , Endo buffer ( 44 . 7 mM K2SO4 , 10 mM MgSO4 , 106 mM sucrose , 5 mM glucose , 20 mM Tris–H2SO4 , 3 . 5 mg/mL BSA , pH 8 . 2 ) . 250 µL of 8E5 parasites/mL in Endo buffer were added to each well of a 24-well flat-bottom plate ( Falcon ) containing a coverslip with a confluent HFF monolayer . The plates were spun at 129 x g for 1 min at 37°C to deposit parasites onto the monolayer . The Endo buffer was gently removed and replaced with invasion permissive medium ( 1% FBS/DMEM ) . Parasites were allowed to invade for 15 min at 37°C , after which the monolayer was gently washed with PBS and fixed with 3% formaldehyde for 15 min at room temperature . Extracellular ( attached ) parasites were stained with mouse anti-Toxoplasma ( TP3 ) ( 1:1000; Abcam ) and goat anti-mouse Alexa Fluor 488 before permeabilisation ( 0 . 2% Triton X-100/PBS ) and detection of intracellular ( invaded ) parasites with rabbit anti-TgCAP ( 1:2000; [Hunt et al . , 2019] ) and goat anti-rabbit Alexa Fluor 594 . For each replicate , at least five random fields were imaged with a 40x objective . Three independent experiments were performed in duplicate . The number of intracellular ( 594+/488- ) and extracellular ( 594+/488+ ) parasites was determined by counting , in a blinded fashion , at least 275 parasites per strain . The parasite counts in the MIC7 iKO and cMut lines were normalised to the cWT , and results were statistically tested with a one-way ANOVA with Dunnett’s multiple comparison test in GraphPad Prism 8 . The data are presented as mean ± SD . For estimation of the parasite attachment efficiency , the number of all ( 594+ ) parasites was used and the results were statistically tested as above . Cell Invasion - HFFs were seeded at a density of 2E4 cells per well into 96-well plate and cultivated in complete medium at 37°C and 5% CO2 for 24 hr to allow for sub-confluence . 5E6 to E7 parasites were collected upon spontaneous egress from synchronously infected HFF monolayers . The supernatant was centrifuged at higher speed ( 900 x g , 7 min ) to collect parasites that were gently suspended in 2 mL of complete medium before counting . 2 . 5E5 parasites were added to each well . To synchronize invasion , the 96-well plate was centrifuged ( 300 x g , 3 min ) and incubated for 30 min at 37°C and 5% CO2 . After gentle aspiration , invasion was stopped by addition of 3 . 2% paraformaldehyde ( PFA ) in PBS , pH 7 . 5 ( 20 min ) . Parasite staining - immunostaining was performed first under conditions that did not permeabilise the HFF cells to allow discriminating between the extracellular and intracellular parasites . Fixed cell samples were incubated in 2%BSA/PBS as a blocking buffer ( BB ) for 20 min . Extracellular tachyzoites were selectively stained using mouse anti-TgSAG1 ( TP3 ) ( 1 mg/mL stock , 1:600 , 40 min; Novocastra ) followed by Alexa Fluor 488-conjugated highly cross-adsorbed ( HCA ) anti-mouse antibody ( 2 mg/mL stock , 1:800 , 1 hr ) . The excess of reagents was washed off with PBS and cells were permeabilised ( 0 . 2% Triton X-100/PBS , 10 min ) prior to incubation in BB and the second staining step using anti-TgSAG1 ( 1 mg/mL stock , 1:600 , 1 hr ) but followed by Alexa Fluor 594-conjugated HCA anti-mouse antibody ( 2 mg /ml stock , 1:800 , 1 hr ) . Cell nuclei were stained with 500 nM DAPI and the 96 well plates were automatically scanned to quantify the average number of cells per well . The nuclei of parasites are detected by blue fluorescence whereas the intracellular tachyzoites by red fluorescence and the extracellular ones by yellow fluorescence ( as the result of green and red fluorescence ) . Quantification - samples were automatically scanned at a magnification of 20x under an Olympus ScanR automated inverted microscope . Images were acquired for five wells per parasite strain for each invasion assay , with 16 randomly scanned fields per well and further processed with ScanR software . ScanRAnalysis includes algorithms to provide automated cell nuclei segmentation following signal-to-noise ratio optimisation and accurate cell surface mask definition . To identify intracellular ( red ) over extracellular ( yellow ) parasites , image subtraction from each channel was automatically obtained . Data collected allowed determining the total number of intracellular tachyzoites over the total number of host cells for each well . Three independent assays were carried out , and data were statistically analysed using a two-tailed Student’s t-test in GraphPad Prism 8 . The data are presented as mean ± SD . Preparation of human cells - HFF and Human Bone OsteoSarcoma cells ( U2OS ) that stably expressed the GFP-GPI plasma membrane reporter , were seeded at a density of 3E5 cells per 18 mm glass coverslip , previously coated with poly-L-lysine ( 50 μg/mL ) . Cells were cultivated in complete medium at 37°C and 5% CO2 for 24 hr to allow for 80% confluence . Coverslips were placed in Chamlide chambers ( LCI Corp . ) and covered with a minimal volume ( i . e . 100 µL ) of motility buffer ( see below ) . Preparation of parasites – 2E5 to 4E5 parasites were typically collected upon spontaneous egress from synchronously infected HFF monolayers . 150 µL of this suspension were mixed with 5 mL of Hanks' Balanced Salt Solution ( HBSS ) supplemented with 0 . 2% FCS . After centrifugation ( 900 x g , 7 min ) , parasites were resuspended in 200 μL of motility buffer ( HBSS supplemented with 1% FCS and 0 . 5 mM CaCl2 to reach about 1 . 6 mM CaCl2 final ) . Typically , 30 to 40 μL of the suspension were added to the cells on the coverslip immobilized in the chamber , to avoid parasite overcrowding during recording . Video recording of the tachyzoite behaviour - the recording chamber that accommodates the coverslips was installed on an Eclipse Ti inverted confocal microscope ( Nikon ) to perform time-lapse video microscopy , with a temperature and CO2-controlled stage ( LCI Corp . ) . The microscope was also equipped with a CMOS camera and a CSU X1 spinning disk ( Yokogawa , Roper Scientific ) . The microscope was piloted using MetaMorph software ( Universal Imaging Corporation , Roper Scientific ) . Similar parameters for image acquisition were used throughout each independent experiment . Time of invasion was estimated for each tachyzoite using MetaMorph time scale between the moment of contact between parasite apex ( i . e . conoid ) and host cell membrane until the tachyzoite has fully passed through the cell-zoite junction . Time of failed invasion was quantified using the same software , once again between the time of apical contact to those of body withdrawal and detachment , or the moment the tachyzoite did not perform any movement . Freshly egressed parasites of each genetic background were prepared as for the video recording assays . About 2 to 4E5 tachyzoites in 300 μL of motility buffer were deposited on 12 mm glass coverslip , previously coated with poly-L-lysine ( as above ) and placed in a 24 well plate . Parasites were gently centrifuged ( 200 x g , 3 min ) to ensure rapid contact with the coverslip and then allowed to glide for 10–15 min at 37°C and 5% CO2 . Motile activity was checked under microscope after a few first minutes . At the end of this period , the samples were fixed after gentle aspiration of the liquid by the addition of 3 . 2% PFA in PHEM pH 7 . 5 ( 20 min ) . Trails left by gliding parasites and parasite surface were stained after a blocking step ( 2% BSA/PBS , 30 min ) with mouse anti-TgSAG1 antibody ( 1 mg/mL stock , 1:600 , 2 hr ) and Alexa Fluor 488-conjugated HCA anti-mouse antibody ( 2 mg/mL stock , 1:800 , 2 hr ) . Cell nuclei were stained with 500 nM DAPI and mounted in Mowiol . Images of trails and tachyzoites were captured under the fluorescent ApoTome two microscope ( Zeiss ) using appropriate set of filters , the Zen software ( Zeiss ) and a z step of 0 . 3 µm . Image stacks were further processed with FIJI ( Schindelin et al . , 2012 ) and Photoshop . MIC7 iKO parasites were treated with DMSO or rapamycin and were prepared as described for the video recording assays . Approximately 8E5 tachyzoites in 300 μL of motility buffer were deposited on 12 mm glass coverslip previously coated with poly-L-lysine ( as above ) and placed in a 24 well plate . Tachyzoites were gently centrifuged ( 300 x g , 3 min ) to ensure rapid contact with the host cell ( non-fluorescent HFF and GFP-GPI expressing U2OS cells ) and allowed to invade for 2 to 4 min periods at 37°C and 5% CO2 . Samples were immediately fixed in 3 . 2% PFA in PHEM pH 7 . 5 ( 20 min ) prior to be processed for IFA . Blocking step and anti-TgSAG1 staining were performed as for the gliding assay except that the incubation with SAG1 primary antibody and Alexa Fluor 633-conjugated secondary antibody was reduced to 30 min . After SAG1 staining , the HFF-tachyzoite samples were permeabilised with ( 0 . 2% Triton X-100/PBS , 5 min ) prior to a second step of blocking . MIC7 labelling was performed using rabbit anti-HA ( clone C29F4 ) , ( 1:800 , 2 hr; Cell Signaling ) followed by Alexa Fluor 488-conjugated HCA anti-rabbit antibody ( 2 mg/mL stock , 1:800 , 2 hr ) . HFF monolayers infected with parasites from the DiCre , MyccWT and MyccMut lines were washed with cold PBS and lysed in IP buffer ( 50 mM Tris , 150 mM NaCl , 0 . 2% Triton-X100 , pH7 . 5 ) supplemented with protease inhibitors ( Roche Diagnostics ) for 30 min on ice . The lysates were then centrifuged ( 5000 x g , 20 min , 4°C ) , the supernatants collected and incubated with 20 µL of α-HA-conjugated agarose beads ( Millipore ) on a rotating wheel at 4°C . After 3 hr the supernatant was removed and beads washed 3x with IP buffer . Protein elution from beads was performed with SDS sample loading buffer and boiling at 95°C for 10 min . Input , IP and supernatant samples for each tested parasite line were then analysed by SDS-PAGE and western blotting . Parasites from MIC7 iKO line were treated with DMSO or rapamycin ( 50 nM , 4 hr ) . After 24 hr incubation parasites were syringe lysed in DMEM at room temperature and collected by centrifugation ( 800 x g , 4°C , 10 min ) . Pellets were resuspended in Ringer’s buffer ( 155 mM NaCl , 3 mM KCl , 2 mM CaCl2 , 1 mM MgCl2 , 3 mM NaH2PO4 , 10 mM HEPES , 10 mM glucose ) supplemented with BIPPO ( 50 µM ) or vehicle and microneme secretion was induced at 37°C for 20 min . Following this incubation step the parasites were placed on ice and pelleted ( 1000 x g , 5 min , 4°C ) . The pellet was kept on ice while the supernatant was re-pelleted ( 2000 x g , 5 min , 4°C ) . The final supernatant , containing the excreted secreted antigens , and pellet fractions were resuspended in sample loading buffer prior to SDS-PAGE and western blotting . Shedding tests during egress for MyccWT and MyccMut lines were performed exactly as described in microneme secretion assay . To test for MIC7 shedding upon invasion , parasites from the iKO , MyccWT and MyccMut lines were treated with DMSO or rapamycin ( 50 nM , 4 hr ) . After 24 hr incubation parasites were syringe lysed in cold DMEM and spun ( 300 x g , 3 min , 4°C ) onto PBS washed HFF monolayers in a 6-well plate ( Falcon ) . The plate was then incubated at 37°C to facilitate invasion . After 1 hr the plate was placed on ice , the supernatant was gently aspirated off and spun down ( 700 x g , 10 min , 4°C ) to remove any aspirated parasites . Proteins were precipitated out by the addition of cold trichloroacetic acid ( 10% v/v ) on ice ( 30 min ) . Samples were centrifuged ( 17 , 000 x g , 20 min , 4°C ) , washed with 300 µL of cold acetone and air dried . The infected monolayers were scraped in 0 . 5 mL cold PBS and collected by centrifugation ( 17 , 000 x g , 20 min , 4°C ) . Both pellet and supernatant samples were resuspended in sample loading buffer prior to SDS-PAGE and western blotting . Parasites from MyccWT and MyccMut lines were treated with rapamycin ( 50 nM , 4 hr ) followed by YnMyr ( 25 µM , 16 hr ) . Parasites were then syringe lysed in DMEM at room temperature and collected by centrifugation ( 800 x g , 10 min , 4°C ) . Pellets were resuspended in 1 . 7 mL cold SoTE buffer ( 0 . 6 M sorbitol , 20 mM Tris–HCl ( pH 7 . 5 ) , and 2 mM EDTA ) and split into three tubes ( 0 . 5 mL each ) per tested parasite line . Tubes 2 and 3 were permeabilised with 0 . 01% cold Digitonin ( Sigma-Aldrich ) in SoTE . Samples were carefully mixed by inversion and incubated on ice ( 10 min ) prior to centrifugation ( 1000 x g , 10 min , 4°C ) . Supernatant was discarded . Pellets were resuspended in 0 . 5 mL cold SoTE and 8 µg of Proteinase K ( Sigma-Aldrich ) were added to tube 3 . All tubes were gently inverted and incubated on ice ( 30 min ) . Proteinase K was inactivated by addition of ice cold trichloroacetic acid to a final concentration of 10% v/v on ice ( 30 min ) . Samples were centrifuged ( 17 , 000 x g , 20 min , 4°C ) , washed with 300 µL of cold acetone and air dried prior to SDS-PAGE and western blotting . Parasites from iKO , MyccWT and MyccMut lines were treated with 50 nM rapamycin or an equivalent volume of vehicle ( DMSO ) for 4 hr after which the medium was replaced and the parasites allowed to grow for 24 hr . Parasites were harvested by syringe lysis , counted , and treated with 1 µM Cytochalasin D ( Sigma ) for 10 min at room temperature . 500 , 000 parasites from each condition were seeded onto confluent HFF monolayers grown in chambered coverslip slides ( ibidi ) and allowed to settle for 10 min on ice . The slides were spun down ( 250 x g , 1 min , 4°C ) then transferred to a 37°C water bath for 20 min to initiate rhoptry secretion . The chambers were washed 3x with PBS then fixed with ice-cold methanol at −20°C for 8 min and washed 3x with PBS . Fixed cells were permeabilised with 0 . 1% Triton X-100 in PBS for 15 min then blocked with 3% BSA in PBS for 1 hr . Cells were then incubated with rabbit anti-phospho-Stat6 ( 1:600; Cell Signaling ) and mouse anti-Toxoplasma ( TP3 ) ( 1:1000; Abcam ) primary antibodies for 1 hr . After 3x washes with PBS , cells were incubated with goat anti-rabbit Alexa Fluor 594 ( 1:2000; Life Technologies ) and goat anti-mouse Alexa Fluor 488 ( 1:2 , 000; Life Technologies ) secondary antibodies and 5 µg/mL DAPI ( Sigma ) for 1 hr followed by 3x washes with PBS . Images were obtained using a Nikon Eclipse Ti-U inverted fluorescent microscope using a 20x objective and analysed using FIJI software . ≥5 fields of view per condition were analysed in three independent experiments . The number of pSTAT6 positive HFFs was normalised to the total number of HFFs , and results were statistically tested with a two-way ANOVA with Sidak’s multiple comparison test in GraphPad Prism 8 . The data are presented as mean ± SD .	A microscopic parasite known as Toxoplasma gondii infects around 30% of the human population . Most infections remain asymptomatic , but in people with a compromised immune system , developing fetuses and people infected with particular virulent strains of the parasite , infection can be fatal . T . gondii is closely related to other parasites that also infect humans , including the one that causes malaria . These parasites have complex lifecycles that involve successive rounds of invading the cells of their hosts , growing and then exiting these cells . Signaling proteins found at specific locations within parasite cells regulate the ability of the parasites to interact with and invade host cells . Sometimes these signaling proteins are attached to membranes using lipid anchors , for example through a molecule called myristic acid . An enzyme called NMT can attach myristic acid to one end of its target proteins . The myristic acid tag can influence the ability of target proteins to bind to other proteins , or to membranes . Previous studies have found that drugs that inhibit the NMT enzyme prevent the malaria parasite from successfully invading and growing inside host cells . The NMT enzyme from T . gondii is very similar to that of the malaria parasite . Broncel et al . have shown that the drug developed against P . falciparum also inhibits the ability of T . gondii to grow . These findings suggest that drugs against the NMT enzyme may be useful to treat diseases caused by T . gondii and other closely-related parasites . Broncel et al . also identified 65 proteins in T . gondii that contain a myristic acid tag using an approach called proteomics . One of the unexpected ‘myristoylated’ proteins identified in the experiments is known as MIC7 . This protein was found to be transported onto the surface of T . gondii parasites and is required in its myristoylated form for the parasite to successfully invade host cells . This was surprising as myristoylated proteins are generally thought to not enter the pathway that brings proteins to the outside of cell . These findings suggest that myristic acid on proteins that are secreted can facilitate interactions between cells , maybe by inserting the myristic acid into the cell membrane .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease" ]	2020	Profiling of myristoylation in Toxoplasma gondii reveals an N-myristoylated protein important for host cell penetration
Bats host virulent zoonotic viruses without experiencing disease . A mechanistic understanding of the impact of bats’ virus hosting capacities , including uniquely constitutive immune pathways , on cellular-scale viral dynamics is needed to elucidate zoonotic emergence . We carried out virus infectivity assays on bat cell lines expressing induced and constitutive immune phenotypes , then developed a theoretical model of our in vitro system , which we fit to empirical data . Best fit models recapitulated expected immune phenotypes for representative cell lines , supporting robust antiviral defenses in bat cells that correlated with higher estimates for within-host viral propagation rates . In general , heightened immune responses limit pathogen-induced cellular morbidity , which can facilitate the establishment of rapidly-propagating persistent infections within-host . Rapidly-transmitting viruses that have evolved with bat immune systems will likely cause enhanced virulence following emergence into secondary hosts with immune systems that diverge from those unique to bats . Bats have received much attention in recent years for their role as reservoir hosts for emerging viral zoonoses , including rabies and related lyssaviruses , Hendra and Nipah henipaviruses , Ebola and Marburg filoviruses , and SARS coronavirus ( Calisher et al . , 2006; Wang and Anderson , 2019 ) . In most non-Chiropteran mammals , henipaviruses , filoviruses , and coronaviruses induce substantial morbidity and mortality , display short durations of infection , and elicit robust , long-term immunity in hosts surviving infection ( Nicholls et al . , 2003; Hooper et al . , 2001; Mahanty and Bray , 2004 ) . Bats , by contrast , demonstrate no obvious disease symptoms upon infection with pathogens that are highly virulent in non-volant mammals ( Schountz et al . , 2017 ) but may , instead , support viruses as long-term persistent infections , rather than transient , immunizing pathologies ( Plowright et al . , 2016 ) . Recent research advances are beginning to shed light on the molecular mechanisms by which bats avoid pathology from these otherwise virulent pathogens ( Brook and Dobson , 2015 ) . Bats leverage a suite of species-specific mechanisms to limit viral load , which include host receptor sequence incompatibilities for some bat-virus combinations ( Ng et al . , 2015; Takadate et al . , 2020 ) and constitutive expression of the antiviral cytokine , IFN-α , for others ( Zhou et al . , 2016 ) . Typically , the presence of viral RNA or DNA in the cytoplasm of mammalian cells will induce secretion of type I interferon proteins ( IFN-α and IFN-β ) , which promote expression and translation of interferon-stimulated genes ( ISGs ) in neighboring cells and render them effectively antiviral ( Stetson and Medzhitov , 2006 ) . In some bat cells , the transcriptomic blueprints for this IFN response are expressed constitutively , even in the absence of stimulation by viral RNA or DNA ( Zhou et al . , 2016 ) . In non-flying mammals , constitutive IFN expression would likely elicit widespread inflammation and concomitant immunopathology upon viral infection , but bats support unique adaptations to combat inflammation ( Zhang et al . , 2013; Ahn et al . , 2019; Xie et al . , 2018; Pavlovich et al . , 2018 ) that may have evolved to mitigate metabolic damage induced during flight ( Kacprzyk et al . , 2017 ) . The extent to which constitutive IFN-α expression signifies constitutive antiviral defense in the form of functional IFN-α protein remains unresolved . In bat cells constitutively expressing IFN-α , some protein-stimulated , downstream ISGs appear to be also constitutively expressed , but additional ISG induction is nonetheless possible following viral challenge and stimulation of IFN-β ( Zhou et al . , 2016; Xie et al . , 2018 ) . Despite recent advances in molecular understanding of bat viral tolerance , the consequences of this unique bat immunity on within-host virus dynamics—and its implications for understanding zoonotic emergence—have yet to be elucidated . The field of ‘virus dynamics’ was first developed to describe the mechanistic underpinnings of long-term patterns of steady-state viral load exhibited by patients in chronic phase infections with HIV , who appeared to produce and clear virus at equivalent rates ( Nowak and May , 2000; Ho et al . , 1995 ) . Models of simple target cell depletion , in which viral load is dictated by a bottom-up resource supply of infection-susceptible host cells , were first developed for HIV ( Perelson , 2002 ) but have since been applied to other chronic infections , including hepatitis-C virus ( Neumann et al . , 1998 ) , hepatitis-B virus ( Nowak et al . , 1996 ) and cytomegalovirus ( Emery et al . , 1999 ) . Recent work has adopted similar techniques to model the within-host dynamics of acute infections , such as influenza A and measles , inspiring debate over the extent to which explicit modeling of top-down immune control can improve inference beyond the basic resource limitation assumptions of the target cell model ( Baccam et al . , 2006; Pawelek et al . , 2012; Saenz et al . , 2010; Morris et al . , 2018 ) . To investigate the impact of unique bat immune processes on in vitro viral kinetics , we first undertook a series of virus infection experiments on bat cell lines expressing divergent interferon phenotypes , then developed a theoretical model elucidating the dynamics of within-host viral spread . We evaluated our theoretical model analytically independent of the data , then fit the model to data recovered from in vitro experimental trials in order to estimate rates of within-host virus transmission and cellular progression to antiviral status under diverse assumptions of absent , induced , and constitutive immunity . Finally , we confirmed our findings in spatially-explicit stochastic simulations of fitted time series from our mean field model . We hypothesized that top-down immune processes would overrule classical resource-limitation in bat cell lines described as constitutively antiviral in the literature , offering a testable prediction for models fit to empirical data . We further predicted that the most robust antiviral responses would be associated with the most rapid within-host virus propagation rates but also protect cells against virus-induced mortality to support the longest enduring infections in tissue culture . We first explored the influence of innate immune phenotype on within-host viral propagation in a series of infection experiments in cell culture . We conducted plaque assays on six-well plate monolayers of three immortalized mammalian kidney cell lines: [1] Vero ( African green monkey ) cells , which are IFN-defective and thus limited in antiviral capacity ( Desmyter et al . , 1968 ) ; [2] RoNi/7 . 1 ( Rousettus aegyptiacus ) cells which demonstrate idiosyncratic induced interferon responses upon viral challenge ( Kuzmin et al . , 2017; Arnold et al . , 2018; Biesold et al . , 2011; Pavlovich et al . , 2018 ) ; and [3] PaKiT01 ( Pteropus alecto ) cells which constitutively express IFN-α ( Zhou et al . , 2016; Crameri et al . , 2009 ) . To intensify cell line-specific differences in constitutive immunity , we carried out infectivity assays with GFP-tagged , replication-competent vesicular stomatitis Indiana viruses: rVSV-G , rVSV-EBOV , and rVSV-MARV , which have been previously described ( Miller et al . , 2012; Wong et al . , 2010 ) . Two of these viruses , rVSV-EBOV and rVSV-MARV , are recombinants for which cell entry is mediated by the glycoprotein of the bat-evolved filoviruses , Ebola ( EBOV ) and Marburg ( MARV ) , thus allowing us to modulate the extent of structural , as well as immunological , antiviral defense at play in each infection . Previous work in this lab has demonstrated incompatibilities in the NPC1 filovirus receptor which render PaKiT01 cells refractory to infection with rVSV-MARV ( Ng and Chandrab , 2018 , Unpublished results ) , making them structurally antiviral , over and above their constitutive expression of IFN-α . All three cell lines were challenged with all three viruses at two multiplicities of infection ( MOI ) : 0 . 001 and 0 . 0001 . Between 18 and 39 trials were run at each cell-virus-MOI combination , excepting rVSV-MARV infections on PaKiT01 cells at MOI = 0 . 001 , for which only eight trials were run ( see Materials and methods; Figure 1—figure supplements 1–3 , Supplementary file 1 ) . Because plaque assays restrict viral transmission neighbor-to-neighbor in two-dimensional cellular space ( Howat et al . , 2006 ) , we were able to track the spread of GFP-expressing virus-infected cells across tissue monolayers via inverted fluorescence microscopy . For each infection trial , we monitored and re-imaged plates for up to 200 hr of observations or until total monolayer destruction , processed resulting images , and generated a time series of the proportion of infectious-cell occupied plate space across the duration of each trial ( see Materials and methods ) . We used generalized additive models to infer the time course of all cell culture replicates and construct the multi-trial dataset to which we eventually fit our mechanistic transmission model for each cell line-virus-specific combination ( Figure 1; Figure 1—figure supplements 1–5 ) . All three recombinant vesicular stomatitis viruses ( rVSV-G , rVSV-EBOV , and rVSV-MARV ) infected Vero , RoNi/7 . 1 , and PaKiT01 tissue cultures at both focal MOIs . Post-invasion , virus spread rapidly across most cell monolayers , resulting in virus-induced epidemic extinction . Epidemics were less severe in bat cell cultures , especially when infected with the recombinant filoviruses , rVSV-EBOV and rVSV-MARV . Monolayer destruction was avoided in the case of rVSV-EBOV and rVSV-MARV infections on PaKiT01 cells: in the former , persistent viral infection was maintained throughout the 200 hr duration of each experiment , while , in the latter , infection was eliminated early in the time series , preserving a large proportion of live , uninfectious cells across the duration of the experiment . We assumed this pattern to be the result of immune-mediated epidemic extinction ( Figure 1 ) . Patterns from MOI = 0 . 001 were largely recapitulated at MOI = 0 . 0001 , though at somewhat reduced total proportions ( Figure 1—figure supplement 5 ) . We next developed a within-host model to fit to these data to elucidate the effects of induced and constitutive immunity on the dynamics of viral spread in host tissue ( Figure 1 ) . The compartmental within-host system mimicked our two-dimensional cell culture monolayer , with cells occupying five distinct infection states: susceptible ( S ) , antiviral ( A ) , exposed ( E ) , infectious ( I ) , and dead ( D ) . We modeled exposed cells as infected but not yet infectious , capturing the ‘eclipse phase’ of viral integration into a host cell which precedes viral replication . Antiviral cells were immune to viral infection , in accordance with the 'antiviral state' induced from interferon stimulation of ISGs in tissues adjacent to infection ( Stetson and Medzhitov , 2006 ) . Because we aimed to translate available data into modeled processes , we did not explicitly model interferon dynamics but instead scaled the rate of cell progression from susceptible to antiviral ( ρ ) by the proportion of exposed cells ( globally ) in the system . In systems permitting constitutive immunity , a second rate of cellular acquisition of antiviral status ( ε ) additionally scaled with the global proportion of susceptible cells in the model . Compared with virus , IFN particles are small and highly diffusive , justifying this global signaling assumption at the limited spatial extent of a six-well plate and maintaining consistency with previous modeling approximations of IFN signaling in plaque assay ( Howat et al . , 2006 ) . To best represent our empirical monolayer system , we expressed our state variables as proportions ( PS , PA , PE , PI , and PD ) , under assumptions of frequency-dependent transmission in a well-mixed population ( Keeling and Rohani , 2008 ) , though note that the inclusion of PD ( representing the proportion of dead space in the modeled tissue ) had the functional effect of varying transmission with infectious cell density . This resulted in the following system of ordinary differential equations: ( 1 ) dPSdt=bPD ( PS+ PA ) −βPSPI−μPS−ρPEPS− εPS+cPA ( 2 ) dPAdt=ρPEPS+ εPS−cPA−μPA ( 3 ) dPEdt=βPSPI-σPE-μPE ( 4 ) dPIdt=σPE-αPI-μPI ( 5 ) dPDdt=μ ( PS+PE+ PI+ PA ) +αPI−bPD ( PS+ PA ) We defined 'induced immunity' as complete , modeling all cells as susceptible to viral invasion at disease-free equilibrium , with defenses induced subsequent to viral exposure through the term ρ . By contrast , we allowed the extent of constitutive immunity to vary across the parameter range of ε > 0 , defining a 'constitutive' system as one containing any antiviral cells at disease-free equilibrium . In fitting this model to tissue culture data , we independently estimated both ρ and ε , as well as the cell-to-cell transmission rate , β , for each cell-virus combination . Since the extent to which constitutively-expressed IFN-α is constitutively translated into functional protein is not yet known for bat hosts ( Zhou et al . , 2016 ) , this approach permitted our tissue culture data to drive modeling inference: even in PaKiT01 cell lines known to constitutively express IFN-α , the true constitutive extent of the system ( i . e . the quantity of antiviral cells present at disease-free equilibrium ) was allowed to vary through estimation of ε . For the purposes of model-fitting , we fixed the value of c , the return rate of antiviral cells to susceptible status , at 0 . The small spatial scale and short time course ( max 200 hours ) of our experiments likely prohibited any return of antiviral cells to susceptible status in our empirical system; nonetheless , we retained the term c in analytical evaluations of our model because regression from antiviral to susceptible status is possible over long time periods in vitro and at the scale of a complete organism ( Radke et al . , 1974; Rasmussen and Farley , 1975; Samuel and Knutson , 1982 ) . Before fitting to empirical time series , we undertook bifurcation analysis of our theoretical model and generated testable hypotheses on the basis of model outcomes . From our within-host model system ( Equation 1-5 ) , we derived the following expression for R0 , the pathogen basic reproduction number ( Supplementary file 2 ) : ( 6 ) R0=βσ ( b-μ ) ( c+μ ) bσ+μα+μc+μ+ε Pathogens can invade a host tissue culture when R0>1 . Rapid rates of constitutive antiviral acquisition ( ε ) will drive R0<1: tissue cultures with highly constitutive antiviral immunity will be therefore resistant to virus invasion from the outset . Since , by definition , induced immunity is stimulated following initial virus invasion , the rate of induced antiviral acquisition ( ρ ) is not incorporated into the equation for R0; while induced immune processes can control virus after initial invasion , they cannot prevent it from occurring to begin with . In cases of fully induced or absent immunity ( ε=0 ) , the R0 equation thus reduces to a form typical of the classic SEIR model: ( 7 ) R0=βσb-μbα+μσ+μ At equilibrium , the theoretical , mean field model demonstrates one of three infection states: endemic equilibrium , stable limit cycles , or no infection ( Figure 2 ) . Respectively , these states approximate the persistent infection , virus-induced epidemic extinction , and immune-mediated epidemic extinction phenotypes previously witnessed in tissue culture experiments ( Figure 1 ) . Theoretically , endemic equilibrium is maintained when new infections are generated at the same rate at which infections are lost , while limit cycles represent parameter space under which infectious and susceptible populations are locked in predictable oscillations . Endemic equilibria resulting from cellular regeneration ( i . e . births ) have been described in vivo for HIV ( Coffin , 1995 ) and in vitro for herpesvirus plaque assays ( Howat et al . , 2006 ) , but , because they so closely approach zero , true limit cycles likely only occur theoretically , instead yielding stochastic extinctions in empirical time series . Bifurcation analysis of our mean field model revealed that regions of no infection ( pathogen extinction ) were bounded at lower threshold ( Branch point ) values for β , below which the pathogen was unable to invade . We found no upper threshold to invasion for β under any circumstances ( i . e . β high enough to drive pathogen-induced extinction ) , but high β values resulted in Hopf bifurcations , which delineate regions of parameter space characterized by limit cycles . Since limit cycles so closely approach zero , high βs recovered in this range would likely produce virus-induced epidemic extinctions under experimental conditions . Under more robust representations of immunity , with higher values for either or both induced ( ρ ) and constitutive ( ε ) rates of antiviral acquisition , Hopf bifurcations occurred at increasingly higher values for β , meaning that persistent infections could establish at higher viral transmission rates ( Figure 2 ) . Consistent with our derivation for R0 , we found that the Branch point threshold for viral invasion was independent of changes to the induced immune parameter ( ρ ) but saturated at high values of ε that characterize highly constitutive immunity ( Figure 3 ) . We next fit our theoretical model by least squares to each cell line-virus combination , under absent , induced , and constitutive assumptions of immunity . In general , best fit models recapitulated expected outcomes based on the immune phenotype of the cell line in question , as described in the general literature ( Table 1; Supplementary file 4 ) . The absent immune model offered the most accurate approximation of IFN-deficient Vero cell time series , the induced immune model best recovered the RoNi/7 . 1 cell trials , and , in most cases , the constitutive immune model most closely recaptured infection dynamics across constitutively IFN-α-expressing PaKiT01 cell lines ( Figure 1 , Figure 1—figure supplements 4–5 , Supplementary file 4 ) . Ironically , the induced immune model offered a slightly better fit than the constitutive to rVSV-MARV infections on the PaKiT01 cell line ( the one cell line-virus combination for which we know a constitutively antiviral cell-receptor incompatibility to be at play ) . Because constitutive immune assumptions can prohibit pathogen invasion ( R0<1 ) , model fits to this time series under constitutive assumptions were handicapped by overestimations of ε , which prohibited pathogen invasion . Only by incorporating an exceedingly rapid rate of induced antiviral acquisition could the model guarantee that initial infection would be permitted and then rapidly controlled . In fitting our theoretical model to in vitro data , we estimated the within-host virus transmission rate ( β ) and the rate ( s ) of cellular acquisition to antiviral status ( ρ or ρ + ε ) ( Table 1; Supplementary file 4 ) . Under absent immune assumptions , ρ and ε were fixed at 0 while β was estimated; under induced immune assumptions , ε was fixed at 0 while ρ and β were estimated; and under constitutive immune assumptions , all three parameters ( ρ , ε , and β ) were simultaneously estimated for each cell-virus combination . Best fit parameter estimates for MOI=0 . 001 data are visualized in conjunction with β – ρ and β – ε bifurcations in Figure 4; all general patterns were recapitulated at lower values for β on MOI=0 . 0001 trials ( Figure 4—figure supplement 1 ) . As anticipated , the immune absent model ( a simple target cell model ) offered the best fit to IFN-deficient Vero cell infections ( Figure 4; Table 1; Supplementary file 4 ) . Among Vero cell trials , infections with rVSV-G produced the highest β estimates , followed by infections with rVSV-EBOV and rVSV-MARV . Best fit parameter estimates on Vero cell lines localized in the region of parameter space corresponding to theoretical limit cycles , consistent with observed virus-induced epidemic extinctions in stochastic tissue cultures . In contrast to Vero cells , the induced immunity model offered the best fit to all RoNi/7 . 1 data , consistent with reported patterns in the literature and our own validation by qPCR ( Table 1; Figure 1—figure supplement 6; Arnold et al . , 2018; Kuzmin et al . , 2017; Biesold et al . , 2011; Pavlovich et al . , 2018 ) . As in Vero cell trials , we estimated highest β values for rVSV-G infections on RoNi/7 . 1 cell lines but here recovered higher β estimates for rVSV-MARV than for rVSV-EBOV . This reversal was balanced by a higher estimated rate of acquisition to antiviral status ( ρ ) for rVSV-EBOV versus rVSV-MARV . In general , we observed that more rapid rates of antiviral acquisition ( either induced , ρ , constitutive , ε , or both ) correlated with higher transmission rates ( β ) . When offset by ρ , β values estimated for RoNi/7 . 1 infections maintained the same amplitude as those estimated for immune-absent Vero cell lines but caused gentler epidemics and reduced cellular mortality ( Figure 1 ) . RoNi/7 . 1 parameter estimates localized in the region corresponding to endemic equilibrium for the deterministic , theoretical model ( Figure 4 ) , yielding less acute epidemics which nonetheless went extinct in stochastic experiments . Finally , rVSV-G and rVSV-EBOV trials on PaKiT01 cells were best fit by models assuming constitutive immunity , while rVSV-MARV infections on PaKiT01 were matched equivalently by models assuming either induced or constitutive immunity—with induced models favored over constitutive in AIC comparisons because one fewer parameter was estimated ( Figure 1—figure supplements 4–5; Supplementary file 4 ) . For all virus infections , PaKiT01 cell lines yielded β estimates a full order of magnitude higher than Vero or RoNi/7 . 1 cells , with each β balanced by an immune response ( either ρ , or ρ combined with ε ) also an order of magnitude higher than that recovered for the other cell lines ( Figure 4; Table 1 ) . As in RoNi/7 . 1 cells , PaKiT01 parameter fits localized in the region corresponding to endemic equilibrium for the deterministic theoretical model . Because constitutive immune processes can actually prohibit initial pathogen invasion , constitutive immune fits to rVSV-MARV infections on PaKiT01 cell lines consistently localized at or below the Branch point threshold for virus invasion ( R0=1 ) . During model fitting for optimization of ε , any parameter tests of ε values producing R0<1 resulted in no infection and , consequently , produced an exceedingly poor fit to infectious time series data . In all model fits assuming constitutive immunity , across all cell lines , parameter estimates for ρ and ε traded off , with one parameter optimized at values approximating zero , such that the immune response was modeled as almost entirely induced or entirely constitutive ( Table 1; Supplementary file 4 ) . For RoNi/7 . 1 cells , even when constitutive immunity was allowed , the immune response was estimated as almost entirely induced , while for rVSV-G and rVSV-EBOV fits on PaKiT01 cells , the immune response optimized as almost entirely constitutive . For rVSV-MARV on PaKiT01 cells , however , estimation of ρ was high under all assumptions , such that any additional antiviral contributions from ε prohibited virus from invading at all . The induced immune model thus produced a more parsimonious recapitulation of these data because virus invasion was always permitted , then rapidly controlled . In order to compare the relative contributions of each cell line’s disparate immune processes to epidemic dynamics , we next used our mean field parameter estimates to calculate the initial ‘antiviral rate’—the initial accumulation rate of antiviral cells upon virus invasion for each cell-virus-MOI combination—based on the following equation: ( 8 ) AntiviralRate=ρPEPs−ϵPswhere PE was calculated from the initial infectious dose ( MOI ) of each infection experiment and PS was estimated at disease-free equilibrium: ( 9 ) PE= 1-e-MOI ( 10 ) PS= ( b-μ ) ( c+μ ) b ( c+μ+ε ) Because ρ and ε both contribute to this initial antiviral rate , induced and constitutive immune assumptions are capable of yielding equally rapid rates , depending on parameter fits . Indeed , under fully induced immune assumptions , the induced antiviral acquisition rate ( ρ ) estimated for rVSV-MARV infection on PaKiT01 cells was so high that the initial antiviral rate exceeded even that estimated under constitutive assumptions for this cell-virus combination ( Supplementary file 4 ) . In reality , we know that NPC1 receptor incompatibilities make PaKiT01 cell lines constitutively refractory to rVSV-MARV infection ( Ng and Chandrab , 2018 , Unpublished results ) and that PaKiT01 cells also constitutively express the antiviral cytokine , IFN-α . Model fitting results suggest that this constitutive expression of IFN-α may act more as a rapidly inducible immune response following virus invasion than as a constitutive secretion of functional IFN-α protein . Nonetheless , as hypothesized , PaKiT01 cell lines were by far the most antiviral of any in our study—with initial antiviral rates estimated several orders of magnitude higher than any others in our study , under either induced or constitutive assumptions ( Table 1; Supplementary file 4 ) . RoNi/7 . 1 cells displayed the second-most-pronounced signature of immunity , followed by Vero cells , for which the initial antiviral rate was essentially zero even under forced assumptions of induced or constitutive immunity ( Table 1; Supplementary file 4 ) . Using fitted parameters for β and ε , we additionally calculated R0 , the basic reproduction number for the virus , for each cell line-virus-MOI combination ( Table 1; Supplementary file 4 ) . We found that R0 was essentially unchanged across differing immune assumptions for RoNi/7 . 1 and Vero cells , for which the initial antiviral rate was low . In the case of PaKiT01 cells , a high initial antiviral rate under either induced or constitutive immunity resulted in a correspondingly high estimation of β ( and , consequently , R0 ) which still produced the same epidemic curve that resulted from the much lower estimates for β and R0 paired with absent immunity . These findings suggest that antiviral immune responses protect host tissues against virus-induced cell mortality and may facilitate the establishment of more rapid within-host transmission rates . Total monolayer destruction occurred in all cell-virus combinations excepting rVSV-EBOV infections on RoNi/7 . 1 cells and rVSV-EBOV and rVSV-MARV infections on PaKiT01 cells . Monolayer destruction corresponded to susceptible cell depletion and epidemic turnover where R-effective ( the product of R0 and the proportion susceptible ) was reduced below one ( Figure 5 ) . For rVSV-EBOV infections on RoNi/7 . 1 , induced antiviral cells safeguarded remnant live cells , which birthed new susceptible cells late in the time series . In rVSV-EBOV and rVSV-MARV infections on PaKiT01 cells , this antiviral protection halted the epidemic ( Figure 5; R-effective <1 ) before susceptibles fully declined . In the case of rVSV-EBOV on PaKiT01 , the birth of new susceptibles from remnant live cells protected by antiviral status maintained late-stage transmission to facilitate long-term epidemic persistence . Importantly , under fixed parameter values for the infection incubation rate ( σ ) and infection-induced mortality rate ( α ) , models were unable to reproduce the longer-term infectious time series captured in data from rVSV-EBOV infections on PaKiT01 cell lines without incorporation of cell births , an assumption adopted in previous modeling representations of IFN-mediated viral dynamics in tissue culture ( Howat et al . , 2006 ) . In our experiments , we observed that cellular reproduction took place as plaque assays achieved confluency . Finally , because the protective effect of antiviral cells is more clearly observable spatially , we confirmed our results by simulating fitted time series in a spatially-explicit , stochastic reconstruction of our mean field model . In spatial simulations , rates of antiviral acquisition were fixed at fitted values for ρ and ε derived from mean field estimates , while transmission rates ( β ) were fixed at values ten times greater than those estimated under mean field conditions , accounting for the intensification of parameter thresholds permitting pathogen invasion in local spatial interactions ( see Materials and methods; Videos 1–3; Figure 5—figure supplement 3; Supplementary file 5; Webb et al . , 2007 ) . In immune capable time series , spatial antiviral cells acted as ‘refugia’ which protected live cells from infection as each initial epidemic wave ‘washed’ across a cell monolayer . Eventual birth of new susceptibles from these living refugia allowed for sustained epidemic transmission in cases where some infectious cells persisted at later timepoints in simulation ( Videos 1–3; Figure 5—figure supplement 3 ) . Bats are reservoirs for several important emerging zoonoses but appear not to experience disease from otherwise virulent viral pathogens . Though the molecular biological literature has made great progress in elucidating the mechanisms by which bats tolerate viral infections ( Zhou et al . , 2016; Ahn et al . , 2019; Xie et al . , 2018; Pavlovich et al . , 2018; Zhang et al . , 2013 ) , the impact of unique bat immunity on virus dynamics within-host has not been well-elucidated . We used an innovative combination of in vitro experimentation and within-host modeling to explore the impact of unique bat immunity on virus dynamics . Critically , we found that bat cell lines demonstrated a signature of enhanced interferon-mediated immune response , of either constitutive or induced form , which allowed for establishment of rapid within-host , cell-to-cell virus transmission rates ( β ) . These results were supported by both data-independent bifurcation analysis of our mean field theoretical model , as well as fitting of this model to viral infection time series established in bat cell culture . Additionally , we demonstrated that the antiviral state induced by the interferon pathway protects live cells from mortality in tissue culture , resulting in in vitro epidemics of extended duration that enhance the probability of establishing a long-term persistent infection . Our findings suggest that viruses evolved in bat reservoirs possessing enhanced IFN capabilities could achieve more rapid within-host transmission rates without causing pathology to their hosts . Such rapidly-reproducing viruses would likely generate extreme virulence upon spillover to hosts lacking similar immune capacities to bats . To achieve these results , we first developed a novel , within-host , theoretical model elucidating the effects of unique bat immunity , then undertook bifurcation analysis of the model’s equilibrium properties under immune absent , induced , and constitutive assumptions . We considered a cell line to be constitutively immune if possessing any number of antiviral cells at disease-free equilibrium but allowed the extent of constitutive immunity to vary across the parameter range for ε , the constitutive rate of antiviral acquisition . In deriving the equation for ε , the basic reproduction number , which defines threshold conditions for virus invasion of a tissue ( R0>1 ) , we demonstrated how the invasion threshold is elevated at high values of constitutive antiviral acquisition , ε . Constitutive immune processes can thus prohibit pathogen invasion , while induced responses , by definition , can only control infections post-hoc . Once thresholds for pathogen invasion have been met , assumptions of constitutive immunity will limit the cellular mortality ( virulence ) incurred at high transmission rates . Regardless of mechanism ( induced or constitutive ) , interferon-stimulated antiviral cells appear to play a key role in maintaining longer term or persistent infections by safeguarding susceptible cells from rapid infection and concomitant cell death . Fitting of our model to in vitro data supported expected immune phenotypes for different bat cell lines as described in the literature . Simple target cell models that ignore the effects of immunity best recapitulated infectious time series derived from IFN-deficient Vero cells , while models assuming induced immune processes most accurately reproduced trials derived from RoNi/7 . 1 ( Rousettus aegyptiacus ) cells , which possess a standard virus-induced IFN-response . In most cases , models assuming constitutive immune processes best recreated virus epidemics produced on PaKiT01 ( Pteropus alecto ) cells , which are known to constitutively express the antiviral cytokine , IFN-α ( Zhou et al . , 2016 ) . Model support for induced immune assumptions in fits to rVSV-MARV infections on PaKiT01 cells suggests that the constitutive IFN-α expression characteristic of P . alecto cells may represent more of a constitutive immune priming process than a perpetual , functional , antiviral defense . Results from mean field model fitting were additionally confirmed in spatially explicit stochastic simulations of each time series . As previously demonstrated in within-host models for HIV ( Coffin , 1995; Perelson et al . , 1996; Nowak et al . , 1995; Bonhoeffer et al . , 1997; Ho et al . , 1995 ) , assumptions of simple target-cell depletion can often provide satisfactory approximations of viral dynamics , especially those reproduced in simple in vitro systems . Critically , our model fitting emphasizes the need for incorporation of top-down effects of immune control in order to accurately reproduce infectious time series derived from bat cell tissue cultures , especially those resulting from the robustly antiviral PaKiT01 P . alecto cell line . These findings indicate that enhanced IFN-mediated immune pathways in bat reservoirs may promote elevated within-host virus replication rates prior to cross-species emergence . We nonetheless acknowledge the limitations imposed by in vitro experiments in tissue culture , especially involving recombinant viruses and immortalized cell lines . Future work should extend these cell culture studies to include measurements of multiple state variables ( i . e . antiviral cells ) to enhance epidemiological inference . The continued recurrence of Ebola epidemics across central Africa highlights the importance of understanding bats’ roles as reservoirs for virulent zoonotic disease . The past decade has born witness to emerging consensus regarding the unique pathways by which bats resist and tolerate highly virulent infections ( Brook and Dobson , 2015; Xie et al . , 2018; Zhang et al . , 2013; Ahn et al . , 2019; Zhou et al . , 2016; Ng et al . , 2015; Pavlovich et al . , 2018 ) . Nonetheless , an understanding of the mechanisms by which bats support endemic pathogens at the population level , or promote the evolution of virulent pathogens at the individual level , remains elusive . Endemic maintenance of infection is a defining characteristic of a pathogen reservoir ( Haydon et al . , 2002 ) , and bats appear to merit such a title , supporting long-term persistence of highly transmissible viral infections in isolated island populations well below expected critical community sizes ( Peel et al . , 2012 ) . Researchers debate the relative influence of population-level and within-host mechanisms which might explain these trends ( Plowright et al . , 2016 ) , but increasingly , field data are difficult to reconcile without acknowledgement of a role for persistent infections ( Peel et al . , 2018; Brook et al . , 2019 ) . We present general methods to study cross-scale viral dynamics , which suggest that within-host persistence is supported by robust antiviral responses characteristic of bat immune processes . Viruses which evolve rapid replication rates under these robust antiviral defenses may pose the greatest hazard for cross-species pathogen emergence into spillover hosts with immune systems that differ from those unique to bats .	Bats can carry viruses that are deadly to other mammals without themselves showing serious symptoms . In fact , bats are natural reservoirs for viruses that have some of the highest fatality rates of any viruses that people acquire from wild animals – including rabies , Ebola and the SARS coronavirus . Bats have a suite of antiviral defenses that keep the amount of virus in check . For example , some bats have an antiviral immune response called the interferon pathway perpetually switched on . In most other mammals , having such a hyper-vigilant immune response would cause harmful inflammation . Bats , however , have adapted anti-inflammatory traits that protect them from such harm , include the loss of certain genes that normally promote inflammation . However , no one has previously explored how these unique antiviral defenses of bats impact the viruses themselves . Now , Brook et al . have studied this exact question using bat cells grown in the laboratory . The experiments made use of cells from one bat species – the black flying fox – in which the interferon pathway is always on , and another – the Egyptian fruit bat – in which this pathway is only activated during an infection . The bat cells were infected with three different viruses , and then Brook et al . observed how the interferon pathway helped keep the infections in check , before creating a computer model of this response . The experiments and model helped reveal that the bats’ defenses may have a potential downside for other animals , including humans . In both bat species , the strongest antiviral responses were countered by the virus spreading more quickly from cell to cell . This suggests that bat immune defenses may drive the evolution of faster transmitting viruses , and while bats are well protected from the harmful effects of their own prolific viruses , other creatures like humans are not . The findings may help to explain why bats are often the source for viruses that are deadly in humans . Learning more about bats' antiviral defenses and how they drive virus evolution may help scientists develop better ways to predict , prevent or limit the spread of viruses from bats to humans . More studies are needed in bats to help these efforts . In the meantime , the experiments highlight the importance of warning people to avoid direct contact with wild bats .	[ "Abstract", "Introduction", "Results", "Discussion" ]	[ "ecology", "epidemiology", "and", "global", "health" ]	2020	Accelerated viral dynamics in bat cell lines, with implications for zoonotic emergence
The coordination of movement across the body is a fundamental , yet poorly understood aspect of motor control . Mutant mice with cerebellar circuit defects exhibit characteristic impairments in locomotor coordination; however , the fundamental features of this gait ataxia have not been effectively isolated . Here we describe a novel system ( LocoMouse ) for analyzing limb , head , and tail kinematics of freely walking mice . Analysis of visibly ataxic Purkinje cell degeneration ( pcd ) mice reveals that while differences in the forward motion of individual paws are fully accounted for by changes in walking speed and body size , more complex 3D trajectories and , especially , inter-limb and whole-body coordination are specifically impaired . Moreover , the coordination deficits in pcd are consistent with a failure to predict and compensate for the consequences of movement across the body . These results isolate specific impairments in whole-body coordination in mice and provide a quantitative framework for understanding cerebellar contributions to coordinated locomotion . Our ability to engage in even the simplest motor tasks requires us to coordinate our movements in space and time across the body . For example , during walking , leg , arm , trunk , and head movements need to be coordinated to achieve a stable , smooth , and efficient gait . The cerebellum is critical for coordinated locomotion; cerebellar damage leads to characteristic gait ataxia across species . Exactly how the cerebellum contributes to motor coordination , however , remains controversial . Mice present several advantages for understanding the role of the cerebellum in locomotor coordination . In addition to their amenability to genetic circuit dissection , their small size makes it possible to analyze even unrestrained , relatively complex whole-body actions within a laboratory setting . Moreover , several spontaneous mutants have been identified based on visible gait ataxia ( Mullen et al . , 1976; Walter et al . , 2006; Lalonde and Strazielle , 2007; Cendelin , 2014 ) . These mutants exhibit abnormal cell patterning within the cerebellum ( Lalonde and Strazielle , 2007; Brooks and Dunnett , 2009; Kim et al . , 2009; Sheets et al . , 2013; Cendelin , 2014 ) . While traditional gait analyses in ataxic mutants have reported a variety of impairments in individual limb movements and interlimb coordination ( Fortier et al . , 1987; Wang et al . , 2006; Vinueza Veloz et al . , 2014 ) , many of these findings are not specific to ataxia and could represent secondary consequences of changes in walking speed ( Batka et al . , 2014 ) . Therefore it has not been clear to what extent exisiting analyses of mouse locomotion capture the essence of the coordination deficits of ataxic mice that are so visible to the human eye ( Cendelin et al . , 2010 ) . Here we describe LocoMouse , a novel system for automated , markerless , 3D tracking of locomotor kinematics in freely walking mice . We used LocoMouse to establish a quantitative framework for locomotor coordination in mice and analyze the deficits of visibly ataxic Purkinje cell degeneration ( pcd ) mutants . Surprisingly , we find that the forward motion of individual paws is normal in pcd—apparent differences in basic stride parameters and forward trajectories disappeared once changes in body size and walking speed were taken into account . In contrast , 3D paw trajectories , interlimb and whole-body coordination were specifically impaired . Moreover , the prominent nose and tail oscillations observed in pcd were successfully modeled as passive consequences of the forward motion of the hind limbs , suggesting the absence of a mechanism that normally predicts and compensates for movements of other parts of the body . Taken together , these results suggest a specific failure to predict the consequences of movement across joints , limbs , and body , and are consistent with the hypothesis that the cerebellum provides a forward model for motor control ( Bastian et al . , 1996; Ebner and Pasalar , 2008; Kennedy et al . , 2014 ) . The noninvasive , markerless LocoMouse system ( Figure 1 ) uses high-speed cameras and machine learning algorithms to automatically detect and track the position of paws , nose , and tail in 3D with high ( 2 . 5 ms ) temporal resolution . 10 . 7554/eLife . 07892 . 003Figure 1 . LocoMouse system for analyzing mouse locomotor coordination . ( A ) LocoMouse apparatus . The mouse walks freely across a glass corridor with mirror below at a 45° angle . A single high-speed camera captures side and bottom views at 400 fps . Infrared ( IR ) sensors trigger data collection . ( B ) Machine learning algorithms identify paws , nose and tail segments and track their movements in 3D . Example ‘paw’ and ‘not paw’ training images for SVM ( Support Vector Machine ) feature detectors are shown for side and bottom views . ( C ) Continuous tracks are obtained by post-processing the feature detections with a Multi-Target Tracking algorithm . ( D ) Continuous forward trajectories ( x position vs time ) for paws , nose , and tail . The inset illustrates the color code used throughout the paper to identify individual features . ( E ) Continuous vertical ( z ) trajectories of the two front paws . ( F ) Side-to-side ( y ) position of proximal ( green ) to distal ( yellow ) tail segments vs time . ( G ) Individual strides were divided into swing and stance phases for further analysis . Further validation of tracking algorithm is presented in Figure 1—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 07892 . 00310 . 7554/eLife . 07892 . 004Figure 1—figure supplement 1 . LocoMouse tracking validation . ( A ) Comparison of manual ( gray ) and automated tracking ( blue ) for front left paw across 3 dimensions . From left to right , plots show normalized x , y , and z paw position aligned to stance onset ( n = 43 strides from 9 movies of 3 mice ) . ( B ) Scatterplots of manual vs automated tracking positions ( x , y , and z ) for all frames ( n = 4194 ) of the same 9 movies . Values where the difference between manual and automated tracking are larger than average paw size are color-coded in green . Correlation coefficients are Pearson's r . ( C ) Example tracking traces demonstrate that stride lengths can differ at similar walking speeds due to pauses in walking . DOI: http://dx . doi . org/10 . 7554/eLife . 07892 . 004 Mice walked across a glass corridor , 66 . 5 cm long and 4 . 5 cm wide ( Figure 1A ) . A mirror was placed at 45 deg under the mouse , so that a single high-speed camera ( AVT Bonito , 1440x250 pixels @400 frames per second ) recorded both bottom and side views . Individual trials consisted of single crossings of the corridor . Mice freely initiated trials by walking back and forth between two dark ‘home’ boxes on each end of the corridor . Data collection was performed in LABVIEW and was automatically triggered by infrared sensors that detected when the mouse entered and exited the corridor . After processing the images to subtract the background and correct for mirror and lens distortions , we applied a machine learning algorithm ( Figure 1B ) to identify and track all four paws , snout , and 15 tail segments in both bottom and side views for each trial ( Figure 1C; Video 1; see ‘Materials and methods’ ) . We then extracted the continuous forward ( x ) , side-to-side ( y ) , and vertical ( z ) trajectories for each feature from each movie ( Figure 1D–F ) . The stride cycles of all four paws were automatically divided into swing and stance phases for subsequent analysis ( Figure 1G ) . Validation of the tracking is provided in Figure 1—figure supplement 1 . 10 . 7554/eLife . 07892 . 005Video 1 . Automated , high-resolution locomotion tracking in freely walking mice . High-speed ( 400 fps ) video of a mouse crossing the LocoMouse corridor , displayed at 30 fps . Side and bottom ( via mirror reflection ) views of the mouse are captured in a single camera . Top: Raw video of a wild-type mouse . Bottom: Same video with the output of the machine learning tracking overlaid: nose ( orange circle ) , paws ( red: front right; blue: front left; magenta: hind right; cyan: hind left ) and tail segments ( green-to-yellow gradient circles ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07892 . 005 We first analyzed basic stride parameters for individual limbs of wildtype control mice ( Figure 2 ) . Parameters such as stride length ( mm ) , cadence ( strides/s ) , swing velocity ( m/s ) , and stance duration ( ms ) , along with the mouse's walking speed , were measured for each stride . The data were highly variable ( Figure 2A; see also Figure 2—figure supplement 1 ) . The walking speed of the mice was similarly variable ( Figure 2B ) . We therefore sorted all strides for individual mice into speed bins in a stridewise manner and analyzed them with respect to the mouse's walking speed . 10 . 7554/eLife . 07892 . 006Figure 2 . Basic stride parameters can be predicted using only walking speed and body size . ( A ) Stride length vs walking speed for 9602 individual strides of the front right paws of 34 wildtype mice are color-coded by weight for each individual animal . More information on the mice can be found in Figure 2—figure supplement 1 and Figure 2—source data 1 . ( B ) Histogram of average walking speeds for each stride from ( A ) . Strides are divided into speed bins of 0 . 05 m/s . ( C–F ) Stride length , cadence ( 1/stride duration ) , swing velocity , and stance duration vs walking speed , respectively . For each parameter , speed-binned median values are shown for each animal ( solid circles , color coded by weight ) . Data for each animal are connected across speeds with a thin dotted line . Thick lines are the output of the linear mixed-effects model , for 3 example weights across walking speeds ( blue: 9g , cyan:19g , red: 33g ) . Marginal R-squared values for each linear mixed model are shown for each parameter . All raw data for Figure 2 and Figure 2—figure supplement 1 can be found in Figure 2—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 07892 . 00610 . 7554/eLife . 07892 . 007Figure 2—source data 1 . Individual limb gait parameters , walking speed , and mouse metrics used to generate the linear mixed effects model in Figure 2 . Data for the four paws ( FR HR FL HL ) are stored in separate columns for all variables; rows are organized stride-wise . The linear mixed model shown in Figure 2 uses speed-binned data ( 0 . 05 m/s speed bins ) from the FR paw , with a minimum stride count criterion of 5 strides per bin . DOI: http://dx . doi . org/10 . 7554/eLife . 07892 . 00710 . 7554/eLife . 07892 . 008Figure 2—figure supplement 1 . Using linear mixed effects models to predict basic stride parameters . ( A ) Properties of wildtype mice used for linear mixed-effects model . A total of 34 wild-type C57BL/6 mice were used for the linear mixed-effects model in Figure 2 . Each individual WT animal is plotted as a circle ( open circles , females , N = 11; closed circles , males , N = 23 ) . Symbols are color-coded by age . The diverse group included a variety of ages ( 30–114 days ) , body lengths ( 61–100 mm ) , and weights ( 7–33g ) . ( B ) Comparison of model fits for basic stride parameters . Speed , gender , age , body length , weight ( fixed terms ) and subject ( random term ) were used as predictor variables in the linear mixed-effects model . Table rows show tested equations for predicting stride parameters and values used for selection criteria of the resulting predictive model . p-values reported for each term are the outcome of a likelihood ratio test comparing indicated equations ( superscripts ) . The last two lines ( f , g ) indicate that age and gender did not improve the predictions beyond the inclusion of speed and body weight . ( C ) Coefficients of speed and weight for basic stride parameters . The final equations for each stride parameter included speed and weight as fixed-term predictor variables; subject was included as a random-term . Coefficient values for fixed terms are represented . These equations can be used to predict stride parameters for a given mouse walking at a particular speed . DOI: http://dx . doi . org/10 . 7554/eLife . 07892 . 008 The median values of stride parameters for each mouse across speed bins are shown in Figure 2C–F ( dots connected by dashed lines ) . Each parameter measured , including stride length , cadence , swing velocity , and stance duration ( Figure 2C–F ) , varied consistently with the walking speed of the mouse . Cadence and stride length increased with walking speed , indicating that faster walking in mice is associated with longer , more frequent strides ( Clarke and Still , 1999; Lalonde and Strazielle , 2007; Batka et al . , 2014 ) . These changes , in turn , resulted from linear increases in swing velocity and steep decreases in stance duration with increasing walking speed . Further subdividing the data by the body weight of each animal ( Figure 2—figure supplement 1 ) revealed that much of the remaining variability in each parameter could be accounted for by the mouse's body size ( Figure 2 , color-coded by weight ) . To quantify the influence of walking speed , body size , and other potential factors on these basic stride parameters , 36 , 369 strides from an average of 1069 ± 266 strides in 34 mice were analyzed ( Figure 2 ) . For each parameter , we first linearized the data by fitting appropriate functions ( e . g . , linear , power ) to the data with respect to walking speed ( Figure 2—figure supplement 1 ) . Then we generated a multilevel linear mixed-effects model that included potential predictor variables speed , weight , body length , age , and gender either alone or in combination and asked to what extent they accurately predicted the measured parameter . This analysis revealed that the value of each stride parameter was readily predicted based solely on walking speed and body weight ( Figure 2—figure supplement 1 ) . While basic stride parameters also varied with gender and age , these effects were related to differences in body size ( Figure 2—figure supplement 1 ) ; adding neither age nor gender improved the predictions once body size was taken into account . The resulting best-fit models are plotted as thick lines in Figure 2C–F . These results indicate that the equations in Figure 2—figure supplement 1 provide quantitative predictions of paw stride parameters for mice of a given size , walking at a particular speed . The systematic analysis of stride parameters across mice presented above provided a starting point for quantifying locomotor deficits of ataxic mice . The Purkinje cell degeneration ( pcd ) mouse is a recessive mutant characterized by complete post-natal degeneration of cerebellar Purkinje cells and subsequent partial loss of cerebellar granule cells ( [Chen et al . , 1996; Le Marec and Lalonde , 1997; Lalonde and Strazielle , 2007; Cendelin , 2014]; The gene affected encodes ATP/GTP binding protein 1 [Fernandez-Gonzalez et al . , 2002] ) . Pcd mice can be easily identified by eye based on their ataxic , uncoordinated movements ( Mullen et al . , 1976; Le Marec and Lalonde , 1997 ) . Pcd mice exhibit impaired rotarod performance and deficits in eyelid conditioning that have been attributed to their cerebellar abnormalities ( Chen et al . , 1996; Le Marec and Lalonde , 1997 ) . Perhaps surprisingly , given the severity of their anatomical phenotype , the motor deficits of pcd mice are relatively mild compared to other spontaneous ataxic mutants ( Lalonde and Strazielle , 2007; Le Marec and Lalonde , 1997 ) . Pcd mice were visibly ataxic when walking on the LocoMouse setup ( Video 2 ) . Consistent with previous studies of cerebellar ataxia in mice ( Fortier et al . , 1987; Wang et al . , 2006; Cendelin et al . , 2010; Vinueza Veloz et al . , 2014 ) , comparing the basic stride parameters of visibly ataxic pcd mice with littermate control mice revealed that the strides of pcd mice were , overall , quite different ( Figure 3A–D ) . Stride lengths were shorter ( Figure 3B , purple shadows ) , even when changes in walking speed ( Figure 3A ) were taken into account . Cadence and stance durations were also altered ( Figure 3C , D , purple shadows ) . 10 . 7554/eLife . 07892 . 009Video 2 . Purkinje cell degeneration mice are visibly ataxic . Purkinje cell degeneration ( pcd ) mouse crossing the LocoMouse corridor . Pcd mice are smaller and walk more slowly than controls . They lift their paws higher and have altered patterns of interlimb coordination . The nose and tail oscillate laterally and vertically . DOI: http://dx . doi . org/10 . 7554/eLife . 07892 . 00910 . 7554/eLife . 07892 . 010Figure 3 . Differences in forward paw trajectories in pcd can be accounted for by walking speed and body size; impairments are restricted to off-axis movement . ( A ) Histogram of walking speeds , divided into 0 . 05 m/s speed bins for pcd ( purple N = 3 mice; n = 3052 strides ) and littermate controls ( green , N = 7; n = 2256 strides ) . ( B–D ) Stride length ( B ) , cadence ( C , 1/stride duration ) and stance duration ( D ) vs walking speed for pcd mice ( purple ) and littermate controls ( green ) . For each parameter , the thin lines with shadows represent median values ±25th , 75th percentiles . Thick lines represent the predictions calculated using the mixed-effect models described in Figure 2 and Figure 2—figure supplement 1 ( including speed and weight as predictor variables ) . ( E ) Average instantaneous forward ( x ) velocity of FR paw during swing phase for pcd ( purple ) and size-matched controls ( black ) . Line thickness represents increasing speed . ( F ) x-y position of four paws relative to the body center during swing . ( G ) y-excursion for front and hind paws , relative to body midline . ( H ) Average vertical ( z ) position of FR paw relative to ground during swing . DOI: http://dx . doi . org/10 . 7554/eLife . 07892 . 01010 . 7554/eLife . 07892 . 011Figure 3—figure supplement 1 . Basic stride parameters for pcd are are not different from their size-matched controls . ( A ) Histogram of walking speeds , divided in 0 . 05 m/s speed bins for pcd ( purple , N = 3 mice , n = 3052 strides ) , littermate controls ( green , N = 7 mice , n = 2256 strides ) and size matched controls ( black , N = 11 mice , n = 3400 strides ) . ( B ) Histogram of stride counts by weight for size-matched controls and pcd . ( C–E ) Basic stride parameters . For each parameter , thick lines represent the prediction , from the mixed-effects models derived from wildtype data in Figure 2 ( including speed and weight as predictor variables ) , for each group . Pcd ( average weight = 12g; purple line ) , control littermates ( average weight = 26g; blue line ) and size-match controls ( average weight = 12g; black line ) . ( C ) Stride length values vs walking speed for pcd , littermate controls and size-matched controls ( median ±25th , 75th percentile ) . Data are represented by thin lines and shadows , thick lines are model predictions . ( D , E ) Temporal measures of the step cycle; cadence ( inverse of stride duration ) and stance duration , respectively ( median ±25th , 75th percentile ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07892 . 01110 . 7554/eLife . 07892 . 012Figure 3—figure supplement 2 . 3D paw trajectories for wildtype controls and pcd . ( A–C ) Average 3D trajectories for front right paw of wildtype control group ( N = 34; n = 9602 strides ) during swing phase . Traces are binned and color coded by walking speed . ( A ) Instantaneous forward ( x ) velocity . ( B ) side-to-side ( y ) excursion ( C ) vertical ( z ) position relative to ground . ( D–F ) Same as above but for hind right paw of wildtype control group . ( G–I ) Paw trajectories for pcd and size-matched controls . ( G ) Hind right paw x-trajectories . ( H ) Hind right paw vertical ( z ) position relative to ground during swing . In pcd , hind paws were lifted higher than forepaws , and their peak positions varied more steeply with speed , ( F[153 . 02 , 1] = 5 . 64 , p < 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07892 . 012 Since pcd mice , like many ataxic animals , are smaller than controls ( Figure 3—figure supplement 1 ) , and given that they walk more slowly ( Figure 3A ) , we asked to what extent the altered stride parameters in pcd could be accounted for simply by changes in body size and walking speed . To do this we used the equations derived from the linear mixed-effects models in Figure 2 to predict stride parameters across walking speeds for mice the size of the pcd mice and their littermates . The models accurately predicted stride parameters for the littermates , which were not visibly ataxic ( Figure 3B–D , green: thick lines represent model predictions ) . Surprisingly , we also found that the models accurately predicted stride parameters of pcd mice ( Figure 3B–D , purple ) . Thus , although stride parameters of pcd mice were different overall from controls ( Figure 3B–D , purple vs green shadows ) , they were comparable to those predicted for control mice of similar body size walking at similar speeds ( Figure 3B–D , the thick lines representing the model predictions fall on top of the data in the shadows ) . Moreover , a direct comparison between stride parameters for pcd mice and size-matched controls walking at the same speeds revealed no difference between the two groups ( Figure 3—figure supplement 1 ) ( stride length F ( 14 , 1 ) = 0 . 70 , p = 0 . 42; cadence F ( 14 , 1 ) = 0 . 004 , p = 0 . 95; stance duration F ( 14 , 1 ) = 1 . 89 , p = 0 . 17 ) . Next we investigated the possibility that there could be changes in variability of stride parameters in pcd that were not apparent in the averaged data . Analysis of the coefficient of variation revealed that swing length variability was unchanged in pcd compared to size matched controls ( F ( 81 , 1 ) =0 . 14 , p = 0 . 0 . 71 ) . Surprisingly , both cadence and stance duration were less variable in pcd ( cadence: F ( 80 , 1 ) =6 . 90 , p < 0 . 05; stance duration: F ( 80 , 1 ) =6 . 90 , p < 0 . 05 ) . Taken together , these results demonstrate that basic stride parameters , although altered in pcd , do not capture the ataxic symptoms of pcd mice , and highlight the importance of accounting for walking speed ( Koopmans et al . , 2007; Cendelin et al . , 2010; Batka et al . , 2014 ) and using size-matched control animals when analyzing locomotor parameters . For this reason pcd animals are compared with size-matched controls from here on ( Figure 3—figure supplement 1 ) . It has been previously hypothesized that detailed analysis of paw trajectories would capture the gait abnormalities of ataxic mice like pcd , but detailed 3D paw kinematics have not been described for mice . We analyzed the continuous 3D paw trajectories for both wildtype and pcd mice ( Figure 3E–H; Figure 3—figure supplement 2 ) . Surprisingly , we found that the instantaneous forward paw velocity profiles of pcd mice were not distinguishable from those of size-matched controls , across speeds ( Figure 3E ) . Paw velocity peaked early during swing and decelerated before stance onset across walking speeds in both control and pcd mice ( Figure 3E; Figure 3—figure supplement 2 ) . Peak swing velocities increased with faster walking speeds but did not vary by genotype ( F ( 10 . 97 , 1 ) = . 092 , p = 0 . 77 ) . There was also no difference in variability of peak swing velocity between genotypes ( F ( 81 , 1 ) =0 . 27 , p = 0 . 60 ) . This surprising result reveals that even detailed forward paw trajectories are normal in pcd mice , once changes in walking speed and body size are taken into account . We next examined the horizontal ( y ) and vertical ( z ) movements of the paws ( Figure 3F–H; Figure 3—figure supplement 2 ) . Consistent with previous findings of ataxic mice and humans , pcd mice exhibited a wider base of support than size-matched control mice ( Figure 3F ) ( F ( 15 . 51 , 1 ) =42 . 87 , p <0 0 . 001 ) . There were also subtle changes in side-to-side ( y ) paw trajectories ( F ( 13 . 88 , 1 ) =20 . 64 , p <0 0 . 001 ) , especially for front limbs ( F ( 152 . 95 , 1 ) =12 . 125 , p <0 0 . 001; Figure 3G ) . Further , analysis of the vertical ( z ) trajectories revealed significantly larger vertical displacement of both front and hind paws of pcd mice , across speeds ( F ( 66 . 84 , 1 ) =17 . 16 , p <0 0 . 001 ) ( Figure 3H; Figure 3—figure supplement 2 ) . The variability of this vertical displacement was not different in pcd ( F ( 81 , 1 ) =2 . 47 , p = 0 . 12 ) . Thus , despite the visibly ataxic walking pattern of pcd mice , the results of the mixed-effects linear models and the trajectory analyses indicate that the forward motion of the paws was remarkably preserved in pcd . Alterations in individual limb movements were restricted to off-axis ( horizontal and vertical ) trajectories . Analyses of mouse locomotion that have focused on quantifying the kinds of basic stride parameters presented in Figure 2 have previously failed to quantitatively capture gait ataxia in visibly ataxic mice ( Cendelin et al . , 2010 ) . We reasoned that this could be because human observers are more sensitive to the patterns of movement across different parts of the body ( Basso et al . , 2006 ) . Therefore we analyzed patterns of interlimb and whole body coordination in both control and pcd mice . In our experiments , wildtype mice walked in a symmetrical trot pattern across speeds—each diagonal pair of limbs moved together and alternated with the other pair ( Figure 4A , left ) . According to the terminology of Hildebrand ( 1989 ) , at slower speeds there was a tendency toward a ‘walking trot’ ( front paws in a diagonal pair touch down just before hind paws and the paws are on the ground more than 50% of the time ) , while at faster speeds a ‘running trot’ was observed ( diagonal paw pairs strike the ground near-simultaneously and paws are on the ground less than 50% of the time ) ( Figure 4—figure supplement 1 ) . There was no abrupt shift between these gait patterns—stance phases varied smoothly with walking speed and duty cycle ( Figure 4A , Figure 4—figure supplement 1 ) ( cf . Bellardita and Kiehn , 2015 ) . We did not observe galloping or bounding even at the highest speeds ( cf . Bellardita and Kiehn , 2015 ) , probably because the mice freely initiated trials in our experiments , rather than being placed in the corridor by the experimenter at the start of each trial ( Figure 4—figure supplement 1 ) . For ease of quantification , and because of a lack of categorical gait boundaries in our data , we analyzed interlimb coordination in terms of phase values and support patterns rather than gait patterns . 10 . 7554/eLife . 07892 . 013Figure 4 . Front-hind limb coordination is specifically impaired in pcd . ( A ) Polar plots indicating the phase of the step cycle in which each limb enters stance , aligned to stance onset of FR paw ( red ) . Distance from the origin represents walking speed . Left , size-matched control mice ( N = 11 ) . Right , pcd ( N = 3 ) . ( B ) Left-right phase ( left ) and front-hind ( right ) phase for individual animals of pcd and size-matched controls . Circles show average values for each animal . Lines show fit of linear-mixed effects model for each variable . ( C ) Average Hildebrand plots aligned to FR stance onset for speeds between 0 . 15 and 0 . 20 m/s . Grayscale represents probably of stance . ( D ) Area plot of average paw support types as % of stride cycle , across speeds for size-matched controls ( left ) and pcd ( right ) . ( E ) 3 paw ( left ) and 2-paw other ( right ) supports for each animal ( circles ) . Lines show fit of linear-mixed effects model . ( F ) Average ±sem percent double support for hind paws of pcd and size-matched controls . ( G ) Coefficient of variation for paw placement distance ( front-hind ) for pcd and size-matched controls . DOI: http://dx . doi . org/10 . 7554/eLife . 07892 . 01310 . 7554/eLife . 07892 . 014Figure 4—figure supplement 1 . Comparison of gait patterns between control and pcd mice . ( A ) Smoothed probability density of diagonal ( FL-HR ) pair stance phase lags and speed obtained by kernel density estimation for all strides of size-matched controls ( left , n = 3400 , N = 11 ) and pcd ( right , n = 3052 , N = 3 ) . Color code is estimated stride density . ( B ) Smoothed probability density of ipsilateral pair ( FL-HL ) stance phase lags and % stance duration for all strides of size-matched controls ( left ) and pcd ( right ) , plotted according to the convention of Figure 5 of Hildebrand ( 1989 ) . Color code is estimated stride density . ( C ) Polar plots of stance to swing phasing aligned to front right paw for controls ( left ) and pcd ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07892 . 014 The normal pattern of interlimb coordination was markedly disrupted in pcd , due to specific and consistent changes in the phase relationship between front and hind limbs ( F ( 77 . 07 , 1 ) = 4 . 11 , p <0 0 . 05; Figure 4A , right; Figure 4B ) . Importantly , in marked contrast to the front-hind limb coupling , left-right alternation was maintained in pcd ( F ( 159 , 1 ) = 0 . 018 , p=0 . 89; Figure 4A , right: red vs blue and cyan vs magenta; Figure 4B , left ) . Thus , as a result of the de-synchronization of front and hind paw movements , the diagonal limbs no longer moved in phase with each other , as illustrated in the Hildebrand plots in Figure 4C . Support patterns , or the configuration of paws on the ground at any given time , vary systematically with walking speed ( Górska et al . , 1999 ) . Typically , wildtype mice had two diagonal paws on the ground at any given time ( 2-paw diagonal support , Figure 4D , left ) , but this ranged from 3 paws on the ground during slow walking to 0 paw supports , or brief periods of flight , during running at higher speeds , due to changes in stance to swing phasing ( Figure 4—figure supplement 1 ) . Pcd mice spent more time with more paws on the ground ( Figure 4D , right; Figure 4E , left ) ( 3-paw support F ( 82 , 1 ) = 83 . 57 , p < 0 . 001 ) . Moreover , while % double paw support was the same for pcd and size-matched control mice walking at comparable speeds ( F ( 167 , 1 ) = 1 . 06 , p = 0 . 31 ) , the upper limit of pcd walking speeds coincided with the transition from positive to negative % double hind limb supports ( Figure 4F , see ‘Materials and methods’ ) . In other words , it appears that the walking speed of pcd mice is limited by the need to have at least one hind paw on the ground , for postural stability ( Stolze et al . , 2002 ) . Despite their slower walking speeds and increased percent of time spent with more paws on the ground , pcd mice also showed an increase in unstable support configurations such as non-diagonal 2-paw support ( Figure 4E , right; Figure 4—figure supplement 1 ) ( F ( 46 . 78 , 1 ) = 7 . 76 , p = 0 . 01 ) , particularly at higher walking speeds ( F ( 67 . 62 , 1 ) = 115 . 82 , p < 0 . 001 ) . This increased instability indicates that pcd mice are not simply switching to a more stable gait pattern as a compensatory mechanism , but rather , are unable to properly time their front-hind limb movements to generate a stable , efficient gait . Further , the changes in interlimb phasing were consistent—front-hind phasing was not more variable in pcd ( F ( 164 , 1 ) = 2 . 88 , p = 0 . 091 ) , and in fact left-right phasing was even less variable ( F ( 166 , 1 ) = 9 . 70 , p = 0 . 0021 ) . Spatial patterns of paw placement were also disrupted in pcd . While the average distance between hind paw placement and previous forepaw position was the same in pcd and controls ( F ( 14 , 1 ) = 0 . 44 , p = 0 . 52 ) , it was more variable in pcd ( F ( 168 , 1 ) = 75 . 30 , p < 0 . 001; Figure 4G ) , across speeds . Taken together , these results reveal that both spatial and temporal measures of interlimb coordination during overground locomotion were altered in pcd mice . Movements not just of the limbs , but of the entire body need to be coordinated during locomotion . In order to characterize whole-body locomotor coordination in both control and pcd mice , we analyzed their head and tail movements while they walked freely across the corridor ( Figure 5; Figure 5—figure supplement 1 ) . 10 . 7554/eLife . 07892 . 015Figure 5 . The tail and nose movements of pcd mice can be modeled as a passive consequence of the forward motion of the hind limbs . ( A , B ) Averaged lateral trajectory of tail segment 8 ( A ) and nose ( B ) , relative to the mid-point between the hind paws , for animals walking at 0 . 25–0 . 30 m/s , for size-matched controls ( black ) , pcds ( purple ) and model ( dashed gray ) . ( C ) Geometric model of the tail and nose . The lateral position of each tail segment at every time step is given by Syi =_ASisinθ , whereas the lateral position of the nose is given by Ny=_DN _sinλ . ( D , E ) Phase of maximum correlation between the forward position of the hind paws and the lateral trajectories of each tail segment ( green-yellow gradient ) and the nose ( orange ) . ( F ) The maximum correlation phases of the passive geometric model ( lines ) are superimposed on the observed values ( circles ) for tail segments 1 ( dark green ) , 8 ( light green ) , 15 ( yellow ) , and nose ( orange ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07892 . 01510 . 7554/eLife . 07892 . 016Figure 5—figure supplement 1 . Nose and tail movements across speed bins . ( A–C ) Average interpolated ( y ) trajectory of segment 1 , 8 , 15 , respectively for wild type mice aligned with stance onset of the hind right paw . ( D ) Average interpolated ( y ) trajectory of nose for wild type mice aligned with stance onset of the front right paw . ( E–G ) . Average interpolated ( y ) trajectory of all tail segments for pcd mice aligned to stance onset of the hind right paw across several speed bins . ( H ) Bode plot of tail and nose phases for pcd . DOI: http://dx . doi . org/10 . 7554/eLife . 07892 . 01610 . 7554/eLife . 07892 . 017Figure 5—figure supplement 2 . Results summary for young pcd and size-matched controls . ( A ) Average vertical ( z ) position of FR paw relative to ground during swing for young ( <P55 ) pcd ( purple ) and size-matched controls ( compare with Figure 3G ) . Line thickness represents increasing speed . ( B–D ) Polar plots indicating the phase of the step cycle in which each limb enters stance , aligned to stance onset of FR paw ( red ) . Distance from the origin represents walking speed . Left , size-matched control mice . Right , young pcd ( compare with Figure 4A ) . ( C ) Averaged lateral trajectory of tail segment 8 relative to the mid-point between the hind paws , for animals walking at 0 . 1–0 . 2 m/s , for size-matched controls ( black ) , and young pcd ( purple ) ( compare with Figure 5A ) . ( D ) Phase of maximum correlation between the forward position of the hind paws and the lateral trajectories of each tail segment ( green-yellow gradient ) for size-matched controls ( left ) and pcd ( right ) ( compare with Figure 5D , E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07892 . 017 In control mice , lateral movements of the nose and tail were small ( Figure 5A , B , black ) . In striking contrast , however , both the nose and the tail of pcd mice exhibited large side-to-side oscillations during the locomotor cycle ( Figure 5A , B , purple; Video 2 ) ( compared to size-matched controls: nose F ( 13 . 48 , 1 ) = 5 . 49 , p < 0 . 05 , tail F ( 415 , 1 ) = 91 . 01 , p < 0 . 001 ) . Not just the amplitude , but also the timing of nose and tail movements relative to paw stride cycles was altered in pcd . In control mice , both the tail and nose were phase-locked to the stride cycle across walking speeds ( Figure 5D ) . However , in pcd both nose and tail ( Figure 5E ) became increasingly phase-lagged at faster speeds ( tail: F ( 401 . 02 , 1 ) = 5 . 55 , p <0 0 . 05; nose: F ( 53 . 61 , 1 ) = 4 . 89 , p <0 0 . 05 for speeds above 0 . 1 m/s ) . Interestingly , the phase relationships for both the nose and the tail in pcd were consistent with their being time- , rather than phase-locked , to the locomotor cycle ( Figure 5—figure supplement 1 ) . The large , sinusoidal amplitude of the tail oscillations and their time-delayed relationship to hind paw movement suggested to us that tail movements in pcd could reflect passive consequences of the forward movement of the hind limbs . To investigate this possibility , we generated a simple geometrical model of a mouse ( Figure 5C; ‘Materials and methods’ ) in which the tail was anchored at a right angle to a point that bisected a line connecting the two hind paws . We used the data from Figures 3 and 4 to model the hind paw alternation and calculated the predicted side-to-side tail trajectories across walking speeds . As illustrated in Video 3 , the orthogonal coupling between the line connecting the hind paws and the tail segments caused the modeled tail to oscillate from side-to-side as the hind paws moved forward . 10 . 7554/eLife . 07892 . 018Video 3 . Passive nose and tail model . Passive model of nose and tail for a mouse walking at 0 . 2 m/s . The forward movements of the paws were modeled according to the data described in Figures 3 and 4 . The lateral movements of the nose and tail segments are predicted from a model in which an orthogonal projection transforms the forward movements of the hind paws ( solid white line ) into lateral movements of the tail ( dashed white line ) and nose with a fixed time-delay for each element . As a result of this orthogonal coupling , the nose and tail oscillate laterally as a passive consequence of the forward motion of the hind paws . DOI: http://dx . doi . org/10 . 7554/eLife . 07892 . 018 The side-to-side tail movements predicted by the passive model were strikingly similar to the tail oscillations of pcd mice ( compare grey dashed lines with purple traces in Figure 5A , B ) . As shown in Figure 5F , the tail oscillations of pcd mice were fit very well , across walking speeds and tail segments , with a passive model that incorporated a 31 ms time delay for the base of the tail ( reflecting the initial inertia of the mouse's rear ) plus a fixed delay per additional tail segment ( see ‘Materials and methods’ ) . A similar model for the nose ( Figure 5 ) was also consistent with a time-delayed ( 96 ms ) side-to-side nose movement in pcd . Importantly , these oscillations cannot be accounted for by differences in hind limbs between control and pcd , because: ( 1 ) hindpaw alternation is unaltered in pcd ( Figure 4A , B ) ; ( 2 ) hindlimb double support is unaltered in pcd ( Figure 4F ) ; and ( 3 ) though the base of support is wider in pcd , the model predicts that this difference alone would result in smaller , not larger tail oscillations in pcd . Thus , in pcd , the tail and nose appear to move as a passive consequence of forward limb motion . Further , our results suggest that this movement must be actively canceled in wildtype mice to keep the body axis aligned for forward movement . To visualize and quantify impairments in whole-body coordination , we compared vertical ( z ) trajectories for each body part , normalized to 100% of the stride cycle . Figure 6 summarizes the trajectories of individual body parts as well as interlimb and whole body coordination of speed- and size-matched control and pcd mice . During locomotion in control mice , the movement of different parts of the body is synchronized , and vertical nose and tail movements are relatively small ( Figure 6A , left ) . In pcd , however , spatial and temporal coordination across the body is dramatically impaired ( Figure 6A , right ) . Finally , a correlation analysis reveals that in control mice , the movement of most body parts is either strongly correlated or anti-correlated ( i . e . they move either in–or out-of phase with each other; Figure 6B , left , red and blue ) . In pcd mice , however , the correlations are weaker and more variable ( Figure 6B , right ) . This lack of correlational structure reflects the failure of pcd mice to synchronize the movements of different parts of the body . 10 . 7554/eLife . 07892 . 019Figure 6 . Visualization of impaired whole-body coordination in pcd . ( A ) Ribbon plots showing average vertical ( z ) trajectories for nose , paw and select tail ( 2 , 5 , 8 , 11 , 14 ) segments for size-matched control ( Left , N = 11 ) ) and pcd ( Right , N = 3 ) mice walking at 0 . 20–0 . 25 m/s . Data are presented relative to 100% of the stride cycle of the FR paw ( x-axis ) . Nose and paw trajectories are z position relative to floor; tail is z relative to floor with mean vertical position of the of base of the tail subtracted for clarity . ( B ) Matrix of correlation coefficients computed for average vertical trajectories of control ( Left ) and pcd ( Right ) . Color bar is value of correlation coefficient . DOI: http://dx . doi . org/10 . 7554/eLife . 07892 . 019 Taken together , the results of Figures 3–6 suggest that while the forward motion of individual paws is largely spared , ataxic pcd mice have specific deficits in coordinating movement in three dimensions across joints , limbs , and body . Because of the significant challenges associated with quantifying whole-body coordination in freely walking animals , assessments of mouse motor coordination phenotypes often rely on indirect measures ( Mullen et al . , 1976; Herbin et al . , 2007; Lalonde and Strazielle , 2007; Guillot et al . , 2008; Brooks and Dunnett , 2009; Cendelin et al . , 2010; Stroobants et al . , 2012; Sheets et al . , 2013; Suidan et al . , 2013; Camera et al . , 2014 ) , such as time to fall from a rotarod ( Walter et al . , 2006; Lalonde and Strazielle , 2007 ) or a fixed bar ( Kim et al . , 2009; Cendelin , 2014 ) , or mis-steps on a ladder ( Vinueza Veloz et al . , 2014 ) . While these can be sensitive markers for global motor dysfunction , they lack specificity . Moreover , performance on coordination tasks , or even on a treadmill ( Hamers et al . , 2001; Hoogland et al . , 2015 ) , does not necessarily correspond to the degree of gait ataxia during overground locomotion ( Herbin et al . , 2007; Lalonde and Strazielle , 2007; Guillot et al . , 2008; Cendelin et al . , 2010; Stroobants et al . , 2012; Suidan et al . , 2013; Camera et al . , 2014 ) . Our goal in developing LocoMouse was to create a system that would be as useful for assessing cerebellar contributions to locomotor coordination as existing systems have been for analyzing spinal cord control of locomotor pattern generation ( e . g . Crone et al . , 2009; Bellardita and Kiehn , 2015 ) . LocoMouse presents several advantages when compared to available systems for analyzing mouse locomotion ( Hamers et al . , 2001; Hoogland et al . , 2015; Leblond et al . , 2003; Kale et al . , 2004; Garnier et al . , 2008; Zörner et al . , 2010 ) . Mice walk freely and naturally across the corridor and throughput is maximized via fully automated data collection and analysis . The high spatiotemporal resolution provides detailed 3D paw kinematics . LocoMouse also tracks nose and tail movements during locomotion , which have not been previously reported in freely walking mice . While the automated tracking algorithm does not measure joint angles , these can be incorporated either through hand-labeling or potentially with the use of additional markers . We exploited the large , multidimensional dataset generated by LocoMouse to establish a quantitative framework for analyzing locomotor coordination in freely walking mice . There were two key features of our approach that allowed us to isolate specific deficits of ataxia . The first was quantitatively accounting for variability across strides and mice . Although individual strides at first appeared quite variable across our diverse set of wildtype mice , we found that wildtype paw kinematics were readily and quantitatively predictable based on walking speed and body size , even down to the level of 3D trajectories ( Figure 2 , Figure 2—figure supplement 1 ) . Second , we placed particular emphasis on quantifying coordination across the body . The statistical models we developed based on a library of data from wildtype mice ( Figure 2 , Figure 2—figure supplement 1 ) accurately predicted the forward paw motion of ataxic pcd mice ( Figure 3 ) . This indicates that observed differences in basic stride parameters were a secondary consequence of differences in body size and walking speed . Further , the forward motion of individual paws was indistinguishable from that of size- and speed-matched controls , down to the level of detailed paw trajectories . This important result highlights that failure to account for differences in walking speed and body size when comparing data across mice and strides can lead to nonspecific effects being misinterpreted as symptoms of ataxia ( Koopmans et al . , 2007; Cendelin et al . , 2010; Wuehr et al . , 2012; Batka et al . , 2014 ) . Moreover , it is likely that by focusing primarily on the forward movement of individual paws , many existing analyses fail to capture the fundamental features of ataxia . While differences in forward paw motion could be fully accounted for by differences in walking speed and body size , in contrast , off-axis paw trajectories , interlimb , and whole-body coordination revealed specific patterns of impairment in pcd ( Figures 4–6 ) . Differences in off-axis movements ( Figure 3 , Figure 3—figure supplement 1 ) suggest that pcd mice , like human cerebellar patients , are unable to coordinate movements across joints within the limb to perform normal strides in 3D ( Earhart and Bastian , 2001; Bastian et al . , 1996 ) . Further , pcd mice exhibited impaired spatial and temporal coordination of movements across the four limbs , nose , and tail . Interestingly , while front-hind paw coupling was dramatically altered in pcd , left-right alternation was preserved entirely , consistent with the idea that such alternation is generated within the spinal cord itself ( Crone et al . , 2009; Kiehn , 2011; Dougherty et al . , 2013 ) . Finally , the large , oscillatory nose and tail movements observed in pcd were not just random , but were successfully modeled as a failure to predict and compensate for the passive consequences of forward motion of the hind limbs . The major neuroanatomical finding in pcd is the complete postnatal degeneration of cerebellar Purkinje cells and subsequent loss of granule cells and related structures ( Mullen et al . , 1976; Lalonde and Strazielle , 2007; Morton and Bastian , 2007 ) . In light of these extensive anatomical defects and the existing body of literature on mouse ataxia , the remarkable preservation of forward paw motion in pcd ( Figure 3E ) was surprising . Moreover , previous studies have associated Purkinje cell modulation with the step cycle of individual limbs ( Armstrong and Edgley , 1984; Edgley and Lidierth , 1988; Udo et al . , 2004 ) and movement kinematics ( Pasalar et al . , 2006; Heiney et al . , 2014 ) . The most likely interpretation of the surprisingly intact forward paw motion in pcd is that it reflects the presence of inevitable compensatory mechanisms resulting from the chronic loss of Purkinje cells . Given this capacity for compensation , the specific and persistent impairments in multi-joint , interlimb , and whole-body coordination in pcd are particularly striking . While many lines of evidence suggest that the cerebellum provides internal models for motor control ( Wolpert et al . , 1998; Ito , 2008 ) , there has been disagreement about the nature of these models ( Ebner and Pasalar , 2008; Medina , 2011 ) . Studies from some systems , including eye movements , have suggested that the cerebellum acts as an inverse model that computes a command to achieve desired movement ( Shidara et al . , 1993 ) . Other work , particularly on reaching movements , the cerebellar-like nuclei of electric fish , and locomotor adaptation , has suggested that the cerebellum provides a forward model that predicts the consequences of movements ( Bastian et al . , 1996; Kennedy et al . , 2014 ) ( Pasalar et al . , 2006 ) ( Morton and Bastian , 2006 ) . These predictions can then be used to optimize joint angle combinations within a limb ( Bastian et al . , 1996 ) , synchronize interlimb coordination ( Figure 4 ) , or to cancel out the unintended passive consequences of movements of other parts of the body ( Figure 5 ) . We found that while on-axis paw kinematics are preserved in the absence of cerebellar cortical output , the ability to predict and actively cancel the passive consequences of movements of other parts of the body appears to be beyond the limits of compensatory mechanisms available to pcd mice . Therefore our results raise the intriguing possibility that the absence of forward models , rather than a failure to execute appropriate movement kinematics per se , could form the basis for impaired coordination associated with gait ataxia in mice . Although cerebellar involvement is a shared feature across ataxic mouse mutants , the details of the motor phenotypes are different for different mutations ( Lalonde and Strazielle , 2007 ) . Similarly , patients with cerebellar damage exhibit varying degrees and features of ataxia ( Morton and Bastian , 2003 , 2007; Yu et al . , 2014 ) . Given the diversity of cerebellar phenotypes , it is likely that the specific features of gait ataxia will vary across mouse models . The novel quantitative framework for mouse locomotion presented here highlights the importance of considering 3D , interlimb and whole-body coordination and dissociating them from the control of individual paw kinematics , particularly when analyzing cerebellar contributions to locomotion . Together with the sophisticated genetic tools available for manipulating neural circuits in mice , the current approach makes mouse locomotion a powerful system for investigating the neural control of coordinated movement and establishing relationships between neural circuit activity and behavior . All procedures were reviewed and performed in accordance with the Champalimaud Centre for the Unknown Ethics Committee guidelines , and approved by the Portuguese Direcção Geral de Veterinária ( Ref . No . 0421/000/000/2015 ) . C57BL/6 mice were housed in institutional standard cages ( 3 animals per cage ) on a reversed 12-hr light/12-hr dark cycle with ad libitum access to water and food . Heterozygous Purkinje cell degeneration mice on a C57BL/6 background were obtained from Jackson labs ( #0537 B6 . BR-Agtpbp1pcd/J ) . Experiments were conducted in two groups: ( a ) wildtype controls ( n = 9602; N = 34 mice; 23 male; 11 female; 7-33g , 30–114 days old ) and ( b ) homozygous pcd mice ( n = 3052; N = 3 mice; 2 female , 1 male; 10–16 g; run at several ages each between 41-154 days old ) and their littermates ( n = 2256; N = 7 mice; 3 female , 4 male; 15–40 g; 34–190 days old ) . Size-matched controls for pcd animals ( n = 3400; N = 11 mice ) were taken from the wildtype data set ( Figure 2—figure supplement 1 and Figure 3—figure supplement 1 ) . Because of extra-cerebellar degeneration in pcd mice after postnatal day 50 , we performed a separate analysis to verify that our main findings held in the youngest pcd mice ( Figure 5—figure supplement 2 ) . A custom-designed setup was developed to assess whole body coordination during overground locomotion in mice ( Figure 1 ) . The LocoMouse apparatus consists of a clear glass corridor , 66 . 5 cm long , 4 . 5 cm wide and 20 cm high . Mice were filmed crossing the corridor with a high-resolution , high-speed camera ( Bonito CL-400B , Allied Vision Technologies , https://www . alliedvision . com ) . A mirror ( 66 cm × 16 cm ) was placed below the corridor at an angle of ∼45° to allow simultaneous collection of side and bottom views in order to generate three-dimensional tracking data . Lighting consisted of a matrix of LEDs that emitted cool white light positioned to maximize contrast and reduce reflection . Infrared sensors positioned along the runway automatically triggered the camera and acquisition software once the mouse entered the corridor and stopped the acquisition once the mouse reached the other end of the corridor or after 25 s . The stride cycles of individual paws were automatically broken down into swing and stance phases based on the first derivative of the paw position trajectories . Individual strides were defined from stance onset to subsequent stance onset . For each stride , average walking speed was calculated by dividing the forward motion of the body center during that stride by the stride duration . All data was sorted into speed bins ( 0 . 05 m/s bin width ) in a stridewise manner , with a minimum stride count criterion of 5 strides per bin , per animal . Individual limb movements and interlimb coordination were calculated as follows: Statistical analyses were done in Matlab and R . For all comparisons , models were selected by comparing equations specifying additive fixed-effects terms with those specifying n-way interaction terms using a likelihood-ratio test and inspection of statistical significance of included terms . Depending on the comparison , fixed-effects terms included a subset of the following variables: speed , genotype and paw . All models were random-intercepts models with subject as a random covariate . Unless otherwise indicated , results are reported as conditional F tests with Satterthwaite degrees of freedom correction . All variability analyses were based on coefficients of variation ( CV ) . The simulations of the lateral movements of the tail and nose were carried out in three steps . ( 1 ) First , we estimated the desired stride parameters for each hind paw . The temporal parameters of the stride ( swing , stance and stride durations ) were taken from Figure 2 , calculated for an animal of 15g ( small animal control ) . We calculated the stride lengths by estimating the final desired position of the paw ( relative to the nose of the animal ) at the end of the stride; this value was also taken from small animal control data . ( 2 ) We calculated the forward trajectory for each paw using a constant swing velocity model given by the stride length divided by the swing duration . We then calculated the periodic forward oscillations of the two hind paws by calculating the difference between their forward trajectories . The phasing of hind paw alternation was taken from the small animal control data ( Figure 4A ) . ( 3 ) Finally , we converted the forward movements of the paws into lateral movements of the tail and nose . We calculated tail and nose trajectories that would be predicted from a purely passive coupling of forward motion of the hind paws with lateral tail/nose motion through the geometric relations BC→⊥ LR→ and BD→⊥ LR→ ( Figure 5A ) . We bisected the line segment that connects the two hind paws , with an orthogonal line segment ( CD ) . For each tail segment Si , at each timestep t , we define a vector with constant length , originating in A , along the direction AC→ ( gray points along the line AC ) . The lateral position of each segment of the tail was then given by Syi = Li sin θ , where Li is the distance between the center of tail oscillation ( A ) and segment i , and θ is the angle between segment ( AC ) and the anterior–posterior axis of the animal . Similarly , for the nose , we defined a vector with constant length , originating in A , along the direction DA→ . The position of the nose was given by Ny = L sin λ , where L is the distance between the center of nose oscillation ( A ) and the nose , and λ is the angle between segment ( DA ) and the anterior–posterior axis of the animal . The final position of each tail segment Si and nose N were then given by each of these vectors delayed by a fixed amount . The time delays for the nose and base of the tail were estimated by linear regression on a plot of phase as a function of stride frequency ( cadence ) ( Figure 5—figure supplement 1H ) . Delays between subsequent tail segments decreased according to the equation delay = −0 . 23 *i + 3 . 97 , where i is the segment number , starting at the base of the tail .	Though it seems simple , walking is a complex activity . The arms , legs , body , and head all need to work together . A part of the brain called the cerebellum helps to coordinate the movements of different body parts allowing both simple and complex tasks to be carried out smoothly . But it is not known exactly how the cerebellum coordinates body movements . Studies of mice have helped shed some light on the coordination of movement . Several mutations that naturally occur in mice can cause them to walk abnormally . These mutations often cause changes in the cerebellum . Neuroscientists studying these mutant mice often use balance beams or other challenging tasks to compare their coordination with typical mice . But studies attempting to measure specific changes in walking movements under natural conditions have yielded conflicting results . Now , Machado , Darmohray et al . have demonstrated that an automated movement-tracking system can capture specific aspects of coordination in freely walking mice . The system , called ‘LocoMouse’ , uses high-speed cameras and computers to document and analyze the paw , nose , and tail movements of mice walking through a glass hallway . In the experiments , the system was used to compare typical mice with mice that have a mutation affecting the cerebellum that causes them to walk abnormally . Unexpectedly , Machado , Darmohray et al . found that forward steps of the mutant mice are comparable to the steps of the typical mice , if you account for the fact that the mutant mice are smaller and slower . Instead , however , the mutants were found to have specific difficulties coordinating movement across the body . The movements of the mutant mice's front and hind paws , for example , did not follow the same coordinated pattern as the typical mice . The mutant mice also swung their head and tail in an exaggerated way . Machado , Darmohray et al . 's analysis revealed that these movements likely resulted from a failure of the cerebellum of the mutant mice to predict and compensate for the motion of the rest of the body . Other scientists will now likely use the LocoMouse system to study mouse movements and how the brain controls them .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2015	A quantitative framework for whole-body coordination reveals specific deficits in freely walking ataxic mice
Metastasis is a major cause of cancer mortality . We generated an autochthonous transgenic mouse model whereby conditional expression of MYC and Twist1 enables hepatocellular carcinoma ( HCC ) to metastasize in >90% of mice . MYC and Twist1 cooperate and their sustained expression is required to elicit a transcriptional program associated with the activation of innate immunity , through secretion of a cytokinome that elicits recruitment and polarization of tumor associated macrophages ( TAMs ) . Systemic treatment with Ccl2 and Il13 induced MYC-HCCs to metastasize; whereas , blockade of Ccl2 and Il13 abrogated MYC/Twist1-HCC metastasis . Further , in 33 human cancers ( n = 9502 ) MYC and TWIST1 predict poor survival ( p=4 . 3×10−10 ) , CCL2/IL13 expression ( p<10−109 ) and TAM infiltration ( p<10−96 ) . Finally , in the plasma of patients with HCC ( n = 25 ) but not cirrhosis ( n = 10 ) , CCL2 and IL13 were increased and IL13 predicted invasive tumors . Therefore , MYC and TWIST1 generally appear to cooperate in human cancer to elicit a cytokinome that enables metastasis through crosstalk between cancer and immune microenvironment . Tumorigenesis is caused by specific oncogenes but tumor progression often involves the acquisition of metastasis ( Chaffer and Weinberg , 2011 ) . Metastasis occur when tumor cells gain the ability to invade , migrate and colonize distant sites , and this accounts for most of the morbidity and mortality associated with cancer ( Mehlen and Puisieux , 2006 ) . Many studies have examined human clinical specimens and/or tumor-derived cell lines to discern mechanisms of metastasis , and identified the role of specific genes ( Ji et al . , 2007; Sun et al . , 2018; Zhu et al . , 2017 ) and also the role of the tumor microenvironment ( Fidler , 2003; Kalluri , 2016; Kim et al . , 2017; Whitfield and Soucek , 2012 ) . However , to date , there are very few mouse models that exhibit spontaneous metastasis , and even fewer in vivo models where the stepwise progression from a non-metastatic to metastatic cancer can be studied . Such a model would provide a tractable approach for studying specific mechanisms of metastasis , particularly the role of the immune microenvironment . Innate immune cells , especially tumor associated macrophages ( TAMs ) , are known to contribute to metastasis through multiple mechanisms including effects on angiogenesis , production of specific cytokines , suppression of the immune system , and induction of epithelial-mesenchymal transition ( EMT ) ( Gonzalez et al . , 2018; Lu et al . , 2011; Qian and Pollard , 2010; Wan et al . , 2014 ) . The specific discrete events in the cancer cell that modulate the tumor immune microenvironment and enable metastasis are not clear . The MYC oncogene is a transcription factor that is one of the most commonly activated oncogenes in the pathogenesis of many types of human cancer including HCC ( Schaub et al . , 2018; Dang , 2012; Gabay et al . , 2014 ) . Previously , we used the Tet System to generate a conditional transgenic mouse model for MYC-induced HCC that we and others have used to study mechanisms of oncogene addiction ( Settleman , 2012 ) and identify potential therapies ( Dhanasekaran et al . , 2018; Kapanadze et al . , 2013; Ma et al . , 2016; Shachaf et al . , 2004 ) . But murine MYC-driven HCC do not metastasize . Twist1 is a transcription factor that is important during embryogenesis for normal cellular migration ( Lee et al . , 1999; Thisse et al . , 1987 ) . Twist1 has been shown to be an important gene product that can enable mouse and human tumor cell lines to acquire the ability to metastasize associated with EMT ( Thiery et al . , 2009; Xu et al . , 2017 ) . Here we used the Tet System to conditionally express Twist1 in combination with MYC to show that their co-expression leads to widely metastatic and invasive HCC . We use this powerful in vivo model to uncover a surprising mechanism by which MYC and Twist1 drive metastasis . Cancer cell-intrinsic properties like proliferation , apoptosis or invasiveness were not different between the non-metastatic MYC-HCC and the metastatic MYC/Twist1-HCC . Instead , metastatic progression was dependent on the ability of MYC and Twist1 to dramatically reprogram the tumor innate immune microenvironment . Together , MYC and Twist1 induce the cancer cell to secrete cytokines like Ccl2 and Il13 that lead to recruitment and polarization of macrophages respectively , thus causing metastasis . Systemically , administering Ccl2 and Il13 is sufficient to cause metastasis of MYC-HCC and , conversely blocking these specific cytokines profoundly inhibits metastasis in MYC/Twist1 HCC . Our results are broadly generalizable to 33 different human cancers and predict invasive cancer in a pilot clinical study . We first generated a transgenic mouse using the Tet system that conditionally expresses Twist1 in a liver specific manner ( LAP-tTA/TRE-Twist1/Luc ) . We crossed TRE-Twist1/Luc mice which harbored the Twist1 and firefly luciferase ( luc ) genes under the control of a bidirectional tetracycline responsive element ( TRE ) , with the LAP-tTA mice which contain the tetracycline-controlled transactivator protein ( tTA ) driven by the liver-enriched activator protein ( LAP ) promoter ( Tran et al . , 2012 ) . Twist1 transgenic mice ( LAP-tTA/TRE-Twist1/Luc ) exhibited no disease nor gross or microscopic pathology for as long as 18 months of observation thus demonstrating that Twist1 did not play a role in autochthonous tumorigenesis when overexpressed in the liver ( Figure 1—figure supplement 1a ) . To examine the influence of Twist1 on tumor progression , LAP-tTA or LAP-tTA/TRE-Twist1/Luc mice were crossed with TRE-MYC ( Shachaf et al . , 2004 ) ( Figure 1a ) to generate transgenic mice that inducibly expressed MYC alone ( MYC mice ) or co-expressed MYC , Twist1 and luciferase ( Luc ) in a liver-specific manner ( MYC/Twist1 mice ) ( Figure 1b ) . We induced transgene expression in adult mice at 6 weeks of age ( Figure 1b ) . In vivo , Twist1 transgene expression was confirmed to be confined to the liver by measuring the luciferase reporter by bioluminescence imaging ( BLI ) ( Figure 1c ) . We followed in vivo tumor progression with serial cross-sectional imaging . Both MYC and MYC/Twist1 mice were observed to develop multifocal liver cancer , while only MYC/Twist1 mice developed lung metastases ( Figure 1d ) . MYC/Twist1 mice were moribund with HCC sooner and had a median survival of 25 months compared to 32 months in MYC mice ( p<0 . 001 , Figure 1e ) . MYC mice rarely exhibited metastasis even after extended observation ( 2% , n = 50 , Figure 1f ) ; whereas , MYC/Twist1 mice regularly exhibited rapid onset of metastasis with high penetrance ( 90% ) -metastases to the lungs ( 70% ) , peritoneum ( 60% ) and lymph nodes ( 20% ) ( n = 50 , Figure 1f ) . Thus , Twist1 combined with MYC expression in liver cells elicits HCC metastasis . A simple explanation for our results is that Twist1 was inducing more rapid onset and thereby progression of tumorigenesis . Against this possibility , the tumor burden in the liver was not statistically different between MYC and MYC/Twist1 mice ( Figure 1g ) . Also , there was no difference in the gross or microscopic appearance of MYC- and MYC/Twist1-HCC ( Figure 1h–i ) . The MYC- and MYC/Twist1-HCC tumors were confirmed to be HCC by a pathologist and by expression of hepatocyte marker glutamine synthetase ( Figure 1—figure supplement 1b ) . We considered that Twist1 could be influencing MYC expression levels , but MYC levels were similar between the two tumor models , while Twist1 was only overexpressed in the MYC/Twist1-HCC ( Figure 1—figure supplement 1c-d ) . Tumor cell proliferative index ( phospho histone three expression ) and apoptosis ( cleaved caspase three ) between MYC- and MYC/Twist1-HCC were not different ( Figure 1j–1k ) . Primary tumor-derived cell lines from MYC- and MYC/Twist1-HCC did not show any difference in migratory capacity ( Figure 1—figure supplement 1e ) . Lastly , Twist1 is a regulator of epithelial-mesenchymal transition ( EMT ) 26 , 37 , but we did not observe significant differences in the expression of multiple epithelial and mesenchymal markers between MYC and MYC/Twist1 tumors ( Figure 1—figure supplement 1f ) . Therefore , Twist1 drives metastasis of MYC-induced HCC without affecting primary tumor burden , MYC expression , tumor cell proliferation , apoptosis , invasiveness or EMT markers . The influence of Twist1 on global gene expression was measured in MYC- and MYC/Twist1-HCC ( n = 5 ) using next generation sequencing ( NGS ) based RNA sequencing . Through unsupervised hierarchical clustering using principal component analysis ( PCA ) , MYC- and MYC/Twist1-HCC were found to have overall distinct , non-overlapping expression profiles that clustered separately ( Figure 2a ) . A comparative analysis identified 514 genes ( 220 up and 294 down ) that were differentially expressed between MYC-HCC and MYC/Twist1-HCC ( p<0 . 001 , q < 0 . 05 , fold change ≥2 ) ( Figure 2b , Supplementary file 1 ) . Functional pathway analysis revealed the top biological processes upregulated in MYC/Twist1-HCC involved inflammatory responses including leukocyte infiltration , myeloid cell and granulocyte recruitment ( Figure 2b , Figure 2—figure supplement 1a , Supplementary file 2 ) . CIBERSORT ( Newman et al . , 2015 ) identified M2 macrophages to be significantly enriched in MYC/Twist1 tumors , of the 22 immune subsets analyzed ( Figure 2c ) . MYC/Twist1-HCC exhibited a 15-fold shift in the ratio of M2 to M1 macrophages when compared to MYC tumors ( Figure 2c ) . No significant differences in other major immune compartments were seen including- B cells , T cells , NK cells , dendritic cells , neutrophils , or mast cells ( Figure 2d ) . Increased macrophage infiltration in MYC/Twist1 primary and metastatic tumors ( Figure 2e ) with no change in neutrophils or CD4 T cells infiltration was confirmed by IHC ( Figure 2—figure supplement 1b ) . TAMs isolated from primary MYC/Twist1-HCC had increased macrophages of the M2 phenotype ( Cd206High/Arg1High ) , more specifically a M2a phenotype ( Figure 2f ) . Induction of MYC and Twist1 expression is associated with tumor initiation and rapid onset of macrophage infiltration in early tumors which increases during tumor progression . Conversely , the inactivation of MYC and Twist1 in tumors shows rapid and complete tumor regression in all observed mice ( n = 20 ) within 2 weeks ( Figure 2—figure supplement 1c ) and also prompt exodus of macrophages ( Figure 2g–2h ) . TAMs have been shown to increase the migratory capacity of tumor cells ( Lin et al . , 2001 ) . Conditioned media from TAMs isolated from MYC/Twist1-HCC but not MYC-HCC increased the invasiveness of both MYC- and MYC/Twist1-HCC tumor cells in vitro ( Figure 3a–3c ) . To discriminate the independent roles of MYC and Twist1 in metastasis we developed cell lines where we could modulate the expression of MYC and Twist1 separately . Primary cell lines from MYC/Twist1 HCC were retrovirally transduced with MYC and/or Twist1 , such that upon inactivation of transgene expression with Doxycycline , they now constitutively expressed MYC and/or Twist1 ( Figure 3d ) . We confirmed that treatment of these cell lines with doxycycline resulted in the continued expression of constitutive MYC and/or Twist1 by qPCR ( Figure 3—figure supplement 1a-b ) . We observed that the inactivation of either MYC or Twist1 abrogated the ability of the cells to develop lung metastasis when injected intravenously in NOD scid gamma ( NSG ) mice , while cells expressing both MYC and Twist1 led to development of extensive lung metastasis with prominent macrophage infiltration ( Figure 3e–3f , Figure 3—figure supplement 1c ) . Thus , MYC and Twist1 cooperate , and are both required to induce metastasis of HCC by a macrophage dependent mechanism . We determined if TAMs are required for Twist1 to drive metastasis on MYC-HCC in vivo . Primary tumor-derived cell lines which conditionally express MYC or MYC/Twist1 ( Figure 4—figure supplement 1a ) were re-introduced in vivo either by orthotopic transplantation into the liver or intravenous injection . Orthotopic implantation ( Figure 4a ) of MYC/Twist1- but not MYC-HCC tumor cells in NSG mice led to pulmonary and intrahepatic metastases with extensive macrophage infiltration ( Figure 4b–c , Figure 4—figure supplement 1b ) . Macrophage depletion with clodronate liposomes ( Moreno , 2018 ) but not control liposomes , in mice orthotopically transplanted with MYC/Twist1-HCC had reduced intrahepatic ( p=0 . 0006 , FC 4 . 4 ) and lung metastases ( p<0 . 0001 , FC 8 . 8 ) ( Figure 4e–f ) . Quantification of BLI signal at the end of treatment did not show statistical difference in densitometry between control treated and clodronate treated mice ( Figure 4e ) . Note , clodronate was confirmed to remove macrophages but not affect tumor cells ( Figure 4—figure supplement 1c-d ) . A reduction in the number of macrophages was confirmed by IHC for F4/80 in normal liver , tumor and lungs ( p<0 . 001 ) ( Figure 4—figure supplement 1e-f ) . To evaluate if macrophages are required for the colonization step of metastasis , we used the lung trap assay . Intravenous injection of MYC/Twist1-HCC but not MYC-HCC cells resulted in pulmonary metastases associated with macrophage infiltration ( Figure 4g–i , Figure 4—figure supplement 1g ) . Clodronate depletion of macrophages , almost completely abrogated pulmonary metastasis ( p=0 . 0003 , FC 4 . 3 ) ( Figure 4j–4l ) . Therefore , Twist1 elicits metastasis of MYC-HCC and promotes the invasiveness and colonization of metastases by a macrophage-dependent mechanism . To evaluate if cytokines secreted by cancer cells mediate MYC and Twist1 driven macrophage recruitment and polarization , we evaluated the impact of tumor cell derived conditioned media on non-polarized macrophage cell lines . Conditioned media derived from MYC/Twist1- but not MYC-HCC cells ( Figure 5a ) was sufficient to promote the migration of macrophages towards cancer cells ( Figure 5b ) . Conditioned media from MYC/Twist1- but not MYC-HCC cells was able to elicit changes in the morphology of macrophages to resemble M2 phenotype ( McWhorter et al . , 2013 ) ( Figure 5c ) and increased expression of M2 markers: Cd206 , Arg1 and Cc3cr1 ( Figure 5c ) , but not M1 markers: iNos , Ccr2 , Ifnar2 ( Figure 5—figure supplement 1a ) . A multiplex ELISA for 38 cytokines was performed ( Figure 5d ) identifying that MYC/Twist1-HCC cells had increased secretion of cytokines: Il13 , Ccl2 , Ccl5 , Ccl7 and Cxcl1 ( p<0 . 05; Fold change ≥2 , mean ≥20 ng/ml ) ( Figure 5e , Figure 5—figure supplement 1b ) . These five cytokines were confirmed to be transcriptionally upregulated in MYC/Twist1- vs . to MYC cells by qPCR ( Figure 5e ) . The role of individual cytokines in macrophage recruitment was assessed . Antibodies that neutralize Ccl2 , Ccl5 , Ccl7 or Cxcl1 inhibited the ability of conditioned media from MYC/Twist1-HCC to promote migration of macrophages ( Figure 5f–h ) . Neutralization of Ccl2 decreased migration 7-fold ( p<0 . 0001 ) , Ccl5 1 . 2-fold ( p=0 . 001 ) , Ccl7 1 . 8-fold ( p=0 . 001 ) and Cxcl1 2 . 1-fold ( p=0 . 0002 ) . Neutralizing Il13 ( p=ns , FC 1 . 2 ) did not affect macrophage migration . Also , we found that the neutralization of Ccl2 inhibited the recruitment of macrophages into MYC/Twist1-spheroids by 3D culture ( Figure 5—figure supplement 1c ) . Next , the role of cytokines on macrophage polarization was determined ( Figure 5h ) . We found that neutralization of Il13 blocked M2 polarization by 50-fold reduction in Cd206 expression ( p<0 . 0001 ) and 7-fold decrease in Arg1 ( p<0 . 0001 ) without any change in M1 markers . Neutralization of Ccl5 led to a 4-fold reduction in Cd206 ( p<0 . 001 ) and 1 . 1-fold decrease in Arg1 ( p<0 . 05 ) and Cxcl1 led to 2 . 4-fold decrease in Cd206 ( p<0 . 001 ) without significant change in Arg1 ( p=ns ) ( Figure 5i ) . Conversely , adding the cytokines Il13 , but not Ccl2 , to co-cultured MYC-HCC cells increased M2 markers expression ( Figure 5—figure supplement 1d ) . MYC and TWIST1 are both transcription factors , so we evaluated if they epigenetically regulated CCL2 and IL13 expression . We identified MYC and TWIST1 promoter binding upstream of human CCL2 and IL13 protein-coding genes in Gene Transcription Regulation Database ( GTRD ) , a meta-analysis of Chip-seq experiments ( Dreos et al . 2017; Yevshin et al . , 2019 ) ( Figure 5—figure supplement 1e ) . Both MYC and Twist1 demonstrated binding at multiple sites in the promoter regions of CCL2 and IL13 in the ChIP-seq data from several different cancer cell lines ( Figure 5—figure supplement 1e ) . We also looked for MYC and Twist1 promoter binding sites in mouse Ccl2 and Il13 promoters using motif finding analysis of the public data from JASPAR ( Bryne et al . , 2008 ) and Eukaryotic promoter database ( EPD ) ( Dreos et al . 2017 ) . Again , we found multiple potential MYC and Twist1 transcription factor binding sites for both Ccl2 and Il13 ( Figure 5—figure supplement 1f ) . These data suggest that MYC and Twist1 cooperate to transcriptionally regulate expression of Ccl2 and Il13 in the cancer cells . We examined if Ccl2 and/or Il13 are sufficient to elicit metastasis in vivo . Orthotopic transplants of non-metastatic MYC-HCC in NSG mice were treated with either PBS ( control ) , or with recombinant Ccl2 alone or Il13 alone or their combination for 4 weeks ( Figure 6a ) . Control mice did not develop metastatic nodules even though scattered , single cells were found in the lungs ( Figure 6b ) . No mice treated with Ccl2 alone ( p=0 . 390 , FC = 3 ) or Il13 alone ( p=0 . 99 , FC = 1 ) exhibited intrahepatic or pulmonary metastases ( Figure 6c , e and g ) . All orthotopic MYC-HCC mice treated with the combination of Ccl2 and Il13 developed intrahepatic metastases ( p=0 . 008 , FC 2 . 5 ) and multifocal pulmonary metastases ( p=0 . 02 , FC = 104 . 4 ) ( Figure 6c , e and g ) . Hence , both Ccl2 and Il13 are sufficient to elicit MYC-HCC to metastasize , even if Twist1 is not expressed in the tumor cells . We evaluated if treatment with cytokines altered tumor macrophage infiltration . The combined treatment with Ccl2 and Il13 increased macrophage infiltration at both the orthotopic primary tumor in the liver and pulmonary metastases ( p<0 . 0001 ) ( Figure 6e and h ) , with enrichment of M2-like Cd206+ TAMs ( Figure 6—figure supplement 1a-b ) . Treatment with Ccl2 alone increased macrophage recruitment 1 . 8-fold ( p=0 . 001 ) but did not induce metastasis ( Figure 6c , d , f and g ) . On the other hand , treatment with Il13 alone did not increase either macrophage infiltration or induce metastasis ( Figure 6e and h ) . Also , in vivo treatment of MYC-HCC with combination of Ccl2 and Il13 stimulated angiogenesis . The most significant increase in angiogenesis , as assessed by Cd31 IHC staining , was noted in MYC-HCC tumors in mice treated with Ccl2 and Il13 ( Figure 6—figure supplement 1c ) . As a negative control , treatment with recombinant Il4 , a cytokine which was not increased in MYC/Twist1-HCC was performed . Il4 did not elicit either macrophage recruitment or increase metastasis ( Figure 6—figure supplement 1d-e ) . Thus , combination of Ccl2 and Il13 induces metastasis of MYC-HCC associated with macrophage recruitment and polarization . We examined if neutralizing antibodies to Ccl2 and/or Il13 influenced metastasis in vivo . Mice with orthotopic transplants of metastatic MYC/Twist1-HCC were treated either with control antibody , or anti-Ccl2 antibody alone or anti-Il13 antibody alone or their combination for 4 weeks ( Figure 7a ) . We first confirmed that there was no statistical difference in orthotopic primary liver tumor burden between the four groups by quantifying primary tumor volume . Quantification of BLI imaging showed a trend towards decrease but no statistical difference in liver tumor burden between the four groups ( Figure 7b–c ) . Control antibody treated mice , as expected , developed multifocal intrahepatic and pulmonary metastases with extensive macrophage infiltration ( Figure 7c–e ) . Inhibition of Ccl2 alone led to 7-fold decrease ( p=0 . 002 ) and Il13 alone to a 2 . 5-fold decrease ( p=0 . 03 ) in lung metastasis ( Figure 7c and e ) . Further , treatment with combination of anti-Ccl2 and anti-Il13 antibodies showed synergism , and led to a 12-fold decrease in liver metastases ( p=0 . 0006 , Figure 7c and d ) and 14-fold decrease in lung metastases ( p=0 . 0009 ) ( Figure 7c and e ) . The combined inhibition of Ccl2 and Il13 was noted to lead to dramatic reduction in macrophage recruitment ( p<0 . 0001 , FC 7 . 1 ) ( Figure 7c and f ) and polarization to M2-like phenotype ( Figure 7—figure supplement 1a ) . The inhibition of Ccl2 alone ( p<0 . 001 , FC 4 . 0 ) , but not Il13 ( p=0 . 435 ) , decreased macrophage recruitment ( Figure 7c and f ) . While inhibition of Il13 , either alone or in combination with Ccl2 , led to loss of M2-like Cd206+ or Arg1+ macrophages in primary or metastatic sites ( Figure 7—figure supplement 1a–b ) . Thus , inhibition of Ccl2 and Il13 decrease macrophage recruitment and polarization respectively . As a control , mice bearing MYC/Twist1-HCC were treated with anti-Il4 antibody , which we show had no effect on metastasis or macrophage recruitment ( Figure 7—figure supplement 1 c–e ) . Therefore , combined inhibition of Ccl2 and Il13 synergistically reduced macrophage recruitment and polarization thus blocking metastases ( Figure 7g ) . We examined if MYC and TWIST1 cooperated in human tumorigenesis by examining 9502 human patients with 33 different cancers from a TCGA study ( Tang et al . , 2017 ) and 144 HCC patients with metastatic HCC ( Ye et al . , 2003 ) . Increased MYC and TWIST1 expression was associated with significantly worse disease-free survival ( DFS ) ( p=4 . 3×10−10 ) in the pan-cancer cohort ( Figure 8a ) . Combined overexpression of MYC , TWIST1 , CCL2 and IL13 predicted a slightly worse disease-free survival than MYC+TWIST1 overexpression alone ( p=2 . 9×10−12 ) ( Figure 8a ) . In another cohort of 144 patients with metastatic HCC , combined overexpression of MYC and TWIST1 was associated with significantly worse prognosis than either MYC or TWIST1 alone ( Figure 8b ) . We determined if MYC and TWIST1 in human tumors cooperated to influence the tumor microenvironment through CIBERSORT analysis of 10 , 366 tumors from the human pan-cancer TCGA study ( Thorsson et al . , 2018 ) . TAMs were the most common infiltrating immune cells in most types of human cancers ( Figure 8—figure supplement 1a ) including HCC ( Figure 8—figure supplement 1b ) . Compared to MYC/TWIST1Low tumors , MYC/TWIST1High tumors were infiltrated with significantly higher proportion of monocytes ( 2 . 9% vs 4 . 4%; p<0 . 001 ) and M2 macrophages ( 22% vs 26% , p<0 . 001 ) , while there was a lower proportion of M0 ( 10% vs 7% , p<0 . 001 ) and M1 macrophages ( 6% vs 4% , p<0 . 001 ) ( Figure 8c ) . This was also true in the HCC cohort ( M2 macrophages 24% vs 28% p=0 . 001; Figure 8c ) . Moreover , combined MYC and TWIST1 expression strongly correlated with the expression of M2 macrophage related genes in the pan-cancer TCGA data ( Figure 8d ) . Increased TAM infiltration in HCC was highly prognostic of overall survival on univariate and multivariate analysis ( p=0 . 01 , HR 16 . 0 , Figure 8—figure supplement 1c-e ) . Also , M2-like TAMs were associated with presence of vascular invasion , advanced stage and poor tumor grade ( Figure 8—figure supplement 1f ) in HCC . Further , combined MYC and TWIST1 expression correlated strongly with CCL2 and IL13 in the pan-cancer cohort ( p=1 . 4×10−109 ) ( Figure 8d ) . Thus , MYC and TWIST1 predict poor survival , CCL2/IL13 expression and M2-like TAM infiltration in human cancers . We screened MYC and TWIST1 expression in four human HCC cell lines-Huh7 , SNU398 , SNU475 and SNU182 . We identified Huh7 as a MYC/TWIST1Low cell line and SNU398 as MYC/TWIST1High cell line ( Figure 8e ) . The other two cell lines had intermediate levels of TWIST1 . The MYC/TWIST1High cells transcriptionally expressed significantly higher levels of CCL2 and IL13 when compared to the MYC/TWIST1Low cells ( Figure 8e ) . We treated macrophages with conditioned media derived from either MYC/TWIST1Low or MYC/TWIST1High cells , in vitro , and assessed their effect on macrophage polarization ( Figure 8f ) . Conditioned media from MYC/TWIST1High cells polarized macrophages to M2-like phenotype while conditioned media from MYC/TWIST1Low cells induced a M1-like phenotype ( Figure 8g ) . Thus , combined MYC and TWIST1 in human HCC cells lines is associated with CCL2 and IL13 secretion and macrophage M2-like polarization . Lastly , a prospective clinical study was performed to measure CCL2 and IL13 in human patients with HCC ( n = 25 ) and in patients with cirrhosis of the liver ( n = 10 ) . Both CCL2 ( p=0 . 006 ) and IL13 ( p<0 . 0001 ) , were significantly elevated in the plasma of patients with HCC but not cirrhosis ( Figure 8h ) . Increased expression of IL13 , but not CCL2 , was associated with the presence of multifocal tumors ( p=0 . 04 ) , suggestive of association with aggressive phenotype ( Figure 8i ) . Also , IL13 levels were higher in the plasma of patients with HCC with vascular invasion , which is a known indicator of metastatic tumor spread in HCC ( p=0 . 01 ) ( Figure 8j ) . Thus , the plasma levels of the cytokines CCL2 and IL13 are elevated in patients with HCC , and higher IL13 levels predict multifocal and invasive HCC . We found that MYC and TWIST1 drive metastasis by eliciting a transcriptional program in cancer cells that induces cytokines that in turn enable crosstalk between tumor and host , thus eliciting the recruitment and polarization of macrophages ( Figure 9 ) . To perform our studies , we have generated the first autochthonous transgenic mouse model of HCC metastasis by hepatocyte-specific expression of MYC and Twist1 that will be useful to study tumor progression and identify new therapies . We demonstrate that constitutive expression of both MYC and Twist1 is required for metastasis . MYC and Twist1 coordinate to regulate a cytokinome , including Ccl2 and Il13 , that both are necessary and sufficient to elicit metastasis . Moreover , in 33 different human cancers , expression of both MYC and TWIST1 predict survival , expression of CCL2/IL13 and M2-like TAM infiltration . In a prospective clinical analysis , human patients with HCC but not cirrhosis had increased plasma levels of CCL2 and IL13 that was associated with invasive and multifocal disease . We conclude that MYC and TWIST1 are general drivers of metastasis . We generated a new conditional transgenic mouse model of metastasis that has some features that are complementary to existing model systems ( Gómez-Cuadrado et al . , 2017 ) . Our models enabled us to interrogate tumor and host interactions during malignant progression in an immunocompetent host . Through the Tet system , we were able to elicit inducible combined MYC and Twist1 expression restricted to the hepatocytes . In our model , greater than 90% of the mice predictably developing extrahepatic metastasis . This allowed us to directly compare the non-metastatic MYC-HCC and the metastatic MYC/Twist1-HCC , to identify specific mechanisms of metastasis . We believe our model will serve as a useful system for developing new therapies that block cancer metastasis . Our work is the first to suggest that MYC and Twist1 generally cooperate to drive metastasis . By analyzing ChIP-seq data , we found MYC and Twist1 bind to both the promoters of human and mouse Ccl2 and Il13 genes . Our results are consistent with a recent report that Twist1 promotes the invasion of MYCN to enhancer sites thus potentiating its pro-proliferative function ( Zeid et al . , 2018 ) . Thus , our results suggest that MYC and Twist1 may contribute to metastasis through a transcriptional mechanism of inducing a cytokinome that activates macrophages . We note that our observations are consistent with a multitude of reports that innate immunity contributes to metastasis ( Kitamura et al . , 2015; Pollard , 2004 ) and Twist1 contributes to metastasis ( Xu et al . , 2017; Yang et al . , 2004 ) . However , our work is the first to suggest that MYC and Twist1 together affect transcription in a manner that modulates innate immunity , thereby driving metastasis . We propose as a possible general explanation for our findings that overexpression of MYC and TWIST1 in a tumor is activating an embryonic program of innate immune cell activation and cellular invasion . MYC and TWIST1 have been reported to cooperate during embryogenesis ( Bellmeyer et al . , 2003; Rodrigues et al . , 2008 ) . The two transcription factors also have been shown to transcriptionally modulate inflammation during embryogenesis ( Hurlin , 2013; Rodrigues et al . , 2008; Spicer et al . , 1996 ) . These microenvironment changes are required to enable mesodermal cells to migrate to their destination ( Šošić et al . , 2003 ) . Both MYC and TWIST1 are overexpressed in multiple human cancers , suggesting there is a common embryonic transcriptional program they regulate in the embryonic microenvironment which is hijacked by cancer cells ( Hendrix et al . , 2007 ) . Hence , MYC and TWIST1 overexpression in cancer may be eliciting tumor invasion by activating embryonic programs that otherwise physiologically enable mesodermal migration . We identified that MYC and Twist1 generally elicit a cytokinome that includes Ccl2 and Il13 that we show are necessary and sufficient to drive metastasis . Our work is consistent with a prior report that Twist1 transcriptionally induces Ccl2 in breast cancer cell lines that leads to macrophage recruitment ( Low-Marchelli et al . , 2013 ) . However , our study further demonstrates Ccl2 induced macrophage recruitment alone was not sufficient to cause metastasis in vivo and Il13 induced macrophage polarization plays an essential and complementary role in promoting angiogenesis and metastasis . Further , our results suggest that Ccl2 and Il13 alone can enable a non-metastatic cancer to become metastatic , cell non-autonomously . In this regard , it is notable that there are anecdotal reports suggesting that metastasis can occur during circumstances that would promote inflammation such as surgery ( Tohme et al . , 2017 ) or during infection ( Smith and Kang , 2013 ) . Our findings have general relevance to human cancer . First , in 9502 patients with 33 different types of human cancer , the subset of tumors with MYC and TWIST1 predicts poor survival , CCL2/IL13 expression and TAM infiltration . Second , in human patients with HCC , we found that CCL2 , and IL13 , were elevated in the plasma of patients with HCC and IL13 levels predicted invasive and aggressive HCC . Stratifying patients based on cytokine expression may help direct therapy . Third , the combined inhibition of CCL2 and IL13 profoundly impeded metastasis in our in vivo experimental model of liver cancer . Inhibition of CCL2 alone has not been effective in clinical trials for solid tumors ( Brana et al . , 2015; Lim et al . , 2016; Pienta et al . , 2013 ) . We suggest that personalized therapy that combines the inhibition of CCL2 and IL13 is more likely to be effective . Our work identifies two transcription factors , MYC and TWIST1 , are key drivers of metastasis . Together , they elicit a cytokinome , that includes CCL2 and IL13 , enabling crosstalk between cancer cells and host macrophages that drives tumor progression . Mouse Twist1 cDNA was PCR cloned into the bidirectional tetO7 vector S2f-IMCg at EcoRI and NotI sites , replacing the eGFP ORF . The resultant construct , Twist1-tetO7-luc ( Tran et al . , 2012 ) , was sequenced , digested with KpnI and XmnI , and used for injection of FVB/N pronuclei by the Stanford Transgenic Facility . Founders were screened by genotyping using PCR . Founders were mated to LAP-tTA mice , and BLI was used to additionally screen for functional Twist1-tetO7-luc founders , subsequently termed LAP-tTA/TRE-Twist1/Luc . The LAP-tTA , and TetO-MYC transgenic lines have been described previously ( Felsher and Bishop , 1999; Kistner et al . , 1996; Shachaf et al . , 2004 ) . LAP-tTA/TRE-Twist1/Luc mice were mated to LAP-tTA/TRE-MYC mice , and progeny were screened by PCR . The final background of the mouse was FVB/N . Doxycycline ( Dox- Sigma ) was administered in the drinking water weekly at 0 . 1 mg/mL during mating and continuing until mice reached 6 weeks of age . Animals were euthanized upon disease morbidity as assessed by tumor burden . Macrometastases were assessed upon necropsy and tissues were collected and stored for further analysis . All procedures were performed in accordance with APLAC protocols and animals were housed in a pathogen-free environment . in vivo bioluminescent imaging ( BLI ) was utilized to confirm oncogene activation in transgenic mice beginning one week before , and continuing each week following , Dox removal . BLI was performed on an IVIS Spectrum ( Caliper Life Sciences , Hopkinton , MA ) . Briefly , mice were injected i . p . with the substrate D-Luciferin ( 150 mg/kg ) and then anesthetized with 2 . 5% isoflurane delivered by the Xenogen XGI-8 5-port Gas Anesthesia System . Animals were then placed into the IVIS Spectrum , and Living Image Software was used to collect , archive , and analyze photon fluxes and transform them into pseudocolor images . MRI scans were performed using a 7T small animal MRI scanner ( Bruker Inc , Billerica , MA , Stanford Small Animal Imaging Facility , CA ) equipped with a 40 mm Millipede RF coil ( ExtendMR LLC , Milpitas , CA ) . Under anesthesia by inhalation of 1–3% isoflurane mixed in with medical-grade oxygen via nose-cone , and acquisitions were gated using the respiratory triggering . For tumor detection , a respiration triggered T2-weighted 3D turbo spin echo sequence was used ( TR/TE 3000/205 ms , voxel size ( 0 . 22 mm3 ) . The isotropic voxel size of 0 . 22 mm in all directions provides a high in plane and across plane resolution . Thereby , the location of one tumor could be defined in all three orientations using specific landmarks , such as major vessels or other tumors . T2-weighted anatomical imaging was performed approximately once weekly . Anatomical and parametric images were analyzed and tumor volumes were measured using Osirix image processing software ( Osirix , UCLA , and Los Angeles , CA ) . Conditional HCC cell lines were derived from LAP-tTA and TetO-MYC or -MYC/Twist1 mice . Cells were grown in DMEM ( Invitrogen ) , supplemented with 10% FBS ( Invitrogen ) , and cultured at 37°C in a humidified incubator with 5% CO2 . Cell lines were confirmed to be negative for Mycoplasma contamination . An orthotopic mouse model was established by transplanting mouse MYC- or MYC/Twist1-HCC tumors ( 1 mm3 ) under the liver capsule of NOD/Scid/Gamma ( NSG ) recipient mice . Bioluminescent ( BLI ) and MRI scans monitoring are used to monitor tumor engraftment and growth . Mice are euthanized once predetermined specific endpoints are met or based on morbidity whichever occurs first . An intravenous transplantation mouse model was established by tail vein intravenous injection of 500 , 000 MYC- or MYC/Twist1-HCC cells of NOD/Scid/Gamma ( NSG ) recipient mice . Bioluminescent ( BLI ) monitoring is used to monitor tumor metastasis in lungs . Mice are euthanized once predetermined specific endpoints are met or based on morbidity whichever occurs first . RNA sequencing of MYC-HCC and MYC/Twist1-HCC was performed at the Beijing Genomics Institute ( BGI ) using their BGIseq 500 platform single end 150 bp , 20 million reads per sample . Genes expression level is quantified by a software package called RSEM . We counted the number of identified expressed genes and calculated its proportion to total gene number in database for each sample RNA sequencing data are deposited in Gene Expression Omnibus ( GEO ) . DEseq software was used to perform differential expression analysis . Ingenuity Pathway Analysis ( IPA , Qiagen ) was used to perform functional pathway analysis . Principal-component analysis ( PCA; Qlucore Omics Explorer v . 2 . 2 ) was used to generate a visually interpretable overview of the transcriptional profile of MYC-HCC and MYC/Twist1 HCC . Qlucore Omics Explorer was used to structure data to verify if tumor subgroups could be identified . Variance filtering was used to reduce the noise , and the projection score to set the filtering threshold . The ‘mean = 0 , var = 1’ setting was used to scale the data . PCA was used to visualize the data set in a three-dimensional space , after filtering out variables with low overall variance to reduce the impact of noise , and centering and scaling the remaining variables to zero mean and unit variance . The projection score was used to determine the optimal filtering threshold , retaining N variables ( Soneson and Fontes , 2011 ) . For antibody treatment , mice were injected i . p . with isotype control IgG or anti-Ccl2 ( BioXcell ) , -Il13 and –Il4 ( Genentech ) antibody ( 10 mg/kg body weight three times per week ) . For recombinant cytokines treatment , mice were injected i . p . three times per week with PBS or Ccl2 ( Peprotech , 500 ng/mouse ) , Il4 and IL4 ( Peprotech , 250 ng/mouse ) . For clodronate liposomes ( CL ) treatment , CL or control liposomes were administered i . p . at 6 . 5 μl/g body weight 3 times per week to NSGs mice . CL and control treatments were administered to NSG mice previously injected i . v . with 0 . 5 × 106 MYC/Twist1-HCC cells or orthotopically transplanted with MYC/Twist1-HCC tumors . Experimental and control mice were killed 4 weeks after tumor were transplanted . Primary tumors and lung metastases were collected for H and E staining and IHC . Paraffin embedded tumor sections were deparaffinized by successive incubations in xylene , graded washes in ethanol , and deionized water . Epitope unmasking was performed by steaming in DAKO antigen retrieval solution for 45 min . Paraffin embedded sections were immunostained with MYC ( 1:150 , Epitomics ) , or cleaved caspase 3 ( 1:100 , Cell Signaling technology ) , phospho histone 3 ( 1:200 , Cell Signaling Technology ) , F4/80 ( 1:50 , ThermoFisher ) , Cd4 ( 1:1000 , Abcam ) , Neutrophil ( 1:100 , Abcam ) , overnight at 4°C . The tissue was washed with PBS and incubated with biotinylated anti-rabbit , anti-rat or anti-mouse for 30 min at room temperature ( 1:300 Vectastain ABC kit , Vector Labs ) . Sections were developed using 3 , 3'- Diaminobenzidine ( DAB , Vector Labs ) , counterstained with hematoxylin , and mounted with Permount . Images were obtained on a Philips Ultrafast Scanner . All IF and IHC experiments were conducted using at least three biological replicates per group . Images are analyzed in Icy ( BioImage Analysis Unit , Paris , http://icy . bioimageanalysis . com ) . Areas of immunopositivity are selected for positive values in the Color Picker threshold tool in the support vector machine ( SVM ) tab . Unstained nuclei and cytoplasm areas are chosen for negative values . Default values are used for the kernel . Immunopositive areas are selected as regions of interest ( ROIs ) . Subsequently , ROIs are separated . Those with interior size <68 pixels are eliminated from the ROI table as they correspond to small specks of non-specific immunopositivity , whereas larger areas corresponded closely to distinct positive cells . Counts were obtained from the ROI tables . 1 × 106 MYC- or MYC/Twist1-HCC cells per well were seeded in 6-well plate and cultured overnight in culture medium . Thereafter , a scratch ( wound ) was introduced in the confluent cell layer using a yellow tip . Cells were washed three times to remove detached cells . Cells were then incubated with supernatant of primary macrophages harvested from MYC- or MYC/Twist1-HCC primary tumors for 24 hr . Pictures of a defined wound spot were made with a Leica DM16000 microscope at t = 0 , 24 , 48 and 72 hr . The area of the wound in the microscopic pictures was measured using Image J software ( National Institutes of Health , MD ) . The percentage wound healing after 72 hr was calculated in relative to the total wound area at t = 0 hr of the same wound spot . To isolate tumor associated macrophages from liver tumors , we first cut the tumor into small pieces , digested with collagenase and incubated in a culture dish at 37°C for 1–2 hr . We removed and discarded all detached cells and/or dead cells at 2 hr . Next , we maintained the remaining cells in culture for 24 hr and then removed any adherent tumor cells by fast trypsinization . Macrophages are highly adherent and are the main population that remains adherent beyond this step . We harvested the macrophages and confirmed their phenotype and polarization states by qPCR ( Figure 2f ) . Cell migration was assessed using a 12 well transwell chamber with 8 μm filter inserts ( Corning ) . Raw macrophages ( 264 . 7 macrophage cell line ( ATCC ) ) were seeded in the upper chamber and MYC/Twist1 cells were seeded in the lower chamber . Neutralizing antibody to Ccl2 , Ccl5 , Ccl7 , Cxcl1 , Il13 or Il4 were added to the lower chamber in triplicates . After 16 hr the migrated cells were fixed in 4% paraformaldehyde ( PFA ) and 100% methanol . The non-migrated cells were gently removed with a swab . Cells in the lower surface of the membrane were stained with 0 . 5% crystal violet for 20 mins . The membranes were imaged and number of macrophages in 10 random fields were counted . The experiment was performed in triplicates . The 24-well plates were precoated with 10 × 103 mouse Raw 264 . 7 macrophages ( ATCC ) resuspended in 200 ul of Matrigel growth factor reduced ( Corning ) for 30 min at 37°C . Then , 80 × 103 MYC or MYC/Twist1-HCC cells resuspended in 200 ul of Matrigel and directly seeded onto the 24-well plate precoated with matrigel+macrophage mixture . The cells were incubated at 37°C for up to 1 week to allow the spheroids to form . Recombinant Ccl2 ( 50 ng/ml ) or Il13 , IL4 ( 25 ng/ml each ) and anti-Ccl2 antibody ( 30 ug/ml ) , or anti–Il13 antibody , anti-Il4 antibody ( 20 μg/ml each ) were added directly to the coculture and refreshed every 48 hr . Spheroids were fixed with 10% PFA overnight , paraffin embedded and sectioned ( 4–5 μm ) as previously described ( Ootani et al . , 2009 ) . Sections were deparaffinized and stained with H and E for the initial histology analysis . For further immunohistochemistry analysis , we used F4/80 antibody as described above . All assays were performed at least 3 times . RNA was isolated using RNeasy plus mini kit according to the manufacturer’s instructions ( Qiagen ) . cDNA was synthesized using SuperScript III ( ThermoFisher ) . qPCR was performed using specific primers ( Key Resources ) and SYBR Green ( Roche ) in an Applied Biosystems Real Time PCR System ( Life Technologies ) . Data were normalized to UBC . A minimum of 3 biological and three technical replicates were used for all qPCR experiments . This assay was performed in the Human Immune Monitoring Center at Stanford University . Human 62-plex or Mouse 38 plex kits were purchased from eBiosciences/Affymetrix and used according to the manufacturer’s recommendations with modifications as described below . Briefly: Beads were added to a 96 well plate and washed in a Biotek ELx405 washer . Samples were added to the plate containing the mixed antibody-linked beads and incubated at room temperature for 1 hr followed by overnight incubation at 4°C with shaking . Cold and Room temperature incubation steps were performed on an orbital shaker at 500–600 rpm . Following the overnight incubation plates were washed in a Biotek ELx405 washer and then biotinylated detection antibody added for 75 min at room temperature with shaking . Plate was washed as above and streptavidin-PE was added . After incubation for 30 min at room temperature wash was performed as above and reading buffer was added to the wells . Each sample was measured in duplicate . Plates were read using a Luminex 200 instrument with a lower bound of 50 beads per sample per cytokine . Custom assay Control beads by Radix Biosolutions are added to all wells . The CIBERSORT gene expression deconvolution package was used to estimate the immune cell composition in the MYC- and MYC/Twist1-HCC . The LM22 signature was used as the immune cell gene signature . We modified it for studying mouse immune subsets by carefully converting the genes in the signature to their respective mouse orthologs . The settings for the run were: 1000 permutations with quantile normalisation disabled . The student T-test was used to infer the statistical significance of the predicted immune cell populations where p<0 . 05 was considered significant . The pan cancer RNAseq data was downloaded from the GDC data portal https://portal . gdc . cancer . gov/ on Oct 15 , 2018 . Spearman test was used for correlation analysis . Kaplan Meier analysis was performed for survival analysis . K means clustering was used to stratify patients into two groups based on MYC and TWIST1 expression . This study was approved by the institutional review board ( IRB ) of Stanford University ( IRB Number: 28374 ) , and all patients provided informed consent before being enrolled in this study . We prospectively collected blood from patients with HCC or cirrhosis alone . Blood samples were obtained at the time of diagnosis of HCC . Plasma was separated , aliquoted and stored at −80C . Luminex assay to measure cytokine levels was performed as mentioned above . Clinical data was gathered from their medical records . Differences between groups were analyzed using Student’s t-test or one-way analysis of variance ( ANOVA ) . Chi square test was used to compare categorical variables . Kaplan Meier analysis with Log Rank test was performed for survival analysis . A P value of less than 0 . 05 was considered to be significant and is indicated by one asterisk ( * ) , a P value of less than 0 . 01 is indicated by two asterisks ( ) , a P value of less than 0 . 001 is indicated by three asterisks ( * ) , and a P value of less than 0 . 0001 is indicated by four asterisks ( **** ) . All graphs are presented as the mean + /- SEM . Analyses were performed with Prism , version 5 ( GraphPad Software , San Diego , CA ) .	Cancer develops when cells in the body gain mutations that allow them to grow and divide rapidly and uncontrollably . As the disease progresses these cancer cells develop the ability to spread around the body . This process of spreading , called metastasis , is responsible for most cancer-related deaths in humans , but no current treatments target it . Mutations that increase the levels of two proteins known as MYC and TWIST1 in cells cause many human cancers . In healthy adult cells , normal levels of MYC and TWIST1 act as key regulators that switch thousands of genes on or off . TWIST1 is known to control the movement and spread of cells in the embryo . However , it is not known how MYC and TWIST1 work together to promote the metastasis of cancer cells . To address this question , Dhanasekaran , Baylot et al . used mice to investigate the roles of MYC and TWIST1 in the metastasis of cancer cells . The experiments showed that these two proteins work together to reprogram mouse cancer cells to release signal molecules known as cytokines . These molecules convert immune cells known as macrophages to a tumor-friendly state that allows cancers cells to spread around the body . Inhibiting two cytokines known as CCL2 and IL13 prevented the cancer cells from moving . Further experiments analyzed tumor samples from around 10 , 000 human patients with 33 different cancers . This revealed that patients that had higher levels of MYC and TWIST1 proteins in their tumors also had increased levels of CCL2 and IL13 , more activated macrophages and were less likely to recover from their cancer . The findings of Dhanasekaran , Baylot et al . suggest that MYC and TWIST1 may instigate metastasis in many human cancers , and therapies targeting specific cytokines may prevent these cancers from spreading around the body . Furthermore , screening blood for the levels of cytokines may help to identify the cancer patients who would benefit from such therapies .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cancer", "biology" ]	2020	MYC and Twist1 cooperate to drive metastasis by eliciting crosstalk between cancer and innate immunity
Plants produce many different specialized ( secondary ) metabolites that function in solving ecological challenges; few are known to function in growth or other primary processes . 17-Hydroxygeranylinalool diterpene glycosides ( DTGs ) are abundant herbivory-induced , structurally diverse and commonly malonylated defense metabolites in Nicotiana attenuata plants . By identifying and silencing a malonyltransferase , NaMaT1 , involved in DTG malonylation , we found that DTG malonylation percentages are normally remarkably uniform , but when disrupted , result in DTG-dependent reduced floral style lengths , which in turn result from reduced stylar cell sizes , IAA contents , and YUC activity; phenotypes that could be restored by IAA supplementation or by silencing the DTG pathway . Moreover , the Nicotiana genus-specific JA-deficient short-style phenotype also results from alterations in DTG malonylation patterns . Decorations of plant specialized metabolites can be tuned to remarkably uniform levels , and this regulation plays a central but poorly understood role in controlling the development of specific plant parts , such as floral styles . Malonylation is a ubiquitous modification of proteins and specialized metabolites . In plant specialized metabolism , the malonyl group is largely transferred from malonyl-coenzyme A ( Malonyl-CoA ) to the C’6 of the glycosyl moiety of glucoconjugates ( Taguchi et al . , 2005 ) . The malonyl residue is thought to confer structural diversity , stability , and solubility to the decorated metabolites and provide a means of detoxifying xenobiotics ( Suzuki et al . , 2002; Taguchi et al . , 2005; Koirala et al . , 2014; Suzuki et al . , 2004 ) . Malonylation of anthocyanins enhances pigment stability and color intensity at the pH of the intracellular milieus ( Suzuki et al . , 2002 ) . In addition , malonylated anthocyanins are preferentially transported to vacuoles in Arabidopsis thaliana ( Zhao et al . , 2011 ) . The malonylation of glycosides is catalyzed by malonyltransferases , members of the biochemically versatile BAHD acyltransferase family ( D'Auria , 2006 ) . Although tens of malonyltransferases have been identified and functionally characterized in vitro ( Luo et al . , 2007; Manjasetty et al . , 2012; Bontpart et al . , 2015 ) , their in vivo functions are largely unknown . 17-hydroxygeranyllinalool diterpene glycosides ( DTGs ) are abundant ( mg/g FW ) secondary metabolites in green tissues of many solanaceous plants , including tobacco ( Nicotiana spp . ) , pepper ( Capsicum annuum ) and wolfberry ( Lycium chinense ) ( Jassbi et al . , 2006; Lee et al . , 2008; Heiling et al . , 2010 ) . DTGs consist of a 17-hydroxygeranyllinalool aglycone that is decorated at the C3 and C17 hydroxyl positions by glucose , which in turn is modified by glucose , rhamnose and malonyl moieties , in various combinations ( Figure 1A and Figure 1—figure supplement 1 ) . So far , 46 different DTGs ( 21 chemical formulas and several structural isomers ) have been characterized in Nicotiana attenuata , an ecological model plant with a rich portfolio of specialized metabolites; however , the glycoside Lyciumoside IV and its malonylated products , Nicotianoside I and Nicotianoside II , constitute more than 80% of the DTG pool ( Poreddy et al . , 2015 ) . In N . attenuata , DTGs function in resistance against the specialist herbivore , tobacco hornworm ( Manduca sexta ) ( Lou and Baldwin , 2003; Jassbi et al . , 2008; Heiling et al . , 2010 ) . The malonylated DTGs are particularly strongly induced by M . sexta feeding and jasmonate signaling ( Heiling et al . , 2010 ) . The malonyl moieties of DTGs are lost from the DTGs soon after their ingestion by M . sexta larvae due to the alkaline environment of M . sexta oral secretions and midgut ( Poreddy et al . , 2015 ) . This observation rules out a central role for malonylation of DTGs in antiherbivore defense , and suggests that other arenas need to be explored for potential functions mediated by this malonylation . Specialized metabolites are primarily thought to function in mediating an organism’s ecological interactions , to help optimize Darwinian fitness . Many are produced in response to particular ecological interactions , such as those that are induced in response to specific attackers . However , many secondary metabolites have effects on growth and development which are more specific than those that might result from resource trade-offs between growth and putative defense metabolites ( Züst and Agrawal , 2017 ) . For example , the insect feeding-induced glucosinolate breakdown product , indole-3-carbinol , arrests growth by interacting with the auxin receptor Transport Inhibitor Response ( TIR1 ) as an auxin antagonist ( Katz et al . , 2015 ) . Arabidopsis glucosinolates are also thought to modulate plant biomass , flowering time and the circadian clock , and inhibit root growth ( Kerwin et al . , 2011; Jensen et al . , 2015; Francisco et al . , 2016; Malinovsky et al . , 2017 ) . Hyper-accumulation of flavonoids is associated with stunted growth and developmental abnormalities , which are assumed to result from effects on auxin transport ( Franke et al . , 2002; Bonawitz et al . , 2014; Steenackers et al . , 2017 ) . With the exception of the glucosinolates and flavonoids , little is known about these potential ‘primary’ roles for other branches of specialized ( also known as ‘secondary’ ) metabolism . The jasmonate signaling pathway is undoubtedly among the most important in plant defense responses to herbivory , regulating a large portion of herbivory-responsive specialized metabolism , including DTGs in N . attenuata ( Kessler et al . , 2004; Kallenbach et al . , 2012; Li et al . , 2017; Li et al . , 2018 ) . However , jasmonates also regulate root growth , anther dehiscence , male fertility , fruit ripening and senescence ( Staswick et al . , 1992; Xie et al . , 1998; Wasternack et al . , 2013; Stitz et al . , 2014 ) . Jasmonate signaling-deficient genotypes of N . attenuata , including RNAi lines targeting ALLENE OXIDE CYCLASE ( irAOC ) and CORONATINE INSENSITIVE 1 ( irCOI1 ) , and ectopic expression of Arabidopsis JASMONIC ACID METHYL TRANSFERASE ( JMT ) and simultaneous silencing of METHYL JASMONATE ESTERASE ( MJE ) ( JMT/mje ) which dramatically decreased JA signaling , are reported to have short styles , resulting in reduced fertility ( Stitz et al . , 2014 ) . Transgenic N . tabacum plants deficient in COI1 also display short styles ( Wang et al . , 2014 ) . However , the mechanism of this seemingly Nicotiana genus-specific jasmonate-regulated phenotype is completely obscure . Auxin plays a pivotal role throughout the entire lifespan of a plant , particularly in determination of floral development . It can influence cell division , cell expansion and cell differentiation , and thereby regulate a wide spectrum of developmental processes ( Benjamins and Scheres , 2008 ) . Mutants of AtPIN1 , a polar auxin transporter , developed naked , pin-shaped inflorescences and abnormalities in all flower parts , confirming that auxin signaling contributes to flower development ( Okada , 1991 ) . Consistent with this , mutants of the auxin biosynthesis genes TRYPTOPHAN AMINOTRANSFERASE OF ARABIDOPSIS ( TAA ) and YUCCA display severe defects in floral patterning , and complete sterility ( Cheng et al . , 2006; Stepanova et al . , 2008 ) . Auxin can also regulate gynoecium morphogenesis , including style elongation , through a concentration gradient from the apical to the basal part of the gynoecium ( Nemhauser et al . , 2000 ) . In this study , we discovered that the malonylation percentage of DTGs is remarkably uniform across development , treatments , and tissue types , and identified the gene involved in this malonylation . We ask whether this uniformity is essential for plant development . Silencing this gene caused strikingly and specifically short styles , and this phenomenon vanished when silencing this gene in DTG-deficient plants . We analyzed phytohormone levels , enzymatic activity , and effects of exogenous application , and all evidence indicated that the short style phenotype is caused by the influence of DTG malonylation status on IAA biosynthesis . Finally , we illustrate that the Nicotiana genus-specific JA-deficient style phenotype is caused by disturbing DTG malonylation patterns . Our work demonstrates that abnormal JA signaling could dysregulate DTG malonylation patterns , thereby affecting plant style development via auxin signaling . To easily describe and understand the malonylation of DTGs , N . attenuata DTGs were classified , based on how many malonyl moieties they contained , into four categories: core , monomalonylated , dimalonylated , and trimalonylated ( Figure 1A and Figure 1—figure supplement 1 ) . In addition , we used a formula to calculate DTG malonylation percentage in each sample , based on numbers of malonyl moieties in each compound ( Figure 1D ) . As previously reported , the biosynthesis of DTGs , especially malonylated DTGs , is strongly induced by mimicking M . sexta larval feeding ( Figure 1B , [Lou and Baldwin , 2003; Jassbi et al . , 2008; Heiling et al . , 2010] ) . Remarkably , although all types of malonylated DTGs increased after M . sexta elicitation , there was no difference in malonylation percentage between treatment and control . To further analyze DTG malonylation patterns over plant development , DTGs were analyzed in flower buds of different stages . Total DTGs decreased dramatically over flower development , whereas the malonylation percentage was very stable ( Figure 1C and F ) . By mining previously published metabolite data sets from different plant tissues ( Li et al . , 2016 ) , we found that although different malonylated DTGs are highly variable across different tissues , with coefficients of variation ( CV ) ranging from 64 to 107 ( Figure 1—figure supplement 2B–F ) , the malonylation percentage was more uniform , with a CV of only 11: a significant outlier ( Figure 1—figure supplement 2G ) . Manipulating gene ( s ) controlling the malonylation process is the most straightforward way to disentangle the function of DTG malonylation and its remarkably uniformity . To do this , N . attenuata malonyltransferase ( MaT ) genes that have high similarity with NtMaT1 ( Taguchi et al . , 2005 ) , were used to conduct a phylogenetic analysis with functionally characterized MaTs ( Figure 2A and Supplementary file 1 ) . There are five putative MaTs aligned in the same clade with other MaTs of the Nicotiana genus . Among these MaTs , NIATv7_g22417 and NIATv7_g34586 shared the highest protein sequence identity with NtMaT1: 91 . 2% and 91 . 4% , respectively; followed by NIATv7_g13429 , which shared 80 . 6% protein sequence identity with NtMaT1 ( Figure 2—figure supplement 1 ) . Protein sequence alignment showed that four of the candidate MaTs contained the two conserved BAHD enzyme motifs HXXXDG and DFGWG , but not NIATv7_g21823 , which had only a HXXXDG motif near the protein’s center portion ( Figure 2—figure supplement 1A ) . Notably , two putative MaTs , NIATv7_ g39356 and g21823 , which share the highest identity with NtMaT1 , also contained the flavonoid acyltransferase conserved motif , YFGNC ( Figure 2—figure supplement 1A ) , indicating that these two MaTs are homologues of NtMaT1 . As gene co-expression network analysis is a powerful way to predict functions of unknown genes ( Serin et al . , 2016; Higashi and Saito , 2013 ) , we performed a cluster analysis of N . attenuata MaTs with known DTG biosynthesis genes , using published multi-tissue and -treatment RNA-seq data ( Brockmöller et al . , 2017 ) ( Supplementary file 2 ) . Among the five putative NaMaTs , three were within the same clade as known DTG biosynthesis genes ( Figure 2B ) . To determine whether any of these candidates were able to transfer the malonyl moiety from Malonyl-CoA to DTGs , we purified the most abundant core DTG , Lyciumoside IV ( Poreddy et al . , 2015 ) , from N . attenuata leaves and performed in vitro enzyme activity assays using purified recombinant GST-tagged NaMaTs . Consistent with the expression patterns , the in vitro recombinant enzyme activity show that the same three genes: NaMaT1 , NaMaT2 and NaMaT3 could catalyze malonylation from Lyciumoside IV to its monomalonylated form , Nicotianoside I , together with a comparatively minor production of the dimalonylated form , Nicotianoside II ( Figure 2C ) . To determine whether these three N . attenuata MaTs transcripts responded to M . sexta feeding as would be expected from the induced dynamics of DTGs , transcript abundance data were extracted from previously published RNA-seq data from leaves attacked by M . sexta larvae ( Ling et al . , 2015 ) . NaMaT1 was strongly induced within 5 hr after the onset of M . sexta larval feeding ( 59-fold ) , and then slightly decreased by 9 hr ( Figure 2D ) . In contrast , both NaMaT2 and NaMaT3 transcripts were suppressed by M . sexta feeding , which is opposite to the dynamics of the M . sexta-induced DTG profile ( Figure 2D and Figure 1B ) . To investigate the function of N . attenuata MaTs , virus-induced gene silence ( VIGS ) was used to silence all three candidate genes . The tobacco rattle virus VIGS vector migrates to growing meristems and thus efficiently silences target genes in new tissues in all parts of plants ( Galis et al . , 2013 ) . VIGS of the carotenoid biosynthetic gene N . attenuata phytoene desaturase ( NaPDS ) , which causes photobleaching where the gene is silenced , demonstrated efficient silencing in floral tissues ( Figure 3—figure supplement 1A ) . Possibly because of low transcript levels ( Figure 2D ) , neither NaMaT2 nor NaMaT3 were successfully silenced ( data not shown ) , and so further work focused on NaMaT1 . The transcript abundance of NaMaT1 in VIGS-NaMaT1 plants ( VIGS-MaT1 ) decreased more than 80% compared with VIGS controls ( empty vector , VIGS-EV ) , without affecting the abundance of NaMaT2 or NaMaT3 transcripts ( Figure 3—figure supplement 1B ) . Notably , we also analyzed transcript abundance of two other reported DTG biosynthesis genes , geranylgeranyl diphosphate synthase ( NaGGPPS ) ( Jassbi et al . , 2008 ) and geranyllinalool synthase ( NaGLS ) ( Falara et al . , 2014 ) in VIGS plants , and found that NaGLS transcript abundance significantly increased in VIGS-MaT1 leaves ( Figure 3—figure supplement 1C ) . NaMaT1 VIGS plants show similar overall growth phenotypes as control plants , including the morphology of shoots , leaves , the floral exterior , corolla limb and stamen ( Figure 3—figure supplement 1D ) . However , we observed that VIGS-MaT1 plants rarely produced capsules . This observation was quantified by counting capsule numbers at the end of seed set , which showed that VIGS-MaT1 plants produced on average only one capsule every two plants ( Figure 3—figure supplement 2A ) . In a second experiment , we determined that styles of VIGS-MaT1 plants were extremely short , less than half the length of VIGS-EV styles ( Figure 3A and E ) . To elucidate whether a decrease of cell number or cell length caused the short style phenotype , we visualized the style cells from freshly opening flowers using a histochemical stain specific for callose . The stylar cell length was strongly reduced , whereas cell number was not ( Figure 3B–D ) . In order to clarify whether the short-style phenotype is due to alteration of DTG malonylation , or to other unknown functions of NaMaT1 , VIGS was conducted on DTG-deficient plants , irGGPPS . The irGGPPS stably transformed line is specifically silenced in the expression of one of three GGPPSs in the N . attenuata genome: the enzyme that controls the flux of substrates into the DTG pathway , and irGGPPS produces only ca . 10 – 15% of the DTGs levels of WT plants ( Heiling et al . , 2010 ) . The results show that the short-style phenotype vanished in irGGPPS-background VIGS-MaT1 plants ( Figure 3 ) , although NaMaT1 was silenced to a similar degree in the styles of irGGPPS and EV plants ( Figure 3—figure supplement 3A ) . To determine whether the sterility of VIGS-MaT1 plants was due to the physical separation of stigma and anthers , or additional effects on the function of male or female parts , we conducted hand-pollinations of VIGS plants . Pollen from either VIGS-MaT1 or VIGS-EV plants applied to the VIGS-EV pistil produced normal capsules and similar numbers of seeds in each capsule ( Figure 3—figure supplement 2B and C ) . However , only withered capsules resulted from hand-pollination of VIGS-MaT1 pistils with VIGS-EV pollen , and no capsules resulted from pollination of VIGS-MaT1 pistils with VIGS-MaT1 pollen . The withered capsules produced dramatically fewer seeds than those of VIGS-EV capsules , although the seed germination rate was not significantly affected ( Figure 3—figure supplement 2C and D ) . To determine whether NaMaT1 controls the induced malonylation of DTGs in planta , we measured DTGs in the leaves of VIGS plants following MeJA treatment , which is known to strongly induce DTG biosynthesis and malonylation . Silencing NaMaT1 significantly reduced the malonylation percentage , primarily through an increase in DTGs with low malonylation degree ( core and monomalonylated DTGs ) , in comparison to VIGS-EV controls ( Figure 4A ) . This effect was more pronounced after leaves were treated with MeJA ( Figure 4B ) . In order to elucidate the short-style phenotype , we analyzed the DTG profile in styles and inflorescences , which are hypothesized to affect the early stages of style development ( Smyth , 1990; Yanofsky , 1995 ) . In addition to increasing core and monomalonylated DTGs , dimalonylated and trimalonylated DTGs significantly decreased in inflorescences and styles with stigmas , leading to a dramatic decrease in malonylation percentage in VIGS-MaT1 in comparison to VIGS-EV ( Figure 4C , D ) : all individual core and monomalonyated DTGs were dramatically increased in VIGS-MaT1 styles , whereas trimalonylated DTGs were no longer detectable ( Figure 4—figure supplement 1A ) . Because the irGGPPS background could rescue the VIGS-MaT1 short style phenotype , we analyzed DTG profiles in the inflorescence of VIGS plants in the irGGPPS and EV backgrounds . Similarly , as for VIGS of EV plants , the malonylation percentage in NaMaT1-silenced irGGPPS plants significantly decreased , but not as much as in NaMaT1-silenced EV plants ( Figure 4—figure supplement 1B; 26% reduction for EV and 16% reduction for irGGPPS ) . The intermediate decrease in malonylation percentage for irGGPPS results from both an increase of core and monomalonylated DTGs , and a decrease of dimalonylated and trimalonylated DTGs; overall , DTG levels are much lower in the irGGPPS background ( Figure 4—figure supplement 1B and C ) . Disturbed phytohormone levels frequently result in serious floral phenotypes ( Okada , 1991; Stitz et al . , 2014 ) . We analyzed JA , JA-Ile and auxin levels in leaves , inflorescences and styles ( Figure 4 ) . In both control and MeJA-treated leaves , JAs and IAA in VIGS-MaT1 were similar to those of controls . VIGS of NaMaT1 reduced JA and JA-Ile levels significantly in inflorescences compared with the VIGS-EV , without affecting IAA contents ( Figure 4C ) . In contrast , VIGS of NaMaT1 reduced IAA levels in styles and stigmas by 40% , but had no significant effect on JAs levels ( Figure 4D ) . Because there is a sophisticated tissue-specific regulatory mechanism for auxin biosynthesis and homeostasis ( Ljung et al . , 2001 ) , and to gain insight into why silencing NaMaT1 specifically affected IAA in styles , we compared the IAA levels among different flower tissues and first stem ( S1 ) leaves . Styles and stigmas contained the highest levels of IAA among all the measured tissues , about 6 . 5-fold more than that of the S1 leaf ( Figure 4—figure supplement 2B ) . Additionally , styles and stigmas also contain relatively large amounts of DTGs ( Figure 4—figure supplement 2C ) . To determine whether decreased auxin caused the short style phenotype , we analyzed IAA and its precursor tryptophan ( Trp ) in styles of both EV and irGGPPS plants inoculated with VIGS-MaT1 . While IAA decreased in VIGS-MaT1 of EV plants , Trp levels increased dramatically ( Figure 5A ) . These differences were eliminated when we silenced NaMaT1 in the irGGPPS background ( Figure 5A and B ) , which again indicated that the effect of NaMaT1 on IAA biosynthesis in styles is DTG-dependent . To elucidate the effect of NaMaT1 on IAA biosynthesis , we extracted crude protein from EV and VIGS-MaT1 styles and performed in vitro enzyme activity assays of IAA biosynthesis . Surprisingly , NaTAA1 activity was similar between EV and VIGS-MaT1 . However , the transformation from IPA to IAA , which is thought to be catalyzed by YUCCA , was significantly impaired in VIGS-MaT1 style ( Figure 5C ) . We then analyzed the transcript abundance of YUCCA-like genes in N . attenuata by RNAseq , and found one YUCCA-like gene , YUC-like 2 ( Machado et al . , 2016 ) , to be highly expressed in styles ( Supplementary file 2 ) . The transcript abundance of YUC-like 2 was similar between VIGS-MaT1 and VIGS-EV in both backgrounds of irGGPPS and EV plants ( Figure 3—figure supplement 3B ) . Exogenous application IAA approximately restored the short styles to their normal lengths ( Figure 5D ) , without affecting the malonylation degree of stylar DTGs ( Figure 5E ) , results which are consistent with the hypothesis that the truncated style resulted from decreased IAA biosynthesis . The effect of flavonoids on auxin transport has been well characterized ( Peer and Murphy , 2007 ) , and flavonoids also could be used as substrate by the homolog of NaMaT1 in N . tabacum , NtMaT1 ( Taguchi et al . , 2005 ) . To test whether the effect of VIGS-MaT1 on auxin could be caused by changes in flavonoid metabolism , we measured the major flavonoids in N . attenuata plants . We found that both leaves and styles contained similar levels of kaempferol-3-O-glucoside and kaempferol-3-O- rhamnosyl glucoside between VIGS-MaT1 and VIGS-EV , and only rutin was significantly decreased in VIGS-MaT1 style compared with EV styles ( Figure 5—figure supplement 1A and B ) . Furthermore , the stably transformed line irGGPPS also displays flavonoid levels indistinguishable from EV in both leaves and inflorescences , consistent with previous data showing that irGGPPS plants have similar rutin contents as WT plants in both greenhouse and field studies ( Heiling et al . , 2010 ) . Brassinosteroids and the transcription factor style2 . 1 were reported to promote cell elongation and thereby affect style length in Primula spp . ( primroses ) and in Solanum lycopersicum ( tomato ) , respectively ( Chen et al . , 2007; Huu et al . , 2016 ) . We identified the homolog of the S . lycopersicum style2 . 1 gene in N . attenuata , but both RNAseq and qRT-PCR failed to detect transcripts in N . attenuata styles ( Supplementary file 2 ) . PveCYP734A50 was reported to degrade brassinosteroids and thereby control style length in Primula veris ( Huu et al . , 2016 ) . Through searching the N . attenuata genome database , we found the three closest homologs of PveCYP734A50 . RT-PCR analysis of those three genes in styles showed that the transcript abundance of one gene , NIATv7_g25593 , matched very well with the short style phenotype , being most abundant in styles after VIGS-MaT1 of EV , and having low abundance in styles after VIGS-EV or VIGS-MaT1 of irGGPPS plants ( Figure 5—figure supplement 2B ) . We then designed a specific construct and silenced NIATv7_g25593 using VIGS , but did not observe any effect on style length ( data not shown ) . To test whether brassinosteroids may contribute to the VIGS-MaT1 style phenotype in other ways , we exogenously applied brassinolide to VIGS-MaT1 flower buds . This treatment did not recover the short style phenotype of VIGS-MaT1 ( Figure 5—figure supplement 2D ) . Thus , we can rule out the possibility of those two mechanisms contributing to the VIGS-MaT1 short style phenotype . The stylar phenotype of VIGS-MaT1 plants was strongly reminiscent of the short styles of plants with JA signaling deficiencies , which is only reported in Nicotiana species , as far as we know . Among species for which JA-deficient phenotypes have been reported , only the genus Nicotiana produces DTGs ( Heiling et al . , 2016 ) . Therefore , we hypothesized that the short styles of JA-deficient plants results from disturbed DTG malonylation patterns , similar to those of VIGS-MaT1 plants . To address this hypothesis , we analyzed DTG profiles using a stably transformed N . attenuata irAOC line as a severely jasmonate-deficient model . In both herbivore-damaged and control leaves , malonylation percentages in irAOC were significantly higher than in EV plants ( Figure 6—figure supplement 1A ) . In flower buds , the irAOC malonylation percentage was also significantly higher than in EV ( Figure 6—figure supplement 1B ) . MeJA treatment could partially restore the irAOC short style phenotype ( Figure 6—figure supplement 1C and [Stitz et al . , 2014] ) , and also partially restored the malonylation percentages towards EV levels ( Figure 6—figure supplement 1D ) . In line with leaves and flower buds , the malonylation percentage of irAOC styles was also significantly higher than in EV ( Figure 6A ) . To further test this hypothesis , we compared style lengths in VIGS-MaT1 and VIGS-EV of both EV and irAOC plants . Silencing NaMaT1 in EV or irAOC plants resulted in similarly truncated styles ( Figure 6B and Figure 3—figure supplement 3C ) . The lack of an additive effect suggests that irAOC short styles may result from the same mechanism as is responsible for the short styles of VIGS-MaT1 EV plants . Consistently , the malonylation percentages after VIGS-MaT1 in both EV and irAOC backgrounds were similar , much lower than the VIGS-EV of both EV and irAOC genetic backgrounds ( Figure 6C ) . This reduction is mainly caused by an increase of core and monomalonylated DTGs and a decrease of trimalonylated DTGs ( Figure 4—figure supplement 1D and E ) . Furthermore , MeJA treatment could partially restore irAOC-EV short styles , but not the short styles of VIGS-MaT1 in irAOC or EV plants ( Figure 6D ) . These data revealed that NaMaT1 functions downstream of NaAOC to control style length . Importantly , when we crossed irAOC with irGGPPS to silence both DTG and JA production , the irAOC short-style phenotype was completely restored to normal style lengths of EV or WT plants ( Figure 6E ) . Because the VIGS-MaT1 short style phenotype results from attenuated IAA levels , we measured IAA contents in irAOC styles . IAA levels in irAOC styles were much lower than in EV ( Figure 6F ) . The protein activity responsible for transforming IPA to IAA was marginally decreased in irAOC styles compared to that in EV ( Figure 6G ) , but the transcript abundance of NaYUC-like 2 was similar between VIGS-MaT1 and VIGS-EV in the backgrounds of irAOC and EV ( Figure 3—figure supplement 3D ) . Finally , exogenous applications of IAA partially restored the truncated styles of irAOC plants to WT lengths ( Figure 6H ) . These results were consistent with the hypothesis that the short styles of JA-deficient N . attenuata are also caused by disturbed DTG malonylation patterns . Specialized , or secondary metabolites are usually thought only to mediate plant responses to specific environmental conditions , and not to be directly involved in plant growth , development and reproduction – the providence of primary metabolism . Here , we report that the decoration of specialized metabolites demonstrated remarkable uniformity across development , tissues , and treatments , and found that one malonyltransferase ( MaT ) contributes significantly to this process . Silencing this gene disturbed 17-hydroxygeranyllinalool diterpene glycoside ( DTG ) malonylation patterns , and stunted elongation of flower styles during development and fertility by affecting auxin biosynthesis . Abnormal jasmonate signaling could also disturb DTG malonylation patterns , causing similar stylar developmental defects . Enzyme assays showed that three MaT proteins , NaMaT1-3 , catalyze the first step of DTG malonylation , from Lyciumoside IV to Nicotianoside I , and the second step , from Nicotianoside I to Nicotianoside II , in vitro ( Figure 2C ) . As Lyciumoside IV only has two glucose moieties ( Figure 1—figure supplement 1 ) and the abundances of tri-glucosylated core DTGs are too low to purify from plants , we were unable to test the ability of those NaMaTs to catalyze the third malonylation reaction . However , in vivo silencing of NaMaT1 dramatically decreased dimalonylated and trimalonylated DTGs in inflorescences and styles ( Figure 4 ) , which indicates that NaMaT1 might also control the third step of the malonylation reaction in planta . Notably , in vivo silencing of NaMaT1 changed individual DTG abundances in complex ways . For example , although the total abundance of dimalonylated DTGs decreased significantly in NaMaT1-silenced styles , the abundance of one dimalonylated DTG , Nicotianoside X , did not change significantly , whereas the abundance of two dimalonylated compounds , Nicotianoside XII and Nicotianoside VII , increased ( Figure 4–Figure supplement 1AFigure 4—figure supplement 1A ) . Currently , we cannot rule out the possibility that some of these changes are not directly mediated by NaMaT1 catalytic activity , but result from systemic influence , such as substrate limitation . Future work should test NaMaT1 substrate specificity for all DTGs , when technology can support the purification of low-abundance components in sufficient amounts . Based on our current results , we infer that NaMaT1 is a malonyl-CoA:DTG malonyltransferase , which may catalyze three consecutive malonylation steps , as reported for Dm3MaT2 , which catalyzes consecutive dimalonyl transfers to anthocyanin ( Suzuki et al . , 2004 ) . Notably , two characterized MaTs in the Nicotiana genus , NbMaT1 and NtMaT1 , accept aromatic glycosides as substrate ( Taguchi et al . , 2005; Liu et al . , 2017 ) . Phylogenic and protein identity analysis showed that NbMaT1 and NtMaT1 share the highest sequence similarity with NaMaT2 and NaMaT3 , about 90% , whereas the identity with NaMaT1 is 80% ( Figure 2A and Figure 2—figure supplement 1B ) . Moreover , cross-tissue comprehensive nontargeted metabolomics analyses did not reveal any malonylated flavonoids in N . attenuata ( Li et al . , 2016 ) , although NaMaT1 is highly expressed in M . sexta-infested leaves . Thus , we assume that NaMaT1 mainly functions in DTG malonylation in planta , and NaMaT2 and NaMaT3 could also accept aromatic glycosides as substrates , like NbMaT1 and NtMaT1 . Specialized metabolites are thought to be strongly regulated by environmental cues , and are often thought to function in a plant’s adaptive responses to environmental changes . Here , our data revealed remarkable uniformity of DTG malonylation status , although individual DTGs are plastic . For the regulation of DTG biosynthesis , feed-back and feed-forward regulation are both described by Heiling et al . ( in preparation ) : silencing a rhamnosyltransferase gene inhibits the transcript accumulation of all upstream genes , and silencing NaGGPPS also suppresses transcript accumulation of downstream biosynthetic genes . Here , we show that VIGS of NaMaT1 tended to increase both NaMaT2 and NaMaT3 expression , and significantly increased NaGLS expression ( Figure 3—figure supplement 1B and C ) . Two or more elements oppositely controlling one process is a powerful homeostatic strategy , as is found in JA-Ile conjugation and JA-Ile hydroxylation which maintains JA-Ile homeostasis ( Kang et al . , 2006; Woldemariam et al . , 2012 ) . Here , the induction of NaMaT1 and suppression of NaMaT2 and NaMaT3 in response to M . sexta feeding may represent a similar mechanism to control DTG malonylation percentage during the response to herbivore attack ( Figure 2D ) . Using feed-back and -forward regulation together with differential expression levels of NaMaTs , plants maintain DTG malonylation percentage within a very narrow range . This is likely important because a malonylation percentage alteration as low as 4 . 1% in irAOC is associated with a drastically shortened style and infertility ( Figure 6A ) . Additionally , the very low variation in the malonylation percentages reported here throughout the results indicates that DTG malonylation is precisely regulated . Although long-term , high-dose MeJA treatments of the leaves of VIGS plants ( 3 days using the same amount of MeJA normally applied to glasshouse-grown plants , with leaves typically 3-5x the size of the leaves of VIGS plants ) reduced DTG malonylation percentages ( VIGS-EV , Figure 4A and B ) , the locally transient endogenous JA bursts elicited by M . sexta larval feeding did not alter malonylation percentages ( Figure 1E ) . Importantly , both high ( irAOC ) and low ( VIGS-MaT1 ) malonylation percentages were accompanied by truncated styles . Notably , malonylation percentage decreased in irGGPPS-MaT1 inflorescences but without apparently affecting style length of irGGPPS-MaT1 plants ( Figure 4—figure supplement 1B ) . We think that this is because irGGPPS plants produce much lower levels of DTGs , and thus the total abundance of DTGs in inflorescences may not be enough to affect style development , regardless of their malonylation . Induction of the jasmonate signaling pathway , including via MeJA treatment and M . sexta attack , strongly induced NaMaT1 expression ( Figure 2D ) and DTG malonylation ( Figure 1B ) , indicating that NaMaT1 is responsible for the increase in malonylated DTGs . However , as noted above , high-dose MeJA treatment suppressed malonylation percentages , and silencing JA biosynthesis ( irAOC ) increased malonylation percentage in many tissues ( Figure 6A and Figure 6—figure supplement 1A and B ) . This apparent contradiction may be explained by the possibility that other MaTs also contribute to the DTG malonylation process , as indicated by the fact that M . sexta feeding suppresses NaMaT2 and NaMaT3 ( Figure 2D ) . Although NaMaT1 transcript accumulation decreased more than 80% in VIGS-MaT1 leaves ( Figure 3—figure supplement 1B ) , the malonylation percentage was only reduced by 19 . 6% ( Figure 4B ) , again consistent with the hypotheses that other MaTs catalyze DTG malonylation . Future research could test NaMaT2 and NaMaT3 functions by manipulating their expression with more robust means , such as CRISPR-mediated genome editing . Our experiments shed light on this phenomenon by demonstrating that the DTG biosynthesis specific NaGGPPS is required for the development of the truncated style phenotype ( Figure 3 ) , and this truncated style phenotype results from the low auxin levels found in the styles , which in turn are due to reduced IAA biosynthesis ( Figure 5 ) . We hypothesize that this style-specific effect occurs because the style synthesizes large amounts of DTGs ( Figure 4—figure supplement 2C ) , and VIGS-MaT1 results in a stronger decrease in malonylation percentage in styles than it does in leaves ( Figure 4B ) . In addition , styles contained the highest levels of IAA of all tissues analyzed ( Figure 4—figure supplement 2 ) and are likely the auxin source for at least the gynoecium , as predicted by the apical-basal gradient of auxin theory ( Nemhauser et al . , 2000 ) . Our data from irGGPPS plants with depleted DTGs supports our hypothesis that the truncated style of VIGS-MaT1 is related to DTGs , but the exact molecular events that lead to this phenomenon remain unknown . Comparing individual DTG patterns in VIGS plants in the irGGPPS , irAOC and EV backgrounds showed that changes of individual compounds are complex . As we do not know how plants perceive DTGs , it is very challenging to figure out which individual compound or combination of individual compounds could be perceived by plants and affect YUC activity . The low abundance and structural similarity of many DTGs , along with the instability of malonylation in solution ex vivo , make purification difficult ( Heiling et al . , 2016 ) , and for the malonylated DTGs it is not feasible to obtain sufficient amounts or sufficient stability for the establishment and duration of bioassays . Furthermore , given the huge number of combinations of candidate DTGs , a gain-of-function test using specific DTGs is currently prohibitive . For example , if we were to try one or two DTGs at once , not accounting for isomers and limiting ourselves to the 16 DTG chemical formulae for which we could present relative quantification here , the total number of possible combinations is 2 . 8 × 1025 . This number of course could be restricted based on comparisons of the dynamics of individual compounds to the dynamics in malonylation percentage of styles in different treatment groups ( EV , irAOC , irGGPPS; VIGS-EV vs . VIGS-MaT1 ) , although the possibility that a ratio rather than a single compound is required could still result in many combinations that require testing . The biosynthesis and malonylation of DTGs , which are diterpenoid derivatives , may interact with other compounds from this pathway , especially some phytohormones , like strigolactones , abscisic acid , cytokinins , gibberellins and brassinosteroids ( Cazzonelli and Pogson , 2010 ) . Notably , flavonoids were thus far the only known secondary metabolites that could inhibit auxin transport , resulting in stunted growth in Arabidopsis , Medicago , tomato and apple ( Brown et al . , 2001; Eckardt , 2006; Besseau et al . , 2007; Schijlen et al . , 2007; Dare et al . , 2013 ) , and malonyl-CoA is a required substrate for flavonoid biosynthesis ( Kreuzaler and Hahlbrock , 1975 ) . It is possible that silencing NaMaT1 may increase flavonoid biosynthesis , thereby affecting auxin and causing the short style phenotype . However , flavonoids did not significantly increase in VIGS-MaT1 plants ( Figure 5—figure supplement 1A and B ) , and levels of rutin even decreased . Rutin has been reported not to affect auxin transport ( Jacobs and Rubery , 1988 ) . In summary , the short style phenotype is a malonylation-dependent phenotype that is based specifically on DTGs . Bioassay-driven metabolite extractions from VIGS-MaT1 tissues may help to discover the specific compound ( s ) inhibiting YUC activity . In conclusion , this study reveals the uniformity of specialized metabolite malonylation and the importance of this uniformity: when this uniformity is disturbed by silencing a malonyltransferase gene and its regulator , JA signaling , the cells of the tissue which accumulates the highest levels of DTGs fail to elongate normally , resulting in a stunted style phenotype ( Figure 7 ) . This stunted style phenotype resulting from either JA signaling deficiencies or silencing NaMaT1 , results from the inhibition of auxin biosynthesis . Elucidating how plants achieve this finely tuned DTG malonylation status , and which specific DTG structure is responsible for the inhibition of auxin biosynthesis in styles are exciting goals for future research . At a functional level , by shortening styles by titrating the degree of malonyl DTG decorations , plants may be able to regulate their outcrossing rates in response to environmental factors that influence JA signaling or DTGs accumulation , such as herbivory rates . The 31 st inbred generation of N . attenuata originating from a collection at the DI ranch in southwestern Utah USA was used as the wild-type background for all transformants . Previously described homozygotes of the third transformed generation of irGGPPS ( A-07-230-5 ) ( Heiling et al . , 2010 ) , irAOC ( A-07-457-1 ) ( Kallenbach et al . , 2012 ) , and an empty vector control line ( EV , A-03-009-1 ) ( Schwachtje et al . , 2008 ) were used . Seeds were germinated on a mixture of plant agar with Gamborg’s B5 medium in sterile petri dishes and seedlings were transferred to pots and grown under 19 – 35°C , 16 hr light ( supplemental lighting by Philips Sun-T Agro 400W and 600W sodium lights ) and 60 – 65% relative humidity as previously described ( Krügel et al . , 2002; Saedler and Baldwin , 2004 ) . The VIGS plants were obtained following the procedures described in ( Galis et al . , 2013 ) . Briefly , leaves of young rosette-stage plants were pressure infiltrated with a mixture of Agrobacterium tumefaciens containing pBINTRA and either pTV-MaT1 or pTV00 ( EV control ) . VIGS experiments were repeated at least three times . For methyl jasmonate ( MeJA ) treatments , leaf petioles of rosette-stage plants were treated with 20 µL lanolin paste containing 150 µg MeJA ( Sigma-Aldrich ) , or with 20 µL of pure lanolin as a control . MeJA treatment of flower buds followed a similar protocol except the lanolin paste volume was 10 µL and applied to the pedicel of each bud . For Manduca sexta W + OS elicitations , rosette-stage leaves were wounded with a pattern wheel , and 20 µL of diluted M . sexta regurgitant ( 1:5 in distilled water ) was gently rubbed into the freshly created puncture wounds using a clean gloved finger as previously described ( Schittko et al . , 2001 ) . Three days after the treatment , leaves , excluding the midvein , were harvested for metabolite and RNA extraction . For flower samples , flowers were harvested following standardized developmental stages ( Li et al . , 2017 ) . For floral tissue samples , five to ten opening flowers from one plant were dissected and each tissue was separately pooled to create one biological replicate . Five replicates were used for each tissue type . For style samples used for metabolites , phytohormones and in vitro enzyme activity assays , ten flower buds were harvested one day before anthesis , dissected and styles were pooled as one biological replicate . All style length measurements were conducted on first-day open flowers . Samples were ground in liquid nitrogen and aliquoted to 10 – 100 mg depending on the tissues and their known DTG and flavonoid concentrations ( precise mass was recorded ) . Approximately 100 mg of aliquoted samples were extracted using 1 mL 80% methanol aqueous buffer and analyzed on a micrOTOF-Q II system ( Bruker Daltonics ) as previously described ( Heiling et al . , 2016; Li et al . , 2016 ) . QuantAnalysis ( Bruker Daltonics ) software was used to integrate the DTG peak areas based on each compound’s diagnostic m/z value and retention time as described in Figure 1—figure supplement 1 . Malonyltransferase candidate genes that had high similarity to NtMaT1 were identified from the N . attenuata data hub ( http://nadh . ice . mpg . de/ ) . Fragments with coding regions of candidate malonyltransferases were amplified from cDNA with gene specific primers as described in Supplementary file 3 . Full-length cDNAs of the candidate genes without stop codons were introduced to Gateway destination vector pDESTTM 15 , through entry vector pENTRTM according to the manufacturer’s instructions . The recombinant proteins were expressed in E . coli BL21 ( DE3 ) , extracted and purified using Glutathione-Sepharose 6B ( GE Healthcare ) in accordance with the manufacturer’s instructions . The in vitro enzyme activity assays were conducted as previously described ( Taguchi et al . , 2010 ) . Briefly , 1 µg purified protein was added to the reaction mixture ( 50 µL ) , which was 50 mM potassium phosphate buffer ( pH 8 . 0 ) with 200 µM malonyl-CoA , 5 mM β-mercaptoethanol , and 8 µg DTGs . The mixture was incubated at 30°C for 1 hr , and the reactions were stopped by adding 10 µL 1M HCl . Methanol ( 40 µL ) was then added to the reaction mixture , and this was subsequently used for DTGs quantification by UPLC-Q-TOF as described above . The complete pistils were harvested , fixed ( ethanol: acetic acid , 3:1 ) and stained with aniline blue as previously described ( Mori et al . , 2006 ) . Stained samples were viewed under a confocal laser scanning microscope ( LSM 880 , Zeiss , Jena , Germany ) in channel mode with a 20x objective ( Plan-Apochromat 20x/0 . 8 ) and a 405 nm laser diode for illumination . Excitation , emission and detection windows were set via a 405 nm main beam splitter and the QUASAR detector range between 480 and 550 nm , respectively . The pinhole size was 41 µm , while the Z-stack step size was 1 µm . Tiled Z stacks were acquired to obtain all necessary details . After image acquisition , the scanned tiles were stitched in ZEN ( black 2012 , Zeiss , Inc . ) . Representative images were obtained with maximum intensity projections in ImageJ 1 . 50e . Cell lengths were measured in ImageJ 1 . 50e . For quantitative RT-PCR , RNA was extracted from 30 mg well-ground tissue using TRIzol reagent ( Invitogen ) , and the RNA quantity was confirmed by the 260/280 nm absorbance ratio using NanoDropTM ( ThermoFisher Scientific ) . One µg RNA was used for reverse transcription by First strand cDNA synthesis kit ( ThermoFisher Scientific ) . RT-qPCR was done via Mastermix ( Eurogentec ) SYBR Green reaction in a Stratagene 500 MX3005P Real-time qPCR machine . The primers used for mRNA detection of target genes by RT-PCR are listed in Supplementary file 3 . The amplification specificity of primers was confirmed by single peaks in a dissociation curve following qPCR . The N . attenuata IF5a-2 mRNA was used as internal control . RNA-seq data for gene expression in all N . attenuata tissues was previously published in NCBI with accession number PRJNA317743 ( Brockmöller et al . , 2017 ) . To readily visualize these data in heatmaps , the raw data was transformed by log2 . RNA-seq data for gene expression after M . sexta larval feeding was previously published in NCBI with accession number PRJNA223344 ( Ling et al . , 2015 ) . Jasmonates , IAA and tryptophan were measured as previously described ( Schäfer et al . , 2016 ) . Briefly , aliquots of ca . 100 mg ( precise mass recorded ) frozen powdered samples were extracted with 800 µL extraction buffer containing the internal standards ( 20 ng D6-JA , 20 ng D6-JA-Ile , 3 ng D5-IAA ) , purified by successive HR-X and HR-XC SPE column chromatography ( MACHEREY-NAGEL ) , and finally analyzed on a EVO-Q EliteTM Triple quadrupole-MS ( Bruker Daltonics ) . Prior to SPE purification , 2 µL of the initial extract was diluted into 98 µL aqueous solution containing 255 fmol µL-1 13C9 , 15N1-phenylalanine as an internal standard for the quantification of tryptophan . Approximately 20 mg frozen powdered samples were extracted with 200 µL cold extraction buffer ( 0 . 1M Tris-C1 , pH 7 . 6; 5% polyvinylpolypyrrolidone; 2 mg/mL phenylthiourea; 5 mg/mL diethyldithiocarbamate; 0 . 05 M Na2EDTA ) . The reaction for TAA1 enzyme activity assays was incubated at 55°C for 20 min , and other procedures were as previously described ( Tao et al . , 2008 ) . The YUC enzyme activity assay was conducted as previously described ( Mashiguchi et al . , 2011 ) . Both reactions were stopped by acidification with 5 µL 3M phosphoric acid , then adding 500 ng D5-IAA as an internal standard before the reaction product was extracted three times with an equal volume of ethyl acetate . The supernatant was dried and the pellet was resuspended in 50 µL of methanol . The methanol-solubilized extracts were analyzed by UHPLC-MS ( impact II , Bruker Daltonics ) in negative ESI mode . All ANOVAs were performed in SPSS statistic 17 . 0 ( SPSS Inc , http://www-01 . ibm . com/software/analytics/spss/ ) . The Student’s t-tests were performed in Microsoft Office Excel 2010 . Homogeneity of variance was evaluated in SPSS using Levene’s test , and outliers were assessed by the function of Explor in SPSS with default parameters . The protein sequences were aligned by CLUSTAL W and phylogenetic trees were constructed using the maximum-likelihood method in MEGA6 ( http://www . megasoftware . net/ ) .	Plants produce tens of thousands of molecules called secondary metabolites that are thought to help them cope with threats from their environment , such as attack by insects or ultraviolet radiation from the sun . Wild coyote tobacco plants produce large amounts of a particular class of secondary metabolite known as DTGs . Insects feeding on tobacco plants containing DTGs cause less damage and produce fewer offspring Plants modify many secondary metabolites by attaching tags known as malonyl groups to them . Enzymes called malonyl transferases take a malonyl group from another substrate and attach it to the secondary metabolite . This process can be repeated so that an individual secondary metabolite molecule may have many malonyl groups attached to it . Previous studies have shown that insects feeding on tobacco plants trigger more malonyl groups to be attached to DTGs , but it is not clear what effect this has on the plants . To simulate attack by an insect , Li et al . punctured holes in the leaves of tobacco plants and applied saliva from tobacco hornworms . The experiments show that more DTGs modified with malonyl groups accumulated in these plants compared to untreated plants . However , there was no change in the average number of malonyl groups added to individual DTGs during the modification process . Further experiments show that a malonyl transferase enzyme called NaMaT1 adds malonyl groups to DTGs in coyote tobacco plants . The flowers of plants that produce less of this protein have shorter styles ( a tube structure that guides pollen to the egg cells at the base of the flower ) and are less fertile than flowers in normal plants . These experiments demonstrate that , along with helping plants to defend themselves from herbivores , DTGs regulate how flowers grow and develop . It was generally thought that secondary metabolites do not play important roles in how plants grow when they are not under stress . Indeed , plant breeders frequently select crops that produce lower levels of secondary metabolites in order to increase their nutritional value . Therefore , the findings of Li et al . may help improve the outcomes of crop breeding programs .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology", "plant", "biology" ]	2018	The decoration of specialized metabolites influences stylar development
During embryonic development , cells of the green alga Oophila amblystomatis enter cells of the salamander Ambystoma maculatum forming an endosymbiosis . Here , using de novo dual-RNA seq , we compared the host salamander cells that harbored intracellular algae to those without algae and the algae inside the animal cells to those in the egg capsule . This two-by-two-way analysis revealed that intracellular algae exhibit hallmarks of cellular stress and undergo a striking metabolic shift from oxidative metabolism to fermentation . Culturing experiments with the alga showed that host glutamine may be utilized by the algal endosymbiont as a primary nitrogen source . Transcriptional changes in salamander cells suggest an innate immune response to the alga , with potential attenuation of NF-κB , and metabolic alterations indicative of modulation of insulin sensitivity . In stark contrast to its algal endosymbiont , the salamander cells did not exhibit major stress responses , suggesting that the host cell experience is neutral or beneficial . All vertebrates have a ‘microbiome’ that includes mutualist ecto-symbionts living in close association with , but not within , their cells ( Douglas , 2010 ) . The most substantial vertebrate ecto-symbioses occur in the colon and small intestine and are implicated in physiological processes such as nutrient absorption from undigested complex carbohydrates ( Ley et al . , 2008; Krajmalnik-Brown et al . , 2012 ) . Known endosymbioses in vertebrates , where microbial cells live within the vertebrate cells , are almost exclusively parasitic , causing diseases such as malaria , toxoplasmosis , and chytridomycosis ( Douglas , 2010; Sibley , 2004; Davidson et al . , 2003 ) . Currently , there is only a single exception . The green alga Oophila amblystomatis enters the cells of the salamander Ambystoma maculatum during early development ( Kerney et al . , 2011 ) , and co-culture experiments show that the algae consistently benefit the salamander embryo hosts ( Small et al . , 2014; Graham et al . , 2013; Pinder and Friet , 1994 ) . There is a long history of experimentation on the ectosymbiotic association between O . amblystomatis and A . maculatum: where the alga populates salamander egg capsules that contain developing embryos ( Small et al . , 2014; Gilbert , 1944 ) . In the ectosymbiosis , the alga appears to benefit from nitrogenous waste excreted by the developing embryo while providing periodic oxygen and photosynthate to the microenvironment of the embryo’s egg capsule , aiding salamander development ( Small et al . , 2014; Graham et al . , 2013; Gilbert , 1944 ) . However , the intracellular association between these two organisms was only recently recognized ( Kerney et al . , 2011 ) , 122 years after the first published description of green salamander egg masses ( Orr , 1888 ) . Perhaps the most intensively studied endosymbiosis between photosynthetic microbes and non-photosynthetic animal hosts is the facultative mutualistic interactions between various invertebrate cnidarians ( i . e . corals and sea anemones ) and dinoflagellate endosymbionts ( Davy et al . , 2012 ) . Such interactions provide the host animals the ability to obtain energy through photosynthesis . Cnidarian-dinoflagellate endosymbioses involve a number of physiological changes in both the host and photo-symbiont on a cellular level . The host animal tends to exhibit a tempered immune response to the ingressing cells ( modulated by the host , symbiont , or both ) ( Detournay et al . , 2012 ) , and to express genes necessary for transferring nutrients to the symbiont ( Lehnert et al . , 2014 ) , receiving nutrients from the symbiont ( Lehnert et al . , 2014 ) , and use of the endosymbiont-derived metabolites ( Lehnert et al . , 2014 ) . Less is known about adaptations of the endosymbiotic cells , but they can include modified osmoregulation ( Mayfield and Gates , 2007 ) , export of nutrients to the host cell ( Lin et al . , 2015 ) , and physical changes such as loss of flagella ( Muller-Parker et al . , 2015 ) . In this study , we used a dual RNA-Seq approach on wild-collected A . maculatum salamander embryos and their endosymbiont alga O . amblystomatis to characterize the transcriptomic changes that occur in both organisms during this unique endosymbiosis . We isolated free-swimming algal cells living within the egg capsule ( ‘intracapsular environment’ , triplicate sampling ) , salamander cells that did not contain algae ( N = 50 cells per replicate , quadruplicate sampling ) , and salamander cells containing intracellular algae ( N = 50 cells per replicate , quadruplicate sampling ) from the same individuals . We identified differentially expressed genes in both organisms attributed to the intracellular association . The algal endosymbiont undergoes drastic changes in metabolism , displaying signs of cellular stress , fermentation , and decreased nutrient transport , while the host salamander cell displays a limited innate immune response and changes to nutrient sensing , but does not appear to invoke cell stress responses such as apoptosis or autophagy . Ectosymbiotic , intra-capsular algal cells were isolated from egg capsules with a syringe ( Figure 1a ) . Individual A . maculatum cells were manually separated into groups of 50 cells with or without intracellular algal symbionts ( Figure 1a , b ) . Total RNA was extracted from A . maculatum cells or from intra-capsular algal samples , and converted to cDNA ( Figure 1c ) . A test for contaminating mRNA from A . maculatum lysed during dissociation was shown to be negative ( Figure 1—figure supplement 1 ) A total evidence assembly contained all reads from all samples ( n = 3 intra-capsular algal samples from three different eggs; salamander cells with and without algae from n = 4 individual salamander embryos ) . This was followed by homology and abundance filtering ( Figure 1—figure supplements 2 , 3 and 4 ) , producing 46 , 549 A . maculatum and 6 , 726 O . amblystomatis genes that were used in differential expression analysis . 10 . 7554/eLife . 22054 . 003Figure 1 . Three populations of cells from A . maculatum egg capsules containing stage 39 embryos were collected and prepared for mRNA extraction , cDNA sequencing , and differential expression analysis revealing several hundred significantly differentially expressed genes detected for the salamander and alga . ( a ) Intracapsular algae ( Population 1 ) were removed from intact eggs using a syringe and hypodermic needle ( photo credit: Roger Hangarter ) . Embryos were decapsulated and washed , and the liver diverticulum region ( dashed line ) , containing high concentrations of algae ( red dots ) , was isolated and dissociated into a single cell suspension ( illustration adapted from Harrison , 1969 ) . The dissociated cells were screened for A . maculatum endoderm cells without alga ( black arrowheads ) and endoderm cells with intracellular alga ( green arrowhead ) . Scale bars on microscope images are 20 µm . ( b ) Isolated endoderm cell , and isolated endoderm cell with intracellular alga . Scale bars on microscope images are 20 µm . ( c ) Representative cDNA distribution ( bioanalyzer trace ) from a population of 50 manually isolated A . maculatum endoderm cells . Peaks at 35 bp and 10380 bp are markers . Due to evidence of lysed A . maculatum cells observed in the cell suspension fluid after dissociation of A . maculatum embryos ( debris seen in dissociated A . maculatum microscope images in ( a ) and ( b ) ) , that fluid was tested for the presence of contaminating mRNA . mRNA was not detected in the surrounding fluid , Figure 1—figure supplement 1 . Lower limit abundance thresholds ( Figure 1—figure supplement 2 ) , and correction for low sequencing depth in intracelluar algal samples ( Figure 1—figure supplement 3 ) were implemented to obtain the final gene sets used for differential expression analysis . Depth of sequencing was not biased for A . maculatum cell with and without alga samples ( Figure 1—figure supplement 4 ) . Library preparation GC bias affected the completeness of the algal transcriptome obtained from intracapsular and intracellular O . amblystomatis ( Figure 1—figure supplement 5 ) . ( d and e ) Dotplots of log2 fold change vs . expression level . The blue horizontal lines are plus and minus 4-fold change in expression between samples . The red dots are genes with FDR adjusted p-values<0 . 05 , indicating a significant difference in expression level between conditions . ( d ) Differentially expressed algal transcripts . ( e ) Differentially expressed salamander transcripts . DOI: http://dx . doi . org/10 . 7554/eLife . 22054 . 00310 . 7554/eLife . 22054 . 004Figure 1—source data 1 . Raw counts matrix with counts for all reads mapped to the total evidence assembly ( the assembly of all salamander and algal reads from wild-collected samples ) . The data in this file ( after filtering and normalization ) was used to generate the dotplots in Figure 1D and E , Figure 1—figure supplements 2–4 , and Figure 3 . This is the raw data that was used for differential expression analysis . Rows are genes . Column names are as follows: S2a-S5a are counts for salamander cells without algae . S2b-S5b are counts for salamander cells with intracellular algae ( samples are paired from the same individuals , such that S2a and S2b came from the same salamander ) . A1-A3 are intracapsular algae samples . RK_* are cultured algal samples . DOI: http://dx . doi . org/10 . 7554/eLife . 22054 . 00410 . 7554/eLife . 22054 . 005Figure 1—source data 2 . List of 6 , 726 algal gene IDs used in differential expression analysis . Use to filter raw counts matrix to get final algal gene list . DOI: http://dx . doi . org/10 . 7554/eLife . 22054 . 00510 . 7554/eLife . 22054 . 006Figure 1—source data 3 . List of 46 , 549 salamander gene IDs used in differential expression analysis . Use to filter raw counts matrix to get final salamander gene list . DOI: http://dx . doi . org/10 . 7554/eLife . 22054 . 00610 . 7554/eLife . 22054 . 007Figure 1—figure supplement 1 . A . maculatum cell lysis during embryo dissociation did not contaminate the cell suspension fluid with significant quantities of mRNA . ( a ) Representative cDNA distribution ( bioanalyzer trace ) from a population of 50 manually isolated A . maculatum endoderm cells . ( b ) No cDNA was produced when the fluid the cells were suspended in was tested indicating that the cDNA populations from manually isolated A . maculatum endoderm cells was specific and not contaminated with cDNAs derived from randomly lysed cells . In both ( a ) and ( b ) , the peaks at 35 bp and 10380 bp are markers . DOI: http://dx . doi . org/10 . 7554/eLife . 22054 . 00710 . 7554/eLife . 22054 . 008Figure 1—figure supplement 2 . Determining lower limit FPKM thresholds for inclusion in differential expression analysis . For pairs of experimental conditions ( i . e . n = 4 A . maculatum samples without intracellular algae , and n = 4 A . maculatum samples with intracellular algae ) , gene expression levels were sorted by the mean FPKM value ( expression level ) in one set of samples ( i . e . in ( a ) expression levels of A . maculatum genes from samples with and without intracellular algae were sorted by mean expression per gene for n = 4 A . maculatum samples without intracellular algae ) . Using a sliding window of 100 genes , starting with the 100 most lowly expressed genes of the sorted set , median expression levels of the 100 gene bins were calculated for both experimental conditions . Those binned values were plotted with the expectation that on average , gene expression from one experimental condition should be positively correlated with gene expression from the other experimental condition . Vertical red dashed lines indicate the level of expression along the x-axis ( in the sorted sample , determined by visual inspection of the plots ) where positively correlated expression between the experimental conditions begins . Those values were used as lower limit thresholds in data pre-filtering steps . ( a ) Salamander cells with endosymbionts vs . salamander cells without endosymbionts; sorted by salamander cells without endosymbionts expression levels . ( b ) Salamander cells without endosymbionts vs . salamander cells with endosymbionts; sorted by salamander cells with endosymbionts expression levels . ( c ) Intracellular algae vs . intracapsular algae; sorted by intracapsular algae expression levels . ( d ) Intracapsular algae vs . intracellular algae; sorted by intracellular algae expression levels . DOI: http://dx . doi . org/10 . 7554/eLife . 22054 . 00810 . 7554/eLife . 22054 . 009Figure 1—figure supplement 3 . Determining a threshold for absence calls in intracellular algal data . Intracapsular algae samples had a higher sequencing depth than the intracellular algae . This filtering determined the lower FPKM limit of expression in intracapsular algae for inclusion in differential expression analysis . ( a ) Algal gene expression levels in intracapsular ( red ) and intracellular ( blue ) algae . The vertical dashed lines represent the median expression level of the respective populations . The large blue bar at −5 ln ( FPKM ) is the overrepresented proportion of genes with no expression in intracellular algal samples due to the low depth of sequencing . ( b ) Genes with low levels of intracapsular algal expression are detected in 100% of the intracellular algal samples due to pre-filtering inclusion of genes that were detected in all four intracellular algal samples . However , as the expression level of genes in intracapsular algal samples increases , the proportion of genes detected in intracellular algae decreases sharply with a minimum of 40% . Following this minimum , the proportion of genes detected in intracellular samples increase proportionally with the intracapsular expression . The red dashed vertical line is the FPKM value in intracapsular algae where 95% or more of the intracellular genes are detected . Below this threshold , a gene’s absence in intracellular genes is possibly due to the low sequencing depth , above this threshold , a gene’s absence in intracellular algae is interpreted as potential under-expression . ( c ) The same plot as in ( a ) , after filtering to remove genes absent in intracellular algae with expression levels in intracapsular algae below threshold . ( d ) The same plot as in ( b ) , after the dependence of detection on expression level was removed . DOI: http://dx . doi . org/10 . 7554/eLife . 22054 . 00910 . 7554/eLife . 22054 . 010Figure 1—figure supplement 4 . Determining threshold for absence calls in salamander data . The algal filtering described in Figure 1—figure supplement 3 was not required for salamander transcripts . ( a ) Salamander gene expression levels in salamander cells without algae ( red ) and salamander cells with algal endosymbionts ( blue ) . Data is plotted on a natural log scale . The vertical dashed lines represent the median expression level of the respective populations ( overlapping in this case ) . ( b ) The proportion of salamander mRNA’s detected in alga-containing cells does not depend on the mRNA expression level in salamander cells without algae . Greater than 95% of all genes are detected in salamander cells plus algal samples for all values of expression in salamander cells without alga samples . DOI: http://dx . doi . org/10 . 7554/eLife . 22054 . 01010 . 7554/eLife . 22054 . 011Figure 1—figure supplement 5 . High GC content algal genes were not detected by the combination of SMARTer cDNA synthesis and Nextera-XT library preparation . ( a ) The GC content distribution of algal transcripts generated using TrueSeq library preparation of total RNA , sequenced on the MySeq platform with approximately 30 million 75 bp paired end reads . 79% of eukaryote BUSCOs were detected in this assembly . The median GC content ( green dashed line ) is 62% . ( b ) The GC content distribution from ( a ) , split by library preparation method . Red bars represent algal transcripts found in transcriptomes generated by both library preparation methods ( SMARTer-Netxtera-XT and TruSeq ) . Blue bars represent transcripts found only in the transcriptome assembly from the TrueSeq library preparation method , that are absent from the transcriptome generated using the SMARTer cDNA synthesis-Nextera-XT library preparation method . There is an apparent bias against high GC content algal transcripts in library prepared using the SMARTer cDNA synthesis-Nextera-XT protocol ( Kolgomorov-Smirnov test , p<2 . 2 × 10−16 ) . Both libraries were sequenced to a similar depth of approximately 30 million reads for the alga-only samples in the total-evidence assembly from the SMARTer-cDNA synthesis-Nextera-XT library and 30 million reads for the TrueSeq library from unialgal cultures . Since sequencing depth was equivalent and GC bias is apparent , the data suggests that GC bias in the SMARTer-cDNA synthesis-Nextera-XT library is what accounts for the low number of detected BUSCOs ( 49% ) in the algal transcriptome generated from wild-collected algal samples associated with salamander eggs and cells . ( C . ) The distribution of GC content in A . maculatum transcripts ( gray bars ) is centered around much lower GC content transcripts ( median GC content of 43% ) compared to that of O . amblystomatis ( green bars , median GC content of 62% ) . The A . maculatum assembly contained 88% of eukaryote BUSCOs . Our evidence points to bias against high GC content transcripts in the SMARTer cDNA synthesis and Nextera-XT library prep method , that becomes significant above 60% GC content . Transcripts with GC content of 60% or greater are in the tail of the salamander GC content distribution , but near the median of the algal GC content distribution . This offers an explanation for the BUSCO results , where the salamander transcriptome from the wild-collected samples is comprehensive , while the algal transcriptome from the same samples and library prep methods is missing around 40% of the algal transcriptome . DOI: http://dx . doi . org/10 . 7554/eLife . 22054 . 011 The salamander and algal transcriptomes were tested for completeness using BUSCO ( Benchmarking Universal Single-Copy Orthologs ) analysis ( Simão et al . , 2015 ) . The final filtered algal assembly contained 31% ( 130/429 ) of eukaryote BUSCOs , reduced , due to limitations of sequencing depth in intracellular algal samples , from 47% ( 199/429 ) for algal genes in the total evidence assembly . For comparison , a de-novo transcriptome assembly from O . amblystomatis cultured in replete media , contained 79% ( 336/429 ) of eukaryote BUSCOs . This is comparable to the Chlamydomonas reinhardtii transcriptome , containing 74% ( 316/429 ) of eukaryote BUSCOs . The algal transcriptome generated from the wild collected samples , however , was prepared using a different library preparation protocol ( SMARTer cDNA synthesis followed by Nextera-XT library preparation ) . This was chosen for the low cell numbers of salamander cells with and without endosymbionts . The transcriptome derived from the cultured alga was sequenced from a TrueSeq library preparation . This was chosen due to relatively large quantities of RNA from lab cultured algal strains . The algal transcriptome from the wild collected total evidence assembly ( SMARTer cDNA synthesis and Nextera-XT library preparation ) was found to be missing as much as about 40% of the total algal transcriptome , likely due to GC-content biases introduced during library preparation ( Figure 1—figure supplement 5a and b ) ( Lan et al . , 2015 ) . The incompleteness of the transcriptome did not affect inference of differentially expressed genes from the set of 6 , 726 found in all algal samples . However , the low-cell count library preparation protocol did limit the sensitivity of our algal analysis in that we could not draw inferences from genes that were not present in the wild-collected algal libraries . The final filtered A . maculatum transcriptome assembly contained 88% ( 375/429 ) of eukaryote BUSCOs and 69% ( 2 , 078/3 , 023 ) of vertebrate BUSCOs . For comparison , the A . mexicanum transcriptome assembly ( Smith et al . , 2005; Voss et al . , 2015; Baddar et al . , 2015; Voss , 2016 ) contained 89% ( 381/429 ) of eukaryote BUSCOs and 65% ( 1 , 953/3 , 023 ) of vertebrate BUSCOs . The SMARTer cDNA synthesis followed by Nextera-XT library prep did not exclude expected salamander transcripts . This is likely due to the low GC content of these RNAs , with a median 43% GC content compared to the algal transcript’s median GC content of 62% ( Figure 1—figure supplement 5c ) . Among the 6 , 726 O . amblystomatis genes available for DE analysis , 277 were significantly differentially expressed with a false discovery rate ( FDR ) adjusted p-value ( Benjamini and Hochberg , 1995 ) of less than 0 . 05 ( Figure 1d ) between intracellular and intracapsular algae . There were 111 genes with higher expression in intracellular algae and 166 genes with lower expression in intracellular algae . Of those , 56 ( 50% ) of the over expressed genes and 91 ( 55% ) of the under expressed genes were assigned putative functions based on homology to known proteins . The genes were grouped into eighteen broad functional categories ( Table 1 ) revealing the response of the alga to the intracellular environment . Intracellular algae exhibit a stress response with over-expression of three heat shock proteins and other indicators of oxidative and osmotic stress , and large metabolic changes compared to freely swimming intracapsular algal cells . The complete list of annotated , differentially expressed alga genes can be found in the file Table 1—source data 1 . 10 . 7554/eLife . 22054 . 012Table 1 . Functional classification of the green alga O . amblystomatis genes that are differentially expressed during intracellular association with the salamander host . DOI: http://dx . doi . org/10 . 7554/eLife . 22054 . 01210 . 7554/eLife . 22054 . 013Table 1—source data 1 . Differentially expressed algal transcripts , annotations , functional groupings , and expression statistics . DOI: http://dx . doi . org/10 . 7554/eLife . 22054 . 013Functional Category# genes#up#downNo Homology904347Conserved Gene with Unknown Function371126Stress Response321418Fermentation17134Electron Transport-Mitochondrial606Photosynthesis1376Ribosomal Proteins11110Nitrogen Transport505Phosphate Transport202Other Transport1266Sulfur Metabolism550Lipid Metabolism752Other Metabolism909Flagellar Apparatus413Signaling514Transposable Element413Glycosylation202Other13211Totals277111166 In A . maculatum , 46 , 549 genes were analyzed for differential expression . A total of 300 genes were identified as differentially expressed with an FDR adjusted p-value less than 0 . 05 ( Figure 1e ) . There were 134 genes with higher expression in salamander cells containing intracellular algae and 166 genes with lower expression in those cells . Of those , 74 ( 55% ) of over expressed genes and 71 ( 43% ) of the under expressed genes were assigned putative functional annotations . The genes were grouped into twelve broad functional categories ( Table 2 ) reflecting the response of A . maculatum cells to the intracellular algae . Transposable elements comprise the largest category of annotated differentially expressed genes ( 18% of over- and 27% of the under-expressed ) . Other functional responses include an immune response to the intracellular alga , modulation of the host cell’s nutrient sensing , and differential expression of genes related to cell survival and interactions with other cells , including cell-cell adhesion and motility . The complete list of annotated , differentially expressed algal genes can be found in the file Table 2—source data 1 . 10 . 7554/eLife . 22054 . 014Table 2 . Functional classification of the salamander , A . maculatum , genes that are differentially expressed when associated with intracellular alga . DOI: http://dx . doi . org/10 . 7554/eLife . 22054 . 01410 . 7554/eLife . 22054 . 015Table 2—source data 1 . Differentially expressed salamander transcripts , annotations , functional groupings , and expression statistics . DOI: http://dx . doi . org/10 . 7554/eLife . 22054 . 015Functional Category# genes#up#downNo Homology1556095Transposable Element692445Immune Response12111Nutrient Sensing1477Metabolism862Adhesion/ECM743Proliferation/Survival/ Apoptosis770Motility532Transcriptional Regulation624Cell-Type Specific330DNA Repair330Others1147Totals300134166 Cultures of the symbiotic alga in AF6 media , allowed in vitro testing of algal inorganic phosphate and nitrogen transporter regulation , in response to availability of relevant nutrient sources . The high affinity phosphate transporter PHT1-2 , was regulated by extracellular inorganic phosphate concentration in cultured O . amblystomatis ( Figure 2a ) . The average qPCR expression difference between high ( 100 µM and above ) and low ( 10 µM and below ) phosphate concentrations was 32-fold ( p=4 . 4 × 10−15 ) , which agrees with both the RNA-seq data ( 25 fold lower expression in the endosymbiotic alga ) , and estimates of phosphate concentrations in vernal pool water ( low micromolar ) ( Brodman et al . , 2003; Carrino-Kyker and Swanson , 2007 ) compared to inside amphibian cells ( low millimolar ) ( Horowitz et al . , 1979; Burt et al . , 1976 ) . A second phosphate transporter , a chloroplast localized sodium dependent phosphate transport protein 1 ( ANTR1 ) , was not regulated by extracellular phosphate levels ( Figure 2b ) . Its low expression level in the endosymbiotic alga is therefore not likely to be related directly to an increased phosphate level of the host cytoplasm . 10 . 7554/eLife . 22054 . 016Figure 2 . An algal phosphate transporter is modulated by inorganic phosphate levels , while nitrogen source transporters are modulated by an organic nitrogen source , glutamine . Normalized measurements from RNAseq data are provided for direct visual comparison of effect sizes in intracellular algae compared to in vitro experiments . Intracapsular alga measurements are ‘caps’ ( filled red circles ) ; intracellular alga measurements are ‘cell’ ( empty red circles ) . ( a ) Expression of high affinity phosphate transporter PhT1-2 mRNA across a range of phosphate concentrations . ( b ) Expression of chloroplast sodium dependent phosphate transporter ANTR1 mRNA across a range of phosphate concentrations . In ( a ) and ( b ) The red dashed line indicates the average expression of the phosphate transporter in the low phosphate range ( 100 pM to 1 µM ) ; the blue dashed line indicates the average expression in the high phosphate range ( 10 µM to 10 mM ) . ( c ) Expression of three algal nitrogen transporters in the absence ( - ) and presence ( + ) of 2 mM L-glutamine . Data is plotted on a log2 scale on the y axis , where more negative values indicate lower expression levels . Circles are individual replicates; bars are the average for each experiment . *p<0 . 05; n . s . indicates no significant difference; the statistical test performed was an ANOVA with contrasts . DOI: http://dx . doi . org/10 . 7554/eLife . 22054 . 01610 . 7554/eLife . 22054 . 017Figure 2—source data 1 . Normalized expression levels of algal phosphate transporters . DOI: http://dx . doi . org/10 . 7554/eLife . 22054 . 01710 . 7554/eLife . 22054 . 018Figure 2—source data 2 . Normalized expression levels of algal nitrogen transporters . DOI: http://dx . doi . org/10 . 7554/eLife . 22054 . 018 Expression of two inorganic nitrogen transporters ( ammonium transporter 1-member 2 , AMT1-2 and high-affinity nitrate transporter 2 . 4 , NRT2 . 4 ) and a urea-proton symporter , DUR3 was repressed by L-glutamine ( Figure 2c ) . Adding 2 mM glutamine , the concentration observed in the cytoplasm of amphibian cells ( Vastag et al . , 2011; Westermann et al . , 2016 , 2012 ) , to algal cultures induced down-regulation of AMT1-2 ( 17-fold , p=3 . 2 × 10−4 ) , NRT2 . 4 ( 7-fold , p=0 . 013 ) , DUR3 ( 278-fold , p=9 . 5 × 10−10 ) . All of these in vitro changes closely match the in vivo expression differences revealed by RNA-seq for the intracellular alga ( Figure 2c ) . This study provides the first transcriptomes for A . maculatum and O . amblystomatis and an in-depth look at gene expression changes of both organisms in their unique endosymbiotic state . The dual-RNA-seq approach has previously been used to investigate intracellular parasitism in vertebrates ( Westermann et al . , 2016 , 2012; Tierney et al . , 2012 ) . However , our analysis represents the first investigation of a vertebrate endosymbiosis where the generalized interaction between the two organisms has consistently been characterized as a mutualism ( Small et al . , 2014; Gilbert , 1944; Bachmann et al . , 1986 ) . Our results also extend dual-RNA-seq methodology to low cell number samples from wild collected , non-model organisms . The transcriptional responses to this cellular association reveals how a vertebrate host responds to an intracellular mutualist and offers insights into the physiological condition of both partners in their endosymbiotic state . In the host salamander , we identified only a small fraction of the analyzed genes ( 300/46 , 549; 0 . 64% ) that are differentially expressed between endosymbiont-bearing vs endosymbiont-free salamander cells . This tempered host response is reminiscent of that of the hosts in coral-dinoflagellate endosymbioses; less than 3% of the analyzed genes were shown to be differentially expressed when the host coral was inoculated with and without a symbiosis competent strain of dinoflagellate ( Voolstra et al . , 2009; Mohamed et al . , 2016 ) . By comparison , the algal response to endosymbiosis from ectosymbiosis was observed to be more pronounced; 4 . 12% ( 277/6 , 726 ) of the algal genes were differentially expressed , proportionally 6 . 4 times more genes than in the host salamander . This level of change , nevertheless , is much more subtle when compared to the changes observed between the endosymbiont algal transcriptome and the cultured free-living alga grown in nutrient replete conditions where 40% ( 2 , 687/6 , 726 ) of the algal genes were differentially expressed . The over-expression of heat shock proteins , autophagy related proteins , and other stress inducible genes reveal hallmarks of stress in the intracellular algae ( Supplementary file 1 ) . These are undergoing multiple metabolic changes compared to their free-swimming intracapsular counterparts . Intracellular algae parallel the response of the closely related green alga Chlamydomonas reinhardtii to low sulfur levels ( Supplementary file 2 ) under hypoxia , including gene expression changes consistent with a switch from oxidative to fermentative metabolism ( Supplementary file 3 ) ( Nguyen et al . , 2008; Grossman et al . , 2011 ) . This response , relative to intracapsular algae , includes under-expression of photosystem II core components ( Supplementary file 4 ) in the chloroplast and complex I of the electron transport chain in the mitochondrion ( Piruat and López-Barneo , 2005 ) ( Supplementary file 5 ) , along with over-expression of fermentative metabolic pathways that would shuttle pyruvate toward acetyl-CoA , organic acids and alcohols [crucially , over-expression of pyruvate-ferredoxin oxidoreductase ( PFOR ) , phosphate acetyltransferase ( PAT ) , and aldehyde-alcohol dehydrogenase ( ADHE ) ] , and potentially produce H2 gas [over-expression of an iron hydrogenase ( HYDA1 ) ] ( Supplementary file 3 ) ( Volgusheva et al . , 2013; Yang et al . , 2013; Catalanotti et al . , 2013 ) . To verify the observed expression differences between intracapsular and intracellular algae , we performed a comparison of expression in the intracellular algae to O . amblystomatis gene expression in unialgal culture in nutrient replete media . A complete analysis of differentially expressed genes between O . amblystomatis cultured in nutrient replete media and intracellular algae revealed 1 , 805 over-expressed transcripts and 882 under-expressed transcripts in the intracellular algae ( indicating 40% of transcripts are differentially expressed , Figure 3 ) . A summary of GO terms enriched among the 1 , 805 genes over-expressed in intracellular algae relative to algae cultured in nutrient replete media confirms an enrichment in fermentation and stress response processes ( Figure 3—figure supplement 1 ) . Processes enriched among the 882 under-expressed genes are also consistent with low oxygen and stress to the intracellular algae relative to algae cultured in nutrient replete media ( Figure 3—figure supplement 2 ) . 10 . 7554/eLife . 22054 . 019Figure 3 . Differentially expressed genes between intracellular algae and cultured algae . Red dots indicate significantly differentially expressed genes ( FDR < 0 . 05 ) . Blue dashed lines represent a plus and minus 2-fold difference in expression . There are 1 , 805 over-expressed genes in intracellular algae and 802 under-expressed genes in intracellular algae in this comparison . DOI: http://dx . doi . org/10 . 7554/eLife . 22054 . 01910 . 7554/eLife . 22054 . 020Figure 3—source data 1 . GC content and length of algal genes . Use as input for normalizing algal count data based on GC content and gene length for algal libraries prepared by different methods . DOI: http://dx . doi . org/10 . 7554/eLife . 22054 . 02010 . 7554/eLife . 22054 . 021Figure 3—figure supplement 1 . REViGO anlysis of GO terms associated with 1805 over-expressed genes in intracellular algae compared to cultured algae . This analysis shows enrichment in fermentation processes such as glycerol-3 phosphate metabolism , 2-oxoglutarate metabolism , the glyoxylate cycle , photosystem II stability , photosystem I , and sulfur assimilation , all of which are consistent with the hypothesis that the intracellular algae are fermenting . Processes such as protein folding , apoptotic cell clearance , and sodium ion homeostasis support the hypothesis that the intracellular algae are stressed . DOI: http://dx . doi . org/10 . 7554/eLife . 22054 . 02110 . 7554/eLife . 22054 . 022Figure 3—figure supplement 2 . REViGO anlysis of 882 under-expressed genes in intracellular algae compared to cultured algae . Under-expressed processes involved in oxidative pathways , photoprotection , and protein refolding are further evidence of an intracellular algal stress response . DOI: http://dx . doi . org/10 . 7554/eLife . 22054 . 022 Specific consideration of the 36 genes demarking the fermentation response in intracellular algae compared to intracapsular algae shows that 21 are similarly significantly differentially expressed when compared to their expression in cultured O . amblystomatis from nutrient replete medium ( Figure 4 ) . Each of these 21 genes are similarly over- or under-expressed in the intracellular-intracapsular and intracellular-cultured algae comparisons . This includes under-expression of 3 components of the mitochondrial electron transport chain , and consistent over-expression of PFOR , PAT , ADHE and HYDA1 . Equivalent photosystem II core components are not significantly under-expressed in intracellular algae compared to cultured algae , suggesting that intracapsular algae over-express photosystem II core components , rather than intracellular algae under-expressing them . This may be due to hyperoxic conditions in the intracapsular environment ( Pinder and Friet , 1994 ) , which could lead to oxidative damage to and rapid turnover of the photosystem II core ( Richter et al . , 1990 ) . 10 . 7554/eLife . 22054 . 023Figure 4 . Differential expression of fermentation genes in intracellular algae compared to cultured algae . Blue dots are genes that were over-expressed in intracellular algae compared to intracapsular algae . Orange dots are genes that were under-expressed in intracellular algae compared to intracapsular algae . Vertical blue lines represent plus and minus two-fold fold change . The horizontal blue line represents FDR adjusted p-value equal to 0 . 05 . Genes above the horizontal blue line are significantly differentially expressed; genes below the blue line are not . Key fermentation genes , PFOR , HYDA1 , ADHE , and PAT are significantly over-expressed in intracellular algae compared to cultured algae , in the same manner as they are over-expressed in intracellular algae compared to intracapsular algae . Several components of complex I of the electron transport chain in the mitochondrion are also significantly under-expressed ( CYB , ND2 , ND4 ) , though ND1 is over-expressed in intracellular algae compared to cultured algae . DOI: http://dx . doi . org/10 . 7554/eLife . 22054 . 023 Overall , these metabolic changes suggest that photosynthesis by the alga does not keep up with respirational demand during the endosymbiotic state . In algal fermentation , some photosynthesis components are still active , notably photosystem II ( even if it is downregulated ) ( Volgusheva et al . , 2013 ) . However , the complement of over- and under-expressed genes in the alga suggests that neither the diffuse oxygen in these tissues nor the oxygen generated by photosynthesis is sufficient to meet the metabolic demands of the algal cell through oxidative phosphorylation . Instead these cells have switched to fermentation . This is potentially attributable in part to the opaque tissues of the embryonic host , which restrict the necessary photons from reaching algal chloroplasts that would allow more oxygen to be generated by the splitting of water . The occurrence of intracellular fermentation is also supported by decreased starch granules in the intracellular algae and transcriptional changes to algal sulfate transport and sulfur metabolism associated genes . Previous analysis of algal ultrastructure revealed intracellular algae had a significant reduction of starch reserves compared to intracapsular algae ( Kerney et al . , 2011 ) . Intracellular algae had approximately 56% of the starch reserves of intracapsular algae by cross sectional area , which corresponds to around 42% of the starch reserves by volume ( Kerney et al . , 2011 ) . Starch consumption during algal fermentation is well characterized in other Chlamydomonad algae ( Zhang et al . , 2002 ) . Two recent studies observed a reduction of intracellular starch reserves to approximately 50% of their peak levels during long term sulfur deprivation ( Zhang et al . , 2002; Chader et al . , 2009 ) . The overexpression of a sulfate transporter and taurine catabolic enzymes ( González-Ballester et al . , 2010 ) , along with other transcripts associated with sulfur metabolism ( Figure 3; Supplementary file 2 ) , indicates that fermentation in intracellular Oophila coincides with sulfur deprivation and closely matches the consumption of starch found in other fermenting algae . One discrete class of algal changes entails modifications to nitrogen and phosphorous transporters ( Supplementary file 6 ) , which are likely attributed to relatively high concentrations of intracellular nutrients compared to the egg-capsule microenvironment ( Brodman et al . , 2003; Carrino-Kyker and Swanson , 2007; Horowitz et al . , 1979; Burt et al . , 1976; Vastag et al . , 2011; Goff and Stein , 1978 ) . Transcriptional responses , closely matching those seen in our DE analysis , were initiated by mimicking intracellular concentrations of phosphate or glutamine in cultured algal stocks . The algal down regulation of inorganic nitrogen transporters in response to glutamine suggests that the algal endosymbiont is using host glutamine as a nitrogen source . This scenario is supported by the use of glutamine as a sole nitrogen source in other related green algal taxa ( Neilson and Larsson , 1980 ) . The changes in transporter expression indicate that metabolite concentration differences in an algal cell’s microenvironment can account for potentially all of the observed transcriptional differences in the DE analysis . This not only validates our dual-RNA-seq experimental design but suggests mechanisms of niche-dependent transcriptional regulation consistent with other green algae ( Fan et al . , 2016 ) , and the potential acquisition of host-derived glutamine for intracellular algal metabolism . There are interesting parallels to both parasitic infections and other known facultative endosymbioses in the salamander transcripts expressed . These include innate immune responses , nutrient sensing , cell motility and apoptosis/survival . The changes in transcript expression reveal a unique cytosolic relationship between these salamander cells and their algal endosymbionts . There are a remarkable number of transposable elements that are differentially expressed between salamander cells with and without algal endosymbionts ( Supplementary file 7 ) . Their differential expression may be controlled by the transcriptional regulation of nearby co-regulated genes ( Batut et al . , 2013 ) . We posit that the observed differential regulation of transposable elements in this study is a function of A . maculatum’s extraordinarily large genome ( at around 31 Gb , which is approximately ten times the size of the human genome ) ( Licht and Lowcock , 1991 ) . This large size is likely attributable to a large number of mobile elements in the salamander genome ( Keinath et al . , 2015 ) , which may share regulatory regions with protein coding genes ( Batut et al . , 2013 ) . With a few exceptions , the genes annotated as transposable elements have few detectable RNA transcripts ( median counts per million of 1 . 54 and 0 . 88 for genes with increased or decreased expression in salamander plus endosymbiont samples , respectively ) compared to other , non-transposon , differentially expressed genes ( median counts per million of 10 . 34 and 14 . 82 , respectively ) . Of the 69 genes with sequence homology to known transposable elements , 32 ( 46% ) have homologs in A . mexicanum transcriptomes ( Voss , 2016Voss , 2016; Stewart et al . , 2013; Wu et al . , 2013 ) . The transposable elements are largely ( 68% ) long interspersed nuclear elements ( LINE retrotransposons ) , which are typically associated with genome expansions in eukaryotes ( Kidwell , 2002 ) . Other differentially expressed transposable elements are PLE retrotransposons ( 6% ) , LTR retrotransposons ( 11% ) , DIRS retrotransposons ( 6% ) , and DNA transposons ( 3% ) . Intracellular invasion by a foreign microbe can lead to apoptosis in animal cells ( e . g . salmonella [Monack et al . , 1996; Kim et al . , 1998] or malaria [Kakani et al . , 2016] ) . However , the salamander cells with algal endosymbionts did not show any clear transcriptional signs of apoptosis . The one gene whose primary functional role may be in apoptosis is the salamander transcript ( Bcl2-like protein 14 , BCL2L14 ) with higher expression in salamander cells with algal endosymbionts , which contains BCL-2 homology ( BH ) domains BH3 and BH2 ( Figure 5 ) . Nonetheless , BCL2L14 has been explicitly shown to not be involved in pro-apoptotic regulation ( Giam et al . , 2012 ) . Other genes with higher expression in salamander cells with algal endosymbionts ( e . g . olfactomedin-4 , OLFM4; TNFAIP3-interacting protein 1 , TNIP1; serine/threonine protein kinase 1 , SGK1 ) have demonstrated anti-apoptotoic functions in other animals ( Liu and Rodgers , 2016; Ramirez et al . , 2012; Lang et al . , 2010 ) . In the lab , intra-tissue , and potentially intracellular , algal cells are detected for prolonged periods during development and post hatching in salamander larvae up to Stage-46 , and algal DNA was detected in adult tissues ( Kerney et al . , 2011 ) . Eventually , the number of detectable algal cells within the larvae decreases ( Kerney et al . , 2011 ) . This may coincide with the development of the salamander’s adaptive immune system , or it could be that the alga stops producing chlorophyll , but are nonetheless maintained within the embryo . There are seven transcripts ( e . g . SGK1; GDNF receptor alpha-4 , GFRA4; thymosin beta 4 , TMSB4 ) ( Supplementary file 8 ) with higher expression in salamander cells with algal symbionts that are linked to cell survival in different physiological contexts including cancer cell survival and proliferation , and neuronal survival during development ( Lang et al . , 2010; Enokido et al . , 1998; Bock-Marquette et al . , 2004 ) . Genes from this category may contribute to building a novel network of gene regulation used to maintain these endosymbionts . 10 . 7554/eLife . 22054 . 024Figure 5 . A . maculatum BCL2L14 protein has both a BH3 and BH2 domain . A multiple alignment of the A . maculatum BCL2L14 protein sequence with other organisms reveals a conserved BH3 and BH2 domains ( boxed ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22054 . 024 Our data reveals a limited immune response of embryonic cells with algal endosymbionts ( Supplementary file 9 ) . The salamander immune response can largely be categorized as an innate immune response , but it also includes components of an adaptive response ( e . g . , interleukin 7 , IL7 ) that precedes the developmental formation of an adaptive immune system ( Charlemagne and Tournefier , 1998 ) . Amphibian immunology has , for the most part , been considered a physiological process of larvae and adults ( Savage et al . , 2014 ) . There are no prior transcriptional datasets on embryonic immune responses , although there is a growing interest in amphibian embryo-microbial interactions ( Gomez-Mestre et al . , 2006 ) . Therefore , our study fills a gap pending the availability of comparative data from embryonic pathogens such as the oomycete Saprolegnia ( Petrisko et al . , 2008 ) . The increased expression of pro-inflammatory interleukins and chemokines ( e . g . interleukins , IL-8; IL-7; IL-1β; and C-X-C motif chemokine 10 , CXCL10 ) parallels the transcriptional response of adult frog skin to the chytrid fungus Batrachochytridium dendrobatidis ( BD ) infection and entry ( Ellison et al . , 2014 ) . Naïve BD infected frogs mount a much more dramatic immune response than these isolated salamander cells ( Savage et al . , 2014; Ellison et al . , 2014; Rosenblum et al . , 2009 ) . However , this may , in part , be due to the tissue-level resolution of these studies on chytrid infection , as opposed to the cellular-level resolution of our study , or the comparison of embryos to adults . Multiple transcriptional differences indicate an increased generation of reactive oxygen species ( ROS ) in salamander cells containing algae . The NADPH-oxidase family member dual oxidase 1 ( DUOX1 ) is involved in the zebrafish intestinal epithelial cell immune response to Salmonella infection ( Flores et al . , 2010 ) and is also more highly expressed in salamander cells with endosymbiotic algae . While this protein is implicated in several physiological processes , its ability to catalyze the generation of reactive oxygen species ( ROS ) reveals a potentially conserved immune response role in these non-phagocytic embryonic cells . Moreover , we observed under-expression of subunit 6b of cytochrome c oxidase ( COX6B1 ) . Reduction of cytochrome c oxidase activity is associated with the generation of ROS through signaling to endoplasmic reticulum NADPH oxidases in yeast ( Leadsham et al . , 2013 ) . Additionally , the downregulation of STEAP4 leads to lower levels of ROS in mouse osteoclasts ( Zhou et al . , 2013 ) , whereas we see upregulation of STEAP4 here . Importantly , the generation of ROS does not rely on the presence of high levels of oxygen in vertebrate cells as ROS can be generated in hypoxic as well as normoxic conditions through a variety of mechanisms ( Kim et al . , 2012; Nathan and Cunningham-Bussel , 2013 ) although there is no indication that the salamander cells are suffering from hypoxia here . Four genes with high expression during the intracellular symbiosis ( OLFM4; TMSB4; TNIP1; and mucin1 , MUC1 ) are associated with negative regulation of the NF-κB response in other systems ( Ueno et al . , 2008; Verstrepen et al . , 2009; Liu et al . , 2010; Sosne et al . , 2007 ) . NF-κB is a transcription factor complex that is expressed in all vertebrate cell types , and is involved in a variety of immune responses ( Tato and Hunter , 2002; Takada et al . , 2005 ) . Of the possible genes transcriptionally regulated by NF-κB ( Ali and Mann , 2004; Gilmore Lab , 2016 ) , only effectors in the cytokine/chemokine group were observed as over-expressed in salamander cells containing intracellular algae . Our salamander transcriptiomes contain 295 genes that are homologous to downstream targets of NF-κB identified in other systems . Only five are over-expressed here , IL8 , IL1b , CXCL10 and TNIP1 ( with TNIP1 also being a negative regulator of NF-κB ) ( Supplementary file 10-sheet 1 , ‘NFkB_Expr’ ) . A fifth NF-κB response gene , which we have annotated as a trypsin , but which also has strong homology to Granzyme B ( associated with apoptosis ) was under-expressed in salamander cells hosting endosymbionts ( Supplementary file 10-sheet 1 , ‘NFkB_Expr’ ) . Increased expression of genes associated with attenuating NF-κB signal transduction has precedent in other intracellular infections ( Tato and Hunter , 2002 ) and possibly symbiotic associations ( McFall-Ngai , 2014 ) . The over-expressed OLFM4 and MUC1 are also implicated in establishing the vaginal microbiome , potentially through their roles in modulating NF-κB signaling ( Doerflinger et al . , 2014; Fields et al . , 2014 ) . How the algal cell entry may be affecting these genes and whether NF-κB by-pass is involved in algal cell entry or maintenance remains to be determined . We did not find modified expression of toll-like receptors ( TLR’s ) , which detect pathogen-associated molecular patterns . These function in activating an innate immune response to both bacterial and protistan pathogens ( Ashour , 2015 ) as well as establishment of gut commensals ( Round et al . , 2011 ) . While TLR’s do not require differential regulation for their normal function , established schistosomiasis , entamoeba , trypanosome , and filarial nematode infections all result in down-regulation of TLR transcripts ( Ashour , 2015 ) , and a resulting NF-κB mediated response . In our data , we found 10 salamander genes with homology to various TLRs . None were differentially expressed between salamander cells without and salamander cells with endosymbionts . Further , we examined expression of 151 additional genes associated with the TLR response , and found only one over-expressed gene downstream of TLRs , TNIP1 , which is a negative regulator of NF-κB , as discussed above ( Supplementary file 10-sheet 2 , ‘TLRs_Expr’ ) . While the lack of differential transcriptional regulation of TLR’s or their regulators does not preclude their involvement in algal entry response , it is notable in comparison to other parasitic infections where TLR expression is often down regulated ( Ashour , 2015 ) . One potential benefit of having an intracellular alga may be to prime the embryo’s immune system , without over-activating it , granting the invaded embryos additional protection against the microbial environment outside of the egg capsule . One mechanism for this immune priming may be revealed by relatively increased hepcidin expression in salamander cells with intracellular algae . Elevated hepcidin levels are protective against multiple infections of malaria parasites in mammalian models ( Portugal et al . , 2011 ) and were shown to enhance resistance to bacterial infection when transgenically over-expressed in zebrafish ( Hsieh et al . , 2010 ) . In established endosymbioses between invertebrates and algae , the transfer of organic molecules synthesized by the algal partner allows otherwise non-photosynthetic animals to become partial or complete autotrophs . The exchange of photosynthate from symbiont to host is mediated by a range of molecules including sugars , sugar alcohols , and lipids ( Burriesci et al . , 2012; Kellogg et al . , 1983; Colombo-Pallotta et al . , 2010 ) . In the endosymbiosis between A . maculatum embryos and O . amblystomatis , the alga does not appear to be using a canonical photosynthetic process of carbon fixation , oxygen evolution , and sugar production , but is rather metabolizing by fermentation . This metabolic state does not , however , preclude the possibility of metabolite transfer from the intracellular alga to the salamander cells . Under fermentation , the alga may still generate ATP from light energy ( Godaux et al . , 2015 ) , fix carbon ( Godaux et al . , 2015 ) or use alternate molecules as a carbon source ( Gibbs et al . , 1986 ) . Fermenting algae are also capable of using anabolic reactions to produce sugars and lipids ( Gibbs et al . , 1986 ) . Indeed , the related alga Chlamydomonas moewusii excretes glycerol , acetate , and ethanol under anoxic conditions ( Klein and Betz , 1978 ) . Release of fermentation byproducts such as formate , acetate , and glycerol , or of energy storage molecules such as sugars or lipids into the host cytoplasm could trigger differential expression of nutrient sensing mechanisms within A . maculatum cells . In A . maculatum cells with an intracellular alga , differentially expressed genes involved in nutrient sensing ( e . g . STEAP4; neurosecretory protein VGF , VGF; resistin , RETN; pyruvate dehydrogenase phosphatase 1 , PDP1; and calcium/calmodulin-dependent protein kinase 1 , CAM-KK 1 ) ( Supplementary file 11 ) ( Wellen et al . , 2007; Petrocchi-Passeri et al . , 2015; Steppan et al . , 2001; Jeoung and Harris , 2010; Witczak et al . , 2007 ) are suggestive of altered metabolic flux through catabolic and anabolic pathways , particularly with respect to insulin production and sensitivity . One differentially expressed algal gene that may be implicated in nutrient exchange with salamander cells is an algal gene with homology to Niemann-Pick type C ( NPC ) proteins , which is more highly expressed in intracellular algae . These proteins are involved in intracellular cholesterol transport , and are potential mediators of lipid transfer in cnidarian-dinoflagellate endosymbioses ( Dani et al . , 2014 ) . Intriguingly , increased expression of this potential sterol sensing gene is observed in our intracellular algal transcripts , whereas in cnidarian-dinoflagellate interactions it is the host that utilizes NPC proteins ( Dani et al . , 2014 ) . Further metabolic changes include lower expression of maltase-glucoamylase ( MGAM ) , an acid phosphatase , and trypsin-like proteins in salamander cells with intracellular algae ( Supplementary file 11 ) . These changes indicate a reduction in glycogen metabolism ( Barbieri et al . , 1967 ) , and a reduction in the degradation and utilization of yolk platelets that has been shown to be mediated in part by acid phosphatases ( Lemanski and Aldoroty , 1977 ) . Collectively , these metabolic changes may be induced by detection of metabolites transferred by the alga , or alternatively , expression changes of these genes might be modulated by autocrine signaling as there is overlap between nutrient sensing and inflammatory responses ( Schenk et al . , 2008 ) . Lipoprotein-related protein 2 ( LRP2 ) , which is expressed 5 . 5 fold higher in invaded salamander cells , is part of a family of multi-ligand receptors that trigger endocytosis ( Supplementary file 11 ) ( Fisher and Howie , 2006 ) . Dual binding of malaria sporozoites to a human LRP receptor and heparin sulfate proteoglycans mediates malaria sporozoite invasion into liver cells ( Shakibaei and Frevert , 1996 ) . All salamander samples also displayed expression of heparin sulfate basement protein ( mean CPM 9 . 82 ) , though it was not differentially expressed in invaded cells . The alga may have surface proteins that interact with the salamander LRP2 receptor however it does not have a recognizable homolog of the malaria circumsporozoite ( CS ) protein , which is implicated in interactions with LRP and heparin sulfate proteoglycans ( Shakibaei and Frevert , 1996 ) . The lack of a recognizable symbiosomal membrane around an intracellular alga ( Kerney et al . , 2011 ) suggests that if they do indeed enter through endocytosis , they must escape from or degrade the host-derived vesicle . Interestingly , there are two algal lipase/esterase genes and one gene with homology to the bacterial virulence factor streptolysin S that are over-expressed in endosymbiotic algae ( 55 , 11 , and 82-fold ) . Lipase/esterases are known virulence factors in bacterial parasites ( Singh et al . , 2010 ) . These algal lipase/esterases may have a role in algal entry or endosome escape . While the LRP connection to malaria entry is an interesting parallel , the over-expression of LRP2 may be attributable to other processes in the host cell . For instance LRP2 may be involved in nutrient sensing as it has been implicated in retinol binding protein ( RBP ) import ( Salamander RBP2 was also more highly expressed in cells with endosymbiotic algae ) and vitamin homeostasis ( Christensen et al . , 1999 ) . The observed increased expression of salamander villin 1 ( VIL1 ) in cells with intracellular algae ( Supplementary file 12 ) may also reveal a pre-disposition of infected cells for algal entry . Similar predispositions exist in hepatocytes infected with a malaria sporozoite . These express high levels of EphA2 prior to parasite entry , which eventually allows a by-pass of host apoptosis ( Kaushansky et al . , 2015 ) . Villin one is an actin modifying protein that has recently been shown to be required for membrane ruffling and closure following Salmonella typhimurium invasion of intestinal epithelial cells ( Lhocine et al . , 2015 ) . This host protein is required for successful pathogen entry and is regulated by the bacterial SptP protein through phosphorylation . Similar membrane ruffling has been observed in regions of algal contact with host alimentary canal epithelium coincident with algal entry ( Kerney et al . , 2011 ) . Whether the alga benefits from this endosymbiotic interaction remains unclear . Similar questions of net ‘mutualism’ persist for the symbiosis between the bobtail squid Eprymna scolopes and the bacterium Aliivibrio fischeri ( McFall-Ngai , 2014 ) although in both systems microbial cells exhibit specific taxic responses to a developing host , suggesting an adaptive origin of these behaviors . In a previous study , we found evidence consistent with vertical transmission . Algal 18S rDNA was amplified from adult oviducts and Wolffian ducts , and encysted algal cells were found inside the egg capsules of freshly laid eggs using transmission electron microscopy ( TEM ) ( Kerney et al . , 2011 ) . However , to date , we have not found conclusive evidence for vertical transmission of the alga from one generation to the next . As such , any benefit to the alga in this endosymbiotic interaction remains unknown . We may speculate that intracellular algae are providing some benefit to its host , as many past light/dark rearing experiments have shown a net benefit to the salamander embryo from their algal symbionts , which presumably included endosymbionts as well ( Pinder and Friet , 1994; Gilbert , 1944; Bachmann et al . , 1986; Gilbert , 1944 ) . Two recent studies have suggested the transfer of photosynthate from intracapsular Oophila to the salamander host ( Graham et al . , 2013 , 2014 ) . However oxygenic photosynthesis and fixed carbon photosynthate transfer does not appear to be a significant contribution from intracellular algae to their hosts ( Graham et al . , 2013 ) . Instead these algae appear to be utilizing fermentation , a common response of chalamydomonad algae to sulfur deprivation and hypoxic conditions ( Yang et al . , 2015 ) . In an intriguing parallel to the metabolic state of intracellular Oophila , some obligate intracellular parasites , such as the apicomplexan Plasmodium sp . , are ‘microaeorophiles:' these generate most of their energy through the incomplete oxidation of glucose to lactate , a fermentative process ( Olszewski and Llinás , 2011 ) . Plasmodium falciparum exhibits increased infection and growth at low partial pressures of oxygen ( Ng et al . , 2014; Scheibel et al . , 1979 ) . Moreover , the observed cell stress response of the alga is reminiscent of that experienced by intracellular apicomplexan parasites ( Bosch et al . , 2015 ) . Mutualist ectosymbionts like the bacteria Aliivibrio fischeri also use anaerobic metabolism , including fermentation , and express genes consistent with a response to oxidative stress during their association with the bobtail squid Eprymna scolopes ( Thompson et al . , 2017; Wier et al . , 2010 ) . The primary fermentation products of chlamydomonad algae include glycerol , ethanol , formate , and acetate , along with smaller amounts of CO2 and H2 ( Catalanotti et al . , 2013 ) . Glycerol is recognized as a major mediator of energy transfer between dinoflagellate photosymbionts and cnidarian hosts ( Davy et al . , 2012 ) . In the context of the salamander-alga endosymbiosis , the related alga C . reinhardtii was shown to export glycerol after osmotic shock ( Léon and Galván , 1995 ) , a condition that intracellular Oophila likely experience upon invasion of salamander cells . Additionally , C . moewusii excretes glycerol during fermentation ( Klein and Betz , 1978 ) . Formate and acetate fermentation by-products in bacterial ectosymbionts are known sources of energy and bases for biosynthesis of more complex molecules in animal intestinal cells ( den Besten et al . , 2013; Karasov and Douglas , 2013 ) . Moderate ethanol concentrations , below 2 mM , appear to be well tolerated by animal cells ( Castilla et al . , 2004 ) , while higher concentrations become cytotoxic , inducing apoptosis and necrosis ( Castilla et al . , 2004 ) . It is unknown whether intracellular algae are indeed releasing significant quantities of ethanol into host cells , however prior TEM observations ( Kerney et al . , 2011 ) , and our results suggest that ethanol is below cytotoxic levels as we do not see indication of a necrotic or apoptotic host response . Although the main comparison in this manuscript was between intracapsular algae and intracellular algae , we also considered differentially expressed genes between algae cultured in nutrient replete medium and intracellular algae . The latter comparison supported fermentation in the intracellular algae , but did not indicate over-expression of additional biosynthetic capabilities including enhanced vitamin biosynthesis or the production and export of other metabolites that might be beneficial to the salamander . Whether the intracellular algae are on the positive end of a net host benefit remains uncertain , however it is clear that the algae have an unconventional ‘photosymbiont’ role . To the best of our knowledge there are only two models where the acquisition of horizontally acquired endosymbionts has mechanistic resolution: dinoflagellates in corals and rhizobial bacteria in root nodules . Starting from wild collected samples in a non-model system , we compiled novel transcriptomes of two organisms and revealed gene expression changes associated with their intracellular symbiosis from low cell number samples . These data reveal that life in a vertebrate’s cytoplasm induces a stress response in the symbiotic alga . While the alga appears to benefit from high concentrations of phosphate and organic nitrogen sources , our data suggests that the alga is limited in oxygen and sulfur , and is osmotically stressed ( Figure 6 ) . The salamander appears to recognize the alga as foreign , but does not mount an immune response comparable to what is seen in amphibian-pathogen interactions , and the salamander may be actively repressing important immune regulators such as NF-κB as well as receiving a nutritive benefit from the endosymbiotic alga ( Figure 6 ) . 10 . 7554/eLife . 22054 . 025Figure 6 . Summary of the major changes in both salamander and algal cells and how they may relate to one another . The inferred salamander responses are broken into four functional categories while algal changes fall within three primary functional categories based on gene annotations . Text indicates hypothetical changes within each category based on the implied roles of under-expressed or over-expressed genes . Major sections are color-coded . Over-expressed genes represented by solid black symbols . Under-expressed gene symbols are white with black outlines . Cellular compartments are in italics . M=mitochondrion , YP=yolk platelet , V=vacuole , N=nucleus , ECM=extracellular matrix , ER=endoplasmic reticulum , Chl=chloroplast . DOI: http://dx . doi . org/10 . 7554/eLife . 22054 . 025 Components of the salamander expression profile are relevant to vertebrate interactions with commensal symbionts . For example , three of the genes with higher expression in cells with endosymbionts in this association , OLFM4 , MUC1 , and IL8 are also up-regulated in human irritable bowel disease ( Gersemann et al . , 2012 ) . Other roles of OLFM4 and MUC1 include negative regulation of NF-κB signal transduction , and interactions with commensal ectosymbionts in humans ( Liu et al . , 2010; Doerflinger et al . , 2014; Fields et al . , 2014 ) . The notable absence of other transcripts ( e . g . TLR’s ) are indicative of endosymbiont tolerance in this system , in sharp contrast to an expected response to vertebrate pathogens . As in other endosymbiotic associations , A . maculatum and O . amblystomatis engage in a unique dialog involving host tolerance of the symbiont and metabolic cross-talk between partners . Distinctive facets of this metabolic cross-talk include fermentation in the endosymbiont as well as phosphate and glutamine acquisition from the host cytoplasm . This study has dramatically expanded our ability to interrogate this endosymbiotic dialogue on a molecular level by co-opting the dual RNA-seq approach established for parasitology research to a non-model mutualism . A . maculatum cells in 90% RNAlater were diluted to a 50:50 solution of cells in RNA later and amphibian-phosphate buffered saline ( APBS; PBS + 25% H2O to match amphibian osmolarity of 225 ± 5 mOsm/L ) because the 90% RNAlater solution was too viscous for single cell isolation . The solution was spread on a glass slide and inspected between the bright field and epi-fluorescence illumination with a Chlorophyll filter set . Fifty salamander cells with intracellular algae and fifty salamander cells without intracellular algae per individual embryo were separated by manual single cell isolation ( mouth pipetting ) from dissociated embryos with a hand-pulled borosilicate pipette connected to a rubber tube . Each cell was collected directly from 45% RNA later into a microcentrifuge tube containing 200 µL lysis buffer ( Extraction Buffer ( XB ) , PicoPure RNA extraction kit; ThermoFisher Scientific ) . Four biological replicates were collected from four different A . maculatum embryos from the same clutch . Three biological replicate samples of intracapsular algae from the intracapsular fluid of three eggs were also collected—by aspiration with a syringe and 23 gauge needle—for RNA extraction . Additional RNA was prepared from three unialgal strains of Oophila including UTEX LB3005 and LB3006 , which were established previously ( Kim et al . , 2014 ) . The third algal strain—isolated from an egg clutch sampled in 2012 from Greenbrook Sanctuary ( Palisades , NY ) —unfortunately , was lost during a laboratory power-outage . Quadruplicate sampling was made for quantitative analyses of the LB3005 strain . RNA was extracted from each sample using the PicoPure RNA extraction kit following the manufacturer’s recommended protocol for ‘RNA Extraction from Cell Pellets’ , starting with incubation of the lysate at 42°C for 30 min with the modification of adding an equal volume of 70% ethanol to the 200 µL of lysis buffer . RNA for the cultured algal strains was prepared by using a combination of TRIzol ( Thermo Fisher Scientific ) and a Qiagen RNeasy kit . Whole cDNA libraries were prepared directly from total RNA using the SMARTer Ultra Low cDNA Kit–HV ( Clontech , Mountain View , CA ) according to the manufacturer’s protocol for a starting sample of 50 cells . Sequencing libraries for the Illumina HiSeq 2500 platform were prepared from the whole cDNA libraries using the Nextera-XT kit ( Illumina , San Diego , CA ) according to the manufacturer’s protocol with an input of 375 pg cDNA per sample and a final clean-up step based on an Ampure-XP ( Beckman Coulter , Brea , CA ) protocol with the modification of adding 0 . 75x volume of bead solution to cDNA sample . Sequencing took place at the New York Genome Center and on an Illumina HiSeq 2500 sequencer with 125 bp paired-end reads . Transcriptome libraries ( Illumina’s TruSeq RNA library ) for the cultured algae were prepared and sequenced at Genome Quebec on HiSeq 2000 with 100 bp paired-end reads or at Cornell Weill Genomics Resources Core Facility on the MySeq platform with 75 bp paired end reads . Eleven paired-end whole cDNA libraries with greater than ten million paired reads per library were processed for assembly . There were four libraries from A . maculatum cells without intracellular algae , and four paired libraries ( from the same individuals ) from A . maculatum cells with intracellular algae , and three libraries from motile intracapsular algae . Quality trimming of the reads was performed with Trimmomatic ( v 0 . 32 ) ( Bolger et al . , 2014 ) to remove low quality bases and adapter sequences . All eleven paired end libraries were used to construct a total evidence assembly using the Trinity algorithm ( version trinityrnaseq_r20140717 ) ( Grabherr et al . , 2011; Haas et al . , 2013 ) . Transcriptomes of cultured algal strains were assembled separately in Trinity . The total evidence assembly returned 1 , 533 , 193 unique RNA-seq contigs that were clustered into 1 , 345 , 464 potential gene level ( isoforms collapsed ) transcript groups . The assembly largely consisted of a mixture of A . maculatum and O . amblystomatis transcripts . There were also 7 , 193 transcript groups ( 0 . 5% ) corresponding to a predatory mite , Metaseiulus occidentalis , and 2 , 641 transcript groups corresponding to a dermal fungus , Malassezia globosa . Sequences corresponding to the mite and fungus were removed by BLASTN homology search ( all BLAST analysis was completed using BLAST+ algorithms , v 2 . 2 . 28+ ( Camacho et al . , 2009 ) using BLAST databases comprised of all known transcript sequences from the genera Metaseiulus and Malassezia . Transcripts corresponding to the alga O . amblystomatis and salamander A . maculatum were recovered by BLASTN against a database consisting of transcript sequences from lab grown cultures of O . amblystomatis and sequences from the model salamander Ambystoma mexicanum ( contributed by Randall Voss—University of Kentucky , and from [Stewart et al . , 2013; Wu et al . , 2013] ) . Transcripts were further filtered by a BLASTX homology search against a database containing the entire protein complement of: Arabidopsis thaliana , Chlamydomonas reinhardtii , Mesostigma viride , Micromonas pusilla , Ostreococcus tauri , Oryza sativa , O . amblystomatis , Chrysemys picta bellii , Xenopus tropicalis , A . mexicanum , Pseudozyma , Saccharomyces cerevisiae , and the genera Melanopsichium and Leptosphaeria . The assortment of species was chosen due to phylogentic proximity to O . amblystomatis or A . maculatum , or due to best hits from those genera/species found when a selection of transcripts was queried against the nr database . Best hits to plant or green algal species were noted as algal sequences and combined with the results of BLASTN against the O . amblystomatis database . Best hits to salamander or other animal species were noted as salamander sequences and combined with the results of BLASTN against the A . mexicanum database . Best hits to fungal sequences were discarded . The remainder with no known homology were retained and included as putative algal or salamander transcripts based on their expression pattern across samples . One expectation of differential expression analysis is that most genes are expressed equally between control and experimental samples . If expression levels are ordered from low to high expression in control samples and binned in a sliding window of 100 genes per bin , the median expression level in each bin will increase as the index increases . Based on the expectation of equal expression , for the same sets of genes , the median expression level of experimental samples should correspondingly increase . For a subset of genes confirmed to belong to A . maculatum by BLAST homology , genes were sorted by Fragments Per Kilobase of transcript per Million mapped reads ( FPKM ) values in the samples of A . maculatum without intracellular algae . Median FPKM values of bins of 100 genes , in a sliding window from the lowest expressed gene through the 100 genes with the highest FPKM values were calculated for the samples of A . maculatum with and without intracellular algae . The median FPKM values for the bins of A . maculatum only samples were plotted against the data from the samples of A . maculatum with intracellular algae ( Figure 1—figure supplement 2a ) . At moderate to high expression levels , A . maculatum with intracellular alga FPKM values increased with A . maculatum without intracellular algae FPMK values ( Figure 1—figure supplement 2a ) . But at very low FPKM values , the data were essentially uncorrelated . Median A . maculatum without intracellular alga FPKM values increased ( since the data was ordered by those values ) , but median A . maculatum with intracellular alga FPKM values stayed the same ( Figure 1—figure supplement 2a ) . Lower limit FPKM values were determined by finding the FPKM value above which A . maculatum without intracellular alga samples and A . maculatum with intracellular alga samples exhibited a positive correlation ( Figure 1—figure supplement 2a ) . Genes with FPKM values below the lower limit of both sets of samples being analyzed for differential expression ( i . e . intracapsular algae and intracellular algae ) were not included in the analysis . The uncorrelated expression pattern of genes with FPKM values below the threshold suggests that there is either insufficient sequencing depth to compare those genes between the two conditions , or those lowly expressed genes are expressed stochastically in these cells and the fluctuations in expression levels of those genes are not indicative of a biological difference between conditions . The same analysis was completed for A . maculatum cells with intracellular algae by ordering the genes based on their expression levels ( Figure 1—figure supplement 2b ) , and for intracapsular ( Figure 1—figure supplement 2c ) and intracellular algal samples ( Figure 1—figure supplement 2d ) . The lower limit FPKM values for A . maculatum genes were 0 . 55 FPKM for A . maculatum cells ( Figure 1—figure supplement 2a , vertical red-dashed line ) and 0 . 61 FPKM for A . maculatum cells with algal endosymbionts ( Figure 1—figure supplement 2b , vertical red-dashed line ) . The lower limit FPKM values for algal genes were 2 FPKM for the intracapsular alga ( Figure 1—figure supplement 2c , vertical red-dashed line ) and 0 . 04 FPKM for the intracellular alga ( Figure 1—figure supplement 2d , vertical red-dashed line ) . The values are reflective of the sequencing depth of each sample , and are close to the widely used FPKM >1 lower limit threshold used in many RNAseq studies ( Fagerberg et al . , 2014; Shin et al . , 2014; Graveley et al . , 2011 ) , except for the intracellular alga samples which suffer from low sequencing depth , but nonetheless display correlated expression with intracapsular alga samples starting at low FPKM values . After determining the lower limit thresholds , the algal gene set consisted of genes with at least one read pair mapping to each of the three intracapsular algal samples or each of the four endosymbiotic cell samples that additionally were not found in the salamander , fungal , or mite BLAST data . Additionally , the genes had to have expression values above the lower limits described above in respective algal samples and below the lower limit for A . maculatum without intracellular alga samples in the salamander only cell samples for those genes . This resulted in a set of 8989 potential algal genes . However , due to a low depth of sequencing of the algal component of the endosymbiotic cell samples , additional filtering was necessary . Genes that were not detected in intracellular algae could have been missing due to the lower depth of sequencing rather than representing an actual biological difference in expression between the algal populations . To determine what level of expression in intracapsular alga would be needed for a complete absence of measured expression in intracellular alga to be meaningful , the 8 , 989 algal genes were first ordered by intracapsular algal FPKM . Then the proportion of genes with no expression in intracellular samples was plotted against the median expression level in high-sequencing-depth intracapsular algae in bins of 100 genes in the ordered data . At low FPKM values in the intracapsular algae , up to 58% of the genes were absent from intracellular samples ( Figure 1—figure supplement 3b ) . As expression in intracapsular alga samples increased , the proportion of genes with measurable expression levels in intracellular algal samples increased as well ( Figure 1—figure supplement 3b ) . The same relationship was not observed for A . maculatum data sets , where the depth of sequencing between samples was approximately equal ( Figure 1—figure supplement 4 ) . The dependence of gene detection in intracellular algal samples on FPKM level in intracapsular algal samples abated at intracapsular alga FPKM values where 95% or more genes could be detected in intracellular algal samples ( Figure 1—figure supplement 3b , vertical red-dashed line ) . That expression level corresponded to 67 . 9 FPKM in intracapsular algal genes . After removing genes with no detectable expression in intracellular algae with expression levels below 67 . 9 FPKM in intracapsular algae , the dependence of gene detection on expression level in intracapsular algae was removed , and the anomalous peak of undetected genes was removed from the expression histogram ( see: Figure 1—figure supplement 3a and c ) . This resulted in a set of 6 , 781 genes . Due to finding some genes with homology to anomalous organisms such as pine and beech trees without homologs in C . reinhardtii in the set of 6 , 781 genes ( perhaps due to pollen in the low cell number samples ) , only genes with homologs in the lab strain Oophila transcriptomes were considered . The final set of algal genes used in differential expression analysis consisted of 6 , 726 genes . For transcripts derived from wild-collected samples , read mapping and transcript count quantification was accomplished using Bowtie2 ( Langmead and Salzberg , 2012 ) and RSEM ( Li and Dewey , 2011 ) using default parameters . For transcripts derived from cultured alga , read mapping and transcript quantification was accomplished using BBmap ( Bushnell , 2016 ) and Salmon ( Patro et al . , 2015 ) ( respectively ) . The read mapping and quantification algorithms used for reads from the cultured alga were implemented due to divergence of the two algal strains . The transcripts common between the two strains were on average 95% similar . BBmap paired with Salmon allowed for relaxed mapping parameters that were able to map the reads to the transcriptome despite differences in sequence composition with increased sensitivity compared to Bowtie2 plus RSEM . Prior to differential expression analysis , data driven abundance and homology filtering was implemented to derive the final gene lists used in differential expression analysis . A detailed account of filtering procedures can be found in Supplementary Materials and methods under the heading ‘total evidence assembly filtering’ . Differential expression analysis was conducted in R ( version 3 . 1 . 2 ) ( R Core Team , 2013 ) using the edgeR package ( Robinson et al . , 2010 ) . Generalized linear models were used for data analysis on normalized count data . Initial normalization of data derived from wild collected samples ( endosymbiont free and endosymbiont containing salamander cells , intracapsular algae , and intracellular algae ) , was performed by trimmed mean of M-values ( TMM ) library size scaling-normalization ( Robinson and Oshlack , 2010 ) . Incorporation of data from unialgal cultures into differential expression analysis required additional GC-content normalization of the libraries , due to differences in GC-bias introduced by the two different library preparation methods ( SMARTer cDNA synthesis followed by Nextera-XT library preparation for the wild collected samples and TrueSeq library preparation for the RNA preparation from unialgal cultures ) ( Figure 7 ) . Normalization of GC-content and transcript length biases were accomplished using conditional quantile normalization ( CQN ) ( Hansen et al . , 2012 ) . CQN resolves GC content and transcript length biases by fitting a model that incorporates observed read counts and a covariate such as GC content , and calculates an offset that is used to remove the covariation of these confounding factors . Following cqn normalization , differential analysis between unialgal cultures and wild collected intracapsular or intracellular algae was completed in edgeR . Salamander-only and salamander-plus-alga samples from the same individual embryo were considered paired samples for statistical analysis . Differentially expressed genes were considered as those with an FDR adjusted p-value less than 0 . 05 . 10 . 7554/eLife . 22054 . 026Figure 7 . GC and transcript length bias in SMARTer-cDNA synthesis-Nextera-XT libraries compared to TrueSeq libraries . Red lines indicate the GC content or transcript length biases in reads obtained from SMARTer-cDNA synthesis-Nextera-XT libraries . Blue lines indicate the GC content or transcript length biases in reads obtained from TrueSeq libraries . ( a ) GC content and length are plotted against ‘QRfit’ which is a measure of fit by quantile regression to the models in Hansen et al . ( 2012 ) . This metric approximates bias in the sequence dataset by comparing read counts to expected models based on quantiles in the distribution of the GC content of the transcripts . The opposing trends in the two sets of lines shows that GC content bias between the two different libraries is vastly different . The reads obtained from SMARTer-cDNA synthesis-Nextera-XT libraries will tend to have more counts for low GC content transcripts , while the reads obtained from TrueSeq libraries will tend to have more counts for high GC content transcripts , systemically . ( b ) There is also some moderate transcript length bias differences between the two library prep methods visualized as the separation between the groups of red and blue lines . The methods implemented by the conditional quantile normalization ( cqn ) package in R handles both types of bias to make the gene count data from both library preparation methods comparable . DOI: http://dx . doi . org/10 . 7554/eLife . 22054 . 026 Functional annotation of A . maculatum and O . amblystomatis transcripts was accomplished by BLASTX of transcripts against the UniProt-SWISSProt curated database ( Gasteiger et al . , 2001 ) . BLASTX results were filtered by ‘homology-derived structure of proteins’ ( HSSP ) score ( Rost , 2002 ) such that annotations were retained for hits with an HSSP_DIST score greater than 5 ( Burns et al . , 2015 ) . UniProt ID annotations were assigned based on the maximum HSSP score for each gene . This led to 14 , 761 functional annotations for A . maculatum transcripts and 3 , 850 functional annotations of O . amblystomatis transcripts . Further annotation of differentially expressed transcripts was accomplished by HHblits ( Remmert et al . , 2012 ) homology detection . Unannotated , differentially expressed transcripts were translated in all six reading frames and the translations were processed by HHblits . Significant hits were determined by manual inspection of the HHblits output . Transposable elements were categorized by homology to known transposons by BLASTX or HHblits , and through the use of the PASTEClassifier tool ( Hoede et al . , 2014 ) . Multiple alignments were created with MUSCLE using default parameters ( Edgar , 2004 ) . Alignments were visualized in SeaView ( Gouy et al . , 2010 ) . Due to the novelty of this symbiosis , the non-model organisms under consideration , and the multi-organism functional annotation obtained , an initial automated functional grouping , gene ontology ( GO ) term analysis , completed as in Burns et al . ( 2015 ) ( Supplementary files 13 and 14 for algal and salamander GO term analysis , respectively ) was determined to be insufficient to understand the likely biological function of over and under-expressed genes . Other tools such as REViGO ( Supek et al . , 2011 ) performed better ( Supplementary files 15 and 16 for algal genes , and Supplementary files 17 and 18 for salamander genes ) , but did not catch large functional modules that were evident upon further manual inspection of the gene lists . For the small sets of differentially expressed genes observed between salamander cells with or without algae , and between intracapsular and intracellular algae , manual curation of each differentially expressed gene was implemented by performing an extensive literature search for each of the differentially expressed genes , based on the SWISSProt or HHblits annotation . Relevant functions associated with each gene in the scientific literature were noted , and those functions were grouped manually to give the final functional categories discussed in the text . For the larger set of differentially expressed genes observed between algae cultured in nutrient replete media and intracellular algae , only automated annotation and GO term grouping ( REViGO ) was used to define functional categories . Hypotheses concerning gene expression patterns associated with endosymbiosis in the alga O . amblystomatis were tested using culture strains of the alga . Validation experiments could not be conducted in the salamander A . maculatum due to the seasonal nature of the association and the lack of any laboratory stock of A . maculatum . O . amblystomatis cultures were maintained at 18°C under a 12 hr light/12 hr dark cycle with an average light intensity of 34 µmol·m−2·sec−1 , in AF-6 medium ( Kato , 1982 ) with modifications as described previously ( et al . , 2004 ) . For examining the dependence of phosphate transporter expression on phosphate levels in the media , AF6 media was formulated without potassium phosphate . Appropriate quantities of concentrated potassium phosphate at a ratio of 1:2 K2HPO4:KH2PO4 plus potassium chloride ( to a final concentration of 130 . 8 µM K+ ions , the concentration of K+ ions present in normal AF-6 media ) were added to the phosphate deficient AF-6 media to make AF-6 with the various phosphate levels used in the experiment . For examining the dependence of nitrogen-related transporters on nitrogen and glutamine concentrations , AF-6 media was formulated with various levels of nitrate and ammonia with the addition of NaCl to balance the loss of Na+ ions from leaving NaNO3 out of the media . For phosphate limitation experiments , O . amblystomatis cells growing in AF-6 media were pelleted ( 1 , 000xg for 5 min ) and washed three times with media completely depleted of phosphate . Following the third wash , cells were re-suspended in phosphate depleted media . Cells were counted and aliquoted into flasks at a concentration of 40 , 000 cells·mL−1 in 5 mL total volume per flask of phosphate deplete AF-6 . For each phosphate concentration , 5 mL of AF-6 media with 2x phosphate was added to the appropriate flask to give the appropriate phosphate concentration for the experiment . Cultures were grown for 5 days prior to harvesting for RNA purification . Three flasks of algae were assayed for each phosphate concentration ( from 100 pM to 10 mM phosphate in 10-fold intervals ) . For nitrogen limitation and glutamine experiments , O . amblystomatis cells were prepared as described for the phosphate limitation experiments using nitrogen depleted AF-6 media . Cultures were grown for 5 days at nitrogen concentrations approximating observed nitrate and ammonia concentrations in salamander egg capsules ( 6 . 6 µM NO3−; 17 µM NH4+ ) prior to addition of glutamine . Glutamine was added to a final concentration of 2 mM . Algal cells were harvested after 6 hr of incubation with or without glutamine and assayed for gene expression . Three flasks of algae were assayed for each condition . To prepare cDNA , O . amblystomatis cells were harvested by centrifugation ( 1 , 000xg for 5 min ) , and 350 µL lysis buffer RLT-Plus with fresh β-mercaptoethanol ( RNeasy mini Plus kit , Qiagen , Valencia , CA ) was added , and the lysate was vortexed for 30s . RNA was purified from the lysate following the manufacturer’s protocol . Purified RNA was converted to cDNA using the Quantitect RT kit ( Qiagen ) following the manufacturer’s protocol . Resultant cDNA was diluted with three volumes of 10mM Tris buffer , pH 7 . 5 , or RNase free water prior to qPCR reactions . Quantitative PCR primers for four reference genes and five response genes ( Table 3 ) were designed using conserved regions in multiple sequence alignments of cDNA sequences from the three O . amblystomatis cultured strains as well as the O . amblystomatis sequences obtained from the field material . Candidate reference genes were selected due to their utility in prior studies in the related chlorophycean alga Chlamydomonas reinhardtii , and were confirmed to have equivalent transcriptome expression levels in intracapsular and intracellular alga in this study . The reference genes were RPL32 & H2B1 ( Liu et al . , 2012 ) , RACK1 ( Mus et al . , 2007 ) , and YPTC1 ( Lake and Willows , 2003 ) . Several primer pairs were designed for each reference and response gene using the software tools GEMI ( Sobhy and Colson , 2012 ) , Primer3 ( Untergasser et al . , 2012 ) , and PrimerQuest ( Owczarzy et al . , 2008 ) . Primer pairs were validated by making standard curves using a cDNA dilution series . Primer pairs with the lowest Cq value for a given gene and PCR efficiencies between 0 . 9 and 1 . 1 in a standard curve of cDNA dilution series were validated for use in gene expression studies ( Table 3 ) . 10 . 7554/eLife . 22054 . 027Table 3 . O . amblystomatis qPCR primer sequences . Primer pairs for four reference genes ( RACK1 , YPTC1 , RPL32 , H2B1 ) , and five response genes ( PhT1 . 2 , NaPhT1 [ANTR1] , AMT1 . 2 , NRT2 . 4 , DUR3 ) used in this study . Efficiency values were measured per amplicon using a standard curve with five two-fold dilutions of cDNA . DOI: http://dx . doi . org/10 . 7554/eLife . 22054 . 027PrimerSequence ( 5ʹ−3ʹ ) EfficiencyOoph_RACK1_L_3CGCACAGCCAGTAGCGGT0 . 94Ooph_RACK1_R_3GGACCTGGCTGAGGGCAAOoph_YPTC1_L_4TTGCGGATGACACCTACACG1 . 09Ooph_YPTC1_R_4TGGTCCTGAATCGTTCCTGCOoph_RPL32_L_2ATAACAGGGTCCGCAGAAAG1 . 03Ooph_RPL32_R_2GTTGGAGACGAGGAACTTGAGOoph_H2B1_L_4CAAGAAGCCCACCATGACCT1 . 04Ooph_H2B1_R_4GGTGAACTTGGTGACTGCCTOoph_PhT1 . 2_L_4TGCCAATGACTTCGCCTTCT1 . 02Ooph_PhT1 . 2_R_4ACGTTCCACTGCTGCTTCTTOoph_NaPhT1_L_4TCCATCATCGGTCTGTCGCT0 . 99Ooph_NaPhT1_R_4GAACCACACGATGCCCAGAGOoph_AMT1 . 2_L_4CGGTCTCCTTCCAATCGCCA0 . 96Ooph_AMT1 . 2_R_4CCAATGGGTGCTGACTGGGAOoph_NRT2 . 4_L_3CGACTACCGCGACCTGAAGA1 . 03Ooph_NRT2 . 4_R_3GAACAAGACCCAGGCCCTGTOoph_DUR3_L_3GCGAATGCCGAGCACTTC1 . 02Ooph_DUR3_R_3CTGTCCCTGGGCTGGGT Quantitative PCR reactions used 1 µL of the diluted cDNA in 20 µL reactions with a 700 nM concentration of each primer using QuantiNova Sybr green ( Qiagen ) for amplification and detection . QPCR reactions were done in duplicate . Reactions were performed on a RotorGeneQ instrument ( Qiagen ) with a 2-step cycling program of 5s at 95°C and 10s at 60°C followed by melting curve analysis . Raw data was exported from the RotorGeneQ and per run-per amplicon efficiency correction was implemented in LinRegPCR ( version 2015 . 3 ) ( Ramakers et al . , 2003; Ruijter et al . , 2009 ) . Differences in expression were analyzed using ANOVA with contrasts in R . The Institutional Animal Care and Use Committee of Gettysburg College approved the research on salamander embryos ( IACUC#2013 F17 ) . Field collection of egg masses was completed under Pennsylvania Fish and Boat Commission permit ( PA-727 Type 1 ) . All transcriptome assemblies and read data are available from the NCBI transcriptome shotgun assembly database under BioProject #PRJNA326420 . Other relevant data are within the paper and its additional files .	Throughout the natural world , when different species form a close association , it is known as a symbiosis . One species can depend on another for food , defense against predators or even for reproduction . Corals , for example , incorporate single-celled algae into their own cells . The algae photosynthesize , harnessing energy from sunlight to make sugars and other molecules that also feed the coral cells . In return , corals protect the algae from the environment and provide them with the materials they need for photosynthesis . This type of relationship where one organism lives inside another species is called an endosymbiosis . In animals with a backbone , endosymbioses with a photosynthetic organism are rare . There is only one known example so far , which is between a green alga called Oophila amblystomatis and the spotted salamander , Ambystoma maculatum . The female spotted salamander deposits her eggs in pools of water , and algae enter the eggs , proliferate , and later invade tissues and cells of the developing embryos . However , it is not understood how similar the interaction between the alga and the salamander is to that in coral-algal symbioses , or whether it is rather more similar to a parasitic infection . Burns et al . now address this question by comparing salamander cells harboring algae to those that lacked algae . A technique called RNA-Seq was used to characterize the changes in gene activity that take place in both organisms during the endosymbiosis . The results show that algae inside salamander cells are stressed and they change the way in which they make energy . Instead of carrying out photosynthesis to produce food for the salamander host – as happens in coral-algal interactions – Oophila amblystomatis is fighting to adapt to its new environment and switches to a less efficient energy producing pathway known as fermentation . Burns et al . found that , in striking contrast to the alga , affected salamander cells do not show signs of stress . Instead several genes that are known to suppress immune responses against foreign invaders are expressed to high levels . This may explain how salamander cells are able to tolerate algae inside them . The next challenge is to understand how the alga enters salamander cells . The current work identified some potential routes of entry , and follow up studies are now needed to explore those possibilities .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology" ]	2017	Transcriptome analysis illuminates the nature of the intracellular interaction in a vertebrate-algal symbiosis
Production of healthy gametes in meiosis relies on the quality control and proper distribution of both nuclear and cytoplasmic contents . Meiotic differentiation naturally eliminates age-induced cellular damage by an unknown mechanism . Using time-lapse fluorescence microscopy in budding yeast , we found that nuclear senescence factors – including protein aggregates , extrachromosomal ribosomal DNA circles , and abnormal nucleolar material – are sequestered away from chromosomes during meiosis II and subsequently eliminated . A similar sequestration and elimination process occurs for the core subunits of the nuclear pore complex in both young and aged cells . Nuclear envelope remodeling drives the formation of a membranous compartment containing the sequestered material . Importantly , de novo generation of plasma membrane is required for the sequestration event , preventing the inheritance of long-lived nucleoporins and senescence factors into the newly formed gametes . Our study uncovers a new mechanism of nuclear quality control and provides insight into its function in meiotic cellular rejuvenation . Aging occurs as an organism loses its ability to maintain homeostasis over time . The cellular changes that accompany aging have been most extensively characterized in the budding yeast , Saccharomyces cerevisiae ( Figure 1A; Denoth Lippuner et al . , 2014; Kaeberlein , 2010; Longo et al . , 2012 ) . Disrupted protein homeostasis results in the accumulation of protein aggregates that contain oxidatively damaged proteins ( Aguilaniu et al . , 2003; Erjavec et al . , 2007 ) . Many organelles exhibit signs of dysfunction: mitochondria fragment and aggregate , mitochondrial membrane potential decreases , and the vacuole becomes less acidic ( Henderson et al . , 2014; Hughes and Gottschling , 2012; Veatch et al . , 2009 ) . Notably , the nucleus also undergoes a number of changes including enlargement of the nucleolus ( Lewinska et al . , 2014; Morlot et al . , 2019; Sinclair et al . , 1997 ) , misorganization of nuclear pore complexes ( Lord et al . , 2015; Rempel et al . , 2019 ) , and accumulation of extrachromosomal ribosomal DNA ( rDNA ) circles ( Denoth-Lippuner et al . , 2014; Sinclair and Guarente , 1997 ) . Many of the cellular changes that accrue with age are conserved across eukaryotes ( Colacurcio and Nixon , 2016; David et al . , 2010; Sun et al . , 2016; Tiku et al . , 2017 ) . In budding yeast mitosis , age-induced damage is asymmetrically retained by the mother cell resulting in the formation of an aged mother cell and a young daughter cell ( Mortimer and Johnston , 1959 ) . In contrast , meiotic cells reset aging symmetrically such that all of the meiotic products are born young , independent of their progenitor’s age ( Unal et al . , 2011 ) . Importantly , senescence factors originally present in the aged precursor cells , including protein aggregates , nucleolar damage , and rDNA circles , are no longer present in the newly formed gametes ( Ünal and Amon , 2011; Unal et al . , 2011 ) . How gametes avoid inheriting age-associated damage and how this event is coupled to the meiotic differentiation program remains unknown . Meiotic differentiation , also known as gametogenesis , is a tightly regulated developmental program whereby a progenitor cell undergoes two consecutive nuclear divisions , meiosis I and meiosis II , to form haploid gametes . Meiotic differentiation requires extensive cellular remodeling to ensure that gametes inherit the necessary nuclear and cytoplasmic contents . In yeast gametogenesis , the nucleus undergoes a closed division , with the nuclear envelope remaining continuous until karyokinesis forms four new nuclei ( Moens , 1971; Moens and Rapport , 1971; Neiman , 2011 ) . Mitochondria and cortical endoplasmic reticulum also undergo regulated morphological changes , separating from the cellular cortex and localizing near the nuclear envelope at the transition between meiosis I and II ( Gorsich and Shaw , 2004; Miyakawa et al . , 1984; Sawyer et al . , 2019; Stevens , 1981; Suda et al . , 2007 ) . Around the same time , new plasma membranes , also known as prospore membranes , grow from the centrosome-like spindle pole bodies embedded in the nuclear envelope . This directed growth of plasma membrane ensures that nascent nuclei and a fraction of the cytoplasmic contents are encapsulated to form gametes ( Brewer et al . , 1980; Byers , 1981; Knop and Strasser , 2000; Moens , 1971; Neiman , 1998 ) . Subsequently , the uninherited cellular contents are destroyed by proteases released upon permeabilization of the progenitor cell’s vacuole , the yeast equivalent of the mammalian lysosome ( Eastwood et al . , 2012; Eastwood and Meneghini , 2015 ) . Whether these cellular remodeling events are integral to the removal of age-induced damage has not been characterized . In this study , we aimed to determine the mechanism by which nuclear senescence factors are eliminated during budding yeast meiosis . Using time-lapse fluorescence microscopy , we found that protein aggregates , rDNA circles , and a subset of nucleolar proteins are sequestered away from chromosomes during meiosis II . Importantly , we show that the core subunits of the nuclear pore complex ( NPC ) also undergo a similar sequestration process in both young and aged cells . The damaged material localizes to a nuclear envelope-bound compartment containing the excluded NPCs that is eliminated upon vacuolar lysis . Finally , we found that the proper development of plasma membranes is required for the sequestration of core NPCs and senescence factors away from the newly forming gametes . Our study defines a key nuclear remodeling event and demonstrates its involvement in the elimination of age-induced cellular damage during meiotic differentiation . To gain a deeper understanding of gametogenesis-induced rejuvenation , we first sought to characterize the meiotic dynamics of age-induced protein aggregates , rDNA circles , and nucleolar damage using time-lapse fluorescence microscopy . To isolate aged cells , we employed a previously established protocol that uses pulse-labeling of cells with biotin followed by harvesting with anti-biotin magnetic beads ( Boselli et al . , 2009; Smeal et al . , 1996 ) . All three types of damage have been reported to localize to the nuclear periphery in aged mitotic cells ( Cabrera et al . , 2017; Denoth-Lippuner et al . , 2014; Saarikangas et al . , 2017; Sinclair et al . , 1997 ) . Therefore , we monitored their meiotic localization relative to chromosomes , marked with a fluorescently tagged chromatin protein: either histone H2A ( Hta1 ) or histone H2B ( Htb1 ) . Similar to mitosis , we observed that protein aggregates , visualized by the fluorescently tagged chaperone Hsp104-eGFP ( Glover and Lindquist , 1998; Saarikangas et al . , 2017 ) , localized to a perinuclear region inside the nucleus prior to the meiotic divisions and in meiosis I ( Figure 1B , right panel; Figure 1—figure supplement 1; Figure 1—figure supplement 1—source data 1 ) . In contrast , during meiosis II , the protein aggregates localized away from chromosomes , a phenomenon we termed sequestration ( Figure 1B and D; Figure 1—source data 1; Video 1 ) . The sequestration was highly penetrant ( >99% ) and occurred with consistent timing shortly after the onset of anaphase II ( Figure 1D; Figure 1—source data 1 ) . Subsequently , the aggregates disappeared late in gametogenesis ( Figure 1B , right panel; Video 1 ) . By comparison , young cells did not contain any Hsp104-associated aggregates but instead displayed diffuse Hsp104 signal throughout meiosis ( Figure 1B , left panel ) . We conclude that age-associated protein aggregates undergo stereotypical sequestration and elimination during meiotic differentiation , suggesting developmentally controlled induction of these events . We next tested whether the extrachromosomal rDNA circles that accumulate in aged cells displayed a similar behavior . To visualize ribosomal DNA in single cells , we used a strain carrying five tandem copies of the tetracycline operator sequence integrated within each rDNA repeat in one of the two chromosome XII homologs ( tetO-rDNA ) . The strain additionally contained a tetracycline repressor protein fused to GFP ( TetR-GFP ) under the control of a meiotic promoter ( Li et al . , 2011 ) . These two modifications , namely the meiosis-restricted expression of TetR-GFP and the heterozygosity of the tetO-rDNA array , did not affect growth rate in vegetative cells . Using this method , we observed that , in aged cells , a substantial fraction of the tetO-rDNA/TetR-GFP signal and a small fraction of the Hta1-mApple signal were sequestered away from the dividing chromosomes after the onset of anaphase II and disappeared during late stages of gamete maturation ( Figure 1C , right panel; Figure 1E; Figure 1—source data 2; Video 2 ) . By comparison , in young cells , the gamete nuclei retained the entire tetO-rDNA array and histone-bound chromatin after completion of anaphase II ( Figure 1C , left panel ) , consistent with previous work ( Fuchs and Loidl , 2004; Li et al . , 2011 ) . In aged cells carrying TetR-GFP without the tetO-rDNA array , the GFP signal remained diffuse throughout meiosis ( Figure 1—figure supplement 2 ) , confirming that the extrachromosomal GFP puncta were due to sequestered rDNA circles as opposed to TetR-GFP aggregation . These findings demonstrate that , similar to age-associated protein aggregates , extrachromosomal rDNA circles also undergo programmed sequestration and destruction during meiotic differentiation . In addition to rDNA circles , other nucleolar aberrations also accumulate during cellular aging . As a mother cell continues to divide mitotically , factors involved in ribosomal biogenesis are upregulated , leading to the formation of enlarged and fragmented nucleoli ( Janssens et al . , 2015; Morlot et al . , 2019; Sinclair et al . , 1997 ) . To visualize nucleoli in more detail , we fluorescently tagged the rRNA processing factor Nsr1 at its endogenous locus ( Lee et al . , 1992 ) . A previous study found that two other rRNA processing factors , the fibrillarin homolog Nop1 and the high mobility group protein Nhp2 , are partially sequestered away from chromosomes during gametogenesis ( Fuchs and Loidl , 2004 ) . Nsr1 similarly demonstrated partial sequestration after the onset of anaphase II in young cells ( Figure 2A ) . In aged cells , Nsr1 foci appeared enlarged and fragmented prior to the meiotic divisions , consistent with previously reported changes in nucleolar morphology ( Figure 2B; Janssens et al . , 2015; Morlot et al . , 2019; Sinclair et al . , 1997 ) . As in young cells , Nsr1 was sequestered away from chromosomes following the onset of anaphase II and subsequently eliminated ( Figure 2B–C; Figure 2—source data 1; Video 3 ) . Interestingly , a significantly higher fraction of the total Nsr1 was sequestered in older cells ( mean = 23% for 0–3 generation-old cells , 36% for 5–8 generation-old cells and 42% for nine or more generation-old cells; Figure 2D; Figure 2—source data 2 ) . A portion of the histone H2B ( Htb1-mCherry ) was also sequestered away from the gamete nuclei , reminiscent of the behavior of histone H2A in the GFP-marked rDNA strain . This chromatin demarcation occurred predominantly in aged cells and always co-localized with the sequestered nucleoli . Since the extrachromosomal histone mass is present in aged cells independent of the GFP-marked rDNA array , the discarded rDNA circles are likely assembled into chromatin , and the extrachromosomal histone signal can be used as a proxy for rDNA circles . Finally , we analyzed the behavior of protein aggregates with respect to nucleoli and found that both the timing and location of the sequestration event were coincident ( Figure 2E–F; Figure 2—source data 3 ) . Taken together , these data reveal that distinct types of age-induced damage all undergo a spatiotemporally linked sequestration and elimination process , suggesting a common mode of meiotic regulation . Since nucleolar constituents localize away from dividing chromosomes even in young cells , we reasoned that the sequestration of age-induced nuclear damage might involve a nuclear remodeling event that takes place generally as part of meiotic differentiation . As a means of assessing nuclear behavior , we sought to characterize the dynamics of nuclear pore complexes ( NPCs ) during meiosis in young cells . Nuclear pore complexes are large protein structures that span the nuclear envelope and primarily function in selective nucleocytoplasmic transport . NPCs contain multiple copies of at least 30 distinct types of proteins termed nucleoporins . Nucleoporins are organized into different subcomplexes with distinct structural roles ( Beck and Hurt , 2017; Kim et al . , 2018 ) . Intriguingly , one nucleoporin , Nsp1 , has been previously shown to localize away from chromosomes in meiosis II ( Fuchs and Loidl , 2004 ) . Using time-lapse microscopy , we surveyed the meiotic dynamics and localization of 17 different endogenously GFP-tagged nucleoporins representing different subcomplexes ( Figure 3A ) . We found that nucleoporins from five of the six tested subcomplexes , including those most integral to the NPC structure , exhibited sequestration and elimination similar to age-induced damage . The nucleoporins localized to the nuclear periphery before the meiotic divisions and during meiosis I , but largely localized away from chromosomes after the onset of anaphase II ( Figure 3B–F; Figure 3—figure supplements 1–5; Video 4 ) . Although a large fraction of the nucleoporins persisted away from the chromosomes , some nucleoporins re-appeared around the gamete nuclei , either by de novo synthesis or return of the pre-existing pool . Several hours after the meiotic divisions , any remaining nucleoporin signal outside of the gamete nuclei abruptly disappeared ( Figure 3B–F; Figure 3—figure supplements 1–5; Video 4 ) . Interestingly , the nucleoporins from one subcomplex , the nuclear basket , exhibited a markedly different behavior: although briefly localizing outside of the developing nuclei during anaphase II along with the nucleoporins from other subcomplexes , they largely returned to the nascent nuclei within 30 min ( Figure 3G–H; Figure 3—figure supplement 6; Video 5 ) . The simplest interpretation of these findings was that the nuclear basket detached from the rest of the NPC during meiosis II . Given that all other NPC subcomplexes tested persist outside of developing nuclei , we propose that intact NPCs without nuclear baskets are left outside of gamete nuclei . Since senescence factors and NPCs were sequestered with similar timing , we next asked whether they were sequestered to a similar location . We monitored the localization of protein aggregates , rDNA circles , and sequestered nucleolar material relative to NPCs and found that they co-localize with the sequestered NPCs after the onset of anaphase II ( Figure 4A–C; Figure 4—figure supplements 1–2 ) . These results suggest that a common nuclear remodeling event is responsible for the spatial separation of various nuclear components from the dividing chromosomes . The nuclear envelope remains continuous during budding yeast meiosis , dynamically changing shape to accommodate the chromosomal divisions ( Moens , 1971; Moens and Rapport , 1971 ) . After the second meiotic division , karyokinesis occurs to form the nascent nuclei of the four gametes . Given the abrupt change in NPC distribution during anaphase II , we sought to determine how other nuclear membrane proteins behave during this time . We found that the integral LEM-domain protein Heh1 ( Gene ID: 854974 ) and a generic inner nuclear membrane ( INM ) marker , eGFP-h2NLS-L-TM , localized to both nascent gamete nuclei and the sequestered NPCs during anaphase II ( Figure 5A–B; Figure 5—figure supplement 1; King et al . , 2006; Meinema et al . , 2011 ) , suggesting the existence of a separate membranous compartment . We next performed serial section transmission electron microscopy ( TEM ) to observe this compartment directly . Reconstructions of individual cells , either during anaphase II or during gamete development , confirmed the existence of nuclear envelope-bound space outside of the four nuclei ( Figure 5C–D; Videos 6–9 ) . The compartment seemed deformed in comparison to the nascent gamete nuclei in that the nuclear envelope membrane structure appeared abnormal and the compartment was often fragmented into multiple nuclear envelope-bound regions ( Figure 5C–5F; Videos 6–9 ) . These regions were located outside of the gamete plasma membranes , also known as prospore membranes ( Figure 5C–D; Videos 6–9 ) . Importantly , individual sections showed that the compartment contained nucleolar material and NPCs ( Figure 5E and F; Figure 5—figure supplement 2; Videos 6 and 8 ) . We conclude that , during meiosis II , the nuclear envelope undergoes a five-way division to form the four nuclei and a separate compartment containing discarded nuclear proteins . The TEM analyses showed that the nuclear envelope-bound compartment localized outside of the developing gamete plasma membranes ( Figure 5C–D; Videos 6–9 ) . It remained unclear , however , how the material was sequestered into this compartment . At least two models could explain how the material was left outside of the nascent gametes: ( 1 ) the material was being ‘extruded , ’ removed from the gamete after initial entry , or ( 2 ) ‘excluded , ’ never entering the nascent gametes . To differentiate between these models , we analyzed the localization of a gamete-specific plasma membrane ( PM ) marker , yeGFP-Spo2051-91 ( Nakanishi et al . , 2004 ) , relative to NPCs and chromosomes . We found that , throughout anaphase II , a sequestered mass of nucleoporins was constrained to a region immediately outside of the nascent plasma membranes and never appeared inside ( Figure 6A ) . The lip of the developing plasma membranes marked by Don1-GFP neatly delineated the boundary of the NPC mass ( Figure 6B ) . Live-cell microscopy confirmed that the NPCs remained outside of nascent plasma membranes throughout their development , supporting ‘exclusion’ as the means by which nuclear material remained outside of the developing gametes ( Figure 6—figure supplements 1–2 ) . To determine if senescence factors were similarly excluded , we monitored the localization of protein aggregates and nucleolar material relative to the gamete plasma membranes . This analysis revealed that age-induced damage almost never entered into newly forming gametes ( Figure 6C; Figure 6—figure supplement 3; Video 10 ) . Only one out of several hundred gametes inherited the Hsp104-associated protein aggregates ( Figure 6D; Video 11 ) ; strikingly , this Hsp104 punctum persisted after gamete maturation , suggesting that the elimination of age-associated damage is dependent on its prior exclusion . These results highlight the existence of an active mechanism in meiotic cells that precludes the inheritance of NPCs and senescence factors by the nascent gametes . Following gamete formation , permeabilization of the precursor cell’s vacuolar membrane causes the release of proteases , which degrade the cellular contents left in the precursor cell cytosol in a process termed mega-autophagy ( Eastwood et al . , 2012; Eastwood and Meneghini , 2015 ) . To determine whether mega-autophagy was responsible for the degradation of the excluded nuclear material , we monitored the disappearance of NPCs and age-associated protein aggregates relative to the lysis of the vacuolar membrane as monitored by either Vph1-eGFP or Vph1-mCherry . We found that both events coincided with the onset of vacuolar lysis ( Figure 7A–D; Figure 7—source data 1–2; Videos 12–13 ) . To further assess nucleoporin degradation , we measured the protein levels of GFP-tagged Nup84 and Nup170 by immunoblotting ( Figure 7E–F ) . Since GFP is relatively resistant to vacuolar proteases , degradation of tagged proteins leads to the accumulation of free GFP ( Kanki and Klionsky , 2008 ) . We found that free GFP accumulated in wild-type cells 12 hours after meiosis induction , consistent with vacuolar proteases driving the elimination of Nup84 and Nup170 ( Figure 7E–F ) . Importantly , we confirmed that the degradation of both nucleoporins depends on the meiotic transcription factor Ndt80 . Ndt80 is a master transcription factor necessary for the meiotic divisions and gamete maturation ( Xu et al . , 1995 ) . In the absence of NDT80 , cells exhibit a prolonged arrest during prophase I and fail to undergo vacuolar lysis ( Eastwood et al . , 2012 ) . Altogether , these analyses highlight mega-autophagy as the probable degradation mechanism for NPCs and nuclear senescence factors . What drives the nuclear remodeling event in meiotic cells ? Given that the boundaries of the excluded NPC mass co-localize with the lips of developing gamete plasma membranes ( Figure 6B ) , we posited that plasma membrane development itself was required for NPC sequestration . To test this hypothesis , we monitored NPC localization in mutants with disrupted plasma membrane formation . Plasma membrane development is initiated from the cytoplasmic face of the spindle pole body , which is converted from a microtubule-nucleation center to a membrane-nucleation center during meiosis II ( Knop and Strasser , 2000 ) . SPO21 ( also known as MPC70 ) is required for this conversion event , and its deletion completely inhibits de novo plasma membrane formation ( Knop and Strasser , 2000 ) . We found that , in spo21Δ cells , nucleoporins remained around chromosomes during anaphase II instead of being sequestered away ( Figure 8A–B; Figure 8—source data 1; Video 14 ) . As an independent test of the role of plasma membrane development in NPC remodeling , we perturbed plasma membrane development by an orthogonal method . The formation of fewer than four plasma membranes can be induced by low carbon conditions , since carbon concentration affects the conversion of the spindle pole body into a membrane nucleator ( Davidow et al . , 1980; Okamoto and Iino , 1981; Taxis et al . , 2005 ) . Under such conditions , we found that the gamete nuclei displayed reciprocal localization of plasma membranes and NPCs: only the nuclei that were devoid of plasma membranes were enriched for NPCs ( Figure 8C–E; Figure 8—source data 2 ) . This was consistent with the observation that , even in high carbon conditions , cells fated to form three or two mature gametes would often have one or two nuclei enriched for NPCs , respectively ( Figure 8B; Figure 8—source data 1 ) . Finally , we examined how defects in leading edge complex formation , the structure that forms at the lip of the developing plasma membranes , affect NPC sequestration . Specifically , the absence of the organizing member Ssp1 or simultaneous deletion of the Ady3 and Irc10 subunits results in the formation of misshapen plasma membranes ( Lam et al . , 2014; Moreno-Borchart et al . , 2001 ) . We found that both ssp1Δ and ady3Δ irc10Δ cells had defective NPC sequestration , with NPCs often remaining partially around anaphase II nuclei ( Figure 8F; Figure 8—figure supplements 1–2 ) . The boundary of NPC removal from the nuclei was marked by constrictions in the DAPI or histone signal and corresponded to the extent of plasma membrane formation ( Figure 8F; Figure 8—figure supplements 1–2 ) . Taken together , these data support the conclusion that NPC sequestration and exclusion are driven by the development of plasma membranes around nascent gamete nuclei . Since the sequestration of NPCs and nuclear senescence factors were spatially and temporally coupled , we reasoned that a common mechanism could mediate both events . We therefore monitored the sequestration of protein aggregates in spo21Δ cells , which are defective in NPC sequestration . In comparison to wild-type cells , we found that spo21Δ mutants exhibited a dramatic increase in the association of protein aggregates with chromosomes during anaphase II ( 48% vs . 0%; Figure 9A–B; Figure 9—source data 1 ) . Regardless of whether or not the protein aggregate was sequestered away from chromosomes in spo21∆ cells , the protein aggregates always co-localized with the nuclear envelope , as marked by NPCs ( Figure 9—figure supplement 1 ) . Thus , without the nascent plasma membranes , protein aggregates appeared randomly distributed along the nuclear periphery . We next assessed how nucleolar sequestration is affected in young spo21Δ cells . We found that in 39% of spo21Δ cells , nucleoli failed to be sequestered in meiosis II and instead co-segregated with chromosomes ( Figure 9C–D; Figure 9—source data 2 ) . In contrast , none of the wild-type cells displayed this behavior . Furthermore , Nsr1 remained co-localized to the nuclear envelope in spo21Δ cells in a similar manner to protein aggregates ( Figure 9—figure supplement 2 ) . Altogether , these findings support the notion that meiotic exclusion of age-induced protein aggregates and nucleolar material is coupled to a nuclear remodeling event that is driven by gamete plasma membrane formation . We found that a subset of nuclear components are sequestered away from chromosomes during anaphase II: core nucleoporins , nucleolar proteins involved in rRNA transcription and processing , extrachromosomal rDNA circles , and protein aggregates . However , other nuclear proteins – including histones , the rDNA-associated protein Cfi1 , and the Ran exchange factor Prp20 – are largely retained with dividing nuclei during anaphase II ( unpublished data ) . A more thorough cataloging of nuclear components is needed to identify parameters that differentiate excluded nuclear material from retained nuclear material . Strong association with chromatin , as in the case for histones , Cfi1 and Prp20 , may be one way to mediate the selective inheritance of nuclear proteins into gametes ( Aebi et al . , 1990; Dilworth et al . , 2005; Li et al . , 2003; Straight et al . , 1999 ) . On the other hand , strong association with NPCs may facilitate sequestration – for example , extrachromosomal rDNA circles have been shown to interact with NPCs in mitotic cells ( Denoth-Lippuner et al . , 2014 ) . It is currently unclear whether a mechanism exists to enrich for damaged proteins in the sequestered pool , such that any proteins remaining with the gamete nuclei are preferentially undamaged . Since some nucleoporins and nucleolar proteins are sequestered and eliminated in young cells , it is likely that undamaged proteins are destroyed during meiosis for reasons that are yet to be determined . However , the observation that the fraction of discarded nucleolar proteins is higher in aged cells than young cells is consistent with the possibility that damaged nucleolar proteins are selectively enriched in the sequestered material . Since both nucleolar proteins and NPCs have been shown to accumulate age-related damage ( Denoth-Lippuner et al . , 2014; Lord et al . , 2015; Rempel et al . , 2019; Sinclair et al . , 1997; Morlot et al . , 2019 ) , selective elimination and subsequent de novo synthesis could be vital to ensuring gamete rejuvenation . Unexpectedly , we found that nuclear basket nucleoporins dissociate from the rest of the nuclear pore complex and remain with nascent nuclei during meiosis II . Consistent with this finding , nuclear basket nucleoporins have been shown to be more dynamic than core nucleoporins ( Denning et al . , 2001; Dilworth et al . , 2001; Niepel et al . , 2013 ) and sub-populations of NPCs without certain nuclear basket nucleoporins are present near the nucleolus ( Galy et al . , 2004 ) . We propose that the nuclear basket segregates with gamete nuclei through re-association with chromatin , which in turn facilitates the formation of new NPCs . In both the fungus Asperilligus nidulans and vertebrates , the nuclear basket nucleoporin Nup2 and its metazoan ortholog Nup50 associate with dividing chromatin during mitosis and contribute to the segregation of NPCs into daughter nuclei ( Dultz et al . , 2008; Markossian et al . , 2015; Suresh et al . , 2017 ) . The nuclear basket nucleoporins Nup1 and Nup60 have innate membrane binding and shaping capabilities , making them attractive candidates to initiate insertion of new NPCs ( Mészáros et al . , 2015 ) . Indeed , deletion of non-essential nuclear basket nucleoporins results in reduced sporulation efficiency and impaired gamete viability , supporting an important functional role during the meiotic program ( Chu et al . , 2017 ) . We found that gamete plasma membrane formation is required for the selective sequestration of nuclear contents . When plasma membrane development is prevented , NPCs are retained and age-induced damage becomes randomly distributed along the nuclear periphery . The mechanism by which the newly forming plasma membrane creates distinct nuclear envelope domains inside and outside of developing gametes remains unclear . A direct physical blockade , while possible , seems unlikely given that large organelles such as mitochondria enter through the lips of developing plasma membranes ( Byers , 1981; Suda et al . , 2007 ) . On the other hand , the sequestration boundary at the leading edge is reminiscent of the outer nuclear envelope lateral diffusion barrier that forms at the bud neck during budding yeast mitosis ( Caudron and Barral , 2009; Clay et al . , 2014 ) . In this context , septins localize to the bud neck and organize a signaling cascade , generating a sphingolipid region in the nuclear envelope that constrains the movement of nuclear envelope proteins ( Clay et al . , 2014 ) . In meiosis , deletion of meiosis-specific septins ( spr3Δ and spr28Δ; De Virgilio et al . , 1996; Fares et al . , 1996; Ozsarac et al . , 1995 ) and leading edge complex components ( ady3Δ , irc10Δ , and don1Δ; Knop and Strasser , 2000; Lam et al . , 2014; Moreno-Borchart et al . , 2001 ) does not grossly alter NPC or protein aggregate sequestration , beyond impacting plasma membrane morphology ( Figure 8F; Supplementary file 5 ) . However , an unidentified scaffold might exist to organize a nuclear envelope diffusion barrier . Determining the mechanism by which gamete plasma membranes sequester nuclear material will reveal important principles of nuclear organization and compartmentalization . After the sequestration event , core nucleoporins begin to re-appear around nascent gamete nuclei , either by de novo synthesis or return from the sequestered mass . This raises the intriguing possibility that some core nucleoporins may be able to overcome the physical or diffusion barrier imposed by the plasma membrane . In mitosis , an active transmission mechanism involving the nucleoporin Nsp1 is required for NPCs to pass the bud neck diffusion barrier and enter into daughter cells ( Colombi et al . , 2013; Makio et al . , 2013 ) . Whether any factors facilitate selective NPC inheritance into gametes during meiosis II is an important direction for future studies . Further , daughter-inherited NPCs are modified by the deacetylase Hos3 as they pass through the bud neck , resulting in the formation of a daughter nucleus with distinct cell-cycle behaviors from the mother nucleus ( Kumar et al . , 2018 ) . Transmission through the leading edge of the gamete plasma membrane might similarly provide an opportunity for any inherited NPCs to acquire gamete-specific modifications and functions . Further characterizing the reintegration of core nucleoporins at the gamete nuclear periphery will improve our understanding of how NPC remodeling contributes to gamete fitness . The sequestration of nuclear damage into a membranous compartment outside of gametes makes it accessible to the degradation machinery active in the progenitor cell cytoplasm during gamete maturation . Due to the strong correlation between the timing of vacuolar lysis and the disappearance of sequestered material as well as the meiosis-specific vacuolar degradation of nucleoporins , we propose that mega-autophagy is responsible for the elimination of nuclear senescence factors ( Eastwood et al . , 2012; Eastwood and Meneghini , 2015 ) . The release of proteases from the vacuole could eliminate protein aggregates and other sequestered nuclear proteins , as has already been observed for unsuccessfully packaged nuclei ( Eastwood et al . , 2012 ) . Another mechanism , however , is necessary for the elimination of rDNA circles . The endonuclease G homolog , Nuc1 , is released from mitochondria during mega-autophagy and therefore could be responsible for the elimination of rDNA circles ( Eastwood et al . , 2012 ) . Our study highlights a mechanism that facilitates the elimination of age-induced damage during meiosis . Given that extensive nuclear remodeling occurs even in young cells , the reorganization of the nuclear periphery appears to be integral to gamete fitness . Importantly , the sequestration of NPCs in budding yeast meiosis is similar to a NPC reorganization event observed in the spermatogenesis of metazoans , including humans ( Fawcett and Chemes , 1979; Ho , 2010; Troyer and Schwager , 1982 ) . In this context , acrosome formation , potentially akin to gamete plasma membrane formation , corresponds to the redistribution of nuclear pores to the caudal end of the nucleus , coincident with chromatin condensation and elimination of un-inherited nuclear material . Whether removal of age-induced damage is also coupled to nuclear remodeling during metazoan spermatogenesis remains to be determined . Elimination of age-induced damage during gamete maturation may also be integral to gamete rejuvenation in other organisms . In C . elegans gametogenesis , oocyte maturation involves the elimination of age-induced protein aggregates by lysosomal activation ( Bohnert and Kenyon , 2017; Goudeau and Aguilaniu , 2010 ) . Further determining the mechanism of age-induced damage sequestration and elimination could aid in the development of strategies to counteract cellular aging in somatic cells . The selective inheritance of distinct types of age-induced damage could provide a means of determining whether a given senescence factor is a cause or consequence of aging . In this manner , meiotic differentiation offers a unique and natural context to uncover quality control mechanisms that eliminate the determinants of cellular aging . All strains in this study are derivatives of SK1 and specified in Supplementary file 1 . Strains UB17338 , UB17509 and UB17532 are derivatives of strain HY2545 ( a gift from Dr . Hong-Guo Yu ) . Deletion and C-terminal tagging at endogenous loci were performed using previously described PCR-based methods unless otherwise specified ( Janke et al . , 2004; Longtine et al . , 1998; Sheff and Thorn , 2004 ) . Deletion of SSP1 was performed by transforming cells with a PCR amplicon of the locus from the SK1 yeast deletion collection ( a gift from Dr . Lars Steinmetz ) . Primer sequences used for strain construction are specified in Supplementary file 2 , and plasmids used for strain construction are specified in Supplementary file 3 . The following strains were constructed in a previous paper: flo8Δ ( Boselli et al . , 2009 ) , Htb1-mCherry ( Matos et al . , 2008 ) , Hsp104-eGFP ( Unal et al . , 2011 ) , ndt80Δ ( Xu et al . , 1995 ) , and spo21Δ ( Sawyer et al . , 2019 ) . To visualize the vacuole , we used either an eGFP-tagged version of Vph1 integrated at the HIS3 locus or a mCherry-tagged version of Vph1 at its endogenous locus . To generate the eGFP-tagged version , we amplified the W303 genomic region from 1000 bp upstream to immediately before the stop codon of VPH1 ( 2520 bp after the ORF start ) and fused it to yeGFP in the HIS3 integrating plasmid pNH603 ( a gift from Leon Chan ) . We then performed integration at the HIS3 locus by cutting the plasmid with PmeI . To generate the mCherry-tagged version , we constructed a new HIS3-selectable mCherry plasmid by replacing eGFP in pYM28 ( Janke et al . , 2004 ) with mCherry . We then tagged the locus via traditional PCR-based methods . To visualize the nuclear envelope , we generated an inner nuclear membrane-localizing reporter ( eGFP-h2NLS-L-TM ) by fusing eGFP and amino acids 93 to 378 of Heh2 ( Gene ID: 852069; Meinema et al . , 2011 ) under control of pARO10 in the LEU2 integrating plasmid pLC605 ( a gift from Leon Chan ) . To visualize the gamete plasma membranes , we used a reporter consisting of amino acids 51 to 91 from Spo20 fused to the C terminus of link-yeGFP under control of pATG8 in a LEU2 integrating plasmid ( Sawyer et al . , 2019 ) . We also constructed a new variant with mKate2 in place of yeGFP . All LEU2 integration constructs were integrated into the genome by cutting the plasmids with PmeI . Sporulation was induced using the traditional starvation method unless otherwise indicated . Diploid cells were first grown in YPD ( 1% yeast extract , 2% peptone , 2% glucose , 22 . 4 mg/L uracil , and 80 mg/L tryptophan ) at room temperature for around 24 hr until the cultures reached a cell density of OD600 ≥10 . The cultures were then diluted in BYTA ( 1% yeast extract , 2% bacto tryptone , 1% potassium acetate , and 50 mM potassium phthalate ) to OD600 = 0 . 25 and grown for 12–16 hr at 30°C . After reaching an OD600 ≥5 , the cells were pelleted , washed in sterile MilliQ water , and resuspended in the sporulation media SPO to OD600 = 1 . 85 . SPO was 0 . 5% potassium acetate alone , 1% potassium acetate alone , or 2% potassium acetate supplemented with amino acids ( 40 mg/L adenine , 40 mg/L uracil , 10 mg/L histidine , 10 mg/L leucine and 10 mg/L tryptophan ) ; the media’s pH was adjusted to seven with acetic acid and 0 . 02% raffinose was sometimes added to improve sporulation . Meiotic cultures were shaken at 30°C for the duration of the experiment . At all stages , the flask size was 10 times the culture volume to ensure proper aeration . To selectively enrich for the formation of dyads and triads , diploid cells were induced to sporulate in reduced carbon media ( Eastwood et al . , 2012 ) . Cells were grown in YPD and BYTA as described above and then resuspended in SPO with reduced potassium acetate ( 0 . 1% potassium acetate ) to an OD600 = 1 . 85 . After 5 hr at 30°C , the cells were then pelleted , washed in sterile MilliQ , and resuspended in 0 . 15% KCl . Aged cells were enriched using a biotin-labeling and magnetic-sorting assay ( Smeal et al . , 1996 ) . Cells were grown in YPD at room temperature or 30°C overnight until saturation ( OD600 ≥10 ) and then diluted to a cell density of OD600 = 0 . 2 in a new YPD culture . Cells were harvested before the cultures reached OD600 = 1 and were labeled with 8 mg/ml EZ-Link Sulfo-NHS-LC-biotin ( ThermoFisher Scientific ) for 30 min at 4°C . Biotinylated cells were grown for 12-16 hours in YPD with 100 μg/ml ampicillin at 30°C . Cells were subsequently harvested and mixed with 100 μl of anti-biotin magnetic beads ( Miltenyi Biotechnology ) for 15 min at 4°C . Cells were washed with PBS pH 7 . 4 , 0 . 5% BSA buffer and sorted magnetically using LS depletion columns with a QuadroMacs sorter following the manufacturer's protocol . A fraction of the flow-through ( biotin-negative ) was kept as young cells and was budscar labeled with eluted aged cells ( biotin-positive ) for 20 min at room temperature using 1 μg/ml Wheat Germ Agglutinin , Alexa Fluor 350 Conjugate ( ThermoFisher Scientific ) . A mixture of aged and young cells was subsequently washed twice in H2O and once with SPO ( 0 . 5% or 1% potassium acetate , 0 . 02% raffinose , pH 7 ) . The cell mixture was resuspended with SPO at a cell density of OD600 = 1 . 85 with 100 μg/ml ampicillin and incubated at 30°C . The number of doublings in subsequent experiments was measured by counting the number of budscars . Images were acquired using a DeltaVision Elite wide-field fluorescence microscope ( GE Healthcare ) . Live cell images were generated using a 60x/1 . 42 NA oil-immersion objective; fixed cell images were generated using a 100x/1 . 40 NA oil-immersion objective . Specific imaging conditions for each experiment are indicated in Supplementary file 4 . Images were deconvolved using softWoRx imaging software ( GE Healthcare ) . Unless otherwise noted , images were maximum intensity z-projected over the range of acquisition in FIJI ( RRID:SCR_002285 , Schindelin et al . , 2012 ) . Live cells were imaged in an environmental chamber heated to 30°C , using either the CellASIC ONIX Microfluidic Platform ( EMD Millipore ) or concanavalin A-coated , glass-bottom 96-well plates ( Corning ) . All live imaging experiments used conditioned sporulation media ( SPO filter-sterilized after five hours of sporulation at 30°C ) , as this was found to enhance meiotic progression . With the CellASIC system , cultures in SPO ( OD600 = 1 . 85 ) were transferred to a microfluidic Y04D plate and were loaded with a pressure of 8 psi for 5 s . Conditioned SPO was subsequently applied with a constant flow rate pressure of 2 psi for 15–20 hr . With the 96-well plates , cells were adhered to the bottom of the wells and 100 μl of conditioned SPO was added to each well . Images were acquired every 15 min for 15–18 hr . Fixed cells were prepared by treating 500–1000 μl of meiotic culture with 3 . 7% formaldehyde for 15 min at room temperature . Cells were permeabilized with either 1% Triton X-100 or 70% ethanol . ( 1 ) For Figure 4—figure supplements 1 and 2 , cells were washed with 0 . 1 M potassium phosphate pH 6 . 4 and subsequently treated with 0 . 05 μg DAPI and 1% Triton in KPi sorbitol ( 0 . 1 M potassium phosphate , 1 . 2 M sorbitol , pH 7 . 5 ) . Cells were then immediately washed with KPi sorbitol before imaging . ( 2 ) For Figure 6A–6B , cells were treated for five minutes with 1% Triton in KPi sorbitol and then resuspended in KPi sorbitol . Cells were then adhered on a poly-lysine treated multi-well slide and mounted with Vectashield Mounting Medium with DAPI ( Vector Labs ) . ( 3 ) For Figures 5A–5B , 8C–8D and F , cells were washed with 0 . 1 M potassium phosphate pH 6 . 4 and then resuspended in KPi sorbitol buffer . Cells were then adhered to a poly-lysine treated multi-well slide , quickly permeabilized with 70% ethanol , and mounted with Vectashield Mounting Medium with DAPI ( Vector Labs ) . Yeast cells were concentrated by vacuum filtration onto a nitrocellulose membrane and then scrape-loaded into 50- or 100- µm-deep high pressure freezing planchettes ( McDonald and Müller-Reichert , 2002 ) . Freezing was done in a Bal-Tec HPM-010 high-pressure freezer ( Bal-Tec AG ) . High pressure frozen cells stored in liquid nitrogen were transferred to cryovials containing 1 . 5 ml of fixative consisting of 1% osmium tetroxide , 0 . 1% uranyl acetate , and 5% water in acetone at liquid nitrogen temperature ( −195°C ) and processed for freeze substitution according to the method of McDonald and Webb ( McDonald , 2014; McDonald and Webb , 2011 ) . Briefly , the cryovials containing fixative and cells were transferred to a cooled metal block at −195°C; the cold block was put into an insulated container such that the vials were horizontally oriented and shaken on an orbital shaker operating at 125 rpm . After 3 hr , the block and cryovials had warmed to 20°C and were transitioned to resin infiltration . Resin infiltration was accomplished by a modification of the method of McDonald ( 2014 ) . Briefly , cells were rinsed 4–5 times in pure acetone and infiltrated with Epon-Araldite resin in increasing increments of 25% over 3 hr plus 3 changes of pure resin at 30 min each . Cells were removed from the planchettes at the beginning of the infiltration series and spun down at 6000 x g for 1 min between solution changes . The cells in pure resin were placed in between 2 PTFE-coated microscope slides and polymerized over 2 hr in an oven set to 100°C . Cells were cut out from the thin layer of polymerized resin and remounted on blank resin blocks for sectioning . Serial sections of 70 nm were cut on a Reichert-Jung Ultracut E microtome and picked up on 1 × 2 mm slot grids covered with a 0 . 6% Formvar film . Sections were post-stained with 1% aqueous uranyl acetate for 10 min and lead citrate for 10 min ( Reynolds , 1963 ) . Images of cells on serial sections were taken on an FEI Tecnai 12 electron microscope operating at 120 kV equipped with a Gatan Ultrascan 1000 CCD camera . Models were constructed from serial sections with the IMOD package ( Kremer et al . , 1996 ) , using 3DMOD version 4 . 9 . 8 . Initial alignment was performed using the Midas tool in the ETomo interface of the IMOD package; afterwards , sections were rotated and minorly warped in Midas to improve alignment . The plasma membrane , nuclear envelope , and nucleoli were segmented in IMOD by manual tracing using the Drawing Tools plugin created by Andrew Noske . If a serial section was missing or unusable , the Interpolator plugin created by Andrew Noske was used to approximate any contours in the missing slice . Movies were made in 3DMOD and assembled in QuickTime Pro Version 7 . 6 . 6; EM movie sizes were compressed to below 10 MB by exporting as HD 720p movies in QuickTime . To quantify the percentage of Nsr1 sequestration , measurements of Nsr1-GFP signal intensity were taken with Fiji ( RRID:SCR_002285 , Schindelin et al . , 2012 ) from maximum intensity z-projection movies of young and aged cells that eventually formed tetrads . Nsr1 signal was measured after nucleolus segregation to the four dividing nuclei , determined by the appearance of four Nsr1 foci in the four nuclei . Percent sequestration was measured by calculating the raw integrated intensity in the fifth compartment and dividing it by the sum of the signal present in the four nuclei and the fifth compartment . The mean intensity measured from non-cellular background was subtracted in each field of view before quantifying Nsr1 levels . For the vacuolar lysis experiments , the timing of vacuolar membrane disruption and either excluded nucleoporin or protein aggregate disappearance were scored in cells that eventually became tetrads . Vacuolar membrane disruption was defined as the time point at which Vph1 signal becomes diffuse , instead of localizing to the membrane . Protein aggregate and NPC disappearance was defined as the time point at which the excluded fluorescence signal was no longer visible . Only cells in which both vacuolar membrane disruption and nucleoporin or protein aggregate disappearance could be confidently called were included in our analysis . In less than 25% of cells , the vacuole appeared to crumple and collapse over more than an hour prior to vacuolar membrane disappearance . Since we were unable to interpret these changes in vacuolar morphology , these cells were not included in our quantification . For protein aggregate and nucleolar sequestration experiments , sequestration was scored in WT cells that formed tetrads and spo21Δ cells that progressed through anaphase II , as tetrad formation cannot be assessed in spo21Δ cells . Protein aggregate sequestration was scored in aged cells and was defined as the aggregate no longer associating with chromatin after the four anaphase II nuclei became distinct . Nucleolar sequestration was scored in young cells and was defined as the presence of a fifth focus that did not associate with a gamete nucleus after the four anaphase II nuclei became distinct . For each meiotic time point , 3 . 7 OD600 equivalents of cells were pelleted and resuspended in 2 mL of 5% trichloroacetic acid and incubated at 4°C for ≥10 min . The cells were subsequently washed with 1 mL 10 mM Tris pH 8 . 0 and then 1 mL of acetone , before being left to dry overnight . Then , ~100 μl glass beads and 100 μl of lysis buffer ( 50 mM Tris-HCl pH 8 . 0 , 1 mM EDTA , 15 mM Tris pH 9 . 5 , 3 mM DTT , 1X cOmplete EDTA-free inhibitor cocktail [Roche] ) were added to each dried pellet . Protein extracts were generated by pulverization using a Mini-Beadbeater-96 ( BioSpec ) . The samples were then treated with 50 μl of 3X SDS sample buffer ( 187 . 5 mM Tris pH 6 . 8 , 6% β-mercaptoethanol , 30% glycerol , 9% SDS , 0 . 05% bromophenol blue ) and heated at 37°C for 5 min . Proteins were separated by polyacrylamide gel electrophoresis using 4–12% Bis-Tris Bolt gels ( Thermo Fisher ) and transferred onto nitrocellulose membranes ( 0 . 45 μm , Bio-rad ) . The Nup84-GFP blot was generated using a semi-dry transfer apparatus ( Trans-Blot Turbo Transfer System , Bio-Rad ) . The Nup170-GFP blot was generated using a Mini-PROTEAN Tetra tank ( Bio-Rad ) filled with 25 mM Tris , 195 mM glycine , and 15% methanol , run at 180 mA ( max 80 V ) for 3 hr at 4°C . The membranes were blocked for at least 30 min with Odyssey PBS Blocking Buffer ( LI-COR Biosciences ) at room temperature . The blots were incubated overnight at 4°C with a mouse anti-GFP antibody ( RRID:AB_2313808 , 632381 , Clontech ) at a 1:2000 dilution in blocking buffer . As a loading control , we monitored Hxk2 levels using a rabbit anti-hexokinase antibody ( RRID:AB_219918 , 100–4159 , Rockland ) at a 1:10 , 000 dilution in blocking buffer . Membranes were washed in PBST ( PBS with 0 . 1% Tween-20 ) and incubated with an anti-mouse secondary antibody conjugated to IRDye 800CW at a 1:15 , 000 dilution ( RRID:AB_621847 , 926–32212 , LI-COR Biosciences ) and an anti-rabbit antibody conjugated to IRDye 680RD at a 1:15 , 000 dilution ( RRID:AB_10956166 , 926–68071 , LI-COR Biosciences ) to detect the GFP epitope and Hxk2 , respectively . Immunoblot images were generated using the Odyssey CLx system ( LI-COR Biosciences ) . Data generated during this study are included in the manuscript and supporting files . Data was deposited to the Image Data Resource ( http://idr . openmicroscopy . org ) under accession number idr0067 .	The cells of living organisms accumulate damage as they age . Some of this age-associated damage is found around the organism’s DNA . However , when genetic material is passed on during sexual reproduction , newly born offspring avoid inheriting this age-induced damage . This ensures that the progeny are ‘re-set’ with a fresh lifespan that is independent from their parents’ age . A lot of what is known about aging has come from studying budding yeast . Yeast cells can undergo a process called meiosis and divide into four cells known as gametes , which are the equivalents of human sperm and egg . During meiosis , the structure that surrounds the cell’s genetic material – known as the nuclear membrane – remains intact , surrounding the DNA as it separates into four distinct parts . As the cell divides , age-associated factors that were originally present in the parent are not inherited by the gametes , but it remains unclear how this occurs . Now , King , Goodman et al . have investigated this process by attaching fluorescent labels to specific aging factors and tracking how they are distributed inside yeast cells undergoing meiosis . This revealed that age-associated factors were physically sequestered away from the inherited genetic material during meiosis . King , Goodman et al . found that as the nuclear membrane remodeled itself around the genetic material of the four gametes , the damage became confined to a fifth previously unknown membrane-bound compartment . Once outside of the gametes , the aging factors were then selectively destroyed by enzymes released from the parent cell . All cells age , and many of the mechanisms underlying these processes are similar across species and cell types . A better understanding of how cells age , and of the process by which gametes are able to sequester and eliminate age-induced damage , may help guide efforts to combat aging in other cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2019	Meiotic cellular rejuvenation is coupled to nuclear remodeling in budding yeast
The molecular codes underpinning the functions of plant NLR immune receptors are poorly understood . We used in vitro Mu transposition to generate a random truncation library and identify the minimal functional region of NLRs . We applied this method to NRC4—a helper NLR that functions with multiple sensor NLRs within a Solanaceae receptor network . This revealed that the NRC4 N-terminal 29 amino acids are sufficient to induce hypersensitive cell death . This region is defined by the consensus MADAxVSFxVxKLxxLLxxEx ( MADA motif ) that is conserved at the N-termini of NRC family proteins and ~20% of coiled-coil ( CC ) -type plant NLRs . The MADA motif matches the N-terminal α1 helix of Arabidopsis NLR protein ZAR1 , which undergoes a conformational switch during resistosome activation . Immunoassays revealed that the MADA motif is functionally conserved across NLRs from distantly related plant species . NRC-dependent sensor NLRs lack MADA sequences indicating that this motif has degenerated in sensor NLRs over evolutionary time . Plants have evolved intracellular immune receptors to detect host-translocated pathogen virulence proteins , known as effectors ( Dodds and Rathjen , 2010; Jones et al . , 2016; Kourelis and van der Hoorn , 2018 ) . These receptors , encoded by disease resistance ( R ) genes , are primarily nucleotide-binding , leucine-rich repeat proteins ( NLRs ) . NLR-triggered immunity ( also known as effector-triggered immunity ) includes the hypersensitive response ( HR ) , a type of programmed cell death associated with disease resistance . NLRs are widespread across eukaryotes and have been described in animals and fungi in addition to plants ( Jones et al . , 2016 ) . In contrast to other taxa , plants express very large and diverse repertoires of NLRs , with anywhere from about 50 to 1000 genes encoded per genome ( Shao et al . , 2016; Steuernagel et al . , 2018 ) . Genome-wide analyses have defined repertoires of NLRs ( NLRome ) across plant species ( Shao et al . , 2016 ) . An emerging paradigm is that plant NLRs form receptor networks with varying degrees of complexity ( Wu et al . , 2018 ) . NLRs have probably evolved from multifunctional singleton receptors—which combine pathogen detection ( sensor activity ) and immune signalling ( helper or executor activity ) into a single protein—to functionally specialized interconnected receptor pairs and networks ( Adachi et al . , 2019a ) . However , our knowledge of the functional connections and biochemical mechanisms underpinning plant NLR networks remains limited . In addition , although dozens of NLR proteins have been subject to functional studies since their discovery in the 1990 s , this body of knowledge has not been interpreted through an evolutionary biology framework that combines molecular mechanisms with phylogenetics . NLRs are multidomain proteins of the ancient group of Signal Transduction ATPases ( STAND ) proteins that share a nucleotide-binding ( NB ) domain . In addition to the NB and LRR domains , most plant NLRs have characteristic N-terminal domains that define three subgroups: coiled-coil ( CC ) , CCR or RPW8-like ( RPW8 ) and toll and interleukin-1 receptor ( TIR ) ( Shao et al . , 2016 ) . In metazoans , NLRs confer immunity to diverse pathogens through a wheel-like oligomerization process resulting in multiprotein platforms that recruit downstream elements , such as caspases ( Qi et al . , 2010; Zhou et al . , 2015; Hu et al . , 2015; Zhang et al . , 2015; Tenthorey et al . , 2017 ) . Plant NLRs have long been thought to oligomerize through their N-terminal domains when they’re activated ( Bentham et al . , 2018 ) . However , the precise molecular mechanisms that underpin NLR activation and subsequent execution of HR cell death have remained largely unknown until very recently . In two remarkable papers , Wang et al . ( 2019a ) and Wang et al . ( 2019b ) have significantly advanced our understanding of both the structural and biochemical basis of CC-NLR activation in plants . They reconstituted the inactive and active complexes of the Arabidopsis CC-NLR ZAR1 ( HOPZ-ACTIVATED RESISTANCE1 ) with its partner receptor-like cytoplasmic kinases ( RLCKs ) ( Wang et al . , 2019a; Wang et al . , 2019b ) . Cryo-electron microscopy ( cryo-EM ) structures revealed that activated ZAR1 forms a resistosome—a wheel-like pentamer that undergoes a conformational switch to expose a funnel-shaped structure formed by the N-terminal α helices ( α1 ) of the CC domains ( Wang et al . , 2019a; Wang et al . , 2019b ) . They propose an engaging model in which the exposed α1 helices of the ZAR1 resistosome mediate cell death by translocating into the plasma membrane and perturbing membrane integrity similar to pore-forming toxins ( Wang et al . , 2019b ) . However , whether the ZAR1 model extends to other CC-NLRs is unknown . One important unanswered question is the extent to which the α1 helix ‘death switch’ occurs in other CC-NLRs ( Adachi et al . , 2019b ) . Although ZAR1 is classified as a singleton NLR that detects pathogen effectors without associating with other NLRs , many plant NLRs are interconnected in NLR pairs or networks ( Wu et al . , 2018; Adachi et al . , 2019a ) . Paired and networked NLRs consist of sensor NLRs that detect pathogen effectors and helper NLRs that translate this effector recognition into HR cell death and immunity . In the Solanaceae , a major phylogenetic clade of CC-NLRs forms a complex immunoreceptor network in which multiple helper NLRs , known as NLR-REQUIRED FOR CELL DEATH ( NRC ) , are required by a large number of sensor NLRs , encoded by R gene loci , to confer resistance against diverse pathogens , such as viruses , bacteria , oomycetes , nematodes and insects ( Wu et al . , 2017 ) . These proteins form the NRC superclade , a well-supported phylogenetic cluster divided into the NRC helper clade ( NRC-helpers or NRC-H ) and a larger clade that includes all known NRC-dependent sensor NLRs ( NRC-sensors or NRC-S ) ( Wu et al . , 2017 ) . The NRC superclade has expanded over 100 million years ago ( Mya ) from an NLR pair that diversified to up to one-half of the NLRs of asterid plants ( Wu et al . , 2017 ) . How this diversification has impacted the biochemical activities of the NRC-S compared to their NRC-H mates is poorly understood . For example , it’s unclear how the ZAR1 conceptual framework applies to more complex NLR configurations such as the NRC network ( Adachi et al . , 2019b ) . This paper originates from use of the in vitro Mu transposition system to generate a random truncation library and identify the minimal region required for CC-NLR-mediated cell death . We applied this method to NRC4—a CC-NLR helper of the NRC family that is genetically required by a multitude of NRC-S , such as the potato late blight resistance protein Rpi-blb2 , to cause HR cell death and confer disease resistance ( Wu et al . , 2017 ) . This screen revealed that the N-terminal 29 amino acids of NRC4 are sufficient to induce cell death . Remarkably , this region is about 50% identical to the N-terminal ZAR1 α1 helix , which undergoes the conformational ‘death switch’ associated with the activation of the ZAR1 resistosome ( Wang et al . , 2019b ) . Computational analyses revealed that this region is defined by a motif , following the consensus MADAxVSFxVxKLxxLLxxEx , which we coined the ‘MADA motif’ . This sequence is conserved not only in NRC4 and ZAR1 but also in ~20% of all CC-NLRs of dicot and monocot species . Motif swapping experiments revealed that the MADA motif is functionally conserved between NRC4 and ZAR1 , as well as between NLRs from distantly related plant species . Interestingly , NRC-S lack N-terminal MADA sequences , which may have become non-functional over evolutionary time . We conclude that the evolutionarily constrained MADA motif is critical for the cell death inducing activity of CC domains from a significant fraction of plant NLR proteins , and that the ‘death switch’ mechanism defined for the ZAR1 resistosome is probably widely conserved across singleton and helper CC-NLRs . The N-terminal CC domain of a subset of CC-NLR proteins can mediate self-association and trigger HR cell death when expressed on its own ( Bentham et al . , 2018 ) . However , to date truncation experiments have been conducted based on educated guesses of domain boundaries ( Maekawa et al . , 2011; Casey et al . , 2016; Cesari et al . , 2016; Wróblewski et al . , 2018 ) . Moreover , one amino acid difference in the length of the assayed truncation can affect cell death inducing activity ( Casey et al . , 2016 ) . Therefore , we designed an unbiased truncation approach using bacteriophage Mu in vitro transposition system to randomly generate a C-terminal deletion library of the helper NLR NRC4 . By using a custom-designed artificial transposon ( Mu-STOP transposon ) that carries staggered translation stop signals at Mu R-end ( Poussu , 2005 ) , we targeted the full-length coding sequence of the NRC4 autoactive mutant , NRC4D478V , ( referred to from here on as NRC4DV ) . We generated a total of 65 truncated NRC4DV::Mu-STOP variants and expressed these mutants in Nicotiana benthamiana leaves using agroinfiltration ( Figure 1A ) . Remarkably , only a single truncate carrying the N-terminal 29 amino acids triggered visible cell death in N . benthamiana leaves ( Figure 1B , Figure 1—figure supplement 1 ) . To validate this phenotype , we expressed NRC4 N-terminal 29 amino acids ( NRC41-29 ) fused with the yellow fluorescent protein ( YFP ) at the C-terminus in N . benthamiana leaves ( Figure 2A ) . NRC41-29-YFP triggered a visible cell death response , although the cell death intensity was weaker than that of the full-length NRC4DV-YFP ( Figure 2B–E ) . To determine whether NRC41-29-YFP requires the endogenous N . benthamiana NRC4 to trigger cell death , we expressed this fusion protein in two independent mutant nrc4a/b plants that carry CRISPR/Cas9-induced mutations in the two NRC4 genes NRC4a and NRC4b ( Figure 2—figure supplement 1 , see Materials and methods ) . In these plants , NRC41-29-YFP still induced cell death indicating that the activity of the N-terminal 29 amino acids of NRC4 is independent of a full-length NRC4 protein ( Figure 2C–D and F–G ) . The CC domains of ZAR1 and maize Rp1 ( RESISTANCE to PUCCINIA 1 ) are autoactive when expressed as a fusion protein with a fluorescent protein tag ( Wang et al . , 2015; Baudin et al . , 2017 ) . Given that YFP and related fluorescent proteins self-oligomerize ( Kim et al . , 2015 ) , we hypothesized that such fluorescent proteins promote self-assembly of the N-terminal 29 amino acids of NRC4 resulting in hypersensitive cell death . To test this hypothesis , we modified YFP with the alanine 206 ( A206 ) to lysine ( K ) mutation that reduces homo-affinity ( Figure 2—figure supplement 2A ) ( Zacharias et al . , 2002 ) . The YFPA206K mutation compromised the cell death intensity of NRC41-29-YFP but not that of full-length NRC4DV ( Figure 2—figure supplement 2B–E ) . This result indicates that YFP-mediated self-assembly is a key step in the capacity of NRC41-29-YFP to trigger hypersensitive cell death . Our finding that the N-terminal 29 amino acids of NRC4 are sufficient to trigger cell death prompted us to investigate the occurrence of this sequence across the plant NLRome . We first compiled a sequence database containing 988 putative CC-NLRs and CCR-NLRs ( referred to from here on as CC-NLR database , Figure 3A , Figure 3—figure supplement 1 ) from six representative plant species ( Arabidopsis , sugar beet , tomato , N . benthamiana , rice and barley ) amended with 23 functionally characterized NLRs . Next , we extracted their sequences prior to the NB-ARC domain ( Figure 3A ) . These sequences were too diverse and aligned poorly to each other to enable global phylogenetic analyses . Therefore , to classify the extracted N-terminal sequences based on sequence similarity , we clustered them into protein families using Markov cluster ( MCL ) algorithm Tribe-MCL ( Enright et al . , 2002 ) ( Figure 3A ) . The 988 proteins clustered into 59 families of at least two sequences ( tribes ) and 43 singletons ( Figure 3B ) . The largest tribe , Tribe 1 , consists of 219 monocot NLRs , including MLA10 , Sr33 , Sr50 , the paired Pik and Pia ( RGA4 and RGA5 ) NLRs , and seven dicot NLRs notably RPM1 ( Figure 3B ) . Tribe 2 , the second largest tribe with 102 proteins , consists primarily of dicot proteins ( 93 out of 102 ) but still includes nine monocot NLRs . Interestingly , Tribe 2 grouped NRC-H proteins , including NRC4 , with well-known CC-NLRs , such as ZAR1 , RPP13 , R2 and Rpi-vnt1 . 3 indicating that these proteins share similarities in their CC domains ( Figure 3B ) . We performed phylogenetic analyses of NLR proteins using the NB-ARC domain because it is the only conserved domain that produces reasonably good global alignments and can inform evolutionary relationships between all members of this family ( Figure 3—figure supplement 2 ) . We mapped individual NLR proteins grouped in Tribe-MCL N-terminal tribes onto a phylogenetic tree based on the NB-ARC domain ( Figure 3C ) . These analyses revealed that the clustering of NLRs into the N-terminal tribes does not always match the NB-ARC phylogenetic clades ( Figure 3C ) . In particular , NLRs in Tribe 1 and Tribe 2 often mapped to distinct well-supported clades scattered throughout the NB-ARC phylogenetic tree . We conclude that there are N-terminal domain sequences that have remained conserved over evolutionary time across distantly related CC-NLRs . Next , we investigated whether N-terminal domains of CC-NLRs carry specific sequence motifs . We used MEME ( Multiple EM for Motif Elicitation ) ( Bailey and Elkan , 1994 ) to identify conserved patterns in each of the N-terminal domain tribes . MEME revealed several conserved sequence patterns in each of the four largest tribes ( Figure 4—figure supplement 1 ) . The previously reported sequence pattern , EDVID motif ( Rairdan et al . , 2008 ) , was as expected predicted in ~87% to 96% in the four largest tribes ( Figure 4—figure supplement 1 ) . Within Tribe 2 , a motif that is conserved at the N terminus of 87 of 102 proteins overlapped with the N-terminal 29 amino acids of NRC4 we identified as sufficient to cause cell death ( Figure 4—figure supplement 1 ) . Remarkably , the conserved sequence pattern of this very N-terminal motif matched the ZAR1 α1 helix that undergoes a conformational switch during activation of the ZAR1 resistosome ( Wang et al . , 2019b ) ( Figure 4A–B ) . In fact , 8 of the first 17 amino acids of ZAR1 are invariant in NRC4 , and the majority of the amino acid polymorphisms between ZAR1 and NRC4 in the α1 helix region are conservative ( Figure 4A ) . We conclude that NRC4 , ZAR1 and numerous other CC-NLRs share a conserved N-terminal motif . We coined this sequence ‘MADA motif’ based on the deduced 21 amino acid consensus sequence MADAxVSFxVxKLxxLLxxEx ( Figure 4A , Figure 4—figure supplement 2 ) . We built a Hidden Markov Model ( HMM ) from a sequence alignment of the MADA motif of 87 NLR proteins from Tribe 2 . To determine whether the MADA motif is primarily found among proteins annotated as NLRs , we used the HMMER software ( Eddy , 1998 ) to query the Arabidopsis and tomato proteomes using the MADA motif HMM . HMMER searches revealed that the MADA motif is mainly found in NLR proteins compared with non-NLR proteins ( Figure 4C ) . An HMM score cut-off of 10 . 0 clearly distinguishes NLR proteins from others with 97 . 1% ( 34 out of 35 ) tomato proteins and 97 . 7% ( 42 out of 43 ) Arabidopsis proteins scoring over 10 . 0 being annotated as NLRs ( Figure 4C ) . We conclude that the MADA motif is a sequence signature of a subset of NLR proteins and that a HMMER cut-off score of 10 . 0 is most optimal for high confidence searches of MADA containing CC-NLR proteins ( MADA-CC-NLRs ) . To what extent does the MADA motif occur in plant NLRomes ? We re-screened the CC-NLR database using HMMER and identified 103 hits ( 10 . 4% ) over the cut-off score of 10 . 0 ( Figure 5A–B , Figure 5—figure supplement 1A ) . We also noted that another 129 NLRs were positive but with a score lower than 10 . 0 , and we tentatively termed these hits as MADA-like CC-NLRs ( MADAL-CC-NLRs ) ( Figure 5B , Figure 5—figure supplement 1A ) . Most of the MADA hits are from dicot plant species whereas MADAL-CC-NLRs are primarily from monocots possibly reflecting a bias in our HMM profile which was built from the dicot enriched Tribe 2 ( Figure 5C , Figure 5—figure supplement 1B ) . Indeed , the majority of MADA hits ( 85 out of 103 ) were from Tribe 2 , which includes NRC4 and ZAR1 , but some MADA hits were also from other Tribes , notably the rice helper NLR Pik-2 from Tribe 1 ( HMM score = 10 . 4 ) ( Figure 5C , Figure 5—figure supplement 1C ) . MADAL-CC-NLRs are mainly from Tribe 1 and Tribe 4 and include the monocot proteins MLA10 and Sr33 , as well as Arabidopsis RPM1 ( Figure 5C , Figure 5—figure supplement 1C ) . Given that the MADA sequence is at the very N-terminus of ZAR1 and NRC4 , and that the N-terminal position of the ZAR1 α1 helix is critical for its function based on the model of Wang et al . ( 2019b ) , we checked the positional distribution of predicted MADA and MADAL motifs ( Figure 5D ) . The majority of the predicted MADA and MADAL motifs ( 199 out of 232 , 85 . 8% ) occurred at the very beginning of the NLR protein . However , 4 of 103 of the predicted MADA- and 29 of 129 MADAL-CC-NLRs have N-terminal extensions over 15 amino acids prior to the motifs ( Figure 5D ) . For example , the MADA motif is located at position 54 to 72 amino acids in the potato NLR Rpi-vnt1 . 3 . Whether these exceptions reflect misannotated gene models or genuinely distinct motif sequences remains to be determined . In summary , our bioinformatic analyses revealed that 199 out 988 ( 20 . 1% ) of the CC-NLRs of six representative dicot and monocot species contain a MADA or MADAL motif at their very N-termini . These MADA sequences have noticeable similarity to NRC4 and ZAR1 . NB-ARC domain phylogenetic trees revealed that the NRC superclade is divided into the NRC clade ( NRC-H ) and a larger clade that includes all known NRC-dependent sensor NLRs ( NRC-S ) ( Wu et al . , 2017 ) . We noted that even though the NRC-H and NRC-S are sister clades based on NB-ARC phylogenetic analyses , they grouped into distinct N-terminal domain tribes in the Tribe-MCL analyses ( Figure 6A ) . Whereas all NRC-H mapped to Tribe 2 , NRC-S clustered into eight different tribes ( Figure 6A ) . This pattern indicates that unlike the NRCs , the N-terminal sequences of their NRC-S mates have diversified throughout evolutionary time . Next , we mapped the occurrence of the MADA motif onto the NB-ARC phylogenetic tree and noted that the distribution of the MADA motif was uneven across the NRC superclade despite their phylogenetic relationship ( Figure 6B ) . Whereas 20 out of 22 NRC-H have a predicted MADA motif at their N-termini , none of the 117 examined NRC-S were predicted as MADA-CC-NLR in the HMMER search ( Figure 6B ) . In fact , 65 of 117 NRC-S , including the well know disease resistance proteins R1 , Prf , Sw5b , Hero , Rpi-blb2 and Mi-1 . 2 , have N-terminal extensions of ~600 amino acids , or more in the case of Prf , prior to their predicted CC domains ( Figure 6B ) . These findings indicate the CC domains of NRCs and their NRC-S mates have experienced distinct evolutionary trajectories even though these NLR proteins share a common evolutionary origin . To experimentally validate our bioinformatic analyses , we performed site directed mutagenesis to determine the degree to which the MADA motif is required for the activity of NRC4 . First , we followed up on the ZAR1 structure-function analyses of Wang et al . ( 2019b ) who showed that three amino acids ( phenylalanine 9 [F9] , leucine 10 [L10] and leucine 14 [L14] ) within the α1 helix/MADA motif are required for ZAR1-mediated cell death and bacterial resistance . We introduced a triple alanine substitution similar to the mutant of Wang et al . ( 2019b ) into the autoactive NRC4DV and found that this L9A/V10A/L14A mutation significantly reduced , but did not abolish , NRC4DV cell death inducing activity ( Figure 7A–C ) . Given that the MADA motif , particularly the mutated L9 , V10 and L14 sites , is primarily composed of hydrophobic residues , we reasoned that substitutions with the negatively charged glutamic acid ( E ) would be more disruptive than hydrophobic alanine . Therefore , we substituted L9 , V10 and L14 with glutamic acid , and observed that the L9E/V10E/L14E mutation resulted in a more severe disruption of the cell death activity of NRC4DV compared to the triple alanine mutant ( Figure 7A–C ) . Both of the NRC4DV triple alanine and glutamic acid mutant proteins accumulated to similar levels as NRC4DV when expressed in N . benthamiana leaves indicating that the observed loss-of-function phenotypes were not due to protein destabilization ( Figure 7D ) . We further introduced the triple alanine mutation to NRC41-29-YFP and ZAR11-144-YFP ( Figure 7—figure supplement 1A ) . ZAR11-144 matches the ZAR1 CC domain and is known to trigger cell death when expressed fused to a YFP tag ( Baudin et al . , 2017 ) . The triple alanine mutation abolished the cell death triggered by both NRC41-29-YFP and ZAR11-144-YFP , supporting the view that MADA motifs are essential for the capacity of the N-termini of NRC4 and ZAR1 to cause cell death ( Figure 7—figure supplement 1B–D ) . Next , we performed single alanine and glutamic acid mutant scans to reveal which other residues in the MADA motif are required for NRC4-mediated cell death . None of the tested single alanine-substituted mutants affected the cell death response of NRC4DV ( Figure 8—figure supplement 1 ) . In contrast , single glutamic acid mutations L9E , L13E and L17E essentially abolished the cell death activity of NRC4DV without affecting the stability of the mutant proteins ( Figure 8 ) . Therefore , we determined that the L9 , L13 , and L17 residues in the MADA motif are critical for cell death induction by NRC4 . Finally , we mapped L9 , L13 and L17 onto a homology model of the CC domain of NRC4 produced based on the ZAR1 resistosome structure of Wang et al . ( 2019b ) ( Figure 8—figure supplement 2 ) . All three residues mapped to the outer surface of the funnel-shaped structure formed by the α1 helices similar to the previously identified residues in positions 9 , 10 and 14 . These results suggest that the outer surface of the funnel-shaped structure formed by N-terminal helices is critical not only for the function of ZAR1 but also for the activity of another MADA-CC-NLR . The ZAR1 model postulates that the resistosome translocates into the plasma membrane through the α1 helix which matches the MADA motif ( Wang et al . , 2019b ) . To investigate the intracellular dynamics of the MADA motif , we analysed the subcellular distribution of NRC41-29-YFP in N . benthamiana leaves ( Figure 9A ) . Interestingly , unlike free YFP which typically shows nucleocytoplasmic distribution , NRC41-29-YFP produced fluorescence signal in punctate structures throughout the cell in addition to relatively weak nucleocytoplasmic signal ( Figure 9A ) . Furthermore , we merged both the z-stack and single plain images of the YFP proteins with the plasma membrane marker RFP-Rem1 . 3 ( Bozkurt et al . , 2014 ) . Although the NRC41-29-YFP puncta did not completely overlap with RFP-Rem1 . 3 signal , we noticed some of the NRC41-29-YFP puncta associated with the plasma membrane ( Figure 9A–B ) . To further study the NRC41-29-YFP puncta , we examined puncta formation of the YFP A206K mutant , which shows reduced cell death by NRC41-29-YFP ( Figure 2—figure supplement 2 ) . In contrast to NRC41-29-YFP , NRC41-29-YFPA206K rarely formed puncta ( Figure 9A , C ) , suggesting that YFP self-assembly is required for NRC41-29-YFP puncta formation . Furthermore , introducing the L9E in NRC41-29-YFP greatly reduced puncta formation ( Figure 9A , C ) . This finding directly connects puncta formation to the activity of full length NRC4 given that L9E also affects NRC4 cell death activity ( Figure 8 ) . Taken together , these results indicate that both an intact MADA motif and YFP oligomerization are required for the capacity of NRC41-29-YFP to form puncta as well as cause cell death in N . benthamiana leaves . Our observation that the ZAR1 α1 helix has sequence similarity to the N-terminus of NRC4 prompted us to determine whether this sequence is functionally conserved between these two proteins . To test this hypothesis , we swapped the first 17 amino acids of NRC4DV with the equivalent region of ZAR1 ( Figure 10A–B ) . The resulting ZAR11-17-NRC4 chimeric protein can still trigger cell death in N . benthamiana leaves indicating that the MADA/α1 helix sequence is functionally equivalent between these two NLR proteins ( Figure 10C , Figure 10—figure supplement 1 ) . Next , we swapped the same 17 amino acids of NRC4 with the matching sequences of the MADA-CC-NLRs NRC2 from N . benthamiana , RPP8 and RPP13 from Arabidopsis , and Pik-2 and Os03g30910 . 1 from rice , all of which gave HMMER scores > 10 . 0 and ranging from 30 . 8 to 10 . 4 ( Figure 10A–B ) . All of the assayed chimeric NRC4DV proteins retained the capacity to trigger cell death in N . benthamiana leaves ( Figure 10C , Figure 10—figure supplement 1 ) . We determined whether the N-termini of MADAL-CC-NLRs Arabidopsis RPM1 and barley MLA10 , which yielded respective HMMER scores of 9 . 3 and 7 . 8 , could also replace the first 17 amino acids of NRC4DV ( Figure 10A–B ) . Both NRC4DV chimeras retained the capacity to trigger cell death indicating that these MADAL sequences are functionally analogous to the NRC4 N-terminus ( Figure 10C , Figure 10—figure supplement 1 ) . These results indicate that the MADA motif is functionally conserved even between distantly related NLRs from dicots and monocots . We further swapped the 17 amino acids of NRC4DV with N-terminal sequences from Arabidopsis LOV1 ( AT1G10920 ) , pepper Bs2 and potato Rx , all of which were not predicted to have a MADA sequences by HMMER searches ( Figure 10A–B ) . LOV1 was among the 13 . 7% of Tribe 2 NLRs that were not predicted to have a MADA/MADAL motif . Bs2 and Rx are NRC-S NLRs that belong to different tribes—Tribe 11 and 25 , respectively ( Figure 6A ) . The N-terminal sequences of Bs2 and Rx are somewhat similar to MADA sequences but were negative in the HMMER analyses ( Figure 10A ) . Interestingly , whereas the N-termini of Bs2 and LOV1 did not complement the cell death activity when swapped into NRC4DV , Rx1-17 could confer cell death activity when swapped into NRC4DV ( Figure 10C , Figure 10—figure supplement 1 ) . This exception indicates that at least one of the N-terminal sequences that are not predicted as having the MADA motif may still functionally complement the N-terminus of NRC4 . We investigated whether the MADA motif of NRC4 is required for disease resistance against the oomycete pathogen Phytophthora infestans . One of the NRC4-dependent sensor NLRs is Rpi-blb2 , an NRC-S protein from Solanum bulbocastanum that confers resistance to P . infestans carrying the matching effector AVRblb2 ( van der Vossen et al . , 2003; Oh et al . , 2009 ) . For this purpose , we set up a genetic complementation assay in which NRC4 is co-expressed with Rpi-blb2 in leaves of the N . benthamiana nrc4a/b_9 . 1 . 3 mutant prior to inoculation with the P . infestans strain 88069 ( Wu et al . , 2017 ) , that carries AVRblb2 ( Figure 11A ) . Unlike wild-type NRC4 , the NRC4 L9A/V10A/L14A and L9E mutants failed to rescue the resistance to P . infestans in the N . benthamiana nrc4a/b_9 . 1 . 3 mutant , indicating that MADA motif mutations not only impair HR cell death as shown above but also affect disease resistance against an oomycete pathogen ( Figure 11B ) . We conducted similar complementation assays with the ZAR11-17-NRC4 chimera in which the first 17 amino acids of NRC4 were swapped with the equivalent region of ZAR1 , and found that ZAR11-17-NRC4 complemented the nrc4a/b_9 . 1 . 3 N . benthamiana mutant to a similar degree as wild-type NRC4 ( Figure 11B ) . These experiments further confirm that the α1 helix/MADA motif of Arabidopsis ZAR1 is functionally equivalent to the N-terminus of NRC4 , and that the chimeric ZAR11-17-NRC4 is not only able to trigger HR cell death but also retains its capacity to function with its NRC-S mate Rpi-blb2 and confer resistance to P . infestans . This study stems from a random truncation screen of the CC-NLR NRC4 , which revealed that the very N-terminus of this protein is sufficient to carry out the HR cell death activity of the full-length protein . It turned out that this region is defined by a consensus sequence—the MADA motif—that occurs in about one fifth of plant CC-NLRs including Arabidopsis ZAR1 . The MADA motif covers most of the functionally essential α1 helix of ZAR1 that undergoes a conformational switch during activation of the ZAR1 resistosome ( Wang et al . , 2019b ) . Our finding that the ZAR1 α1 helix/MADA motif can functionally replace its matching region in NRC4 indicates that the ZAR1 ‘death switch’ mechanism may apply to NRCs and other MADA-CC-NLRs from dicot and monocot plant species . We recently proposed that NLRs may have evolved from multifunctional singleton receptors to functionally specialized and diversified receptor pairs and networks ( Adachi et al . , 2019a ) . In this study , a striking finding from the computational analyses is that all NRC-S lack the MADA motif even though they are more closely related to NRC-H than to ZAR1 and other MADA-CC-NLRs in the NB-ARC phylogenetic tree ( Figure 6 ) . These observations led us to draw the evolutionary model of Figure 12 . In this model , we propose that MADA-type sequences have emerged early in the evolution of CC-NLRs and have remained conserved from singletons to helpers in NLR pair and network throughout evolution . In sharp contrast , MADA sequences appear to have degenerated over time in sensor CC-NLRs as these proteins specialized in pathogen detection and lost the capacity to execute the immune response without their helper mates . Consistent with this view , NRC-H are known to be more highly conserved than their NRC-S partners within the Solanaceae ( Wu et al . , 2017; Stam et al . , 2019 ) . Future analyses will determine whether MADA-CC-NLRs are generally more evolutionarily constrained than non-MADA containing NLRs . In addition , about half of the NRC-S proteins have acquired N-terminal extensions ( N-terminal domains ) before their CC domain , which would preclude a free N-terminal α1 helix essential for a ZAR1 type ‘death switch’ mechanism ( Figure 6 ) . In fact , the N-terminal domains of Prf and Sw5b function as baits that sense pathogen effectors , suggesting functional analogy to integrated effector detection motifs found in some NLRs , and are not known to be involved in executing the immune response ( Saur et al . , 2015; Li et al . , 2019a ) . Here , we hypothesize that the CC domains of these and other sensor NLRs have extensively diversified over evolutionary time and are losing the capacity to function as HR cell death executors . This could be a consequence of relaxed selection given that these proteins rely on their MADA-CC-NLR partners to execute the immune response as discussed above . Additional structure-function experiments will be needed to determine the extent to which this ‘use-it-or-lose-it’ evolutionary model applies to the sensor sub-class of NLR immune receptors . Understanding the precise nature of the N-terminal sequences that can functionally replace the α1 helix requires further investigation . In the MADA motif swap experiments , we found one exception to the correlation between MADA predictions and functional complementation of NRC4 . The N-terminal sequence of the NRC-S NLR Rx , which was negative in the MADA HMMER searches , complemented the cell death activity of NRC4 MADA motif ( Figure 10 , Figure 10—figure supplement 1 ) . Nonetheless , previously the NB domain of Rx was reported to be capable of triggering cell death ( Rairdan et al . , 2008 ) , suggesting that the CC domain of Rx is dispensable for activation of HR . Therefore , in our ‘use-it-or-lose-it’ model , the N-termini of some NRC-S may not have fully degenerated into non-functional sequences and may have residual ability to functionally complement MADA . In the future , it would be fascinating to determine resistosome configurations of NLR sensor and helper hetero-complexes . As discussed elsewhere ( Adachi et al . , 2019a; Jubic et al . , 2019 ) , one hypothesis is that sensor NLRs associate with the resistosome as functional equivalents of RLCKs in the ZAR1 resistosome . Another is that sensor NLRs form one of the wheel spokes in a hetero-oligomeric resistosome as in the mammalian NAIP/NLRC4 inflammasome ( Tenthorey et al . , 2017 ) . It is possible that in this configuration , the N-terminus of a sensor NLR such as Rx remains evolutionarily constrained in terms of length and sequence composition . Future structural analyses of NLR sensor/helper heterocomplexes are needed to address these questions . Already , our evolutionary model appears to be consistent with some paired NLR configurations in addition to the NRC-H/NRC-S network . One example is rice Pik-1 and Pik-2 , which are a well-established NLR pair that detects the AVRPik effector of the rice blast fungus M . oryzae ( Maqbool et al . , 2015; Białas et al . , 2018 ) . AVRPik binding to the integrated heavy metal associated ( HMA ) domain of Pik-1 results in HR cell death and blast fungus resistance only in the presence of its helper Pik-2 protein ( Maqbool et al . , 2015 ) . In our computational analyses only Pik-2 was detected to carry an N-terminal MADA motif ( Figure 5 , HMM score = 10 . 4 ) even though the CC domains of both proteins grouped into Tribe 1 ( Figure 3 ) . The Pik-2 MADA motif could substitute for the N-terminus of NRC4 in our cell death assays despite having six additional amino-acids at its N-terminus ( Figure 10 ) . These results are consistent with our Figure 11 model and imply that the helper NLR Pik-2 may execute HR cell death via its N-terminal MADA motif whereas its paired sensor NLR Pik-1 does not have the capacity to carry this activity on its own . In addition to ZAR1 , RPP8 is another Arabidopsis MADA-CC-NLR with high similarity to the N-terminus of NRC4 with nine invariant amino acids out of 17 ( 53%; HMMER score = 30 . 8 ) . This RPP8 MADA motif could substitute for the N-terminus of NRC4 indicating that it is functional ( Figure 10 ) . In Arabidopsis , RPP8 ( AT5G43470 ) and its paralogs occur at dynamic genetic loci that exhibit frequent sequence exchanges as deduced from comparative genomic analyses ( Kuang et al . , 2008 ) . Four of the five RPP8 paralogs in the Arabidopsis ecotype Col-0 were deemed to have a MADA motif based on our HMMER searches , whereas a fifth paralog LOV1 ( AT1G10920 ) was negative and did not complement NRC4 autoactivity in the MADA motif swap experiments ( Figure 10 , Figure 10—figure supplement 1 ) . LOV1 confers sensitivity to the victorin effector produced by the necrotrophic fungus Cochliobolus victoriae by binding the defense-associated thioredoxin TRX-h5 when it is complexed with victorin ( Lorang et al . , 2012 ) . Interestingly , LOV1 binds TRX-h5 via its CC domain indicating that this region has evolved a pathogen sensor activity in this NLR protein ( Lorang et al . , 2012 ) . How the sensor activity of the CC domain of LOV1 relates to the absence of a detectable MADA motif and whether this protein relies on other MADA-CC-NLRs to execute the cell death response are unanswered questions that are raised by these observations . In activated ZAR1 resistosome , a funnel-shaped structure formed by five α1 helices is thought to directly execute hypersensitive cell death by forming a toxin-like pore in the plasma membrane ( Wang et al . , 2019b ) . To what extent do activated MADA-CC-NLRs function according to this ZAR1 model ? Structure informed mutagenesis of ZAR1 revealed that F9 , L10 and L14 on the outer surface of the funnel-shaped structure are required for immunity ( Wang et al . , 2019b ) . Here , our Ala and Glu scans of the MADA motif revealed that the NRC4 L9 , L13 and L17 residues are essential for HR cell death activity . All three residues mapped to the outer surface of NRC4 α1 helices as predicted from a homology model based on the ZAR1 resistosome ( Figure 8—figure supplement 2 ) . We also found that mutations that perturb the MADA motif and prevent YFP self-association impair the capacity of NRC41-29-YFP to cause cell death and form puncta in N . benthamiana leaf cells ( Figure 2—figure supplement 2 , Figure 7—figure supplement 1 , Figure 9 ) . Our current interpretation of these results is that NRC41-29-YFP forms high-order complexes to cause cell death . However , direct support for this hypothesis is still missing . In addition , we lack detailed analyses of the cellular dynamics of the NRC41-29-YFP puncta and the degree to which they associate with membrane compartments in living plant cells . In the future , further biochemical , structural and cellular analyses are needed to determine the precise nature of the broadly conserved MADA motif and address the extent to which the ZAR1 ‘death switch’ model occurs in CC-NLRs . As discussed by Wang et al . ( 2019b ) , the interior space of the funnel structure is also important because the ZAR1 double mutant E11A/E18A is impaired in cell death and disease resistance activities . However , in our Glu mutant scan , we failed to observe a reduction in HR cell death activities with single site mutants in these residues or other amino acids that are predicted to line up the interior space of the funnel-shaped structure . Whether or not this reflects genuine biological differences between ZAR1 and NRC4 remains to be studied . A subset of CC-NLRs of the RPW8/HR family of atypical resistance proteins have a distinct type of coiled-coil domain known as CCR ( Barragan et al . , 2019; Li et al . , 2019b ) . We failed to detect any MADA type sequences in these CCR-NLR proteins . Indeed , the CCR domain has similarity to mixed lineage kinase domain-like ( MLKL ) proteins and fungal HeLo/HELL domains , which form multi-helix bundles and act as membrane pore forming toxins ( Barragan et al . , 2019; Li et al . , 2019a; Mahdi et al . , 2019 ) . Whether the CCR domains function as a distinct cell death inducing system in plants compared to MADA-CC-NLRs remains to be determined . Interestingly , Arabidopsis HR4 , a CCR containing protein , interacts in an allele-specific manner with the genetically unlinked CC-NLR RPP7b to trigger autoimmunity in the absence of pathogens ( Barragan et al . , 2019 ) . Recently , Li et al . ( 2019b ) showed that RPP7b forms higher-order complexes of six to seven subunits only when activated by the matching autoimmune HR4Fei-0 protein in a biochemical process reminiscent of activated ZAR1 resistosome ( Li et al . , 2019a ) . In our HMMER searches , RPP7b and its four Arabidopsis paralogs were all classed as carrying the MADA motif . Thus , findings by Li et al . ( 2019b ) directly link a MADA-CC-NLR to the formation of resistosome type structures consistent with our view that the ZAR1 model widely applies to other NLRs with the MADA α1 helix . It will be fascinating to determine whether or not RPP7b and HR4 are both capable of executing cell death , especially as two-component systems of NLR and HeLo/HELL proteins are common in fungi and mammals ( Barragan et al . , 2019 ) . Plant NLRs can be functionally categorized into singleton , sensor or helper NLRs based on their biological activities ( Adachi et al . , 2019a ) . However , it remains challenging to predict NLR functions from the wealth of unclassified NLRomes that are emerging from plant genome sequences . It has not escaped our attention that the discovery of the MADA motif as a signature of NLR singletons and helpers—but missing in sensor NLRs—enables the development of computational pipelines for predicting NLR networks from naïve plant genomes . Such in silico predictions can be tested by co-expression of paired NLRs in N . benthamiana . In addition , MADA motif predictions can be validated using our straightforward functional assay of swapping the NRC4 N-terminus , with the readouts consisting of both HR cell death ( Figure 10 ) and resistance to P . infestans ( Figure 11 ) . Dissecting the NLR network architecture of plant species is not only useful for basic mechanistic studies but has also direct implications for breeding disease resistance into crop plants and reducing the autoimmune load of NLRs ( Chae et al . , 2016; Wu et al . , 2018; Adachi et al . , 2019a ) . Wild type and mutant N . benthamiana were propagated in a glasshouse and , for most experiments , were grown in a controlled growth chamber with temperature 22–25°C , humidity 45–65% and 16/8 hr light/dark cycle . Constructs for generating NRC4 knockout N . benthamiana were assembled using the Golden Gate cloning method ( Weber et al . , 2011; Nekrasov et al . , 2013; Belhaj et al . , 2013 ) . sgRNA4 . 1 and sgRNA4 . 2 were cloned under the control of the Arabidopsis ( Arabidopsis thaliana ) U6 promoter ( AtU6pro ) [pICSL90002 , The Sainsbury Laboratory ( TSL ) SynBio] and assembled in pICH47751 ( Addgene no . 48002 ) and pICH47761 ( Addgene no . 48003 ) , respectively as previously described ( Belhaj et al . , 2013 ) . Primers sgNbNRC4 . 1_F ( tgtggtctcaATTGAAAAACGGTACATACCGCAGgttttagagctagaaatagcaag ) , sgNbNRC4 . 2_F ( tgtggtctcaATTGAGTCAGGAATCTTGCAGCTGgttttagagctagaaatagcaag ) and sgRNA_R ( tgtggtctcaAGCGTAATGCCAACTTTGTAC ) were used to clone sgRNA4 . 1 and sgRNA4 . 2 . pICSL11017::NOSpro::BAR ( TSL SynBio ) , pICSL11021::35Spro::Cas9 ( Addgene no . 49771 ) , pICH47751::AtU6p::sgRNA4 . 1 , pICH47761::AtU6pro::sgRNA4 . 2 , and the linker pICH41780 ( Addgene no . 48019 ) were assembled into the vector pICSL4723 ( Addgene no . 48015 ) as described ( Weber et al . , 2011 ) resulting in construct pICSL4723::BAR::Cas9::sgRNA4 . 1::sgRNA4 . 2 that was used for plant transformation . Transgenic N . benthamiana were generated by TSL Plant Transformation team as described before ( Nekrasov et al . , 2013 ) . Genomic DNA of selected T2 N . benthamiana transgenic plants nrc4a/b_9 . 1 . 3 and nrc4a/b_1 . 2 . 1 was extracted using DNeasy Plant DNA Extraction Kit ( Qiagen ) . Primers NRC4_1_F ( GGAAGTGCAAAGGGAGAGTT ) , NRC4_1_R ( TCGCCTGAACCACAAACTTA ) , NRC4_2_F ( GGCAAGAATTTTGGATGTGG ) and NRC4_2_R ( CGAGGAACCCTTTTTAGGCAG ) were used in multiplex polymerase chain reaction ( PCR ) assays to amplify the region targeted by the two sgRNAs . Multiplex amplicon sequencing was performed by the Hi-Plex technique ( Lyon et al . , 2016 ) . Sequence reads were aligned to the reference N . benthamiana draft genome Niben . genome . v0 . 4 . 4 [Sol Genomics Network ( SGN ) , https://solgenomics . net/] , and NRC4a ( on scaffold Niben044Scf00002971 ) and NRC4b ( on scaffold Niben044Scf00016103 ) were further analysed . T3 lines from the selected T2 plants were used for the experiments . To generate NRC41-29-YFP expression construct , NRC41-29 coding sequence was amplified by Phusion High-Fidelity DNA Polymerase ( Thermo Fisher ) , and the purified amplicon was directly used in Golden Gate assembly with pICH85281 [mannopine synthase promoter+Ω ( MasΩpro ) , Addgene no . 50272] , pICSL50005 ( YFP , TSL SynBio ) , pICSL60008 [Arabidopsis heat shock protein terminator ( HSPter ) , TSL SynBio] into binary vector pICH47742 ( Addgene no . 48001 ) . Primers used for NRC41-29 coding sequences are listed in Supplementary file 1 . To generate an autoactive mutant of N . benthamiana NRC4 , the aspartic acid ( D ) in the MHD motif was substituted to valine ( V ) by site-directed mutagenesis using Phusion High-Fidelity DNA Polymerase ( Thermo Fisher ) . pCR8::NRC4WT ( Wu et al . , 2017 ) was used as a template . Primers NRC4_D478V_F ( 5’-Phos/ATGTTGCATCAGTTCTGCAAAAAGGAGGCT ) and NRC4_D478V_R ( 5’-Phos/GACGTGAAGACGACATGTTTTTATTTGACC ) were used for introducing the mutation in the PCR . The mutated NRC4 was verified by DNA sequencing of the obtained plasmid . pCR8::NRC4WT ( Wu et al . , 2017 ) or pCR8::NRC4DV without its stop codon were used as a level 0 modules for the following Golden Gate cloning . NRC4DV-3xFLAG was generated by Golden Gate assembly with pICH51266 [35S promoter+Ω promoter , Addgene no . 50267] , pICSL50007 ( 3xFLAG , Addgene no . 50308 ) and pICH41432 ( octopine synthase terminator , Addgene no . 50343 ) into binary vector pICH47732 ( Addgene no . 48000 ) . NRC4WT-6xHA and NRC4DV-6xHA were generated by Golden Gate assembly with pICH85281 ( MasΩpro ) , pICSL50009 ( 6xHA , Addgene no . 50309 ) , pICSL60008 ( HSPter ) into the binary vector pICH47742 . NRC4WT-YFP and NRC4DV-YFP were generated by Golden Gate assembly with pICH85281 ( MasΩpro ) , pICSL50005 ( YFP ) , pICSL60008 ( HSPter ) into binary vector pICH47742 . For free YFP expression construct , pAGM3212 ( YFP , TSL SynBio ) was assembled with pICH85281 ( MasΩpro ) and pICSL60008 ( HSPter ) into the binary vector pICH47742 by Golden Gate reaction . To reduce homo-affinity of YFP , YFP alanine ( A ) 206 was substituted to lysine ( K ) ( Zacharias et al . , 2002 ) , by site-directed mutagenesis using Phusion High-Fidelity DNA Polymerase ( Thermo Fisher ) . pAGM3212 ( YFP , TSL SynBio ) was used as a template . Primers used for mutagenesis are listed in Supplementary file 1 . The amplicons were directly used in Golden Gate assembly with pICH41308 ( Addgene no . 47998 ) or pAGM1301 ( Addgene no . 47989 ) . pICH41308::YFPA206K was assembled with pICH85281 ( MasΩpro ) and pICSL60008 ( HSPter ) into the binary vector pICH47742 by Golden Gate reaction . pAGM1301::YFPA206K was assembled with pCR8::NRC4DV or NRC41-29 amplicon , pICH85281 ( MasΩpro ) and pICSL60008 ( HSPter ) into the binary vector pICH47742 by Golden Gate reaction . To generate MADA motif mutants and chimeras of NRC4 , the full-length sequence of NRC4WT or NRC4DV was amplified by Phusion High-Fidelity DNA Polymerase ( Thermo Fisher ) with the forward primers listed in Supplementary file 1 . Purified amplicons were cloned into pCR8/GW/D-TOPO ( Invitrogen ) as a level 0 module . The level 0 plasmids were then used for Golden Gate assembly with pICH85281 ( MasΩpro ) , pICSL50009 ( 6xHA ) and pICSL60008 ( HSPter ) into the binary vector pICH47742 . To generate pTRBO::YFP , pTRBO::ZAR11-144-YFP , pTRBO::ZAR11-144F9A/L10A/L14A-YFP , pTRBO::NRC41-29-YFP and pTRBO::NRC41-29L9A/V10A/L14A-YFP plasmids , we used GENEWIZ Standard Gene Synthesis with custom vector cloning service into the pTRBO vector ( Lindbo , 2007a ) . To generate the Mu-STOP transposon ( Poussu , 2005 ) , entranceposon M1-KanR ( Mutation Generation System Kit , Thermo Fisher ) was used as a PCR template , and three translational stop signals were added to each transposon end by Phusion High-Fidelity DNA Polymerase and Mu-STOP primer ( GGAAGATCTGATTGATTGAACGAAAAACGCGAAAGCGTTTC ) . The 3’ A overhang was then introduced to the Mu-STOP amplicon by DreamTaq DNA polymerase ( Thermo Fisher ) , and the resulting Mu-STOP amplicon was cloned into pGEM-T Easy ( Promega ) . Mu-STOP transposon was then released from pGEM::Mu-STOP by BglII digestion and purified by GeneJET Gel Extraction Kit ( Thermo Fisher ) . 100 ng of the purified Mu-STOP transposon was mixed with 500 ng of the target plasmid , pICH47732::35SΩpro::NRC4DV-3xFLAG , and MuA transposase from the Mutation Generation System Kit ( Thermo Fisher ) . The in vitro transposition reaction was performed according to the manufacturer’s procedure and carried out at 30°C for 6 hr . The NRC4DV::Mu-STOP library was transformed into Agrobacterium tumefaciens Gv3101 by electroporation . Mu-STOP insertion sites were determined by colony PCR using DreamTaq DNA polymerase ( Thermo Fisher ) and PCR amplicon sequencing . For the PCR , we used a forward primer ( GAACCCTGTGGTTGGCATGCACATAC ) matching pICH47732 and a reverse primer ( CAACGTGGCTTACTAGGATC ) matching Mu-STOP transposon . Transient expression of NRC wild-type and mutants , as well as other genes , in N . benthamiana were performed by agroinfiltration according to methods described by Bos et al . ( 2006 ) . Briefly , four-weeks old N . benthamiana plants were infiltrated with A . tumefaciens strains carrying the binary expression plasmids . A . tumefaciens suspensions were prepared in infiltration buffer ( 10 mM MES , 10 mM MgCl2 , and 150 μM acetosyringone , pH5 . 6 ) and were adjusted to OD600 = 0 . 5 . For transient expression of NRC4WT-YFP , NRC4DV-YFP , NRC41-29-YFP , free YFP and the YFPA206K variants , the A . tumefaciens suspensions ( OD600 = 0 . 25 ) were mixed in a 1:1 ratio with an A . tumefaciens expressing p19 , the suppressor of posttranscriptional gene silencing of Tomato bushy stunt virus that is known to enhance in planta protein expression ( Lindbo , 2007b ) . HR cell death phenotypes were scored according to the scale of Segretin et al . ( 2014 ) modified to range from 0 ( no visible necrosis ) to 7 ( fully confluent necrosis ) . In Figure 2—figure supplement 2 and Figure 7—figure supplement 1 , cell death was visualized with Odyssey Infrared Imager ( 800 nm channel , LI-COR ) . Protein samples were prepared from six discs ( 8 mm diameter ) cut out of N . benthamiana leaves at 1 day after agroinfiltration and were homogenised in extraction buffer [10% glycerol , 25 mM Tris-HCl , pH 7 . 5 , 1 mM EDTA , 150 mM NaCl , 2% ( w/v ) PVPP , 10 mM DTT , 1x protease inhibitor cocktail ( SIGMA ) , 0 . 2% IGEPAL ( SIGMA ) ] . The supernatant obtained after centrifugation at 12 , 000 xg for 10 min was used for SDS-PAGE . Immunoblotting was performed with HA-probe ( F-7 ) HRP ( Santa Cruz Biotech ) or anti-GFP antibody ( ab290 , abcam ) in a 1:5000 dilution . Equal loading was checked by taking images of the stained PVDF membranes with Pierce Reversible Protein Stain Kit ( #24585 , Thermo Fisher ) . We used NLR-parser ( Steuernagel et al . , 2015 ) to identify NLR sequences from the protein databases of tomato ( SGN , Tomato ITAG release 2 . 40 ) , N . benthamiana ( SGN , N . benthamiana Genome v0 . 4 . 4 ) , Arabidopsis ( https://www . araport . org/ , Araport11 ) , sugar beet ( http://bvseq . molgen . mpg . de/index . shtml , RefBeet-1 . 2 ) , rice ( http://rice . plantbiology . msu . edu/ , Rice Gene Models in Release 7 ) and barley ( https://www . barleygenome . org . uk/ , IBSC_v2 ) . The obtained NLR sequences , from NLR-parser , were aligned using MAFFT v . 7 ( Katoh and Standley , 2013 ) , and the protein sequences that lacked the p-loop motif were discarded from the NLR dataset . The gaps in the alignments were deleted manually in MEGA7 ( Kumar et al . , 2016 ) and the NB-ARC domains were used for generating phylogenetic trees ( Figure 3—figure supplement 1—source data 1 ) . The neighbour-joining tree was made using MEGA7 with JTT model and bootstrap values based on 100 iterations ( Figure 3—figure supplement 1 ) . We removed TIR-NLR clade members from the final database , and retained all CC-NLR sequences , including the CCR-NLR ( RPW8-NLR ) , that possess N-terminal domains longer than 30 amino acids ( 988 protein sequences , Figure 3—source data 1 ) . The NB-ARC domain sequences from 988 proteins ( Figure 3—figure supplement 2—source data 2 ) were used to construct the CC-NLR phylogenetic tree in Figure 3—figure supplement 2 . The neighbour-joining tree was constructed as described above . For the tribe analyses , we extracted the N-terminal domain sequences , prior to NB-ARC domain , from the CC-NLR database ( Figure 3—source data 2 ) , and used the Tribe-MCL feature from Markov Cluster Algorithm ( Enright et al . , 2002 ) to cluster the sequences into tribes with BLASTP E-value cutoff <10−8 . NLRs in each tribe were subjected to motif searches using the MEME ( Multiple EM for Motif Elicitation ) v . 5 . 0 . 5 ( Bailey and Elkan , 1994 ) with parameters ‘zero or one occurrence per sequence , top five motifs’ , to detect consensus motifs conserved in ≥ 70% of input sequences . We used the most N-terminal motif detected in Tribe 2 from the MEME analysis to construct a hidden Markov model ( HMM ) for the MADA motif . Sequences aligned to the MADA motif were extracted in Stockholm format and used in hmmbuild program implemented in HMMER v2 . 3 . 2 ( Eddy , 1998 ) . The HMM was then calibrated with hmmcalibrate . This MADA-HMM ( Supplementary file 2 ) was used to search the CC-NLR database ( Figure 3—source data 1 ) with the hmmsearch program ( hmmsearch --max -o < outputfile > <hmmfile > <seqdb > ) . To estimate the false positive rate , hmmsearch program was applied to full Arabidopsis and tomato proteomes ( Araport11 and ITAG3 . 2 ) with the MADA-HMM and the output is displayed in Figure 4—source data 1 and discussed in the results section . P . infestans infection assays were performed by applying droplets of zoospore suspension on detached leaves as described previously ( Song et al . , 2009 ) . Briefly , leaves of five-weeks old wild-type and nrc4a/b N . benthamiana plants were infiltrated with A . tumefaciens solutions , in which each Agrobacterium containing a plasmid expressing RFP::Rpi-blb2 ( Wu et al . , 2017 ) was mixed in a 1:1 ratio ( OD600 = 0 . 5 for each strain ) with Agrobacterium containing either the empty vector , wild type NRC4 , or NRC4 variant . At 24 hr after agroinfiltration , the abaxial side of the leaves were inoculated with 10 µL zoospore suspension ( 100 zoospores/μL ) of P . infestans strain 88069 prepared according to the methods reported by Song et al . ( 2009 ) . The inoculated leaves were kept in a moist chamber at room temperature ( 21–24°C ) for 7 days , and imaged under UV light ( UVP Blak-Ray B-100AP lights – 365 nm ) with Wratten No . 8 Yellow Filter for visualization of the lesions . The camera setting was ISO 1600 , White Balance 6250K , F11 and 10 s exposure . We used the cryo-EM structure of activated ZAR1 ( Wang et al . , 2019b ) as template to generate a homology model of NRC4 . The amino acid sequence of NRC4 was submitted to Protein Homology Recognition Engine V2 . 0 ( Phyre2 ) for modelling ( Kelley et a . , 2015 ) . The coordinates of ZAR1 structure ( 6J5T ) were retrieved from the Protein Data Bank and assigned as modelling template by using Phyre2 Expert Mode . The resulting model of NRC4 comprised amino acids Val-5 to Glu-843 and was illustrated in CCP4MG software ( McNicholas et al . , 2011 ) . For localization analyses , leaf discs ( 6 mm in diameter ) of N . benthamiana leaves were made 2 days after agroinfiltration and were used for imaging . Images were captured with Leica SP8 resonant inverted confocal microscope ( Leica Microsystems ) . For excitation , Argon laser and Helium-Neon laser wer set to 514 nm and 633 nm , respectively . Hybrid detectors were used with 517–575 and 584–638 nm bandpass filters to capture YFP and RFP signals , respectively . Gain , laser intensities and zoom were kept the same for all images . Images were processed in FIJI ( Fiji Is Just ImageJ ) . The NRC4 sequences used in this study can be found in the Solanaceae Genomics Network ( SGN ) or GenBank/EMBL databases with the following accession numbers: NbNRC4 ( NbNRC4 , MK692737; NbNRC4a , Niben044Scf00002971; NbNRC4b , Niben044Scf00016103 ) .	Just like humans , plants get sick . They can be infected by parasites as diverse as fungi , bacteria , viruses , nematode worms and insects . But , also like humans , plants have an immune system that helps them defend against disease . Their first line of defence are disease resistance genes . Many of these genes encode so-called immune receptors , which are proteins that detect parasites and kick-off the immune response . Plant genomes may encode anywhere between 50 and 1000 immune receptors; some of which work solo as singletons , while others operate in pairs or as complex networks . Understanding how immune receptor genes have evolved would give fundamental knowledge about how they work , which in turn would set the stage for researchers to be able to use them to protect agricultural crops from disease . One driving force behind the evolution of many genes is gene duplication . Genes duplicate and afterwards the two copies can evolve in different ways . The original immune receptors are multi-tasking proteins that both detect parasites and trigger the immune response . Yet , following gene duplication , evolution has led to some immune receptors becoming dedicated to detection and losing the ability to trigger a defence response on their own . Now , Adachi et al . have discovered a molecular signature – named the MADA motif – that defines the subset of immune receptors that can trigger the immune response in plants . This motif is made of just 21 amino acids ( the building blocks of proteins ) at one end of the receptor and , remarkably , a short fragment of the protein containing this motif is enough to trigger a defence response when produced in plants . In contrast , the immune receptors that have specialized to only detect parasites have lost this molecular signature throughout evolution , presumably because they do not need it as they rely on their receptor partners to trigger defences instead . Every year , billions of dollars’ worth of food is lost to plant diseases . These new findings will enable the research community to classify disease resistance genes into categories to help deduce the network architecture of the plant immune system . A better understanding of this , and how networks of plant immune receptor evolve , should set the stage for breeding crop plants that are more able to resist diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "plant", "biology" ]	2019	An N-terminal motif in NLR immune receptors is functionally conserved across distantly related plant species
Diarrheal illness is a leading cause of antibiotic use for children in low- and middle-income countries . Determination of diarrhea etiology at the point-of-care without reliance on laboratory testing has the potential to reduce inappropriate antibiotic use . This prospective observational study aimed to develop and externally validate the accuracy of a mobile software application ( ‘App’ ) for the prediction of viral-only etiology of acute diarrhea in children 0–59 months in Bangladesh and Mali . The App used a previously derived and internally validated model consisting of patient-specific ( ‘present patient’ ) clinical variables ( age , blood in stool , vomiting , breastfeeding status , and mid-upper arm circumference ) as well as location-specific viral diarrhea seasonality curves . The performance of additional models using the ‘present patient’ data combined with other external data sources including location-specific climate , data , recent patient data , and historical population-based prevalence were also evaluated in secondary analysis . Diarrhea etiology was determined with TaqMan Array Card using episode-specific attributable fraction ( AFe ) >0 . 5 . Of 302 children with acute diarrhea enrolled , 199 had etiologies above the AFe threshold . Viral-only pathogens were detected in 22% of patients in Mali and 63% in Bangladesh . Rotavirus was the most common pathogen detected ( 16% Mali; 60% Bangladesh ) . The present patient+ viral seasonality model had an AUC of 0 . 754 ( 0 . 665–0 . 843 ) for the sites combined , with calibration-in-the-large α = −0 . 393 ( −0 . 455––0 . 331 ) and calibration slope β = 1 . 287 ( 1 . 207–1 . 367 ) . By site , the present patient+ recent patient model performed best in Mali with an AUC of 0 . 783 ( 0 . 705–0 . 86 ) ; the present patient+ viral seasonality model performed best in Bangladesh with AUC 0 . 710 ( 0 . 595–0 . 825 ) . The App accurately identified children with high likelihood of viral-only diarrhea etiology . Further studies to evaluate the App’s potential use in diagnostic and antimicrobial stewardship are underway . Funding for this study was provided through grants from the Bill and Melinda GatesFoundation ( OPP1198876 ) and the National Institute of Allergy and Infectious Diseases ( R01AI135114 ) . Several investigators were also partially supported by a grant from the National Institute of Diabetes and Digestive and Kidney Diseases ( R01DK116163 ) . This investigation was also supported by the University of Utah Population Health Research ( PHR ) Foundation , with funding in part from the National Center for Advancing Translational Sciences of the National Institutes of Health under Award Number UL1TR002538 . The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health . The funders had no role in the study design , data collection , data analysis , interpretation of data , or in the writing or decision to submit the manuscript for publication . Diarrheal diseases remain a leading cause of morbidity and mortality in children younger than five years worldwide , with approximately one billion episodes and 500 , 000 deaths annually . James et al . , 2018; Troeger et al . , 2018 . While a significant problem in all countries , the greatest burden of pediatric diarrhea exists in low- and middle-income countries ( LMICs ) , primarily in South Asia and sub-Saharan Africa . Troeger et al . , 2018 . Although the majority of diarrhea episodes are self-limiting and the mainstay of diarrhea treatment is rehydration , clinicians must also make decisions regarding appropriate use of diagnostics and for antibiotic prescribing . Guidelines from the World Health Organization ( WHO ) recommend against antibiotic use for the treatment of pediatric diarrhea , except for specific presentations of diarrhea such as suspicion of Vibrio cholerae ( V . cholerae ) with severe dehydration , blood in stool , or concurrent illness such as severe malnutrition World Health Organization , 2005a . For the majority of diarrhea etiologies , antibiotics are not recommended , particularly for viral causes of diarrhea in which antibiotics have no benefit . Bruzzese et al . , 2018 . Viral pathogens such as rotavirus , sapovirus , and adenovirus , are among the top causes of diarrhea in young children in LMICs , as shown in two large multi-center studies from LMICs , the Global Enteric Multicenter Study ( GEMS ) and the Malnutrition and Enteric Disease ( MAL-ED ) study Kotloff et al . , 2013; Platts-Mills et al . , 2018 . Laboratory testing by culture or molecular assays are often impractical when treating children with diarrhea in the majority of LMIC clinical settings due to time and resource constraints Bebell and Muiru , 2014 . As a result , clinicians often make decisions regarding antibiotic use based on non-evidence-based assumptions or broad syndromic guidelines Kotloff , 2017 . Unfortunately , physician judgment has been shown to poorly predict etiology and need for antibiotics in diarrheal infections . For example , patients presenting to Kenyan hospitals with diarrhea showed that syndrome-based guidelines for Shigella infection led to the failure to diagnose shigellosis in nearly 90% of cases Pavlinac et al . , 2016 . Accurate and cost-effective tools to better determine diarrhea etiology at the point-of-care without relying on laboratory tests are greatly needed to reduce antibiotic overuse while conserving scarce healthcare resources . Electronic clinical decision support systems ( CDSS ) incorporating prediction models may offer a solution to the challenges of determining diarrhea etiology in low-resource settings . CDSSs have been used in high-income country ( HIC ) settings to improve the accuracy of diagnosis and reduce costs by avoiding unnecessary diagnostic tests at the point-of-care Bright et al . , 2012 . CDSSs , especially as mHealth applications on smartphone mobile devices , hold great potential for implementing sophisticated clinical prediction models that would otherwise be impossible for providers to calculate manually . These tools can also enable flexibility by clinician choice or automation to optimize the clinical algorithm based on epidemiologic and clinical factors dominant in a given location . Despite opportunities to improve clinical care in a cost-aware mindset , there remains a paucity of data on the use of CDSS for infectious disease etiology determination and to support appropriate antibiotic use in LMICs Tuon et al . , 2017 . Our team previously derived and internally validated a series of clinical prediction models using data from GEMS , integrating characteristics of the present patient’s diarrhea episode ( patient-specific factors including age , blood in stool , vomiting , breastfeeding status , and mid-upper arm circumference ) with external data sources ( such as characteristics of recent patients , historical prevalence , climate , and seasonal patterns of viral diarrhea ‘viral seasonality’ ) in a modular approach Brintz et al . , 2021 . The best-performing model , which used ‘present patient’+ location-specific viral seasonality data , had an internally validated area under the receiver-operating characteristic curve ( AUC ) of 0 . 83 Brintz et al . , 2021 . The objective of this study was to prospectively externally validate the models for the prediction of viral-only etiology of diarrhea in children aged 0–59 months in Bangladesh and Mali and demonstrate a proof-of concept for the incorporation of the primary model ( ‘present patient’+ location-specific viral seasonality ) into a mobile CDSS software application ( ‘App’ ) for use in LMIC settings with high diarrheal disease burden . A prospective , observational cohort study was conducted in Dhaka , Bangladesh and Bamako , Mali . Enrollment was conducted in Bangladesh at the Dhaka Hospital of the International Center for Diarrhoeal Disease Research , Bangladesh ( icddr , b ) rehydration ( short stay ) unit and in Mali at the Centres de Santé de Référence ( CSREF ) and the Centres de Santé Communautaires ( CSCOM ) in Commune V and VI in Bamako , Mali . These locations were selected because of their geographic proximity to GEMS study sites from which the clinical prediction models were trained , without using the same sites . Participants were enrolled in Bangladesh during November and December 2019 and Mali during January and February 2020 . The Dhaka Hospital of the icddr , b provides free clinical services to the population of the capital city of Dhaka , Bangladesh and surrounding rural districts and cares for over 100 , 000 patients with acute diarrhea each year . The CSREF and CSCOM of communes V and VI in Mali serve a catchment area of 2 million people . CSCOM provides basic care such as family planning , vaccination and outpatients visits , while patients with severe illness are referred to the CSREF where there is capacity for hospital admission for medical conditions and for basic and intermediate surgeries . Patients under five years of age ( 0–59 months ) with symptoms of acute diarrhea were eligible for enrollment . Acute diarrhea was defined as three or more loose stools per day for less than seven days . Patients were excluded using the following criteria: no parent or primary caretaker available for consent , diarrhea lasting seven days or longer , fewer than three loose stools in the prior 24 hr , or having a diagnosis of severe pneumonia , severe sepsis , meningitis , or other condition aside from gastroenteritis . General practice nurses were hired specifically to collect data at both study sites , and study nurses had no other patient care responsibilities during the study period . Nurses received training in study procedures under the guidance of the research investigators . Training topics included: screening procedures , obtaining informed consent , collecting clinical data and laboratory procedures . Study nurses also received practical hands-on training regarding the use of the App to ensure all nurses were comfortable with entering data and using the devices during clinical workflow . ‘Modular’ clinical prediction models for the outcome of viral etiology of pediatric diarrhea were previously derived and internally validated with full details previously published by Brintz et al in 2021 Brintz et al . , 2021; Brintz et al . , 2020 . Briefly , a series of five models predicting viral etiology were independently developed based on the hypothesis that including location-specific ‘external’ data sources ( i . e . characteristics such as recent patients or climate data in addition to the present patient’s characteristics ) , may improve predictive performance . This study team’s prior work described the development of these predictive models that integrates multiple sources of data in a principled statistical framework using a ‘post-test odds formulation’ . This method incorporates observed prior knowledge of a prediction , typically using prior known location-specific prevalence ( e . g . historical prevalence of viral diarrhea in Mali ) , as pre-test odds and then updates the pre-test odds using a single model or series of models based on current data . It also enables electronic real-time updating and flexibility in a ‘modular’ fashion , such that the component models can be flexibly included or excluded according to data availability , an important consideration for LMIC settings in which prior epidemiologic data may be unavailable . The post-test odds formulation combines the likelihood ratios derived from these independent models along with pre-test odds into a single prediction . In order to externally validate the predictions from the post-test odds , we processed the data from this study to match the variables used in previously trained models as closely as possible . Table 1 shows the terminology used to refer to each model and the features included in each model . The derived models used clinical , historical , anthropometric and microbiologic data from the GEMS study , a large case-control study conducted at seven sites in Asia and Africa ( The Gambia , Kenya , Mali , Mozambique , Bangladesh , India , and Pakistan ) which enrolled 22 , 568 children under 5 years , including 9439 children with moderate/severe diarrhea and 13 , 129 controls Kotloff et al . , 2013 . Demographics , predictors and viral-only outcome data from the development datasets from GEMS in Bangladesh and Mali are shown in Supplementary file 1 . Additional location-specific sources of data used for model development included local climate ( i . e . weather ) data , and site-specific viral diarrhea seasonality modeled using sine and cosine curves ( ‘viral seasonality’ ) . Pre-test odds were generated using epidemiologic data based on historical prevalence from the same study site ( ‘historical patient’ ) and from the past 4 weeks ( ‘recent patient’ ) at the same study site . More specifically , local weather data proximate to each site’s health centers was obtained using the National Oceanic and Atmospheric Administration ( NOAA ) Integrated Surface Database . Climate model features include rain and temperature averages based on a 2-week aggregation of the inverse-distance weighted average of the nearest five NOAA-affiliated weather stations to the hospital sites . Weather stations at a distance of greater than 200 km were excluded . Standardized seasonal sine and cosine curves with a periodicity of 1 year , sin ( 2πt365 . 25 ) and cos ( 2πt365 . 25 ) , where t is based on the date , were used to model the location-specific seasonal patterns of viral etiology of diarrhea Brintz et al . , 2021 . The seasonal sine and cosine values as well as temperature and rain averages ( climate ) were calculated for the dates in this study as described previously Brintz et al . , 2021 . In Bangladesh , due to the high volume of potentially eligible patients presenting daily to icddr , b Dhaka Hospital , study staff randomly selected participants for enrollment on arrival 9am-5pm Sunday to Thursday . Random selection was accomplished using a black pouch filled with white and blue marbles in a preset ratio . Study nurses drew a marble each time a patient presented to the rehydration unit . If a blue marble was pulled , the patient was screened for inclusion and exclusion criteria as described above . After each marble was pulled , it was returned to the bag and shaken . An average of approximately eight patients were enrolled per working day which allowed for high-quality data collection and integrity of all study protocols to be maintained . In Mali , a consecutive sample of patients presenting with acute diarrhea were enrolled . After initial assessment by the facility doctor , children with acute diarrhea were referred to the study team for screening . Study staff were located in the intake area and potential participants’ information was recorded in a screening log . All patients presenting with diarrhea were assessed for eligibility . After screening , research staff provided the parent or guardian with information about the purpose of the study , risks and benefits in Bangla ( Bangladesh ) and Bambara ( Mali ) language . Research staff then obtained written consent if the parent or guardian agreed to participate on behalf of the child . In cases where the parent or guardian was illiterate , the consent form was marked with a thumbprint for signature , based on standards for informed consent at icddr , b and CVD-Mali . In these cases , a witness ( other than study staff ) also signed the consent form . If a child arrived without a parent or guardian in attendance , they were not considered for enrollment in the study . After enrollment , study staff collected demographic , historical and clinical information from the parent or guardian . All information was collected on a paper case report form ( CRF ) . The ‘present patient’ clinical variables were entered into the App on a mobile device ( Bangladesh: Samsung Galaxy A51; Mali: Samsung Galaxy Note 10 ) by two different study nurses independently to ensure reliability; variables were age in months for 0–23 months and in years for 2–4 years , blood in stool since illness began , history of vomiting ( three or more times a day since illness started ) , breastfeeding status ( ‘currently’ ) , and mid-upper arm circumference ( MUAC ) . During data collection , study investigators noted that nearly all participants in Bangladesh had reported ‘yes’ to the question regarding history of vomiting . Given vomiting was not expected in all patients especially those with non-viral diarrhea etiology , it was determined after speaking with the study nurses that the phrasing of the question sometimes led patients to respond ‘yes’ if there had been regurgitation with feeding or ORS administration , rather than actual vomiting . The question format was then revised for the remainder of the study enrollment in Bangladesh , and prior to any patient enrollment in Mali , to clarify the definition of vomiting . The App calculated the probabilities specific to each data source ( present patient , recent patient , historical patient , climate , and viral seasonality ) . The model deployed on the App used the present patient and location-specific viral seasonality components along with the location-specific viral diarrhea alone prevalence from GEMS as pre-test odds . The App results were not used for clinical decision-making to allow first for the external validation and second to iterate the software in response to engineering challenges exposed from ‘live deployment’ . All patients were treated according to standard local clinical protocols , and the clinicians caring for patients were blinded to any study data collected in order to prevent any undue influence in clinical care . Study procedures were not allowed to delay any immediately necessary patient care , such as the placement of an intravenous line or delivery of intravenous fluid to the patient . As the primary outcome was determined only after all clinical enrollment and procedures concluded and laboratory analysis was not linked to predictors , all assessments of predictors and the outcome were blinded . The first available stool specimen after enrollment of the participant was collected . Study participation concluded after a stool sample was obtained . Nurses were unaware of the etiology of diarrhea at the time of clinical assessment as microbiological testing was conducted only after the study period concluded . Stool samples were collected in a sterile plastic container and then transferred to two separated 2 mL cryovials – one vial with 1 mL stool only and one vial for storage in 70% ethanol ( Bangladesh ) or 95% ethanol ( Mali ) . Samples were stored at –20°C or –80°C freezer for processing . At the conclusion of the study , samples were thawed , underwent bead beating and nucleic acid extraction using the QIAamp Fast DNA Stool Mini Kit . Total Nucleic acid was mixed with PCR buffer and enzyme and loaded onto custom multiplex TaqMan Array Cards ( TAC ) containing compartmentalized probe-based quantitative real-time PCR ( qPCR ) assays for 32 pathogens at the icddr , b or CVD-Mali laboratories ( see Supplementary file 2 for full list of pathogen targets ) . The outcome ( dependent ) variable was defined as the presence or absence of viral-only etiology . Diarrheal etiology was determined for each patient using qPCR attribution models developed previously by Liu et al . , 2016 Viral-only diarrhea was defined as a diarrhea episode with at least one viral pathogen with an episode-specific attributable fraction ( AFe ) threshold of ≥0 . 5 and no bacterial or parasitic pathogens with an AFe ≥0 . 5 . This clinically relevant outcome measure was selected because patients with viral-only diarrhea should not receive antibiotics . Other etiologies were defined as having a majority attribution of diarrhea episode by at least one other non-viral pathogen . Patients without an attributable pathogen ( unknown final etiology for diarrheal episode ) were excluded from this analysis since the cause of the diarrheal episode could not be definitively determined . However , prior studies by this research team have shown that these cases have a similar distribution of viral predictions using a model with presenting patient information as those cases with known etiologies and this study had similar results Brintz et al . , 2020 . For the primary analysis , MUAC measurements collected by the two study nurses were averaged . Patients were considered to have ‘bloody stool’ only if report from both nurses agreed on bloody stool . For children older than 2 years , age in months was rounded down to the nearest year in months ( i . e . 42 months was rounded to 36 months ) to match the user interface on the software . Using clinical information gathered from the data sources ( present patient , recent patient , historical patient , climate , viral seasonality ) , predictions using post-test odds formulation with the developed models were made . The primary model deployed in the App , selected based on the best-performing model from the derivation and internal validation , used the present patient data and viral seasonality components . Model performance for the prediction of viral-only diarrhea was calculated using AUC for each model to evaluate discrimination; calibration was assessed using calibration-in-the-large and calibration slope Steyerberg and Vergouwe , 2014 . The target for calibration slope is 1 , where < 1 suggests predictions are too extreme and >1 suggests predictions are too moderate . The target for calibration intercept is 0 , where negative values suggest overestimation and positive values suggest underestimation Van Calster et al . , 2019 . We estimated the calibration coefficients by regressing predicted values versus the observed proportion of viral cases , calculated using the observed proportion of viral cases within 0 . 05 plus or minus the predicted probability . For the primary analysis , data from the time period using the original vomiting question in Bangladesh was excluded; however , site specific results incorporate all Bangladesh data for the purpose of highlighting the potential impact of misunderstanding of the required App input questions in real-world scenarios . The reliability of the predictor data entered independently by the two study nurses was assessed using Cohen’s kappa coefficient ( κ ) which is a calculation of inter-observer agreement for categorical data . The performance of additional models integrating other available components was assessed in post-hoc secondary analysis for comparison . These alternate models included the following components: ( 1 ) ‘present patient’ data only ( 2 ) ‘present patient’+ climate data ( climate ) ( 3 ) ‘present patient’+ historical prevalence pre-test odds ( historical ) ( 4 ) ‘present patient’+ recent patient pre-test odds ( recent ) as described in Table 1 . For all analyses , a two-tailed p value of 0 . 05 was considered statistically significant . R ( R Foundation for Statistical Computing , Vienna , Austria ) were used for all analyses . Standard guidelines from the transparent reporting of a multivariable prediction model for individual diagnosis ( TRIPOD ) Checklist for Prediction Model Validation were used . The de-identified dataset is available online in Supplementary file 5 . The previously derived clinical prediction model had an internally validated AUC of 0 . 83 Brintz et al . , 2021 . In order to ensure that the confidence intervals around the estimates in the validation study would not cross 0 . 75 ( generally considered to be a marker of adequate accuracy for a clinical prediction model ) a margin of error around the AUC of 0 . 08 was required . Using the approximate variance estimate for AUC from the literature , and assuming a prevalence of a viral only etiology of approximately 30% in our sample , a sample size of 300 patients ( 150 patients per site ) was required . A total of 302 patients were recruited from the two study sites with 152 patients in Bangladesh and 150 patients in Mali . All patients except two in Bangladesh had a stool sample collected for TAC testing , for a total of 300 patients with TAC results ( Figure 2 ) . Diarrhea etiology was assigned for a total of 199 patients ( 66%; 130 in Bangladesh and 69 in Mali ) for inclusion in the final analysis ( Table 2 ) . The median [IQR] age of included patients was 12 months; there was a predominance of male patients at both study sites ( 61 . 8% overall ) with 59 . 2% in Bangladesh and 66 . 7% in Mali . Sociodemographic and clinical characteristics of the study participants are shown in Table 2 . In Bangladesh , 73 participants were asked about vomiting history using the original question format and 57 using the revised question format . Antibiotic use prior to health facility presentation was common with 66% of participants in Bangladesh reporting antibiotic use for the current illness ( most commonly azithromycin , ciprofloxacin , and metronidazole alone or in combination ) and 13 . 1% of participants Mali ( most commonly cotrimoxazole , amoxicillin , metronidazole ) . Viral-only etiologies of diarrhea were the predominant cause of illness among patients with assigned etiology overall ( 42% ) although viral-only etiologies were more common in Bangladesh ( 63% ) compared to Mali ( 22 . 0% ) . Rotavirus was the most common viral pathogen detected in both study sites ( 38% overall ) while the rate was higher in Bangladesh ( 60% ) compared to Mali ( 16% ) . Other viral causes ( e . g . adenovirus , astrovirus , norovirus ) were less common in both sites . Diarrhea etiology was unable to be assigned in a substantial proportion of the patients in Mali ( 54% ) compared to Bangladesh ( 13 . 3% ) . The prevalence of the various pathogens detected are listed in Table 3 . When applied to both sites , the primary model which included present patient+ viral seasonality component performed better than secondary models at discriminating a viral-only etiology from all other known etiologies with an AUC of 0 . 754 ( 95% CI 0 . 665–0 . 843 ) ( Table 4 ) . However , this same model was not well calibrated with an estimate of calibration-in-the-large of α = −0 . 393 ( 95% CI -0 . 455–-0 . 331 ) and a calibration slope of β = 1 . 287 ( 95% CI 1 . 207 , 1 . 367 ) . The present patient+ historical patient model having low discrimination AUC 0 . 702 ( 95%CI 0 . 603–0 . 800 ) was the best calibrated with estimates of 0 . 036 ( −0 . 031–0 . 102 ) and 1 . 063 ( 0 . 943–1 . 184 ) for α and β , respectively . Combinations of more than two components including the present patient information reduced the discriminatory performance of the model , likely due the positive correlation between components . When looking at the sites individually , the present patient+ recent patient model performed best in Mali with an AUC of 0 . 783 ( 95% CI 0 . 705–0 . 86 ) while the primary model with the present patient+ viral seasonality component data performed best in Bangladesh with an AUC of 0 . 71 ( 95% CI 0 . 595–0 . 825 ) ( Table 5 ) . When the dates where the vomiting question was asked incorrectly are included in the analysis , all models performed less well with the viral seasonality and climate models performing best in Bangladesh with an AUC of 0 . 610 ( 95% CI: 0 . 523–0 . 697 ) and 0 . 621 ( 95% CI: 0 . 510–0 . 732 ) , respectively ( Table 5 ) . While the aim of the study was not to provide a binary classification decision but rather a continuous predicted risk of viral etiology , the numbers of false positives and false negatives ( i . e , bacteria/protozoal etiologies misidentified as viral and vice versa ) at various viral probability thresholds for ‘present patient’ and the ‘present patient+ viral seasonality’ models are shown in Figure 3 . Notably , the ‘present patient+ viral seasonality’ model tended to have fewer false negatives while the ‘present patient’ model fewer false positives at various thresholds . However , the sensitivity and specificity of the ‘present patient’ and ‘present patient+ viral seasonality’ models were similar ( Figure 3 ) . The agreement between the two study nurses’ independent assessments of the required predictor variables and estimated viral-only etiology risk calculated by the App for each nurse’s assessment was evaluated and showed excellent reliability and agreement between the nurses; results are shown in Supplementary file 3 . Data from the App were recorded through the cellular network to a database as well as on the paper CRF . To monitor integrity of the data transfer to the App database from a technical perspective , data were compared between the two records . The agreement between the App database and study nurse’s paper CRF are shown in Supplementary file 4 and Figure 4—figure supplement 1 . The Bangladesh study exposed that one field ( ‘breastfeeding’ ) did not sync successfully which was addressed at the midpoint between the Bangladesh and Mali study phases . The Mali study used an updated software version ( v2 . 2 . 5 ) ; the reliability and agreement between the App prediction of viral etiology with the post-hoc predictions are shown in Figure 4 . Our findings have several limitations . The study was conducted at study sites located in close proximity to sites from which the prediction model data were trained on . The prediction models may not therefore be generalizable to other locations , sites that have poor epidemiologic data on diarrhea etiologies , or locations that have initiated new vaccination campaigns ( e . g . rotavirus ) . Rotavirus vaccination has not yet been introduced in Bangladesh , although a cluster randomized trial was conducted in rural Matlab in 2008–2011 , and rotavirus vaccine introduction into the routine immunization program has been planned; thus , we do not expect any effect of the vaccine on the present study results for the Bangladesh study site . However , in Mali , rotavirus vaccination was introduced into the routine immunization program in 2014 , as of 2019 official coverage rate was 63% in infants 12–23 months Zaman et al . , 2017; Rota Council , 2021; Mali , 2019 . We recognize that the App may not perform as well in places where the rotavirus vaccine has been introduced given the change in viral diarrhea epidemiology in the years since the data were collected from which the models were trained . However , there remain many countries without access to the rotavirus vaccine , notably in Asia where only a few countries have introduced the vaccine nationally or sub-nationally Rota Council , 2021 . For this reason , the App may be most useful in these settings . Interestingly , the models performed better at the Mali study site compared to Bangladesh despite fairly high coverage rates of rotavirus vaccine with subsequently much lower rates of rotavirus detected on TAC PCR . These findings may indicate that the clinical features associated with viral diarrhea may be pathogen-independent , and that seasonal variations in viral diarrhea may still be relevant even when overall rates of rotavirus are substantially lower than in prior time periods . Lastly , as the rotavirus vaccine is planned to be introduced in an increasing number of countries globally , we also recognize that the model may need to be updated to account for changes in epidemiology as rotavirus vaccine is increasingly introduced and rotavirus becomes a less prominent cause of viral diarrhea . Another limitation is that approximately one-third of patients enrolled in this study did not have diarrhea etiology assigned using the predefined threshold AFe >0 . 5 from the TAC testing data . The much higher rate of diarrhea etiology assignment in Bangladesh compared to Mali may be partially attributable to the study being conducted during the annual rotavirus season in Bangladesh ( typically November to February each year ) with rotavirus strongly associated with diarrhea and therefore easier to attribute Dey et al . , 2020 . However , the predominance of rotavirus may limit the generalizability of this study’s findings . Additionally , given the known seasonal cycles of viral diarrhea infections , ideally , at least one full annual cycle of data would have been collected at each study site to account for seasonal variations in diarrhea epidemiology . However , funding and resource limitations , and the prioritization of including two diverse study sites precluded a longer duration of study enrollment . Notably , the occurrence of a large seasonal rotavirus peak during the data collection period in Bangladesh may be a reason why the models performed sub-optimally at this site in comparison to Mali . We believe that , had resources been available for a longer duration of study enrollment that accounted for greater seasonal variability , our models would have shown better performance characteristics than in the present analysis . Studies to further validate these models using additional study sites with different seasonal cycles of viral diarrhea over at minimum one full annual cycle are recommended . Changes over time in the epidemiology of diarrhea at the study sites as well as the internal validation being conducted in a very large dataset ( > 20 , 000 patients ) from the GEMS study compared to this study with 300 patients over a short period during a surge of rotavirus likely also contributed to the drop in AUC between the internal validation and external validation . Lastly , there is currently no existing gold standard for threshold of AFe that should be used to assign etiology and the threshold of 0 . 5 was set a priori based on expert consultation; however , the effect of using this cut-off has not yet been explored . A lower AFe threshold may still indicate likely etiology of diarrhea . Notably , there were no norovirus infections detected using these pre-determined thresholds . This may have been due to the particularly high prevalence of other pathogens especially rotavirus in Bangladesh during this time period , the previously documented lower prevalence of norovirus in Bangladesh in November-January compared to other months , as well as the relatively short period of participant enrollment , although using a different AFe cutoff may have detected possible cases of norovirus Satter et al . , 2021 . Despite these limitations , we have externally validated , in a prospective multicenter study , a smartphone-based clinical decision-support system that dynamically implements clinical prediction models based on real-time data streams . A clinical trial led by this study team is currently underway to evaluate the impact of this tool on antibiotic prescribing behaviors , and further research is needed to better understand how these tools could be best adapted for use by practicing clinicians in busy LMIC settings .	Diarrhea is one of the most common illnesses among children worldwide . In low- and middle-income countries with limited health care resources , it can be deadly . Diarrhea can be caused by infections with viruses or bacteria . Antibiotics can treat bacterial infections , but they are not effective against viral infections . It can often be difficult to determine the cause of diarrhea . As a result , many clinicians just prescribe antibiotics . However , since diarrhea in young children is often due to viral infections , prescribing unnecessary antibiotics can cause children to have side effects without any benefit . Excessive use of antibiotics also contributes to the development of bacteria that are resistant to antibiotics , making infections harder to treat . Scientists are working to develop mobile health tools or ‘apps’ that may help clinicians identify the cause of diarrhea . Using computer algorithms to analyze information about the patient and seasonal infection patterns , the apps predict whether a bacterial or viral infection is the likely culprit . These tools may be particularly useful in low- or middle-income country settings , where clinicians have limited access to testing for bacteria or viruses . Garbern , Nelson et al . previously built an app to help distinguish cases of viral diarrhea in children in Mali and Bangladesh . Now , the researchers have put their app to the test in the real-world in a new group of patients to verify it works . In the experiments , nurses in Mali and Bangladesh used the app to predict whether a child with diarrhea had a viral or non-viral infection . The children’s stool was then tested for viral or bacterial DNA to confirm whether the prediction was correct . The experiments showed the app accurately identified viral cases of diarrhea . The experiments also showed that customizing the app to local conditions may further improve its accuracy . For example , a version of the app that factored in seasonal virus transmission performed the best in Bangladesh , while a version that factored in data from recent patients in the past few weeks performed the best in Mali . Garbern and Nelson et al . are now testing whether their app could help reduce unnecessary use of antibiotics in children with diarrhea . If it does , it may help minimize antibiotic resistance and ensure more children get appropriate diarrhea care .	[ "Abstract", "Introduction", "Materials", "and", "methods", "Results", "Discussion" ]	[ "epidemiology", "and", "global", "health", "medicine" ]	2022	External validation of a mobile clinical decision support system for diarrhea etiology prediction in children: A multicenter study in Bangladesh and Mali
The age of large-scale genome-wide association studies ( GWAS ) has provided us with an unprecedented opportunity to evaluate the genetic liability of complex disease using polygenic risk scores ( PRS ) . In this study , we have analysed 162 PRS ( p<5×10−05 ) derived from GWAS and 551 heritable traits from the UK Biobank study ( N = 334 , 398 ) . Findings can be investigated using a web application ( http:‌//‌mrcieu . ‌mrsoftware . org/‌PRS‌_atlas/ ) , which we envisage will help uncover both known and novel mechanisms which contribute towards disease susceptibility . To demonstrate this , we have investigated the results from a phenome-wide evaluation of schizophrenia genetic liability . Amongst findings were inverse associations with measures of cognitive function which extensive follow-up analyses using Mendelian randomization ( MR ) provided evidence of a causal relationship . We have also investigated the effect of multiple risk factors on disease using mediation and multivariable MR frameworks . Our atlas provides a resource for future endeavours seeking to unravel the causal determinants of complex disease . Developing our understanding of how modifiable social , behavioural and physiological factors influence risk of disease is of vital importance to improve effective medical treatment and preventative interventions ( Abraham et al . , 2016 ) . Genetic factors may also contribute substantially to disease susceptibility , as demonstrated by recent large-scale genome-wide association studies ( GWAS ) which have uncovered thousands of trait-associated single nucleotide polymorphisms ( SNPs ) throughout the human genome . However , typically the magnitude of effect and variance explained by one of these common genetic variants is small ( Visscher et al . , 2017 ) . Polygenic risk scores ( PRS ) , commonly defined as the sum of trait-associated SNPs weighted by their effect sizes , harness findings from GWAS to provide an overall measure of an individual’s genetic liability to develop disease ( Torkamani et al . , 2018 ) . Although early applications of PRS were found to be underwhelming in terms of disease prediction ( Ripatti et al . , 2010 ) , breakthroughs in the scale of GWAS and accessibility to biobank scale datasets have substantially improved their performance ( Khera et al . , 2018; Lee et al . , 2018 ) . As such , they hold considerable potential to improve early disease prognosis and treatment plan formulation ( Lewis and Vassos , 2017 ) . Along with the emerging utility of PRS to predict disease , they have also been previously used to evaluate putative causal relationships ( Davies et al . , 2018; Palmer et al . , 2012 ) . For example , instead of using a coronary heart disease ( CHD ) PRS to predict incidence of this disease , studies have investigated whether scores for known risk factors , such cholesterol and lipid levels ( Holmes et al . , 2015 ) , are also strongly associated with CHD incidence . One such approach in this paradigm is Mendelian randomization ( MR ) , a method by which genetic variants are leveraged as instrumental variables to investigate causal relationships between modifiable risk factors and disease outcomes ( Davey Smith and Ebrahim , 2003; Davey Smith and Hemani , 2014 ) . MR is typically limited to using SNPs which survive conventional GWAS corrections ( i . e . p<5×10−08 ) , which may lack statistical power if these variants do not explain a large proportion of trait variance . In contrast , PRS derived using a more lenient threshold ( e . g . p<5×10−05 ) can help recover some of this missing heritability due to a larger number of SNPs being included . This may help improve detection rates for causal relationships , which can be particularly useful when evaluating associations between genetic liability for a given trait and hundreds of diverse health outcomes . Such endeavours are commonly referred to as phenome-wide association studies ( Denny et al . , 2013; Fritsche et al . , 2018; Krapohl et al . , 2016; Millard et al . , 2015 ) . To investigate this we undertook a preliminary simulation study to compare the performance of using a PRS to detect causal relationships with a popular MR approach ( the inverse variance weighted ( IVW ) method ( Burgess et al . , 2013 ) ) ( Figure 1 ) . Results indicated that , although using a PRS provides higher statistical power , it also suffers from substantive false positive rates due to horizontal pleiotropy , the phenomenon whereby a gene influences multiple traits via independent biological pathways ( Davey Smith and Hemani , 2014 ) . SNPs which are known to be pleiotropic with large effects on different and diverse traits have been found to distort findings from PRS analyses ( Felsky et al . , 2018 ) . As a consequence , findings from phenome-wide association studies using a PRS may be useful in terms of highlighting putative causal associations , although robust evaluations are necessary to investigate results . We therefore propose using various sensitivity analyses developed in the field of MR to discern whether PRS associations represent causal relationships or not . To facilitate such future analyses , an accessible resource to evaluate associations between disease genetic liability and complex traits from across the human phenome should prove to be of considerable value . In this study , we have constructed 162 different PRS ( based on p<5×10−05 ) using findings from large-scale GWAS and evaluated their association with 551 traits in up to 334 , 398 individuals enrolled in the UK Biobank study ( Bycroft et al . , 2018; Sudlow et al . , 2015 ) . To disseminate these findings , we have developed a web application to examine and visualise this derived atlas of associations . We have also undertaken follow-up analyses to demonstrate the usefulness of this resource to help identify putative causal relationships . Firstly , we have interpreted findings from a hypothesis-free scan of associations between the schizophrenia PRS and each of the 551 traits . We demonstrate that amongst these findings are associations which may likely reflect underlying causal relationships . We have also showcased the utility of evaluating the association between all 162 PRS and a single outcome using our atlas . Using gout susceptibility as an example , we demonstrate how recently developed methodology ( mediation MR and multivariable MR ) can be applied to evaluate the effects of multiple risk factors on disease risk . Overall , we undertook 89 , 262 tests to investigate the association between 162 different PRS derived from GWAS ( Supplementary file 1a ) and 551 complex traits from the UK Biobank study ( Supplementary file 1b ) . PRS were constructed using independent SNPs for each GWAS ( p<5×10−05 ) based on r2< 0 . 001 using genotype data from European individuals ( CEU ) from phase 3 ( version 5 ) of the 1000 Genomes project ( Abecasis et al . , 2012 ) . As opposed to the conventional GWAS cut-off of p<5×10−08 , the threshold of p<5×10−05 was selected to incorporate additional SNPs into scores which may explain additional heritability for GWAS traits . Furthermore , this allowed us to create PRS for traits which had no SNPs surviving conventional GWAS corrections , as well as increasing the number of SNPs used in scores for traits with only a small number of GWAS hits . Our final sample size for analysis consisted of 334 , 398 individuals . This was determined using a strict exclusion criterion to reduce false positive associations , removing individuals with withdrawn consent , evidence of genetic relatedness or who were not of ‘white European ancestry’ based on a K-means clustering ( K = 4 ) . Of the 162 GWAS we identified , 11 reported that they included UK Biobank participants in their analysis . As this may lead to overfitting , the PRS for these 11 traits were not weighted to reduce this source of bias . To demonstrate this , we evaluated the association of the sleep duration PRS in the UK Biobank study , weighting SNPs based on a GWAS involving the interim release of this dataset ( Jones et al . , 2016 ) ( Supplementary file 1c ) . However , this only mitigates this limitation , and as such these scores in particular require extensive follow-up analyses . In case they are still useful for follow-up analyses despite overlapping with UK Biobank , these scores have been clearly flagged in Supplementary file 1a by being allocated to the ‘unweighted’ subcategory . In this study we have only interpreted findings from associations with PRS derived using the p<5×10−05 threshold . However , analyses have been repeated using scores derived using the conventional GWAS threshold of p<5×10−08 for future studies that wish to evaluate these results . Complex traits from the UK Biobank study were selected based on p<0 . 05 from previously undertaken heritability analyses within this study ( Neale Lab , 2017 ) . This threshold was chosen as a heuristic to highlight associations worth pursuing in further detail . A web app to query and visualise these results can be found at http://mrcieu . mrsoftware . org/PRS_atlas/ . Stratifying the UK Biobank sample into deciles based on their PRS supported previous findings in the literature demonstrating the ability of PRS to predict risk of disease . For example , comparing the highest and lowest deciles of the coronary heart disease ( CHD ) PRS found that individuals had increased odds of 3 . 64 to develop this disease ( based on the ICD10 code ‘I25’ ) . A recent study by Khera and colleagues ( Khera et al . , 2018 ) reported a similar odds ratio for CHD in their analysis ( OR:>3 . 0 for the highest 8% of individuals based on their PRS ) . However , we note that they identified a higher area under curve in their analysis ( 0 . 806 ) , which is likely attributed to tuning parameters such as LD clumping , along with covariates adjusted for in their analysis . Combining this PRS with scores for established causal risk factors for CHD suggested that they can help improve polygenic prediction ( namely low density lipoprotein ( LDL ) cholesterol and myocardial infarction ) , although integrating any associated scores in a hypothesis-free manner may hinder prediction ( Figure 2 ) . This could potentially due to the increase in variance incorporated into prediction analyses from scores that do not directly influence CHD , or alternatively may indicate that they are spurious associations . Additional research is required to evaluate the contribution of multiple PRS as predictors of a single outcome . Doing so may help develop a greater understanding regarding which traits can help predict disease outcomes using PRS . Amongst other findings , we observed that participants had increased odds of 2 . 43 in terms of obtaining a University or College degree when comparing top and bottom deciles for the years of schooling PRS . Other noteworthy examples included a 3 . 48 fold increase in odds of taking atorvastatin as medication when comparing the extreme deciles for the LDL PRS . We also observed that participants in the highest decile for the ulcerative colitis PRS had increased odds of 5 . 36 in terms of developing this disease in comparison to those in the lowest decile ( based on the ICD10 code ‘K51’ ) . To demonstrate the value of this atlas of results , we have investigated some of the strongest associations detected between the schizophrenia PRS and all 551 complex traits analysed in the UK Biobank study ( Figure 3 , Supplementary file 1d ) . Associations within our atlas could potentially be identified due to underlying epidemiological relationships , although there are various other possible explanations such as a shared genetic aetiology between traits . To investigate this for our associations with the schizophrenia PRS , we have used various methods in two-sample MR as an example of how future studies could evaluate findings from our atlas . For these analyses we only used SNPs with p<5×10−08 as instrumental variables to reduce the likelihood of weak instrument bias in our analysis ( Davies et al . , 2015 ) . Furthermore , in these analyses we model liability to schizophrenia as our exposure within an MR framework with associated complex traits as outcomes ( unless stated otherwise ) . Our systematic approach involved the following: The top association with the schizophrenia PRS suggests that individuals with high schizophrenia genetic liability have increased odds of seeing a psychiatrist at some point in their lives due to nerves , anxiety , tension of depression ( OR = 1 . 09 per standard deviation increase in PRS , 95% CI = 1 . 08 to 1 . 10 , p=1 . 55×10−50 ) . The schizophrenia PRS was also strongly associated with various neurological traits , such as neuroticism ( Beta = 0 . 066 , SE = 0 . 006 , p=8 . 17×10−27 ) , being ‘tense or highly strung’ ( OR = 1 . 07 , 95% CI = 1 . 07 to 1 . 08 , p=2 . 25×10−47 ) and self-reported depression ( OR = 1 . 07 , 95% CI = 1 . 06 to 1 . 08 , p=4 . 91×10−18 ) . We identified strong evidence that schizophrenia genetic liability influences this set of neurological traits ( Supplementary file 1e ) , except for self-reported depression where strong evidence was only detected using the inverse variance weighted ( IVW ) method ( Beta = 0 . 004 , SE = 0 . 001 , p=0 . 009 ) . There was also no strong evidence of directional horizontal pleiotropy for these results based on the MR Egger intercept term and associations were detected after repeating analyses using MR directionality filtering . Along with using MR to investigate the effect of PRS traits on outcomes , we recommend investigating the converse direction of effect where possible ( also known as ‘bi-directional’ MR ( Timpson et al . , 2011 ) . For example , for the associations detected with the schizophrenia PRS , associated traits in the UK Biobank were modelled as our exposure in an MR setting and schizophrenia was treated as our outcome . Results suggested that neuroticism liability influences schizophrenia risk ( Supplementary file 1f ) , although we detected evidence of directional horizontal pleiotropy based on the MR Egger intercept term ( Beta = 0 . 043 , SE = 0 . 018 , p=0 . 018 ) . After applying MR directionality filtering , we also identified evidence of association between being ‘tense or highly strung’ and schizophrenia risk . Therefore , the most parsimonious explanation for these findings could be that they have been observed due to a shared genetic aetiology between schizophrenia and other neurological traits . This is also likely to be a plausible explanation for other associations within our atlas . In particular , caution is advised when interpreting findings between autoimmune traits which are known to be influenced by highly correlated genes residing in the HLA region of the genome ( Gough and Simmonds , 2007 ) . Although these findings could still be of interest in terms of genetic correlations between traits , they may not reflect underlying causal relationships ( O'Connor and Price , 2018 ) . Amongst other findings , there were associations which suggested individuals with high schizophrenia genetic liability had a lower fluid intelligence score ( Beta = −0 . 083 , SE = 0 . 006 , p=1 . 49×10−39 ) . We also observed evidence that these individuals performed worse than others in an assessment of cognitive function concerning memorising pairs of cards ( Beta = 0 . 020 , SE = 0 . 002 , p=6 . 66×10−34 for ‘number of incorrect matches’ ) . Follow-up MR analyses provided evidence from multiple methods that schizophrenia genetic liability ( i . e . our exposure ) influences both of these outcomes ( Supplementary file 1g ) . These results were robust to sensitivity analyses using MR directionality filtering and MR Egger intercepts did not indicate that findings were prone to directional horizontal pleiotropy . In contrast , we did not detect strong evidence of a causal effect in the opposite direction for these associations ( i . e . evaluating the effect on measures of cognition and memory on schizophrenia risk ) , in particular after applying MR directionality filtering and when evaluating results from the weighted median and mode methods ( Supplementary file 1h ) . We also conducted a leave-one out analysis which suggested that no individual SNPs were responsible for driving observed effects ( Appendix 1—Figure 1 , Appendix 1—Figure 2 ) . Taken together , these analyses support evidence that schizophrenia genetic liability may lead to reduced cognitive function . Elsewhere , there were associations indicating that participants with a high schizophrenia PRS were more likely to be unsuccessful when attempting to quit smoking ( Beta = 0 . 028 , SE = 0 . 003 , p=3 . 87×10−22 ) and , accordingly reduced odds of being a past smoker ( OR = 0 . 97 , 95% CI = 0 . 97 to 0 . 98 , p=9 . 71×10−17 ) . We observed strong evidence of association that schizophrenia genetic liability influences these outcomes ( Supplementary file 1i ) , whereas the converse direction of effect provided weak evidence of an effect ( Supplementary file 1j ) . However , the ‘number of unsuccessful smoking attempts’ outcome could only be instrumented using a single variant which limits our ability to investigate this effect . Moreover , a recent study has uncovered a large number of SNPs robustly associated with smoking cessation and provided evidence of a bi-directional relationship between smoking and schizophrenia using MR ( Wootton , 2018 ) . Leave-one out analyses suggested that no individual SNP was responsible for driving observed associations ( Appendix 1—Figure 3 , Appendix 1—Figure 4 ) . We also observed a strong inverse association between the schizophrenia PRS and various anthropometric traits . However , evaluating the relationship between schizophrenia liability and body mass index ( BMI ) provided weak evidence of a causal effect in both directions ( Supplementary files 1k & 1l ) . This result reinforces our recommendation that all findings within our atlas require in-depth evaluation to discern whether they represent potential causal associations . Another strength of our atlas is that findings can be evaluated by selecting an outcome of interest and evaluating which of the 162 PRS are most strong associated with it . Doing so may motivate future endeavours to investigate the effect of multiple risk factors on disease risk . As a demonstration of this , we have evaluated the associations between all PRS and self-reported gout in the UK Biobank study ( Supplementary file 1m ) . In this analysis , there was strong evidence of association using the PRS for gout itself ( OR = 1 . 16 , 95% CI = 1 . 13 to 1 . 19 ) , although we also observed a much larger magnitude of effect using the urate PRS ( OR = 1 . 75 , 95% CI = 1 . 72 to 1 . 78 ) . Although many of the PRS in our analysis may be the best polygenic predictors for their target disease/trait , there may be other examples similar to this where the strongest association for an outcome is not the corresponding PRS . For example , the strongest association for birth weight as an outcome in our atlas was with the height PRS ( Beta = 0 . 080 , SE = 0 . 002 , p=1 . 31×10-249 ) . A receiver operating characteristic plot ( Figure 4 ) illustrates this point , where the area under curve for the gout PRS was 0 . 54 in comparison to the urate PRS which had a value of 0 . 65 . This may be attributed to gout being a binary outcome heavily influenced by the number of cases analysed in its corresponding GWAS ( N = 2 , 115 ) . In comparison , urate is a continuous trait measured in all participants for its respective GWAS ( N = 110 , 347 ) . After urate , the next strongest positive associations with self-reported gout were triglycerides ( TG ) and body mass index ( BMI ) ( OR = 1 . 14 , 95% CI = 1 . 11 to 1 . 16 and OR = 1 . 09 , 95% CI = 1 . 06 to 1 . 12 respectively ) . However , it is unclear whether these risk factors influence gout risk independently of one and other or if they reside on the same causal pathway to disease . We investigated this by firstly using an MR mediation framework which involved evaluating bi-directional relationships for each risk factor in turn . As before , only SNPs with p<5×10−08 for each PRS were used as instrumental variables in MR analyses . There was strong evidence that BMI ( i . e . our exposure ) had a causal effect on each other trait in turn ( TG , urate and gout ) , where effect estimates appeared to be consistent between different MR methods ( Supplementary file 1n ) . Repeating this analysis for TG as our exposure provided evidence of a causal effect on urate and gout risk , but not BMI ( Supplementary file 1o ) . We then modelled urate as our exposure variable , which suggested that increased urate positively influences gout risk , although there was weak evidence of an effect on either BMI or TG ( Supplementary file 1p ) . In all analyses there was no strong evidence of horizontal pleiotropy based on the MR-Egger intercept terms and findings were robust to sensitivity analyses using MR directionality filtering ( Supplementary file 1n-1p ) . We also undertook leave-one out analyses which found that no single SNP was driving observed effects ( Appendix 1—Figure 5 , Appendix 1—Figure 6 , Appendix 1—Figure 7 , Appendix 1—Figure 8 ) . In conclusion , as illustrated in Figure 5a , findings from the mediation MR analysis suggests that BMI influences TG levels ( Figure 5a ( 1 ) ) , which has an effect of urate ( Figure 5a ( 2 ) ) , and this subsequently influences gout risk ( Figure 5a ( 3 ) ) . Using the effect estimates from our IVW analysis , we estimated that 77% of the overall effect of BMI on gout risk ( Figure 5a ( 4 ) ) is mediated through this causal pathway . We also used a related approach to investigate the effect of these multiple risk factors on gout susceptibility , known as multivariable MR ( Sanderson et al . , 2018 ) . In this analysis genetic instruments for all exposures ( i . e . BMI , TG and urate ) are modelled simultaneously to investigate whether these risk factors influence our outcome ( i . e . gout ) independently of one and other . We observed the effects of BMI and TG on gout risk attenuate when analysed in the same model as urate ( Supplementary file 1q ) . Furthermore , in subsequent analyses we applied multivariable MR to investigate each pairwise combination of these risk factors on gout risk . There was evidence of an attenuation of the effect of BMI on gout risk when accounting for either the TG or urate effect ( Supplementary files 1r and 1s ) . We also observed the effect of TG on gout risk attenuate when accounting for urate levels ( Supplementary file 1t ) . These findings therefore support the same direction of effect observed using the mediation framework ( Figure 5b ) . In this study we have developed an atlas of associations between PRS and complex traits across the human phenome . Along with contributing to mounting evidence that PRS can be valuable in predicting later life disease outcomes , we have provided examples of how this resource can be harnessed to help identify potential risk factors in disease which warrant further investigation . We envisage that the inferences we have made in this study are just the beginning of potential findings which can be uncovered using such catalogues of associations . Multiple lines of evidence from robust follow-up studies of putative causal risk factors will help improve our understanding of disease susceptibility ( Munafò and Davey Smith , 2018 ) . Large-scale biobank datasets provide an unparalleled opportunity to undertake hypothesis-free causal inference . Such efforts can help identify evidence supporting established causal relationships , as well as potentially implicating novel ones ( Davey Smith and Hemani , 2014; Cai et al . , 2018 ) . We have illustrated this approach in our study by evaluating the results of a phenome-wide association study of schizophrenia genetic liability . This identified strong associations with measures of cognitive function and smoking behaviour which MR follow-up analyses suggested may be due to putative causal relationships with schizophrenia genetic liability . There is long standing evidence from the literature that cognitive impairment is a recognised characteristic of schizophrenia ( Mohamed et al . , 1999 ) . Although PRS may prove useful in determining lifelong risk of developing schizophrenia , based on currently available data they may be less effective in terms of predicting age of schizophrenia onset as well as the severity of its progression . Characterization of cognitive decline in individuals with a high schizophrenia PRS may therefore help improve elucidation of its neurological basis , and ultimately improvement in therapeutic approaches to it ( Green , 1996 ) . There is also a wealth of evidence in the literature from observational studies that individuals diagnosed with schizophrenia smoke more frequently compared to the general population ( Sacco et al . , 2005 ) . Our results indicate that UK Biobank participants with a high schizophrenia genetic liability are more likely to be unsuccessful in their attempts to stop smoking . This may therefore suggest that the high frequency of schizophrenia patients who smoke could be attributed to their inability to quit smoking . However , we were unable to support recent evidence which suggests that smoking is a risk factor for schizophrenia which could be attributed to weak instruments in our analysis ( Wootton , 2018 ) . The positive association with smoking behaviour may also provide a possible explanation for the inverse association we observed between schizophrenia genetic liability and anthropometric traits . In this study we have also provided an example of how investigating various PRS associations with the same outcome may help motivate studies evaluating the effect of multiple risk factors on disease risk . Our analysis detected evidence of an association between body mass index and gout risk , putatively mediated by triglycerides and urate levels . The findings from this analysis therefore appear to recapitulate known biology regarding the established causal pathway to gout ( Matsubara et al . , 1989 ) , ( Li et al . , 2017 ) . Speculatively , a diet including high calorie and alcohol consumption , which are known risk factors for increased body mass index and triglyceride levels , may result in elevated circulating uric acid level and in turn increase gout risk . A recent study has suggested that genetic factors may have a greater impact on serum urate levels than environmental factors such as diet ( Major et al . , 2018 ) . Our findings suggest that genetic drivers of appetite which may influence higher BMI levels are likely to predominantly influence gout risk via increased urate levels . We hope this illustration will motivate creative hypotheses for future endeavours to investigate the effect of multiple risk factors on disease risk . The application of PRS is a topic which has sparked considerable recent debate , particularly concerning whether scores are relevant for clinical decision making ( Warren , 2018 ) . Although resources such as the UK Biobank provide an unparalleled opportunity to investigate the determinants of complex disease as we have done in this study , findings regarding genetic liability may not be generalizable to individuals who are not of European descent . As such , there is likely to be an emphasis in the forthcoming years on efforts to establish disease-specific datasets for a diverse range of ancestries . We also note that , although we have adjusted all analyses in our study using the top 10 principal components from the UK Biobank , there may still be an influence of geographic clustering which remains unaccounted for Abdellaoui et al . ( 2018 ) . Furthermore , although we have flagged the PRS traits in our study derived using GWAS which have overlapping samples with the UK Biobank , we are unable to assess this for scores whose GWAS predate this cohort . Future efforts to link anonymous identifiers between the UK Biobank and UK cohorts would be of helpful in terms of ascertaining this information to prevent overfitting . Lastly , certain complex traits in our study may benefit from being combined to improve statistical power . For instance , a more powerful approach to identify associations between genetic liability and statin medication could involve deriving a combined measure of all the different types of statins reported . Investigating these results in a hypothesis-free manner as we have described in this study may also prove useful for drug repurposing efforts . Polygenic risk scores hold huge promise in the era of large-scale genetic epidemiology to identify individuals who are at high risk of disease . Associations detected between these scores and outcomes undertaken by large-scale analyses should prove powerful for future studies that wish to unravel causal relationships between complex traits . Doing so will help improve disease prevention by developing a stronger understanding of complex epidemiological pathways . Our simulation study concerned two different models; the causal model ( simulating a risk factor which has a causal effect on the simulated outcome ) and the null model ( where there is no causal effect between the simulated exposure and outcome ) . We ran 1000 simulations using each model to compare the PRS approach with the IVW method using a dataset comprising of 10 , 000 samples and 50 SNPs . Further details and all the code used to conduct these simulations can be found at https://‌github . com/‌explodecomputer/‌prs-vs-mr . We have used the MR-Base platform ( Hemani et al . , 2018 ) to identify SNPs from large-scale GWAS to include in our PRS . Our inclusion criteria for selected GWAS was having a sample size of more than 1000 participants , over 100 , 000 SNPs measured on genotyping arrays and based on European/mixed populations . If multiple studies were found for the same trait , we selected the most recent study or the one with the largest sample size . PRS were constructed using SNPs for each GWAS trait based on p<5×10−05 . A threshold of r2 <0 . 001 was selected to identify independent SNPs using genotype data from European individuals ( CEU ) from phase 3 ( version 5 ) of the 1000 genomes project ( Abecasis et al . , 2012 ) . When a GWAS SNP was not available from the UK Biobank study genotype data , we used a proxy SNP instead based on r2 ≥0 . 8 using the same reference panel . Scores were then calculated as the sum of the effect alleles for all SNPs weighted by their reported regression coefficients . However , a small subset of PRS were left unweighted to reduce the likelihood of overfitting . This was due to their GWAS including participants from the initial release of the UK Biobank study . As such , additional caution should be exercised when interpreting findings from these unweighted PRS . Prior to analysis , each PRS was normalised to have a mean of zero and a standard deviation ( SD ) of one . Our PRS construction pipeline was also applied using a more stringent threshold of p<5×10−08 . Although we have not interpreted any of the results using these more stringent scores in this report , they are available within our atlas for future use . We selected traits from the UK Biobank study ( Sudlow et al . , 2015 ) which had p<0 . 05 in the heritability analyses conduct by the Neale lab ( Neale Lab , 2017 ) . Genotype data were available for approximately 490 , 000 individuals enrolled in the study . Phasing and imputation of these data are explained elsewhere ( Bycroft et al . , 2018 ) . Individuals with withdrawn consent , evidence of genetic relatedness or who were not of ‘white European ancestry’ based on a K-means clustering ( K = 4 ) were excluded from analysis . After exclusions there were up to 334 , 398 individuals with both genotype and complex trait data who were eligible for analysis . We evaluated the association between each combination of PRS and complex trait in the UK Biobank study using linear regression ( for continuous traits ) , logistic regression ( for case/control traits ) , ordinal logistic regression ( for ordered categorical traits ) and multinomial logistic regression ( for unordered categorical traits ) . All analyses were adjusted for age , sex , the first 10 genetic principle components ( to adjust for population stratification ) and genotyping chip used to measure genetic data in participants . Only female participants were included in the ‘Age at menarche’ and ‘Age at menopause’ PRS analyses . We also calculated R2 coefficients for continuous traits and McFadden pseudo R2 coefficients for other models by repeating analyses unadjusted for covariates . McFadden’s R2 is defined as: R2McF= 1 – ln ( Lm/ln ( L0 ) where ln is the natural logarithm , L0 is the value of the likelihood function of the model with no predictors and Lm is the likelihood of the model being estimated . We note that pseudo R2 coefficients should not be interpreted in a similar manner to those derived using linear regression ( Hu et al . , 2006 ) . We used various two-sample MR methods to evaluate associations detected in the PRS analysis . This involved using the observed effects of the genetic variants used in the PRS on both the GWAS trait that the score was based on ( treated as the exposure in our MR analysis ) as well as the UK Biobank trait ( treated as the outcome in our MR analysis ) . For all MR analyses we only selected SNPs with p<5×10−08 based on GWAS findings as instrumental variables to reduce the likelihood of weak instrument bias ( Davies et al . , 2015 ) . In terms of MR methods , we applied the inverse variance weighted ( IVW ) ( Burgess et al . , 2013 ) , weighed median ( Bowden et al . , 2016 ) and weighted mode approaches . We also conducted several different sensitivity analyses to evaluate findings . We derived Cochran’s Q statistic when undertaking the IVW approach as an indicator of heterogeneity , as well as repeating all analyses after filtering out SNPs which the MR directionality test suggested did not influence the outcome of interest through the analysed exposure . The intercept of the MR-Egger approach ( Bowden et al . , 2015 ) was used to investigate directional horizontal pleiotropy and leave-one-out analyses ( i . e . reapplying the IVW method after removing each SNP in turn with replacement ) were conducted to discern whether any individual SNPs were driving observed associations . These types of analyses are particularly important when assessing findings from our atlas , as one possible explanation is that they could be attributed to a single pleiotropic SNP which has a large effect size ( e . g . the APOE locus which is associated with Alzheimer’s disease and lipid levels ) . To investigate the direction of effect for associations identified in the PRS analysis we undertook bi-directional MR ( Timpson et al . , 2011 ) . This involves firstly modelling our PRS trait as our exposure and complex trait as our outcome , and subsequently the complex trait as our exposure and PRS trait as our outcome in a separate analysis . Lastly , we have incorporated two recent developments within the field of MR; mediation MR and multivariable MR ( Sanderson et al . , 2018 ) . These methods can be used to investigate the effect of multiple risk factors on a single outcome , as well as uncover potential mediators in disease . In this study we have evaluated findings from the PRS analysis based on the p<5×10−05 threshold . We note however that it is only advisable to apply techniques in MR using this threshold as long as in-depth sensitivity analyses ( e . g . leave-one out , MR-Egger intercept ) are also undertaken to robustly evaluate findings . When undertaking our example of mediation MR in this study , we also calculated the proportion mediated along the causal pathway from exposure to outcome using effect estimates derived using the IVW method , where: Proportion mediated = direct effect - indirect effectdirect effect The direct effect here is the IVW effect estimate derived for the association between the exposure ( i . e . BMI ) and our outcome ( i . e . gout ) . The indirect effect was calculated as the product of all IVW effect estimates derived for all relationships along the causal pathway of interest ( i . e . the effect of BMI on triglycerides , the effect of triglycerides on urate and the effect of urate on gout ) . All analyses were undertaken using R ( version 3 . 5 . 1 ) . The R package ‘shiny’ v1 . 1 was used to develop the web application and ‘highcharter’ v0 . 5 was used to generate interactive plots . Figures in this manuscript were generated using ‘ggplot2’ v2 . 2 . 1 . All summary statistics for the analyses undertaken in this study can be downloaded using our web application ( http://mrcieu . mrsoftware . org/PRS_atlas/ ) . Our dataset was derived from the UK Biobank study as part of projects 8786 and 15825 . The same dataset can be created with an application to use data from the UK Biobank study ( http://biobank . ctsu . ox . ac . uk/crystal/ ) .	An individual’s risk of developing many diseases , including heart disease and schizophrenia , is influenced by a complex combination of lifestyle factors and the genes they inherit at birth . The total number of genetic variants that an individual has that increases their risk of developing a particular disease can be measured as their ‘polygenic risk score’ . These scores allow researchers to predict whether it is likely that someone will develop a disease during their lifetime . Polygenic risk scores can also be used to link different conditions or traits to each other . For example , if high blood pressure can be caused by obesity , then genetic variants linked to obesity will also influence blood pressure . As a result , individuals with a high polygenic risk score for obesity will , on average , have a higher blood pressure than those with a low score . Comparing associations between polygenic risk scores and traits can therefore suggest whether one trait causes another . Richardson et al . have developed an ‘atlas’ that uses data from the UK Biobank study – which contains genetic data from over 300 , 000 people – to investigate how shared characteristics and risk factors in individuals relate to their genetic likelihood of developing a disease . The data currently includes 162 different polygenic risk scores and 551 traits . Richardson et al . used the atlas to evaluate which traits are most strongly linked to the polygenic risk score for schizophrenia . Analyses of these traits suggested that individuals with a high genetic risk of developing schizophrenia tend to perform worse in IQ and short-term memory tests , and that they are less likely to successfully quit smoking . These characteristics have previously been observed in studies of individuals with schizophrenia . In the future , the atlas could be used to identify possible relationships between a wide range of individual traits and diseases . This could help to prioritise which relationships should be investigated further as part of studies to understand the causes and consequences of disease . In the long term , such studies should improve our ability to prevent and treat many different medical conditions .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "tools", "and", "resources", "genetics", "and", "genomics" ]	2019	An atlas of polygenic risk score associations to highlight putative causal relationships across the human phenome
In the heart , reliable activation of Ca2+ release from the sarcoplasmic reticulum during the plateau of the ventricular action potential requires synchronous opening of multiple CaV1 . 2 channels . Yet the mechanisms that coordinate this simultaneous opening during every heartbeat are unclear . Here , we demonstrate that CaV1 . 2 channels form clusters that undergo dynamic , reciprocal , allosteric interactions . This ‘functional coupling’ facilitates Ca2+ influx by increasing activation of adjoined channels and occurs through C-terminal-to-C-terminal interactions . These interactions are initiated by binding of incoming Ca2+ to calmodulin ( CaM ) and proceed through Ca2+/CaM binding to the CaV1 . 2 pre-IQ domain . Coupling fades as [Ca2+]i decreases , but persists longer than the current that evoked it , providing evidence for ‘molecular memory’ . Our findings suggest a model for CaV1 . 2 channel gating and Ca2+-influx amplification that unifies diverse observations about Ca2+ signaling in the heart , and challenges the long-held view that voltage-gated channels open and close independently . L-type Ca2+ channels are composed of a pore-forming α1 subunit and four additional accessory subunits ( α2 , β , γ , δ ) ( Catterall , 1995 ) . Four different α1 subunits have been identified to date , one of which , CaV1 . 2 , is expressed in neurons as well as cardiac and arterial smooth muscle ( Koch et al . , 1990; Navedo et al . , 2007; Zhang et al . , 2007 ) . Four distinct genes encode L-type Ca2+ channel β-subunits , each with multiple splice variants . In addition , four α2δ genes have been identified . Both cell- and tissue-specific combinations of these CaV1 . 2 subunits endow the channels with distinct functional properties ( Catterall , 2000 ) . A prominent characteristic of CaV1 . 2 channels is the tight regulation of their activity by the Ca2+ signals they produce ( Ben-Johny and Yue , 2014 ) . For example , increases in [Ca2+]i have been implicated in CaV1 . 2 facilitation; this Ca2+-dependent facilitation ( CDF ) is a form of positive feedback that amplifies Ca2+ influx . An increase in intracellular Ca2+ concentration ( [Ca2+]i ) has also been proposed to exert the opposite effect—Ca2+-dependent inactivation ( CDI ) . Thus , the balance between CDF and CDI of CaV1 . 2 channels plays a key role in regulating the magnitude of Ca2+ influx . The general consensus is that CDF and CDI involve Ca2+ binding to calmodulin ( CaM ) in the IQ domain in the C-terminal tail of these channels . During excitation-contraction ( EC ) coupling , membrane depolarization opens CaV1 . 2 channels in the sarcolemma of ventricular myocytes . This allows a small amount of Ca2+ to enter ventricular myocytes that can be detected optically in the form of a ‘CaV1 . 2 sparklet’ , raising local [Ca2+]i beyond the threshold for activation of ryanodine receptors ( RyRs ) in the sarcoplasmic reticulum ( Wang et al . , 2001 ) . Synchronous activation of multiple RyRs by CaV1 . 2 channels produces a global rise in [Ca2+]i that initiates myocardial contraction ( Cheng et al . , 1993 ) . EC coupling in ventricular myocytes is remarkably reproducible , with each action potential ( AP ) invariably evoking a whole-cell [Ca2+]i transient that results in contraction . At the membrane potentials reached during the plateau of the ventricular AP ( approximately +50 mV ) , the driving force for Ca2+ entry at physiological Ca2+ levels ( ∼2 mM ) is so low that opening of a single CaV1 . 2 channel is not sufficient to raise local [Ca2+]i beyond the RyR activation threshold . However , the probability of RyR activation during this phase of the AP is very high ( >0 . 9 ) . This degree of reliability would presumably require 5–10 CaV1 . 2 channels to open simultaneously ( Inoue and Bridge , 2003; Sobie and Ramay , 2009 ) . However , because the maximum open probability ( Po ) of CaV1 . 2 channels at physiological [Ca2+]o is ∼0 . 3 ( Josephson et al . , 2010 ) , the probability of 5–10 independently gating channels opening simultaneously is extremely low ( i . e . , 0 . 35 to 0 . 310 ) . This raises a fundamental question: if the probability of coincident openings of the requisite number of CaV1 . 2 channels is so low , why is the probability of RyR activation during the cardiac AP so high ? Answering this question is critical for understanding the mechanistic basis of reliable cardiac performance . A potential answer to this conundrum lies in the recently proposed concept that clusters of Cav1 . 2 channels can be functionally coupled to one another through physical interactions between their C-terminal tails ( Dixon et al . , 2012 ) . This interaction enables physically linked channels to coordinate their gating , leading to amplification of Ca2+ influx , Ca2+ current facilitation , and EC coupling in ventricular myocytes . Importantly , this model challenges the long-held assumption that Cav1 . 2 channels , like other classes of voltage-gated channels , function exclusively as monomers that gate independently of one another . To date , however , the mechanism underlying functional CaV1 . 2 channel coupling has remained unknown . Here , using a combination of super-resolution nanoscopy , Ca2+-imaging , electrophysiology and two methodologically distinct assays of protein–protein interaction , we have assembled a body of evidence that significantly alters our current understanding of Ca2+/CaM regulation of CaV1 . 2 channels and reconciles diverse observations about Ca2+ signaling in ventricular myocytes . We discovered that binding of Ca2+ to CaM induces C-terminal-to-C-terminal CaV1 . 2 channel interactions that increase the activity of adjoined channels , facilitating Ca2+ currents and increasing Ca2+ influx . Notably , we found that functional coupling outlasts the Ca2+ current that evokes it , providing evidence for a type of ‘molecular memory’ that could transiently shape the response of the cell to subsequent APs . We propose that cooperative gating of CaV1 . 2 channels is a new general mechanism for the regulation of excitability and Ca2+ influx in cardiac myocytes and suggest that this concept can be extended to other excitable cells . We began our study by expressing CaV1 . 2 channels in tsA-201 cells and recording elementary CaV1 . 2 currents from cell-attached patches . Currents were elicited with a step depolarization to −30 mV with Ba2+ or Ca2+ as the charge carrier ( Figure 1A , B ) . The mean amplitudes of iCa and iBa were 0 . 50 ± 0 . 02 ( n = 6 ) and 1 . 45 ± 0 . 01 pA ( n = 6 ) , respectively . All-points histograms revealed that multi-channel openings were more likely with Ca2+ than with Ba2+ . Accordingly , the activity ( nPo ) , defined as the number of channels ( n ) times the open probability ( Po ) , of CaV1 . 2 channels within a patch , was significantly higher with Ca2+ ( 0 . 24 ± 0 . 10 ) than Ba2+ ( 0 . 02 ± 0 . 01; Figure 1C ) . Closer inspection of the multi-channel openings revealed that , with Ca2+ as the charge carrier , multiple channels frequently opened together instantaneously and subsequently closed together . For example , in the enlarged trace in Figure 1B , four channels opened simultaneously , followed by the opening of four additional channels . The eight channels remained open for a time ( ∼11 . 5 ms ) , then all closed simultaneously . This apparent coordinate gating of multiple CaV1 . 2 channels was not observed when Ba2+ was used as the charge carrier . These results challenge the long-held and generally accepted view that individual CaV1 . 2 channels gate independently , and instead strongly suggest that these channels frequently exhibit ‘cooperative gating’ . An additional implication is that Ca2+ itself increases the probability of cooperative CaV1 . 2 channel gating . 10 . 7554/eLife . 05608 . 003Figure 1 . Single-channel electrical and optical recordings of CaV1 . 2 channel coupling . ( A and B ) Representative iBa ( A ) and iCa ( B ) traces recorded from CaV1 . 2-expressing tsA-201 cells during step depolarizations from −80 to −30 mV . Amplitude histograms ( constructed from n = 6 cells each ) were fit with multi-component Gaussian functions ( solid black lines ) . A portion of each trace ( gray box ) is shown enlarged below , showing that the resulting L-type Ca2+ current reflects the simultaneous opening and closing of multiple channels with Ca2+ as the charge carrier , but not with Ba2+ as the charge carrier . ( C ) Bar chart of iBa and iCa single-channel activity ( nPo ) . Data are presented as means ± SEM ( *p < 0 . 01 ) . ( D ) Calibrated TIRF image of an adult ventricular myocyte dialyzed with the Ca2+ indicator dye Rhod-2 via the patch pipette ( see also Video 1 ) . Time courses of [Ca2+]i from each sparklet site ( indicated by green circles on TIRF image ) and their κ values are shown in panels a-e . ( E ) All-points histogram of Ca2+ sparklet data recorded from adult ventricular myocytes . The data were fit with a multi-component Gaussian function ( solid black line ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05608 . 003 To test whether this hypothesis holds true in primary cells , we recorded CaV1 . 2 Ca2+ sparklet activity in freshly isolated adult ventricular myocytes ( Figure 1D and Video 1 ) . A multi-Gaussian fit to the all-points histogram of the calibrated Ca2+ signal obtained with 20 mM [Ca2+]o revealed quantal amplitudes of 36 . 8 ± 4 . 2 nM ( Figure 1E ) , in excellent agreement with the ∼38 nM reported previously for CaV1 . 2 Ca2+ sparklets in arterial smooth muscle cells and tsA-201 cells ( Navedo et al . , 2005 , 2006 ) . Consistent with our single-channel data , we frequently observed multi-quantal Ca2+ sparklets corresponding to the simultaneous opening of several CaV1 . 2 channels . We next applied a coupled Markov chain model to determine whether these Ca2+-influx events were solely attributable to stochastic , independently gating CaV1 . 2 channels or instead reflected cooperative channel gating . The model assigns a coupling coefficient ( κ ) value to each Ca2+ sparklet site , ranging from 0 for independently gating channels to 1 for channels that gate exclusively in a cooperative manner ( Chung and Kennedy , 1996; Navedo et al . , 2010 ) . Using κ > 0 . 1 as a threshold for cooperative gating ( Navedo et al . , 2010 ) , we found that the majority of Ca2+ sparklet sites displayed cooperative or ‘coupled’ gating behavior , consistent with our hypothesis . Together , these data suggest that functional coupling of CaV1 . 2 channels is a Ca2+-dependent phenomenon . 10 . 7554/eLife . 05608 . 004Video 1 . Cardiomyocyte Ca2+ sparklets . Stack of 2D images acquired at 100 Hz from a whole-cell patch-clamped adult ventricular myocyte held at −80 mV and dialyzed with Rhod-2 via the patch pipette . DOI: http://dx . doi . org/10 . 7554/eLife . 05608 . 004 If the trigger for CaV1 . 2 channel coupling were a local elevation in [Ca2+]i resulting from the opening of a single channel , then the efficacy of this signal in recruiting a nearby channel would be critically dependent on the distance separating the channels . Using super-resolution nanoscopy ( Folling et al . , 2008 ) , we examined the spatial organization of endogenous CaV1 . 2 channels in ventricular myocytes ( Figure 2A–C ) and heterologously expressed CaV1 . 2 channels in tsA-201 cells ( Figure 3A–C ) . Our data clearly show that CaV1 . 2 channels formed clusters along the sarcolemmal Z-lines of ventricular myocytes ( Figure 2A , B ) and throughout the plasma membrane ( PM ) of tsA-201 cells ( Figure 3A , B ) . The average area occupied by a CaV1 . 2 channel cluster was 2555 ± 82 nm2 in ventricular myocytes ( n = 5; Figure 2C ) and 2190 ± 20 nm2 in tsA-201 cells ( n = 9; Figure 3C ) . 10 . 7554/eLife . 05608 . 005Figure 2 . CaV1 . 2 channels form clusters in the ventricular myocyte PM . ( A ) TIRF image of a fixed , adult mouse ventricular myocyte immunolabeled with an antibody specific for CaV1 . 2 channels . ( B ) Super-resolution GSD image of the same cell . Channels are located along the t-tubule network with the characteristic 1 . 8-μm separation . Yellow boxes denote location of higher-magnification images of channel clusters ( right ) . ( C ) Distribution of cluster areas in ventricular myocytes ( n = 5 myocytes ) . ( D ) An average of the first five frames of a TIRF image time series taken of a myocyte isolated from mice expressing CaVβ2a-PA-GFP . Yellow boxes indicate spots selected for analysis . Scale bars = 2 μm . ( E ) Examples of bleaching steps for CaVβ2a-PA-GFP associated with CaV1 . 2 channels . ( F ) Distribution of bleaching steps obtained from 435 spots selected from n = 11 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 05608 . 00510 . 7554/eLife . 05608 . 006Figure 3 . CaV1 . 2 channels form clusters in tsA-201 cell membranes . ( A and B ) TIRF and GSD images of immunolabeled CaV1 . 2 channels in a transfected tsA-201 cell ( A ) . Yellow boxes in ( B ) indicate the location of each higher-magnification image ( right ) . ( C ) Distribution of cluster areas in tsA-201 cells ( n = 9 cells ) . ( D ) An average of the first five frames of a TIRF image time series for a tsA-201 cell expressing CaV1 . 2-EGFP . ( E ) Examples of bleaching steps for CaV1 . 2-EGFP . Scale bars = 2 μm . ( F ) Distribution of bleaching steps obtained from 484 spots selected from n = 10 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 05608 . 00610 . 7554/eLife . 05608 . 007Figure 3—figure supplement 1 . CaV1 . 2 channel clusters are not co-localized to tsA-201 ER structures . ( A ) TIRF ( top ) and GSD ( bottom ) images of immunolabeled CaV1 . 2 channels ( left ) and mCherry-Sec61β ( middle ) in a transfected tsA-201 cell . The image on the bottom right was generated by merging CaV1 . 2 and mCherry-Sec61β GSD images . ( B ) Top: TIRF images of immunolabeled CaV1 . 2 channels ( left ) , mCherry-Sec61β ( middle ) , and JPH2 ( right ) in a transfected tsA-201 cell . Bottom: GSD images from the same cell showing CaV1 . 2 channels ( left ) , mCherry-Sec61β ( middle ) , and a merge of the two ( right ) . Scale bars = 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05608 . 007 Since CaV1 . 2 channel clusters were localized to the t-tubule regions of ventricular myocytes where the junctional SR ( jSR ) comes into close apposition to the myocyte PM , one might predict that the channel clusters would similarly localize to PM-adjacent ER structures in tsA-201 cells . To test this idea , we co-expressed mCherry-sec61β ( a general ER marker ( Zurek et al . , 2011 ) ) in tsA-201 cells together with CaV1 . 2 channels . Super-resolution imaging revealed that CaV1 . 2 channel cluster distribution was not restricted to regions of the surface membrane where the ER was located ( Figure 3—figure supplement 1A ) . Instead , these channels were broadly distributed along the PM surface . It is possible that the lack of organization of CaV1 . 2 channels along ER junctions in tsA-201 cells reflects a lack of contact points or excessive distance between the ER and the PM . In ventricular myocytes , the membrane binding protein junctophilin-2 ( JPH2 ) has been suggested to tether the jSR to the t-tubule membrane ( Takeshima et al . , 2000 ) . Thus , we attempted to anchor the ER to the PM by co-transfecting tsA-201 cells with JPH2 . However , even in the presence of JPH2 , the CaV1 . 2 channel cluster distribution was not limited to PM-ER junctions in the manner that they are in cardiomyocytes ( Figure 3—figure supplement 1B ) . These data suggest that CaV1 . 2 channel clustering occurs independently of SR/ER microdomains . To determine the number of channels within CaV1 . 2 clusters , we injected mice with an adeno-associated virus serotype 9 ( AAV9 ) designed to express photo-activatable-GFP–tagged Cavβ2a , and examined ventricular myocyte CaV1 . 2 clusters 5 wk later using single-particle photobleaching ( Ulbrich and Isacoff , 2007 ) . CaVβ2a is a palmitoylated peripheral membrane protein that binds to the α1 pore-forming subunit of CaV1 . 2 with a 1:1 stoichiometry ( Dalton et al . , 2005 ) ; therefore , photo-activation of this protein with 405-nm light provides a fluorescent marker of CaV1 . 2 channels . Single CaV1 . 2 were identified and excited using total internal reflection fluorescence ( TIRF ) microscopy . The number of channels in each cluster was determined by continuous photobleaching and counting of stepwise decreases in fluorescence intensity ( Figure 2D–F ) . A preponderance ( 47% ) of CaV1 . 2 clusters displayed 1 to 6 stepwise decreases in fluorescence ( Figure 2F ) . A single photobleaching step was observed in only 1% of the spots analyzed . Indeed , the mean number of bleaching steps per cluster was 7 . 91 ± 0 . 23 ( n = 25 ) . This suggests that CaV1 . 2 channels preferentially cluster in groups of about 8 channels in adult ventricular myocytes . We did not discriminate cluster size based on location; thus , our estimate of 8 channels/cluster combines dyadic and extra-dyadic populations . Similar stepwise photobleaching experiments were performed on enhanced green fluorescent protein ( EGFP ) -tagged CaV1 . 2 channels heterologously expressed in tsA-201 cells ( Figure 3D–F ) . TIRF imaging showed that CaV1 . 2 clusters displayed a mean of 5 . 07 ± 0 . 15 ( n = 10 ) discrete bleaching steps . Taken together with the data from ventricular myocytes , these findings suggest that the formation of multi-channel clusters is a fundamental property of CaV1 . 2 channels with important implications for Ca2+ signaling . To investigate the mechanisms regulating CaV1 . 2-CaV1 . 2 interactions in living cells ( Figure 4 ) , we applied a bimolecular fluorescence complementation approach using CaV1 . 2 channels fused with either the N- or C-terminus of the split-Venus fluorescent protein system to yield CaV1 . 2-VN155 ( I152L ) and CaV1 . 2-VC155 , respectively . In isolation , VN155 ( I152L ) and VC155 are non-fluorescent; however , when brought into close proximity by interacting proteins , they can reconstitute a full , fluorescent Venus protein . Thus , Venus fluorescence can be used to report spontaneous interactions between adjacent CaV1 . 2 channels , as depicted in Figure 4A . In cells expressing CaV1 . 2-VN155 ( I152L ) and CaV1 . 2-VC155 channels , Venus fluorescence at −80 mV was very low , suggesting that CaV1 . 2-CaV1 . 2 channel interactions are rare at this hyperpolarized membrane potential . To determine whether an increase in [Ca2+]i is sufficient to induce CaV1 . 2-CaV1 . 2 channel interactions , we loaded tsA-201 cells with DMNP-EDTA ( caged Ca2+ ) via the patch pipette and induced flash photolysis of DMNP-EDTA with pulses of 405-nm light while holding cells at −80 mV . Photolysis of DMNP-EDTA induced a transient increase in [Ca2+]i and a concomitant increase in Venus fluorescence ( Figure 4—figure supplement 1 ) , demonstrating that an elevation in [Ca2+]i is indeed sufficient to induce CaV1 . 2-CaV1 . 2 interactions . 10 . 7554/eLife . 05608 . 008Figure 4 . Interactions between CaV1 . 2 channel C-termini occur spontaneously and in a Ca2+/CaM-dependent manner . ( A ) Illustration of the bimolecular fluorescence complementation strategy for assaying interactions between CaV1 . 2 channel C-termini . Non-interacting channels tagged at their C-terminus with either the N- or C-terminal half of split Venus are non-fluorescent ( left ) . Spontaneous interactions between channel C-termini result in reconstitution of Venus and emission of fluorescence ( right ) . ( B–E ) TIRF images obtained from whole-cell patch-clamped tsA-201 cells expressing CaV1 . 2-VN and CaV1 . 2-VC over 9-s voltage steps to the indicated potentials . Images were median-filtered , smoothed , pseudo-colored with a ‘red-hot’ LUT , and divided by the initial −60 mV image to obtain calibrated Venus F/F0 . Experiments were performed with 2 mM Ba2+ ( B ) or 20 mM Ca2+ ( C–E ) in the perfusing solution . Scale bars = 3 μm . ( See also Figure 4—figure supplement 2 ) ( D ) Images obtained during dialysis with MLCKp ( 0 . 1 μM ) . ( E ) Images from a cell in which CaM1234 was co-expressed with CaV1 . 2-VN and CaV1 . 2-VC ( see also Figure 4—figure supplement 3 ) . ( F ) Relationship between membrane voltage and Venus reconstitution for each experimental condition . ( G ) Bar chart showing mean Venus fluorescence ( F/F0 ) ± SEM for each condition at −40 and +40 mV ( p < 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05608 . 00810 . 7554/eLife . 05608 . 009Figure 4—figure supplement 1 . Flash photolysis of caged Ca2+ stimulates CaV1 . 2 interactions . ( A ) Illustration of flash photolysis-induced bimolecular fluorescence complementation . Prior to uncaging of Ca2+ , cells were bathed in a zero Ca2+ solution; under these conditions , split-Venus–tagged channels do not interact ( left ) . Upon application of a UV flash to uncage Ca2+ , interactions between the channel C-termini results in reconstitution of Venus protein and emission of fluorescence ( right ) . ( B ) Confocal images of Venus fluorescence emission ( F/F0 ) before and after flash photolysis of caged Ca2+ . ( C ) The bar chart shows mean Venus and Rhod-2 fluorescence emission ( F/F0 ) ± SEM before and after Ca2+ uncaging ( *p < 0 . 001 , for comparison of CaV1 . 2interactions [Venus reconstitution] and intracellular Ca2+ concentration [Rhod-2 emission] before and after flash photolysis of caged Ca2+; paired t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05608 . 00910 . 7554/eLife . 05608 . 010Figure 4—figure supplement 2 . CaV1 . 2 interactions are Ca2+ dependent . ( A–C ) Venus fluorescence and CaV1 . 2 currents were recorded in whole-cell mode from tsA-201 cells expressing CaV1 . 2-VN and Cav1 . 2-VC during depolarizing voltage steps to −40 , −20 , 0 and +20 mV with Ba2+ as the charge carrier ( 2 mM [Ba2+]o ) . Voltage dependencies were fit with a Boltzmann sigmoidal function ( red solid line ) except in ( C ) , where Venus F/F0 decayed exponentially during the voltage protocol in a manner reminiscent of photobleaching . ( A ) Calibrated Venus F/F0 TIRF images ( top ) and IBa ( bottom ) from a representative tsA-201 cell . ( B ) Voltage dependence of normalized conductance . ( C ) Venus fluorescence ( F/F0 ) plots . ( D–F ) Venus fluorescence and CaV1 . 2 currents were recorded as above ( A–C ) with Ca2+ as the charge carrier ( 20 mM [Ca2+]o ) . ( D ) Calibrated Venus F/F0 TIRF images ( top ) and IBa ( bottom ) from a representative tsA-201 cell . ( E ) Voltage dependence of normalized conductance . ( F ) Venus fluorescence ( F/F0 ) plots . Voltage dependencies were fit with a Boltzmann sigmoidal function ( red solid line ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05608 . 01010 . 7554/eLife . 05608 . 011Figure 4—figure supplement 3 . Ca2+ binding to distinct CaM pools regulates CDI . ( A–C ) tsA-201 cells expressing CaV1 . 2 were subjected to a 300-ms depolarization from a holding potential of −80 mV to a test potential of +20 mV in the presence of 20 mM [Ca2+]o . ( A ) ICa elicited under control conditions ( 20 mM [Ca2+]o ) , with 0 . 1 μM MLCKp dialyzed via the patch pipette , and in cells co-expressing Ca2+-insensitive CaM1234 . ( B ) Mean r300 ratios ± SEM ( p = 0 . 002 ) . ( C ) Current-voltage relationships . DOI: http://dx . doi . org/10 . 7554/eLife . 05608 . 01110 . 7554/eLife . 05608 . 012Figure 4—figure supplement 4 . Ca2+/CaM binding to the pre-IQ domain , and not the IQ domain , mediates channel coupling . ( A ) Illustration of the pore-forming subunit of a CaV1 . 2 channel embedded in the PM , with intracellular N and C-termini . The C-terminus contains pre-IQ ( green ) and IQ ( blue ) motifs , which are known to bind CaM . The amino acid sequence of the pre-IQ and IQ segments of the C-terminus of WT , pre-IQ swap , and I1654E mutant channels used in this study are shown on the right . For the pre-IQ swap , a 33-amino-acid segment was replaced with 33 non-identical amino acids . For the I1654E mutation , a point mutation was made replacing I1654 with E . ( B ) ICa elicited by a depolarizing step from −80 mV to a test potential of +20 mV under control conditions ( 20 mM [Ca2+]o ) in tsA-201 cells expressing WT ( black ) , I1624E ( blue ) , or pre-IQ swap ( green ) CaV1 . 2 channels . ( C ) Venus fluorescence ( F/F0 ) plots for the two mutant channels . CaV1 . 2 ( I1624E ) channels exhibited a voltage-dependent increase in Venus reconstitution that fit a Boltzmann sigmoidal function ( blue solid line; n = 5 ) . Venus F/F0 decayed over the course of the voltage protocol in cells expressing CaV1 . 2 ( pre-IQ swap ) ( n = 8 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05608 . 01210 . 7554/eLife . 05608 . 013Figure 4—figure supplement 5 . CaV1 . 2 channel clustering is necessary but not sufficient for functional coupling . ( A ) TIRF ( top ) and super-resolution GSD ( bottom ) images of immunolabeled CaV1 . 2 ( pre-IQ swap ) channels ( left ) and mCherry-Sec61β ( middle ) in an exemplary transfected tsA-201 cell . The image on the bottom right was generated by merging CaV1 . 2 ( pre-IQ swap ) and mCherry-Sec61β GSD images . ( B ) Top: TIRF images of immunolabeled CaV1 . 2 ( pre-IQ swap ) channels ( left ) , mCherry-Sec61β ( middle ) , and JPH2-GFP ( right ) in a transfected tsA-201 cell . Bottom: GSD images from the same cell showing CaV1 . 2 ( pre-IQ swap ) channels ( left ) , mCherry-Sec61β ( middle ) , and a merge of the two ( right ) . Scale bars = 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 05608 . 013 To determine if Ca2+ influx via CaV1 . 2 channels is required for channel interactions , we depolarized cells and recorded Venus fluorescence and membrane currents in the presence of Ba2+ or Ca2+ . Cells were dialyzed with an intracellular solution containing 10 mM EGTA to restrict the local [Ca2+]i signal to about 1 μm from the channel and maintain very low global [Ca2+]i . With Ba2+ in the external solution , depolarization evoked currents ( IBa ) over a wide range of potentials , but Venus fluorescence was very low at all membrane potentials ( Figure 4B , F , G and Figure 4—figure supplement 2A–C ) . After switching to a Ca2+-containing external solution , application of the same voltage protocol activated currents ( ICa ) and induced graded increases in Venus fluorescence ( Figure 4C , F , G and Figure 4—figure supplement 2D–F ) . The fluorescence-voltage and ICa conductance-voltage relationships were sigmoidal . The normalized conductance and Venus fluorescence exhibited similar voltage dependencies ( Figure 4—figure supplement 2E , F ) . Taken together with super-resolution , photobleaching and DMNP-EDTA data , these findings suggest that local and global [Ca2+]i signals produced by Ca2+ influx via CaV1 . 2 channels are required for physical interactions between adjacent channels within a cluster . We next investigated the mechanisms underlying Ca2+-dependent coupling of CaV1 . 2 channels , focusing on CaM since this protein binds Ca2+ , associates with CaV1 . 2 C-terminal pre-IQ and IQ domains , and is involved in CDI and CDF of CaV1 . 2 channels ( Eckert and Tillotson , 1981 ) . Dialysis with an intracellular solution containing the CaM inhibitory peptide MLCKp ( 0 . 1 μM ) prevented Venus reconstitution during membrane depolarization ( Figure 4D , F , G ) , without altering the rate of inactivation of ICa ( Figure 4—figure supplement 3 ) . Expression of a mutant CaM that does not bind Ca2+ ( CaM1234 ) also prevented CaV1 . 2-VN-CaV1 . 2-VC fusion upon membrane depolarization ( Figure 4E , F , G ) . However , unlike MLCKp , CaM1234 did slow the rate of ICa inactivation ( Figure 4—figure supplement 3 ) , suggesting that two functionally distinct CaM molecules are involved in CDI and CaV1 . 2-to-CaV1 . 2 channel interactions . Given the apparently essential role of Ca2+/CaM , we next examined the importance of IQ and pre-IQ motif CaM-binding sites for CaV1 . 2-CaV1 . 2 channel interactions . To do so , we mutated critical residues in each segment and used bimolecular fluorescence complementation to evaluate changes in channel coupling ability . CaV1 . 2-VN and CaV1 . 2-VC channels with an I1654E mutation in their IQ domains were able to fuse upon membrane depolarization ( Figure 4—figure supplement 4 ) . This isoleucine residue is crucial for CaM binding to the IQ motif . Previous in vitro studies have reported that the I1654E ( or the human homolog I1624E ) mutation decreases the affinity of the IQ motif for CaM by ∼100-fold ( Zühlke et al . , 1999 , 2000 ) . Thus , our results suggest that CaM binding to the IQ-motif is not required for CaV1 . 2 channel coupling . To investigate the role of the pre-IQ motif in channel interactions , we exchanged a 33-amino-acid segment of the pre-IQ domain for 33 non-identical amino acids ( Figure 4—figure supplement 4A ) . A similar segment exchange performed on human CaV1 . 2 channels expressed in tsA-201 cells was previously shown to impair channel clustering on the PM and reduce Po compared to wild-type ( WT ) channels ( Kepplinger et al . , 2000 ) . Interestingly , the pre-IQ ‘swap’ mutation rendered CaV1 . 2-VN and CaV1 . 2-VC channels incapable of functional coupling ( Figure 4—figure supplement 4C ) . Collectively , these results suggest that pre-IQ domains are required for Ca2+/CaM-dependent CaV1 . 2 channel oligomerization . Additional credence is given to this finding by the previously published crystal structure of dimeric cardiac L-type Ca2+ channels showing two pre-IQ helices bridged by two Ca2+/CaMs ( Fallon et al . , 2009 ) . We next examined the spatial distribution of CaV1 . 2 ( pre-IQ swap ) channels in tsA-201 cells co-transfected with mCherry-sec61β ( to permit visualization of the ER ) . Surprisingly , CaV1 . 2 ( pre-IQ swap ) channels still formed clusters in tsA-201 cells despite their inability to functionally interact ( Figure 4—figure supplement 5A ) . Indeed , under identical imaging conditions ( i . e . , using the same fixative and TIRF penetration depth ) , CaV1 . 2 ( pre-IQ swap ) channel cluster areas were not significantly different from those of WT channels . As noted above for WT channels , the CaV1 . 2 ( pre-IQ swap ) channel cluster distribution was not limited to PM-ER junctions , even in the presence of JPH2 ( Figure 4—figure supplement 5A , B ) . These results suggest that , while the physical proximity of Cav1 . 2 channels is necessary for channel interactions , it is not sufficient for functional coupling of the channels . An important prediction of our findings is that the functional coupling produced by Ca2+-induced CaV1 . 2-CaV1 . 2 channel interactions manifests as enhanced channel activity . To test this , we exploited the fact that Venus reconstitution is irreversible , recording CaV1 . 2 sparklets ( Wang et al . , 2001; Navedo et al . , 2005 ) before and after Ca2+-induced CaV1 . 2 coupling during membrane depolarization . Prior to membrane depolarization , CaV1 . 2 sparklet activity ( nPs ) was low ( 0 . 002 ± 0 . 001 ) . However , after depolarization , nPs increased ∼40-fold ( 0 . 085 ± 0 . 022 ) and CaV1 . 2 sparklet density increased almost 10-fold ( Figure 5A , C , D and Video 2; n = 5 cells ) . Membrane depolarization failed to significantly alter nPs or CaV1 . 2 sparklet density in cells co-expressing the Ca2+-insensitive CaM1234 mutant ( Figure 5B–D; n = 6 cells ) . These data indicate that the post-depolarization augmentation of CaV1 . 2 sparklet activity resulted from a Ca2+/CaM-dependent increase in the activity of previously active sites as well as the emergence of new , high-activity Ca2+ sparklet sites that appear to lack CDI . These findings suggest that CaV1 . 2 channel interactions increase Ca2+ influx and , consequently , total conductance by increasing the extent to which the channels are functionally coupled . 10 . 7554/eLife . 05608 . 014Figure 5 . Effects of interactions between CaV1 . 2 channel C-termini on channel activity . ( A and B ) Calibrated TIRF images ( see also Video 2 ) of representative control tsA-201 cells ( A ) and tsA-201 cells expressing the Ca2+-insensitive CaM1234 mutant ( B ) . In both cases , cells expressed CaV1 . 2-VN and CaV1 . 2-VC and were loaded with Rhod-2 via the patch pipette . Cells were held at −80 mV during sparklet recordings before ( left ) and after ( right ) depolarization to +60 mV . The time course of [Ca2+]i for each sparklet site ( denoted by green circles ) before and after depolarization is shown to the right of each image . ( C and D ) Scatter plots of sparklet activity ( C; nPs; **p < 0 . 001 ) and sparklet site density ( D; p = 0 . 03 ) , before and after depolarization in control ( n = 5 ) and CaM1234-expressing ( n = 6 ) cells . ns , not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 05608 . 01410 . 7554/eLife . 05608 . 015Video 2 . Ca2+ sparklet activity and site density are augmented by CaV1 . 2 channel interactions . Stacks of 2D images acquired at a frame rate of 100 Hz from a representative , Rhod-2–dialyzed tsA cell expressing CaV1 . 2-VN and CaV1 . 2-VC , held at −80 mV , before ( left ) and after ( right ) depolarization . DOI: http://dx . doi . org/10 . 7554/eLife . 05608 . 015 Having demonstrated that physical CaV1 . 2-to-CaV1 . 2 interactions functionally couple adjoining channels to enhance channel activity , we investigated the dynamics of these interactions . Bimolecular fluorescence complementation experiments are unable to provide information about channel-interaction dynamics , since Venus reconstitution is irreversible ( Kerppola , 2006 ) . Instead , we sought to detect fluorescence resonance energy transfer ( FRET ) between EGFP and red fluorescent protein ( RFP ) -tagged CaV1 . 2 channels as a function of [Ca2+]i . Flash photolysis of DMNP-EDTA ( at −80 mV ) induced a transient increase in [Ca2+]i and enhanced the CaV1 . 2-RFP/CaV1 . 2-EGFP FRET ratio ( FRETr; Figure 6A ) . The time courses of [Ca2+]i and FRETr were similar ( Figure 6B ) , with a time-to-peak of 605 . 8 ± 98 . 8 ms for [Ca2+]i and 531 . 5 ± 57 . 5 ms for FRETr . The [Ca2+]i-FRETr relationship was sigmoidal , yielding a FRETr1/2 at a [Ca2+]i of approximately 250 nM ( Figure 6C ) . Both [Ca2+]i and FRETr traces followed triple exponential decay kinetics; for [Ca2+]i , the decay time constants ( τ ) for fast , intermediate and slow components were 0 . 62 ± 0 . 15 , 2 . 19 ± 0 . 40 and 3 . 15 ± 0 . 26 s , respectively , and the corresponding values for FRETr were 0 . 64 ± 0 . 48 , 1 . 09 ± 0 . 97 and 3 . 71 ± 0 . 73 s ( n = 5 ) . The time to decay to 50% of the peak ( T50% ) was 2 . 65 ± 0 . 22 s for [Ca2+]i and 0 . 63 ± 0 . 23 s for FRETr . The complex multi-component decay kinetics of each trace reflects the multiple Ca2+ binding sites and affinities of the two lobes of CaM for Ca2+ ( Faas et al . , 2011; Faas and Mody , 2012 ) . 10 . 7554/eLife . 05608 . 016Figure 6 . Interactions of CaV1 . 2 C-termini occur dynamically with Ca2+ influx and are transiently persistent . ( A ) Time course of the percent change in FRETr ( bottom ) evoked by flash photolysis of caged Ca2+ ( purple box ) . Experiments were performed at a holding potential of −80 mV with zero EGTA or BAPTA; thus , the change in [Ca2+]i ( F/F0; top ) was global . Averaged traces ( n = 5 cells ) and error bars showing SEM at each sampling point . ( B ) Correlation between changes in FRETr and global [Ca2+]i produced by caged Ca2+ photolysis , showing that increases in [Ca2+]i were accompanied by an increase in FRETr ( p < 0 . 0001 ) . ( C ) Plot of the percent change in FRETr vs [Ca2+]i ( nM ) . Data ( n = 8 cells ) were fit to a Boltzmann Sigmoidal function ( solid red line ) with V50 = 254 . 3 nM ( black dashed line ) . ( D ) Step depolarization ( top ) in the presence of 10 mM EGTA produced local [Ca2+]i elevation and inactivating ICa ( middle ) . Increased channel interactions , represented as the percent change in FRETr ( bottom; averaged from n = 6 cells ) , were detected at the onset of depolarization and repolarization during the peak and tail currents . Dashed line shows FRETr baseline level . DOI: http://dx . doi . org/10 . 7554/eLife . 05608 . 016 Concurrent recordings of currents and FRETr during membrane depolarization in the presence of Ca2+ ( with 10 mM EGTA in the pipette solution ) showed that depolarization to +10 mV evoked a transient increase in FRETr ( Figure 6D ) . The increase in FRETr appeared biphasic , with a large amplitude spike followed by a lower level plateau that was sustained for the duration of the increase in [Ca2+]i . These data suggest that CaV1 . 2-CaV1 . 2 interactions are dynamic and regulated by local and global changes in [Ca2+]i . Notably , the increase in FRETr persisted after the Ca2+ current had decayed to baseline , indicating that channels remained coupled for a time in the absence of a stimulus . An explicit implication of our results is that physical CaV1 . 2-CaV1 . 2 interactions are critical for CDF in ventricular myocytes . To test this hypothesis , we investigated whether Ba2+ permeation and inhibition of CaM with MLCKp—both of which prevent CaV1 . 2-CaV1 . 2 interaction ( see above ) —decreases or eliminates CDF in ventricular myocytes . Because Ca2+/CaM-dependent kinase II ( CaMKII ) has been implicated in Ca2+ current facilitation ( Anderson et al . , 1994; Xiao et al . , 1994; Yuan and Bers , 1994 ) , we included 10 mM EGTA in the patch pipette to maintain global [Ca2+]i < 50 nM , which is below the threshold ( >200 nM ) for activation of this kinase ( Miller and Kennedy , 1986 ) . Experiments were performed in cells dialyzed with 100 nM or 1 μM MLCKp . The rationale for using these two concentrations of MLCKp is that , whereas 100 nM MLCKp inhibits CaM ( Török and Trentham , 1994; Török et al . , 1998 ) , we found that it does not change CaMKII activity ( p < 0 . 05 ) . However , we found that 1 μM MLCKp inhibits CaM and eliminates CaMKII activity . Thus , using these two MLCKp concentrations , we can selectively dissect the contribution of CaM and locally activated ( i . e . , near the channel pore ) CaMKII . We used a three-step protocol ( Figure 7A ) to record facilitated CaV1 . 2 currents as previously described ( Poomvanicha et al . , 2011 ) . Briefly , cells were held at −80 mV and ICa was elicited with a 200-ms control pulse ( V1 ) to 0 mV . Cells were held again at −80 mV for 10 s followed by a 200-ms prepulse to +80 mV ( Vpre ) . In light of our FRET results suggesting that CaV1 . 2 channels remain coupled for a time in the absence of a stimulus , we held the cells at −80 mV for a variable interpulse interval of 0 . 1–1 . 6 s before beginning the second 200-ms test pulse ( V2 ) to 0 mV . Fast Na+ currents were inactivated with a 50-ms step to −40 mV applied prior to each control ( V1 ) or test pulse ( V2 ) . The ratio of ICa resulting from test and control pulses ( I2/I1 ) was used as a measure of facilitation ( I2/I1 > 1 ) and recovery from inactivation ( I2/I1 ≈ 1 ) . 10 . 7554/eLife . 05608 . 017Figure 7 . CaV1 . 2-to-CaV1 . 2 channel coupling is critical for ICa facilitation in cardiomyocytes . ( A ) Voltage protocol used to evoke I1 and I2 in ventricular myocytes . ( B ) Line chart summarizing the current amplitude ratio ( I2/I1 ) at 0 mV for each condition over the range of interpulse intervals from 0 . 1 to 1 . 6 s . ( C ) Bar chart summarizing the current-amplitude ratio ( I2/I1 ) with a fixed interpulse interval of 300 ms for each condition . ( D ) Normalized whole-cell currents evoked by the protocol in ( A ) , with 2 mM [Ca2+]o as the charge carrier without ( left ) or with ( middle ) intracellular dialysis of 0 . 1 μM MLCKp ( middle ) . The currents on the right were recorded with 2 mM [Ba2+]o as the charge carrier . Data are shown as means + SEM ( p < 0 . 05 , *p < 0 . 01 vs control ( 2 mM [Ca2+]o ) ) . ( E ) Simulated time-course of ICa recovery from inactivation ( left ) , CaV1 . 2-CaV1 . 2 channel cluster disassembly ( middle ) , and ICa facilitation ( solid black line; right ) . Recovery and channel decoupling curves are single exponential functions with time constants ( τ ) of 375 ms and 333 ms , respectively . The ICa facilitation curve is the product of the recovery and decoupling functions . For comparison , the experimental facilitation data ( orange circles ) collected with 2 mM [Ca2+]o as the charge carrier is plotted alongside the simulated ICa facilitation data . The amplitude of the decoupling function was scaled to fit the facilitation data . DOI: http://dx . doi . org/10 . 7554/eLife . 05608 . 017 In control cells ( no peptide , 2 mM external Ca2+ ) , this protocol induced ICa facilitation ( i . e . , I2/I1 > 1 ) ( Figure 7B ) . Note , however , that the magnitude of ICa facilitation varied with interpulse duration , reaching a peak at a Vpre-V2 interval of 0 . 3 s . Longer Vpre-V2 intervals induced progressively less ICa facilitation . Indeed , I2/I1 was statistically indistinguishable from 1 at interpulse durations longer than 1 . 2 s , suggesting no ICa facilitation at stimulation rates >0 . 8 Hz . In the presence of extracellular Ca2+ , intracellular dialysis with 0 . 1 μM MLCKp suppressed ICa facilitation at interpulse intervals of 0 . 1–0 . 3 s ( n = 10 for 0 . 1 s intervals; n = 5 for 0 . 2 and 0 . 3 s intervals ) , but control levels of facilitation returned with intervals of 0 . 4–1 s ( n = 5–9 , p < 0 . 05; Figure 7B ) . Dialysis with a 10-fold higher concentration of MLCKp ( 1 μM ) eliminated ICa facilitation at all interpulse intervals ( n = 4; Figure 7B ) . Finally , superfusion of ventricular myocytes with Ba2+ was as effective as dialysis with 1 μM MLCKp in preventing ICa facilitation ( Figure 7B–D ) . We performed a detailed analysis of our data to gain more insights into the relationship between CaV1 . 2 channel coupling and ICa facilitation during high frequency stimulation . The amplitude of I2 in the paired pulse facilitation would depend , in part , on the degree of recovery from inactivation . A previous study has determined that the time course of ICa recovery from inactivation follows a single exponential function with a τrecovery of about 375 ms ( Blaich et al . , 2010 ) . Our FRET data in Figure 6D suggest that the number of coupled CaV1 . 2 channels will fade exponentially with an estimated τdecoupling of 333 ms upon repolarization . In Figure 7E ( left ) , we show a simulation of the time-course of ICa recovery and CaV1 . 2 channel decoupling ( middle ) during repolarization using exponential functions with these τdecoupling and τrecovery values . Note that the time-course of ICa facilitation data obtained from myocytes superfused with Ca2+ is well described by the product of the recovery and decoupling functions ( Figure 7E; right ) . This analysis is consistent with a model in which ICa facilitation during high frequency stimulation is directly proportional to the number of coupled channels and the number of channels available for activation . These data support the view that ICa facilitation in ventricular myocytes depends on Ca2+ influx and CaM , but has CaMKII-dependent and independent components . In combination with our FRET and split-Venus data , these findings suggest that Ca2+ augments CaV1 . 2 channel activity at least in part by increasing the number of functionally coupled channels . Our results support a new model for CaV1 . 2 channel function . An illustration of our proposed model for coupled gating of CaV1 . 2 channels is shown in Figure 8A–D . In our formulation , CDI and CDF are interrelated processes , both dependent on the coupling state of adjacent CaV1 . 2 channels . The C-terminal tail of the channel serves a dual role: inducing the functional coupling of adjacent channels via protein-to-protein interactions and regulating channel open probability . The physical interaction between clustered CaV1 . 2 channels is tightly regulated by local and global [Ca2+]i . The cascade of events that culminates in the coupling of CaV1 . 2 channels during an action potential begins with the gating of an individual channel within a cluster . The resulting CaV1 . 2 sparklet induces the binding of Ca2+ to CaM in the pre-IQ domain of the channel , which promotes physical interactions between contiguous channels . This increases the activity of adjoined channels , elevating local [Ca2+]i . As individual channels within a cluster undergo VDI and CDI and close , [Ca2+]i decreases and coupled channels disassemble . This , in turn , decreases channel opening probability and terminates Ca2+ flux . Thus , the overall activity of CaV1 . 2 channels within a cluster depends on the number of channels that form dimers or higher-order oligomers . 10 . 7554/eLife . 05608 . 018Figure 8 . Mechanism and proposed model for the functional coupling of CaV1 . 2 channels . ( A ) CaV1 . 2 channels are arranged into clusters in the PM of excitable cells; for simplicity , a cluster of two channels is shown . At the resting membrane potential ( e . g . , −80 mV ) , [Ca2+]i and CaV1 . 2 Po are low; hence , the majority of CaV1 . 2 channels are non-interacting . ( B ) During an action potential , the PM becomes depolarized , increasing the Po of independently gating CaV1 . 2 channels . Ca2+ flows into the cell through these active channels , producing an elevation in local [Ca2+]i and increasing Ca2+ binding to CaM . ( C ) Ca2+/CaM binding to the C-terminal pre-IQ domain of the CaV1 . 2 channel promotes physical interactions between adjacent channels . This functional coupling increases the activity of adjoined channels and thus amplifies Ca2+ influx . ( D ) CaV1 . 2 channels undergo VDI and CDI , and [Ca2+]i declines once more . However , the channels remain coupled in a ‘primed’ , non-conducting state for a finite time . If the membrane is depolarized again when the channels are still primed , the amplification of Ca2+ influx will be immediate; otherwise , if [Ca2+]i remains at resting levels beyond the lifetime of the primed state , the coupling dissolves and the cycle begins again . ( E ) and ( F ) show proposed rate-dependent changes in CaV1 . 2 channel coupling in ventricular myocytes and neurons , respectively . Top: Simulated ventricular and neuronal action potentials are depicted at low , intermediate , and high firing rates . Bottom: The accompanying dynamic change in CaV1 . 2 channel coupling ( reflected by FRET changes between adjacent channels ) . bpm , beats per minute; ips , impulses per second . DOI: http://dx . doi . org/10 . 7554/eLife . 05608 . 018 Our results have profound implications for current models of cardiac EC coupling as they provide an answer to a long-standing question in the field: What are the mechanisms that allow the simultaneous , coordinated opening of multiple CaV1 . 2 channels near the jSR ? EC coupling starts with membrane depolarization . According to our model , membrane depolarization increases the probability of CaV1 . 2 sparklet occurrence . CaV1 . 2 sparklets elevate local [Ca2+]i , thereby increasing the number of channels that form dimers or higher-order oligomers . This physical coupling increases the probability of synchronous , coincident openings of channels within a cluster ( ∼5–10 channels ) to reliably activate nearby RyRs through Ca2+-induced Ca2+ release ( Inoue and Bridge , 2003; Sobie and Ramay , 2009 ) . Accordingly , the strength of cardiac contraction would depend , at least in part , on the number of physically and functionally coupled CaV1 . 2 channels . Our data further suggest that the degree of CaV1 . 2-CaV1 . 2 coupling varies within the physiological range of [Ca2+]i reached in ventricular myocytes during a cardiac cycle . We found that CaV1 . 2 channel coupling is dynamic and has an apparent Kd of ∼250 nM . While the model assumes that CaV1 . 2-to-CaV1 . 2 coupling is initiated by a Ca2+ sparklet , it could also be induced by a Ca2+ spark resulting from the opening of a small cluster of closely apposed RyRs . This could induce local CaV1 . 2-to-CaV1 . 2 coupling , even if transiently , priming these channels for opening during the action potential . Note , however , that because Ca2+ spark activity is very low during diastole , the number of coupled CaV1 . 2 channels would likely be low . However , at the peak level of the [Ca2+]i transient ( ∼700 nM ) during EC coupling ( Santana et al . , 2002 ) , the cell would have reached a maximal level of CaV1 . 2-to-CaV1 . 2 coupling . An important finding in our study is that , while CaV1 . 2 channel coupling fades as [Ca2+]i decreases , it persists longer than the current that evoked it . This is important because CaV1 . 2 channel activity remains elevated for as long as the channels remain coupled . Thus , by outlasting the [Ca2+]i signal that evoked it , CaV1 . 2 channel coupling acts as a type of ‘molecular memory’ that might serve to augment Ca2+ influx during repetitive membrane depolarization . As a consequence , an increase in action potential frequency can enhance Ca2+ influx via two mechanisms: first , it can increase [Ca2+]i , thereby increasing CaV1 . 2 channel coupling; and second , if an AP arrives while a subpopulation of CaV1 . 2 channels remain coupled , which could occur even if [Ca2+]i had decreased to basal levels owing to the molecular memory phenomenon , it would encounter a cell with a higher number of coupled—that is , more active—channels ( see Figure 8E , F ) . This molecular memory might also manifest itself as CaV1 . 2 current facilitation ( Marban and Tsien , 1982; Argibay et al . , 1988; Gurney et al . , 1989 ) . In the case of cardiac muscle , CaV1 . 2 current facilitation augments contractile force during increases in heart rate ( Pieske et al . , 1999 ) . In neurons , facilitation of CaV1 . 2 channels has been suggested to contribute to an intrinsic amplification of synaptic current and the enhancement of neuronal excitability ( Powers and Binder , 2001; Striessnig et al . , 2014 ) . Our data support the view that Ca2+ current facilitation in ventricular myocytes requires Ca2+ influx and CaM , but has CaMKII-dependent and -independent components . The latter likely involve Ca2+/CaM-induced CaV1 . 2-CaV1 . 2 channel coupling . Future experiments should investigate if CaMKII facilitates Ca2+ current in neurons and cardiac muscle by a similar mechanism . Dynamic , [Ca2+]i-dependent CaV1 . 2 channel coupling could have pathological repercussions . For example , coupling of a mutant channel with a higher intrinsic open probability to a WT channel could increase Ca2+ influx . One condition associated with aberrantly high CaV1 . 2 channel gating is Timothy syndrome ( TS ) , also known as long-QT syndrome 8 . TS is an autosomal-dominant , multi-organ disorder caused by de novo gain-of-function missense mutations in exon 8 or an alternatively spliced exon 8A of Cav1 . 2 ( Splawski et al . , 2004 , 2005 ) . Forced physical interactions between Cav1 . 2 channels have an intriguing effect on adjoined channels: fusion of intrinsically hyperactive CaV1 . 2-TS channels to WT channels induces these latter channels to function like TS channels . These findings have important implications . If TS channels can form stable interactions with neighboring WT channels in TS patients , then these mutant channels , which constitute only ∼23% of the total cardiac Cav1 . 2 population , could have a disproportionally large effect on Ca2+ influx . We previously tested this idea in ventricular myocytes expressing our fusible Cav1 . 2 channels and found that fusion of TS channels with WT channels led to the development of arrhythmogenic spontaneous SR Ca2+ release events in addition to increasing the amplitude of [Ca2+]i transients and contractions ( Dixon et al . , 2012 ) . These findings support the hypothesis that physical interactions between the C-termini of TS and WT channels produce a disproportionally large Ca2+ influx that ultimately induces arrhythmogenic changes in [Ca2+]i . Consistent with this , we found that the relationship between the level of CaV1 . 2-TS channel expression and the probability of a Ca2+ wave is non-linear , suggesting that even low levels of these channels are sufficient to induce maximal changes in [Ca2+]i ( Drum et al . , 2014 ) . Mutations in CaM have recently been linked to severe forms of long-QT syndrome , which are associated with life-threatening arrhythmias that occur very early in life ( Limpitikul et al . , 2014 ) . Expression of these CaMs increases Cav1 . 2 activity and open times , effects similar to those produced by the TS mutation . Because CaM regulates many other Ca2+ channel subtypes , including those that predominate in neurons , these mutant CaMs could lead to a multi-system disorder similar to TS . An important direction for future experiments will be to investigate whether Cav1 . 2 channel dysfunction is associated with aberrant TSTS or TS-WT Cav1 . 2 channel coupling or coupling of channels with mutant CaM to channels with WT CaM . In summary , our study demonstrates that dynamic , Ca2+-driven physical interactions among clustered CaV1 . 2 channels lead to cooperative gating of adjacent channels and enhanced Ca2+ influx . The physical proximity afforded by clustering of CaV1 . 2 channels is necessary , but not sufficient , for functional coupling of the channels and occurs whether the channels are functionally coupled or not . Future studies should investigate the mechanisms that dictate the clustered arrangement of CaV1 . 2 channels . It is likely that cooperative CaV1 . 2 channel gating also plays an important role in physiological functions as diverse as neuronal excitability and rate-dependent increases in cardiac contraction , as well as pathological conditions such as long-QT syndrome . WT C57BL/6 mice were euthanized with a single lethal dose of sodium pentobarbital delivered via intraperitoneal injection , as approved by the University of Washington Institutional Animal Care and Use Committee ( IACUC ) . Ventricular myocytes were isolated as previously described ( Dixon et al . , 2012 ) . Briefly , the heart was excised and rinsed with cold 150 μM EGTA digestion buffer containing 130 mM NaCl , 5 mM KCl , 3 mM Na-pyruvate , 25 mM HEPES , 0 . 5 mM MgCl2 , 0 . 33 mM NaH2PO4 , and 22 mM glucose . The aorta was cannulated for Langendorff perfusion , and the coronary arteries were subsequently perfused with warmed ( 37°C ) 150 μM EGTA digestion buffer until they were cleared of blood . The perfusate was then switched to digestion buffer ( no EGTA ) supplemented with 50 μM CaCl2 , 0 . 04 mg/ml protease ( XIV ) , and 1 . 4 mg/ml collagenase ( type 2; Worthington Biochemical , Lakewood , NJ ) for 8–10 min . The ventricles were then cut away from the atria , sliced , and placed in 37°C digestion buffer supplemented with 0 . 96 mg/ml collagenase , 0 . 04 mg/ml protease , 100 μM CaCl2 , and 10 mg/ml bovine serum albumen ( BSA ) . Gentle agitation was applied using a transfer pipette until the ventricles dissociated . The cells were then allowed to pellet by gravity for 15–20 min , after which they were washed in enzyme-free digestion buffer supplemented with 10 mg/ml BSA and 250 μM CaCl2 , pelleted once more , and finally resuspended at room temperature in Tyrode's solution containing 140 mM NaCl , 5 mM KCl , 10 mM HEPES , 10 mM glucose , 2 mM CaCl2 , and 1 mM MgCl2; pH was adjusted to 7 . 4 with NaOH . tsA-201 cells ( Sigma–Aldrich , St . Louis , MO ) were cultured in Dulbecco's Modified Eagle Medium ( DMEM; Gibco-Life Technologies , Grand Island , NY ) supplemented with 10% fetal bovine serum ( FBS ) and 1% penicillin/streptomycin at 37°C in a humidified 5% CO2 atmosphere and passaged every 3–4 day . Cells were transiently transfected at ∼70% confluence using jetPEI ( Polyplus Transfection , New York , NY ) transfection reagent and plated onto the appropriate coverglass ∼12 hr before experiments . For super-resolution imaging and photobleaching experiments , cells were plated onto 22 mm no . 1 . 5 coverslips ( Thermo Fisher Scientific , Waltham , MA ) . For all other experiments , cells were plated onto 25 mm no . 1 coverslips ( Thermo Fisher Scientific ) . Plasmids used in this study include pcDNA clones of the pore-forming subunit of rabbit CaV1 . 2 ( α1c ) and rat auxillary subunits CaVα2δ and CaVβ3 ( kindly provided by Dr Diane Lipscombe; Brown University , Providence , RI ) . Standard PCR techniques were used to fuse the carboxyl tail of CaV1 . 2 to different proteins depending on the experimental approach: for FRET experiments , to EGFP or tagRFP; for bimolecular fluorescence complementation , to either the N-fragment ( VN ) or the C-fragment ( VC ) of the Venus protein ( 27097 , 22011; Addgene , Cambridge , MA ) ( Kodama and Hu , 2010 ) ; for photobleaching experiments , to the monomeric GFP variant GFP ( A206K ) ( Zacharias et al . , 2002 ) , kindly provided by Dr Eric Goaux ( Vollum Institute , Portland , OR ) . Ca2+-insensitive CaM1234 was a gift from Dr Johannes Hell ( UC Davis , CA ) . Mutant rabbit CaV1 . 2 ( I1654E ) , analogous to human I1624E ( Zühlke et al . , 1999 ) , was used in experiments designed to elucidate the role of the IQ motif in channel interactions . This single point mutation ( I1672E ) was introduced using the QuikChange II XL Site-Directed Mutagenesis kit ( Agilent Technologies , Santa Clara , CA ) . The role of the pre-IQ motif in channel interactions was examined by exchanging a 33-amino-acid segment , as illustrated in Figure 4—figure supplement 4A . The resultant CaV1 . 2 ( pre-IQ swap ) was used in whole-cell patch-clamp and bimolecular fluorescence complementation experiments . The general ER marker mCherry-Sec61β was a gift from Gia Voeltz ( Addgene plasmid # 49155 ) . Finally , JPH2-GFP was used to tether the ER to the PM in tsA-201 cells . Ca2+ currents were recorded in the whole-cell voltage-clamp or cell-attached patch configurations using borosilicate patch pipettes with resistances of 3–6 μΩ for tsA cells and 2–3 μΩ for cardiomyocytes . Currents were sampled at a frequency of 20 kHz , low-pass–filtered at 2 kHz using an Axopatch 200B amplifier , and acquired using pClamp 10 . 2 software ( Molecular Devices , Sunnyvale , CA ) . All membrane potentials referred to herein have been corrected for liquid junction potential . All experiments were performed at room temperature ( 22–25°C ) . For whole-cell current recordings from tsA-201 cells , pipettes were filled with a Cs-based internal solution containing 87 mM Cs-aspartate , 20 mM CsCl , 1 mM MgCl2 , 10 mM HEPES , 10 mM EGTA and 5 mM MgATP , adjusted to pH 7 . 2 with CsOH . In experiments requiring dialysis of the calmodulin inhibitory peptide MLCKp ( EMD Millipore , Darmstadt , Germany; 208735 ) , this chemical was added to the pipette solution . Cells were continuously superfused throughout experiments with our regular external solution containing 5 mM CsCl , 10 mM HEPES , 10 mM glucose , 113 mM NMDG , 1 mM MgCl2 and 20 mM CaCl2 , adjusted to pH 7 . 4 with HCl . For experiments in which Ba2+ was used as the charge carrier in place of Ca2+ , the perfusate contained 5 mM CsCl , 10 mM HEPES , 10 mM glucose , 140 mM NMDG , 1 mM MgCl2 and 2 mM BaCl2 , adjusted to pH 7 . 4 with HCl . Current-voltage relationships were obtained by subjecting cells to a series of 300-ms depolarizing pulses from a holding potential of −80 mV to test potentials ranging from −60 to +80 mV . The voltage dependence of conductance was obtained by converting the resultant currents to conductances using the equation , G = ICa/[test pulse potential − reversal potential of ICa] , normalizing ( G/Gmax ) , and plotting conductance vs the test potential . Cell-attached patch , single-channel currents ( iCa ) were recorded from tsA-201 cells superfused with high K+ solution to fix the membrane potential at ∼0 mV . The solution had the following composition: 145 mM KCl , 2 mM MgCl2 , 0 . 1 mM CaCl2 , 10 mM HEPES and 10 mM glucose; pH was adjusted to 7 . 3 with KOH . Pipettes were filled with a solution containing 10 mM HEPES and either 110 mM CaCl2 or 110 mM BaCl2; pH was adjusted to 7 . 2 with CsOH . The dihydropyridine agonist BayK-8644 ( 500 nM ) was included in the pipette solution to promote longer channel open times . A voltage-step protocol from a holding potential of −80 mV to a depolarized potential of −30 mV was used to elicit currents . The single-channel event-detection algorithm of pClamp 10 . 2 was used to measure single-channel opening amplitudes and nPo , and to construct all-points histograms . To record ICa from isolated ventricular myocytes , cells were initially perfused with Tyrode's solution . Once the whole-cell configuration was successfully established , the external solution was replaced with one containing 5 mM CsCl , 10 mM HEPES , 10 mM glucose , 140 mM NMDG , 1 mM MgCl2 and either 2 mM CaCl2 or 2 mM BaCl2 , adjusted to pH 7 . 4 with HCl . The pipette was filled with the Cs-based internal solution described above . Facilitation of ICa was measured using a triple-pulse protocol consisting of two identical test pulses ( V1 and V2 ) , separated by a conditioning pulse to +80 mV ( Vpre ) , as previously described ( Poomvanicha et al . , 2011 ) and illustrated in Figure 7A . The currents elicited by V1 and V2 were referred to as I1 and I2 , and the ratio between their peaks ( I2/I1 ) was used as a measure of facilitation . To measure Ca2+ sparklets in tsA-201 cells or ventricular myocytes , cells were patch-clamped in whole-cell mode and held at a hyperpolarized potential of −80 mV to increase the driving force for Ca2+ entry . Sub-sarcolemmal Ca2+ signals were monitored by dialyzing cells with 200 μM Rhod-2 via the patch pipette and continuously perfusing them with the external solution described above containing 20 mM CaCl2 and 10 mM EGTA , a relatively slow Ca2+ buffer . With this buffer/indicator combination , Ca2+ entering the cell via membrane Cav1 . 2 channels binds to the relatively fast Ca2+ indicator Rhod-2 to generate a fluorescent signal , and the excess EGTA rapidly buffers Ca2+ , restricting the signal to the point of entry . Sub-sarcolemmal Ca2+ signals ( Ca2+ sparklets ) were captured using a through-the-lens TIRF microscope built around an Olympus IX-70 inverted microscope equipped with an oil-immersion ApoN 60×/1 . 49 NA TIRF objective and an Andor iXON CCD camera . Images were acquired at 100 Hz using TILLvisION imaging software ( TILL Photonics , FEI , Hillsboro , OR ) . Sparklets were detected and analyzed using custom software written in MATLAB ( Source code 1 ) . Rhod-2 fluorescence signals were converted to Ca2+ concentration units using the Fmax equation ( Maravall et al . , 2000 ) . The activity of Ca2+ sparklets was determined by calculating the nPs of each Ca2+ sparklet site , where n is the number of quantal levels and Ps is the probability that a quantal Ca2+ sparklet event is active . A detailed description of this analysis can be found in Navedo et al . ( Navedo et al . , 2005 , 2006 ) . The degree of coupling between single CaV1 . 2 channels or Ca2+ sparklet sites was assessed by further analyzing single-channel and sparklet recordings using a binary coupled Markov chain model ( Source code 2 ) , as first described by Chung and Kennedy ( 1996 ) and previously employed by our group ( Navedo et al . , 2010; Cheng et al . , 2011; Dixon et al . , 2012 ) . The custom program , written in the MATLAB language , assigns a coupling-coefficient ( κ ) to each record , where κ can range from 0 ( purely independently gating channels ) to 1 ( fully coupled channels ) . Elementary event amplitudes were set at 0 . 5 pA for iCa , 1 . 5 pA for iBa , and 38 nM for Ca2+ sparklets . Transfected tsA-201 cells expressing the relevant CaV1 . 2 channel constructs were plated onto poly-L-lysine–coated #1 . 5 coverslips ( Thermo Fisher Scientific ) the day before fixation . For immunostaining , cells were fixed by incubating for 10 min in ice-cold methanol , then washed with PBS and blocked for 1 hr at room temperature in 50% SEA BLOCK ( Thermo Fisher Scientific ) and 0 . 5% vol/vol Triton X-100 in PBS ( blocking buffer ) . The pore-forming subunit of CaV1 . 2 was probed with a rabbit polyclonal primary antibody ( anti-CNC1; kindly provided by Drs William Catterall and Ruth Westenbroek [Hulme et al . , 2003] ) , diluted to 5 μg/ml in diluted blocking buffer ( 20% SEA BLOCK , 0 . 5% Triton X-100 ) , by incubating overnight at 4°C . The following morning , cells were washed extensively , receiving three washes with PBS and five washes with diluted blocking buffer . Cells were then incubated for 1 hr at room temperature with Alexa Fluor 647-conjugated donkey anti-rabbit secondary antibody ( 2 μg/ml; Molecular Probes–Life Technologies ) in diluted blocking buffer . Cells were finally washed thoroughly with PBS and mounted for imaging . Native Cav1 . 2 channels in freshly isolated adult ventricular myocytes were immunostained in an identical manner except that plating procedures differed . Specifically , myocytes were plated onto laminin and poly-L-lysine–coated #1 . 5 coverslips and allowed to adhere for 1 hr before fixation . For double-labeling experiments used to examine the co-localization of CaV1 . 2 channels and the ER , tsA-201 cells transfected with CaV1 . 2 channels ( WT or pre-IQ swap mutants ) and mCherry-Sec61β were fixed in 3% paraformaldehyde and 0 . 1% glutaraldehyde in PBS for 10 min followed by extensive washing in PBS and reduction for 5 min in ∼0 . 1% sodium borohydride in water to reduce background fluorescence . Cells were washed and blocked as described above and were then incubated in primary antibody solution containing 5 μg/ml rabbit anti-CNC1 and 2 μg/ml rat monoclonal anti-mCherry ( Life Technologies ) for 1 hr at room temperature . Excess primary antibody was removed by three washes with PBS and five washes with diluted blocking buffer . Cells were then incubated for 1 hr at room temperature with Alexa Fluor 647-conjugated donkey anti-rabbit and Alexa Fluor 568-conjugated chicken anti-rat secondary antibodies ( 2 μg/ml each; Molecular Probes–Life Technologies ) in diluted blocking buffer . Cells were finally washed thoroughly with PBS and mounted for imaging . In some experiments , tsA-201 cells co-expressing JPH2-GFP were fixed and stained as per the double-staining protocol described above . JPH2-GFP was not immunostained , but simply imaged in TIRF mode by exciting the GFP tag . Coverslips were mounted with MEA-GLOX ( for double staining ) or β-ME-GLOX imaging buffer on glass depression slides ( neoLab , Heidelberg , Germany ) and sealed with Twinsil ( Picodent , Wipperfürth , Germany ) . The imaging buffers contained TN buffer ( 50 mM Tris pH 8 . 0 , 10 mM NaCl ) , GLOX oxygen scavenging system ( 0 . 56 mg/ml glucose oxidase , 34 μg/ml catalase , 10% wt/vol glucose ) , and either 100 mM MEA ( cysteamine ) or 142 mM 2-mercaptoethanol ( β-ME ) . Excess imaging buffer was blotted away before application of the Twinsil sealant . This is particularly important with the β-ME-GLOX imaging buffer as the Twinsil will not set if it contacts this buffer . GSD super-resolution images of CaV1 . 2 channels in fixed ventricular myocytes or CaV1 . 2 channels and mCherry-Sec61β in fixed tsA-201 cells were generated using a Leica SR GSD 3D system . The system is built around a Leica DMI6000 B TIRF microscope and is equipped with a Leica oil-immersion HC PL APO 160×/1 . 43 NA super-resolution objective , four laser lines ( 405 nm/30 mW , 488 nm/300 mW , 532 nm/500 mW , and 642 nm/500 mW ) , and an Andor iXon3 EM-CCD . Images were collected in TIRF mode at a frame rate of 100 Hz for 20 , 000–100 , 000 frames using Leica Application Suite ( LAS AF ) software . CaV1 . 2 cluster area sizes were determined using binary masks of the images in ImageJ/Fiji . tsA-201 cells expressing CaV1 . 2 channels tagged at their C-terminus with monomeric GFP were fixed in 4% paraformaldehyde ( 10 min ) and imaged in TIRF mode on the Leica 3D-GSD system described above using a 160×/1 . 43 NA objective . The core Leica DMI6000 B TIRF microscope in this system is capable of functioning outside of GSD SR imaging mode as a conventional TIRF microscope . Cells were illuminated with 488-nm laser light , and image stacks of 2000 frames were acquired at 30 Hz . The first five images after the shutter was opened were averaged , and a rolling-ball background subtraction was applied using ImageJ/Fiji . This image was then low-pass filtered with a 2-pixel cut-off and high-pass filtered with a 5-pixel cut-off ( see Figures 2D and 3D ) . Thresholding was then applied to identify connected regions of pixels that were above threshold . The ImageJ/Fiji plugin ‘Time Series Analyzer v2 . 0’ was then used to select 4 × 4 pixel regions of interest ( ROIs ) centered on the peak pixel in each spot . Next , z-axis intensity profiles ( where z is time ) from these ROIs were examined over the entire image stack . To facilitate the identification of bleaching steps , the signal-to-noise ratio was improved by applying a 5-pixel rolling-ball background subtraction , a median filter ( 1 pixel radius ) , and a 10 frame moving average . Bleaching steps were then manually counted . Identical procedures were performed on cardiomyocytes expressing photo-activatable , GFP-tagged CaVβ2 auxiliary subunits . This subunit binds to the pore-forming α1 subunit of the channel with a 1:1 stoichiometry; thus , expression of this protein in cardiomyocytes represents a photo-activatable fluorescent marker of CaV1 . 2 channels . Since adult cardiomyocytes are impervious to chemical transfection , we used adeno-associated virus serotype 9 ( AAV9 ) to transfer this gene into mice via retro-orbital injection , a strategy that has been successfully used by others to transfer cardiac genes in mice ( Fang et al . , 2012 ) . The AAV9-packaged CaVβ2-PA-GFP was engineered from CaVβ2-PA-GFP pcDNA by Vector Biolabs . Mice were sacrificed 5 week after retro-orbital injection . Successful gene transfer was confirmed by photo-activating the CaVβ2-PA-GFP with 405 nm laser light . Prior to photo-activation , no GFP fluorescence emission was detected upon excitation with 488 nm laser light , but after photo-activation , robust GFP fluorescence emission was observed in the z-lines of isolated ventricular myocytes ( data not shown ) . For stepwise photobleaching experiments , GFP was photo-activated prior to starting movie recordings . Spontaneous interactions of CaV1 . 2 channels were assayed using bimolecular fluorescence complementation . In these experiments , CaV1 . 2 channels were tagged at their C-terminus with non-fluorescent N- ( VN ( 1-154 , I152L ) ) or C-terminal ( VC ( 155-238 , A206K ) ) halves of a ‘split’ Venus fluorescent protein . When CaV1 . 2-VN and CaV1 . 2-VC are brought close enough together to interact , the full Venus protein can fold into its functional , fluorescent confirmation . The magnitude of Venus fluorescence emission therefore provides an indicator of CaV1 . 2 interactions . Venus fluorescence was monitored in tsA-201 cells expressing CaV1 . 2-VN and CaV1 . 2-VC using TIRF microscopy , as described above ( ‘recording of Ca2+ sparklets’ ) . Transfected cells were identified by co-expression of tagRFP or , in experiments requiring Ca2+ imaging with Rhod-2 , by weak initial Venus expression . The relationship between membrane voltage and CaV1 . 2 interactions was obtained by subjecting patch-clamped cells in whole-cell mode to a series of 9-s depolarizations from a holding potential of −80 mV to test potentials ranging from −60 to +80 mV . Maturation of newly reconstituted Venus protein takes some time , hence the long depolarizing pulse ( Nagai et al . , 2002 ) . Cells were superfused throughout with the 20 mM Ca2+ external solution described above . A TTL pulse generated by the TILL imaging system was used to trigger the onset of each voltage sweep . The cell TIRF footprint was illuminated using 491-nm light throughout each voltage sweep , and PM-localized Venus fluorescence emission was monitored in a TIRF movie acquired at a rate of 100 Hz . The final six frames of each movie were averaged to generate a single image for each voltage sweep . These images were median filtered ( 1 pixel radius ) then divided by the −60 mV image to obtain Venus F/F0 images . Finally , the images were smoothed and pseudo-colored using the ‘red hot’ lookup table in ImageJ/Fiji . The Ca2+ dependence of CaV1 . 2 interactions was tested by performing the aforementioned procedure first with 2 mM Ba2+ as the conducting ion , then switching the perfusate to our regular external solution described above ( 20 mM Ca2+ ) and running the procedure again on the same cell ( see Figure 4—figure supplement 2 ) . Reconstitution of the Venus protein is irreversible; thus , once channels spontaneously interact , they remain fused together . We exploited this feature of the bimolecular fluorescence complementation assay in experiments designed to investigate the physiological effects of CaV1 . 2 interactions . In these experiments , transfected tsA-201 cells expressing the split-Venus–tagged channels were patch-clamped in whole-cell mode and dialyzed with 200 μM Rhod-2 via the pipette . Ca2+ sparklets were recorded while holding the cell at −80 mV . The depolarizing step protocol described above was performed , and Venus fluorescence was monitored as before . Increases in Venus fluorescence emission during the protocol provide an indication of Cav1 . 2 interactions . Holding cells once more at −80 mV , Ca2+ sparklet activity was recorded from the irreversibly fused CaV1 . 2 channels . For this set of experiments , tsA-201 cells were transfected with expression constructs for CaV1 . 2-EGFP and CaV1 . 2-tagRFP . FRET from CaV1 . 2-EGFP ( donor ) to CaV1 . 2-tagRFP ( acceptor ) was measured on an Olympus Fluoview 1000 ( FV1000 ) confocal laser-scanning microscope equipped with an Olympus APON 60×/1 . 49 NA oil-immersion objective . A 473-nm diode laser was used to excite the sample . Emitted light was separated with an SDM560 beam splitter , collected with BA490-540 and BA575-675 emission filters , and detected by a photomultiplier tube . The resultant raw donor and acceptor images were corrected for background and bleed-through of GFP emission into the RFP channel . In separate experiments using cells expressing only CaV1 . 2-tagRFP , bleed-through was determined to be ∼16% . Intensity measurements from the corrected images , referred to as GFPg and RFPg respectively ( where g refers to excitation of GFP with 473 nm light ) , were extracted using Metamorph or ImageJ software . FRET was expressed as the ratio , FRETr = RFPg/GFPg . In some experiments , the time course of FRETr was monitored during photolysis of caged Ca2+ . In others , the time course of FRETr was recorded simultaneously with whole-cell CaV1 . 2 currents . A TTL pulse generated by the FV1000 system at the onset of imaging was used to trigger a voltage protocol consisting of a 500-ms depolarizing pulse from a holding potential of −80 to +10 mV . Finally , the Ca2+ dependence of FRET between CaV1 . 2 channels was monitored while perfusing cells with external solutions containing 5 mM CsCl , 10 mM HEPES , 10 mM glucose , 140 mM NMDG , 1 mM MgCl2 and increasing concentrations ( 0 , 25 , 50 , 100 , 200 , 300 , 400 , 800 and 5000 nM ) of CaCl2 . The solutions were adjusted to pH 7 . 4 , and 1 μM ionomycin was added to equilibrate intracellular and extracellular Ca2+ . tsA cells were perfused with a ‘zero calcium’ external solution containing 5 mM CsCl , 10 mM HEPES , 10 mM glucose , 140 mM NMDG , 1 mM MgCl2 , and 2 mM BaCl2 ( pH . 7 . 4 ) . Cells were patch-clamped in whole-cell mode as previously described ( Tadross et al . , 2013 ) using an internal solution containing 5 mM CsCl , 40 mM HEPES , 135 mM CsMeSO3 , 1 mM citrate and 1 . 6 mM CaCl2 , adjusted to pH 7 . 4 and frozen into aliquots . On the day of experiments , the Ca2+ cage DMNP-EDTA ( 2 mM; Invitrogen , D6814 ) was added . Mg2+ was omitted from the pipette solution since DMNP-EDTA can also cage Mg2+ . Internal solutions were supplemented with 0 . 25 μM PI ( 4 , 5 ) P2 ( Avanti Polar Lipids , Alabaster , AL; 840046P ) and 0 . 5 μM okadaic acid ( LC Laboratories , Woburn , MA; O-2220 ) to stabilize ICa ( Tadross et al . , 2013 ) . This enabled us to record stable currents and achieve successful uncaging of Ca2+ , as assayed with the Ca2+-sensitive dyes , Fluo-4 AM or Rhod-2 AM ( see Figure 6A and Figure 4—figure supplement 1 ) . Uncaging experiments were performed on an Olympus FV1000 confocal microscope equipped with a SIM scanner unit that permits simultaneous laser light uncaging ( 100% 405 nm for two frames ) and recording of Venus , Fluo-4 , Rhod-2 or EGFP/tagRFP FRET fluorescence emission . CaMKII activity was measured using the ‘SignaTECT Calcium/Calmodulin-Dependent Protein Kinase Assay System’ ( Promega Corporation , Madison , WI ) . Mouse hearts were homogenized in a lysis buffer solution containing 20 mM Tris–HCl ( pH 8 . 0 ) , 2 mM EDTA , 2 mM EGTA , 2 mM DTT , PhosSTOP phosphatase inhibitor cocktail ( Roche Diagnostics GmbH , Mannheim , Germany ) and cOmplete , Mini protease inhibitor cocktail ( Roche Diagnostics ) . Hearts were subjected to three 5 s pulses at 12 , 000–17 , 000 rpm using a PRO200 Homogenizer Unit ( Pro Scientific , Oxford , CT ) . The homogenate was centrifuged at 2000 rpm for 10 min and the resultant supernatant was collected and assayed for CaMKII activity as per the manufacturers instructions . To determine the effect of MLCKp on CaMKII activity , 0 . 1 or 1 μM MLCKp was added to the reaction .	To pump blood around the body , the muscle cells within the heart must contract and relax together with a regular rhythm . A contraction begins when proteins called CaV1 . 2 channels embedded in the cell membranes of heart cells open to allow calcium ions to enter the cells . The calcium ions that enter through these CaV1 . 2 channels trigger the release of calcium ions from storage compartments within the cells , which leads to the heart contracting . However , to trigger this release of calcium ions , many CaV1 . 2 channels have to open at the same time and we do not yet know how this is co-ordinated . Dixon et al . studied CaV1 . 2 channels in heart muscle cells from mice . The experiments show that these proteins are arranged in clusters of eight , on average , in the cell membrane . When calcium ions enter the cell they bind to a protein called calmodulin , which in turn binds to a CaV1 . 2 channel . This allows the CaV1 . 2 channels within a cluster to interact with each other . The physical association between CaV1 . 2 channels within clusters enables them to work cooperatively; they open at the same time to allow more calcium ions to enter and then close together to allow the cell to relax . Dixon et al . found that even when levels of calcium ions in the cells declined , the CaV1 . 2 channels within clusters remained open for a little while longer before they closed . This suggests that the interactions between the CaV1 . 2 channels act as a type of ‘molecular memory’ that may alter how the cells respond to future activity . These results challenge the previously held view that the CaV1 . 2 channels open and close independently of one another . Future studies will seek to understand the molecular details of how these channels cluster together , and how this clustering affects changes in heart rate and heart abnormalities like long QT syndrome .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics" ]	2015	Graded Ca2+/calmodulin-dependent coupling of voltage-gated CaV1.2 channels

Subsets and Splits

No community queries yet

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